Cassandra 2.1 Used by DataStax Enterprise 4.7

Apache Cassandra™ 2.
1
Documentation
July 10, 2015
© 2015 DataStax. All rights reserved.

Apache, Apache Cassandra, Apache Hadoop, Hadoop and the
eye logo are trademarks of the Apache Software Foundation
Contents
Contents
About Apache Cassandra............................................................................................... 9
What's new in Cassandra 2.1....................................................................................... 11
CQL.................................................................................................................................. 12
Understanding the architecture................................................................................... 13

Architecture in brief........................................................................................................................... 13
Internode communications (gossip).................................................................................................. 15
Failure detection and recovery............................................................................................... 15
Data distribution and replication........................................................................................................16
Consistent hashing................................................................................................................. 16
Virtual nodes...........................................................................................................................17
Data replication.......................................................................................................................18
Partitioners......................................................................................................................................... 19
Murmur3Partitioner..................................................................................................................20
RandomPartitioner.................................................................................................................. 20
ByteOrderedPartitioner........................................................................................................... 21
Snitches............................................................................................................................................. 21
Dynamic snitching...................................................................................................................21
SimpleSnitch........................................................................................................................... 22
RackInferringSnitch................................................................................................................. 22
PropertyFileSnitch................................................................................................................... 22
GossipingPropertyFileSnitch................................................................................................... 23
Ec2Snitch................................................................................................................................ 23
Ec2MultiRegionSnitch............................................................................................................. 24
GoogleCloudSnitch................................................................................................................. 26
CloudstackSnitch.................................................................................................................... 27
Client requests...................................................................................................................................27
Planning a cluster deployment.................................................................................... 28

Selecting hardware for enterprise implementations.......................................................................... 28
Planning an Amazon EC2 cluster..................................................................................................... 31
Calculating usable disk capacity....................................................................................................... 32
Calculating user data size.................................................................................................................33
Anti-patterns in Cassandra................................................................................................................34
Installing..........................................................................................................................37
Installing on RHEL-based systems................................................................................................... 37
Installing on Debian-based systems................................................................................................. 38
Installing the binary tarball................................................................................................................ 39
Installing on Windows systems......................................................................................................... 40
Installing prior releases..................................................................................................................... 41
Uninstalling Cassandra......................................................................................................................41
Installing on cloud providers............................................................................................................. 42
3
Contents
Installing on Amazon EC2......................................................................................................42

Installing on GoGrid................................................................................................................50
Installing Oracle JRE.........................................................................................................................52
Installing the JRE on RHEL-based systems.......................................................................... 52
Installing the JRE on Debian-based systems........................................................................ 53
Recommended production settings...................................................................................................54
Initializing a cluster....................................................................................................... 57
Initializing a multiple node cluster (single data center)..................................................................... 57
Initializing a multiple node cluster (multiple data centers)................................................................ 59
Security........................................................................................................................... 63
Securing Cassandra.......................................................................................................................... 63
SSL encryption.................................................................................................................................. 63
Client-to-node encryption........................................................................................................63
Node-to-node encryption........................................................................................................ 64
Using cqlsh with SSL encryption............................................................................................65
Preparing server certificates...................................................................................................66
Internal authentication....................................................................................................................... 67
Internal authentication............................................................................................................ 67
Configuring authentication...................................................................................................... 67
Logging in using cqlsh............................................................................................................68
Internal authorization......................................................................................................................... 68
Object permissions................................................................................................................. 69
Configuring internal authorization...........................................................................................69
Configuring firewall port access........................................................................................................ 70
Enabling JMX authentication.............................................................................................................71
Database internals......................................................................................................... 73
Storage engine.................................................................................................................................. 73
Separate table directories................................................................................................................. 73
Cassandra storage basics.................................................................................................................73
The write path to compaction.................................................................................................74
How Cassandra stores indexes............................................................................................. 76
About index updates...............................................................................................................77
The write path of an update............................................................................................................. 77
About deletes.....................................................................................................................................77
About hinted handoff writes.............................................................................................................. 77
Reads.................................................................................................................................................80
About reads............................................................................................................................ 80
How off-heap components affect reads................................................................................. 81
Reading from a partition.........................................................................................................82
How write patterns affect reads............................................................................................. 82
How the row cache affects reads.......................................................................................... 82
About transactions and concurrency control.....................................................................................82
Atomicity..................................................................................................................................83
Consistency.............................................................................................................................83
Isolation................................................................................................................................... 83
Durability................................................................................................................................. 84
Lightweight transactions......................................................................................................... 84
Data consistency............................................................................................................................... 85
About data consistency.......................................................................................................... 85
About built-in consistency repair features.............................................................................. 85
4
Contents
Configuring data consistency................................................................................................. 85

Read requests........................................................................................................................ 89
Write requests.........................................................................................................................93
Configuration.................................................................................................................. 95
cassandra.yaml configuration file...................................................................................................... 95
Configuring gossip settings............................................................................................................. 111
Configuring the heap dump directory..............................................................................................112
Configuring virtual nodes.................................................................................................................113
Enabling virtual nodes on a new cluster.............................................................................. 113
Enabling virtual nodes on an existing production cluster..................................................... 113
Using multiple network interfaces................................................................................................... 114
Configuring logging..........................................................................................................................116
Commit log archive configuration....................................................................................................118
Generating tokens........................................................................................................................... 119
Hadoop support............................................................................................................................... 120
Operations.................................................................................................................... 123
Monitoring Cassandra..................................................................................................................... 123
Monitoring a Cassandra cluster............................................................................................123
Tuning Bloom filters........................................................................................................................ 128
Data caching....................................................................................................................................128
Configuring data caches.......................................................................................................128
Monitoring and adjusting caching.........................................................................................131
Configuring memtable throughput................................................................................................... 132
Configuring compaction................................................................................................................... 132
Compression.................................................................................................................................... 133
When to compress data....................................................................................................... 134
Configuring compression...................................................................................................... 134
Testing compaction and compression.............................................................................................135
Tuning Java resources.................................................................................................................... 135
Purging gossip state on a node......................................................................................................137
Repairing nodes.............................................................................................................................. 138
Adding or removing nodes, data centers, or clusters..................................................................... 142
Adding nodes to an existing cluster..................................................................................... 142
Adding a data center to a cluster.........................................................................................144
Replacing a dead node or dead seed node........................................................................ 145
Replacing a running node.................................................................................................... 147
Moving a node from one rack to another.............................................................................148
Decommissioning a data center........................................................................................... 148
Removing a node................................................................................................................. 148
Switching snitches................................................................................................................ 149
Edge cases for transitioning or migrating a cluster..............................................................150
Adding or replacing single-token nodes............................................................................... 151
Backing up and restoring data.................................................................................. 154

About snapshots..............................................................................................................................154
Taking a snapshot...........................................................................................................................154
Deleting snapshot files.................................................................................................................... 155
Enabling incremental backups........................................................................................................ 155
Restoring from a snapshot..............................................................................................................156
Node restart method.............................................................................................................157
Restoring a snapshot into a new cluster........................................................................................ 157
5
Contents
Recovering using JBOD.................................................................................................................. 158
Cassandra tools........................................................................................................... 160

The nodetool utility.......................................................................................................................... 160
cfhistograms.......................................................................................................................... 160
cfstats.................................................................................................................................... 161
cleanup..................................................................................................................................167
clearsnapshot........................................................................................................................ 168
compact.................................................................................................................................168
compactionhistory................................................................................................................. 169
compactionstats.................................................................................................................... 171
decommission....................................................................................................................... 172
describecluster...................................................................................................................... 172
describering...........................................................................................................................173
disableautocompaction......................................................................................................... 175
disablebackup....................................................................................................................... 175
disablebinary......................................................................................................................... 176
disablegossip........................................................................................................................ 176
disablehandoff....................................................................................................................... 177
disablethrift............................................................................................................................ 177
drain...................................................................................................................................... 178
enableautocompaction.......................................................................................................... 179
enablebackup........................................................................................................................ 179
enablebinary..........................................................................................................................180
enablegossip......................................................................................................................... 180
enablehandoff....................................................................................................................... 181
enablethrift............................................................................................................................ 181
flush.......................................................................................................................................181
getcompactionthreshold........................................................................................................ 182
getendpoints..........................................................................................................................183
getlogginglevels.................................................................................................................... 183
getsstables............................................................................................................................ 184
getstreamthroughput............................................................................................................. 184
gossipinfo.............................................................................................................................. 185
help....................................................................................................................................... 185
info........................................................................................................................................ 186
invalidatekeycache................................................................................................................ 187
invalidaterowcache................................................................................................................187
join.........................................................................................................................................188
listsnapshots......................................................................................................................... 188
move..................................................................................................................................... 189
netstats..................................................................................................................................190
pausehandoff........................................................................................................................ 191
proxyhistograms.................................................................................................................... 191
rangekeysample.................................................................................................................... 192
rebuild................................................................................................................................... 193
rebuild_index......................................................................................................................... 193
refresh................................................................................................................................... 194
reloadtriggers........................................................................................................................ 194
removenode.......................................................................................................................... 195
repair..................................................................................................................................... 196
resetlocalschema.................................................................................................................. 198
resumehandoff...................................................................................................................... 199
ring........................................................................................................................................ 200
scrub..................................................................................................................................... 201
6
Contents
setcachecapacity...................................................................................................................202
setcachekeystosave.............................................................................................................. 202
setcompactionthreshold........................................................................................................ 203
setcompactionthroughput...................................................................................................... 204
sethintedhandoffthrottlekb..................................................................................................... 205
setlogginglevel...................................................................................................................... 205
setstreamthroughput............................................................................................................. 206
settraceprobability................................................................................................................. 207
snapshot................................................................................................................................208
status.....................................................................................................................................210
statusbackup......................................................................................................................... 211
statusbinary........................................................................................................................... 212
statusgossip.......................................................................................................................... 212
statushandoff.........................................................................................................................213
statusthrift..............................................................................................................................213
stop....................................................................................................................................... 214
stopdaemon.......................................................................................................................... 214
tpstats....................................................................................................................................215
truncatehints..........................................................................................................................219
upgradesstables.................................................................................................................... 219
version...................................................................................................................................220
Cassandra bulk loader (sstableloader)........................................................................................... 221
The cassandra utility....................................................................................................................... 223
The cassandra-stress tool............................................................................................................... 226
Using the Daemon Mode..................................................................................................... 230
Interpreting the output of cassandra-stress..........................................................................231
The sstablescrub utility....................................................................................................................232
The sstablesplit utility...................................................................................................................... 233
The sstablekeys utility..................................................................................................................... 234
The sstableupgrade tool..................................................................................................................234
References.................................................................................................................... 236
Starting and stopping Cassandra....................................................................................................236
Starting Cassandra as a service.......................................................................................... 236
Starting Cassandra as a stand-alone process..................................................................... 236
Stopping Cassandra as a service........................................................................................ 236
Stopping Cassandra as a stand-alone process................................................................... 237
Clearing the data as a service............................................................................................. 237
Clearing the data as a stand-alone process........................................................................ 237
Install locations................................................................................................................................ 238
Tarball installation directories............................................................................................... 238
Package installation directories............................................................................................ 238
Cassandra include file..................................................................................................................... 239
Cassandra-CLI utility (deprecated)..................................................................................................239
Table attributes..................................................................................................................... 240
Moving data to/from other databases....................................................................... 243
Troubleshooting........................................................................................................... 244
Peculiar Linux kernel performance problem on NUMA systems.....................................................244
Reads are getting slower while writes are still fast.........................................................................244
Nodes seem to freeze after some period of time........................................................................... 244
Nodes are dying with OOM errors..................................................................................................245
7
Contents
Nodetool or JMX connections failing on remote nodes.................................................................. 245

Handling schema disagreements.................................................................................................... 245
View of ring differs between some nodes.......................................................................................246
Java reports an error saying there are too many open files...........................................................246
Insufficient user resource limits errors............................................................................................ 246
Cannot initialize class org.xerial.snappy.Snappy............................................................................ 248
Firewall idle connection timeout causing nodes to lose communication......................................... 248
Release notes...............................................................................................................250
Using the docs.............................................................................................................252
8
About Apache Cassandra

Documentation for developers and administrators on installing, configuring, and using the features and
capabilities of Apache Cassandra scalable open source NoSQL database.
Apache Cassandra™ is a massively scalable open source NoSQL database. Cassandra is perfect for
managing large amounts of structured, semi-structured, and unstructured data across multiple data centers
and the cloud. Cassandra delivers continuous availability, linear scalability, and operational simplicity
across many commodity servers with no single point of failure, along with a powerful dynamic data model
designed for maximum flexibility and fast response times.
How does Cassandra work?

Cassandra’s built-for-scale architecture means that it is capable of handling petabytes of information and
thousands of concurrent users/operations per second.
Cassandra is a Cassandra's architecture allows any authorized user to connect to any node
partitioned row store in any data center and access data using the CQL language. For ease of
database use, CQL uses a similar syntax to SQL. The most basic way to interact
with Cassandra is using the CQL shell, cqlsh. Using cqlsh, you can create
keyspaces and tables, insert and query tables, plus much more. If you prefer
a graphical tool, you can use DataStax DevCenter. For production, DataStax
supplies a number drivers so that CQL statements can be passed from client
to cluster and back. Other administrative tasks can be accomplished using
OpsCenter.
Automatic data Cassandra provides automatic data distribution across all nodes that
distribution participate in a ring or database cluster. There is nothing programmatic that
a developer or administrator needs to do or code to distribute data across a
cluster because data is transparently partitioned across all nodes in a cluster.
Built-in and Cassandra also provides built-in and customizable replication, which stores
customizable redundant copies of data across nodes that participate in a Cassandra ring.
replication This means that if any node in a cluster goes down, one or more copies of that
node’s data is available on other machines in the cluster. Replication can be
configured to work across one data center, many data centers, and multiple
cloud availability zones.
Cassandra supplies Cassandra supplies linear scalability, meaning that capacity may be easily
linear scalability added simply by adding new nodes online. For example, if 2 nodes can handle
100,000 transactions per second, 4 nodes will support 200,000 transactions/
sec and 8 nodes will tackle 400,000 transactions/sec:
9
10
What's new in Cassandra 2.1
What's new in Cassandra 2.1

An overview of new features in Cassandra.
New features:
• User-defined types
• Collection indexes
• Better implementation of counters that makes them safer, simpler, and typically faster
• New listsnapshots and reloadtriggers nodetool commands
• Improved metrics reporting through the use of the metrics-core library
Performance improvements:
• Faster reads and writes than previous releases
• Improved row cache
• Reduced heap used by memtables
• New counters implementation
Compaction and repair improvements:
• Post-compaction read performance
• A configurable percentage of cold SSTables can be ignored
• Incremental node repair
Other notable changes:
• Improved Hadoop support
• Unique table IDs
• Improved logging using logback
• New configuration options for allocating and managing memtable memory
• Improvements to bootstrapping a node that ensure data consistency
• Bundled JNA
• A number of other CQL and cqlsh changes
Release notes cover key changes in Cassandra 2.1. For a detailed list of changes, see the change log.
11
CQL
CQL
Cassandra Query Language (CQL) is the default and primary interface into the Cassandra DBMS.
Cassandra Query Language (CQL) is the default and primary interface into the Cassandra DBMS. Using
CQL is similar to using SQL (Structured Query Language). CQL and SQL share the same abstract idea
of a table constructed of columns and rows. The main difference from SQL is that Cassandra does not
support joins or subqueries. Instead, Cassandra emphasizes denormalization through CQL features like
collections and clustering specified at the schema level.
CQL is the recommended way to interact with Cassandra. Performance and the simplicity of reading and
using CQL is an advantage of modern Cassandra over older Cassandra APIs.
The CQL documentation contains a data modeling section, examples, and command reference. The cqlsh
utility for using CQL interactively on the command line is also covered.
12
Understanding the architecture

Important topics for understanding Cassandra.
Architecture in brief
Essential information for understanding and using Cassandra.
Cassandra is designed to handle big data workloads across multiple nodes with no single point of failure.
Its architecture is based on the understanding that system and hardware failures can and do occur.
Cassandra addresses the problem of failures by employing a peer-to-peer distributed system across
homogeneous nodes where data is distributed among all nodes in the cluster. Each node exchanges
information across the cluster every second. A sequentially written commit log on each node captures
write activity to ensure data durability. Data is then indexed and written to an in-memory structure, called
a memtable, which resembles a write-back cache. Once the memory structure is full, the data is written to
disk in an SSTable data file. All writes are automatically partitioned and replicated throughout the cluster.
Using a process called compaction Cassandra periodically consolidates SSTables, discarding obsolete
data and tombstones (an indicator that data was deleted).
Cassandra is a row-oriented database. Cassandra's architecture allows any authorized user to connect
to any node in any data center and access data using the CQL language. For ease of use, CQL uses
a similar syntax to SQL. From the CQL perspective the database consists of tables. Typically, a cluster
has one keyspace per application. Developers can access CQL through cqlsh as well as via drivers for
application languages.
Client read or write requests can be sent to any node in the cluster. When a client connects to a node with
a request, that node serves as the coordinator for that particular client operation. The coordinator acts as
a proxy between the client application and the nodes that own the data being requested. The coordinator
determines which nodes in the ring should get the request based on how the cluster is configured. For
more information, see Client requests.
Key structures
• Node
Where you store your data. It is the basic infrastructure component of Cassandra.
• Data center
A collection of related nodes. A data center can be a physical data center or virtual data center.
Different workloads should use separate data centers, either physical or virtual. Replication is set by
data center. Using separate data centers prevents Cassandra transactions from being impacted by
other workloads and keeps requests close to each other for lower latency. Depending on the replication
factor, data can be written to multiple data centers. However, data centers should never span physical
locations.
• Cluster
A cluster contains one or more data centers. It can span physical locations.
• Commit log
All data is written first to the commit log for durability. After all its data has been flushed to SSTables, it
can be archived, deleted, or recycled.
• Table
A collection of ordered columns fetched by row. A row consists of columns and have a primary key. The
first part of the key is a column name.
• SSTable
13
A sorted string table (SSTable) is an immutable data file to which Cassandra writes memtables
periodically. SSTables are append only and stored on disk sequentially and maintained for each
Cassandra table.
Key components for configuring Cassandra

• Gossip
A peer-to-peer communication protocol to discover and share location and state information about the
other nodes in a Cassandra cluster. Gossip information is also persisted locally by each node to use
immediately when a node restarts.
• Partitioner
A partitioner determines how to distribute the data across the nodes in the cluster and which node to
place the first copy of data on. Basically, a partitioner is a hash function for computing the token of a
partition key. Each row of data is uniquely identified by a partition key and distributed across the cluster
by the value of the token. The Murmur3Partitioner is the default partitioning strategy for new Cassandra
clusters and the right choice for new clusters in almost all cases.
You must set the partitioner and assign the node a num_tokens value for each node. The number
of tokens you assign depends on the hardware capabilities of the system. If not using virtual nodes
(vnodes), use the initial_token setting instead.
• Replication factor
The total number of replicas across the cluster. A replication factor of 1 means that there is only one
copy of each row on one node. A replication factor of 2 means two copies of each row, where each
copy is on a different node. All replicas are equally important; there is no primary or master replica.
You define the replication factor for each data center. Generally you should set the replication strategy
greater than one, but no more than the number of nodes in the cluster.
• Replica placement strategy
Cassandra stores copies (replicas) of data on multiple nodes to ensure reliability and fault tolerance.
A replication strategy determines which nodes to place replicas on. The first replica of data is simply
the first copy; it is not unique in any sense. The NetworkTopologyStrategy is highly recommended for
most deployments because it is much easier to expand to multiple data centers when required by future
expansion.
When creating a keyspace, you must define the replica placement strategy and the number of replicas
you want.
• Snitch
A snitch defines groups of machines into data centers and racks (the topology) that the replication
strategy uses to place replicas.
You must configure a snitch when you create a cluster. All snitches use a dynamic snitch layer,
which monitors performance and chooses the best replica for reading. It is enabled by default and
recommended for use in most deployments. Configure dynamic snitch thresholds for each node in the
cassandra.yaml configuration file.
The default SimpleSnitch does not recognize data center or rack information. Use it for single-data
center deployments or single-zone in public clouds. The GossipingPropertyFileSnitch is recommended
for production. It defines a node's data center and rack and uses gossip for propagating this information
to other nodes.
• The cassandra.yaml configuration file
The main configuration file for setting the initialization properties for a cluster, caching parameters for
tables, properties for tuning and resource utilization, timeout settings, client connections, backups, and
security.
By default, a node is configured to store the data it manages in a directory set in the cassandra.yaml
file.
14
• Package installations: /var/lib/cassandra

• Tarball installations: install_location/data/data
In a production cluster deployment, you can change the commitlog-directory to a different disk drive
from the data_file_directories.
• System keyspace table properties
You set storage configuration attributes on a per-keyspace or per-table basis programmatically or using
a client application, such as CQL.
Internode communications (gossip)

Cassandra uses a protocol called gossip to discover location and state information about the other nodes
participating in a Cassandra cluster.
Gossip is a peer-to-peer communication protocol in which nodes periodically exchange state information
about themselves and about other nodes they know about. The gossip process runs every second and
exchanges state messages with up to three other nodes in the cluster. The nodes exchange information
about themselves and about the other nodes that they have gossiped about, so all nodes quickly learn
about all other nodes in the cluster. A gossip message has a version associated with it, so that during a
gossip exchange, older information is overwritten with the most current state for a particular node.
To prevent problems in gossip communications, use the same list of seed nodes for all nodes in a cluster.
This is most critical the first time a node starts up. By default, a node remembers other nodes it has
gossiped with between subsequent restarts. The seed node designation has no purpose other than
bootstrapping the gossip process for new nodes joining the cluster. Seed nodes are not a single point
of failure, nor do they have any other special purpose in cluster operations beyond the bootstrapping of
nodes.
Attention: In multiple data-center clusters, the seed list should include at least one node from each
data center (replication group). More than a single seed node per data center is recommended for
fault tolerance. Otherwise, gossip has to communicate with another data center when bootstrapping
a node. Making every node a seed node is not recommended because of increased maintenance
and reduced gossip performance. Gossip optimization is not critical, but it is recommended to use a
small seed list (approximately three nodes per data center).
Failure detection and recovery

A method for locally determining from gossip state and history if another node in the system is down or has
come back up.
Failure detection is a method for locally determining from gossip state and history if another node in the
system is down or has come back up. Cassandra uses this information to avoid routing client requests to
unreachable nodes whenever possible. (Cassandra can also avoid routing requests to nodes that are alive,
but performing poorly, through the dynamic snitch.)
The gossip process tracks state from other nodes both directly (nodes gossiping directly to it) and indirectly
(nodes communicated about secondhand, third-hand, and so on). Rather than have a fixed threshold
for marking failing nodes, Cassandra uses an accrual detection mechanism to calculate a per-node
threshold that takes into account network performance, workload, and historical conditions. During gossip
exchanges, every node maintains a sliding window of inter-arrival times of gossip messages from other
nodes in the cluster. Configuring the phi_convict_threshold property adjusts the sensitivity of the failure
detector. Lower values increase the likelihood that an unresponsive node will be marked as down, while
higher values decrease the likelihood that transient failures causing node failure. Use the default value
for most situations, but increase it to 10 or 12 for Amazon EC2 (due to frequently encountered network
congestion). In unstable network environments (such as EC2 at times), raising the value to 10 or 12 helps
prevent false failures. Values higher than 12 and lower than 5 are not recommended.
15
Node failures can result from various causes such as hardware failures and network outages. Node
outages are often transient but can last for extended periods. Because a node outage rarely signifies a
permanent departure from the cluster it does not automatically result in permanent removal of the node
from the ring. Other nodes will periodically try to re-establish contact with failed nodes to see if they are
back up. To permanently change a node's membership in a cluster, administrators must explicitly add or
remove nodes from a Cassandra cluster using the nodetool utility or OpsCenter.
When a node comes back online after an outage, it may have missed writes for the replica data
it maintains. Once the failure detector marks a node as down, missed writes are stored by other
replicas for a period of time providing hinted handoff is enabled. If a node is down for longer than
max_hint_window_in_ms (3 hours by default), hints are no longer saved. Nodes that die may have stored
undelivered hints. Run a repair after recovering a node that has been down for an extended period.
Moreover, you should routinely run nodetool repair on all nodes to ensure they have consistent data.
For more explanation about hint storage, see Modern hinted handoff.
Data distribution and replication

How data is distributed and factors influencing replication.
In Cassandra, data distribution and replication go together. Data is organized by table and identified by a
primary key, which determines which node the data is stored on. Replicas are copies of rows. When data is
first written, it is also referred to as a replica.
Factors influencing replication include:
• Virtual nodes: assigns data ownership to physical machines.
• Partitioner: partitions the data across the cluster.
• Replication strategy: determines the replicas for each row of data.
• Snitch: defines the topology information that the replication strategy uses to place replicas.
Consistent hashing
Consistent hashing allows distribution of data across a cluster to minimize reorganization when nodes are
added or removed.
Consistent hashing allows distribution of data across a cluster to minimize reorganization when nodes are
added or removed. Consistent hashing partitions data based on the partition key. (For an explanation of
partition keys and primary keys, see the Data modeling example in CQL for Cassandra 2.0.)
For example, if you have the following data:
name age car gender

jim 36 camaro M
carol 37 bmw F
johnny 12 M
suzy 10 F
Cassandra assigns a hash value to each partition key:
Partition key Murmur3 hash value

jim -2245462676723223822
carol 7723358927203680754
johnny -6723372854036780875
16
Partition key Murmur3 hash value

suzy 1168604627387940318
Each node in the cluster is responsible for a range of data based on the hash value:
Hash values in a 4 node cluster
- 9223372036854775808
to
A
- 4611686018427387903
- 4611686018427387904
to
-1
D Data Center Alpha B
4611686018427387904
to
9223372036854775807 C
-1
to
4611686018427387903
Cassandra places the data on each node according to the value of the partition key and the range that
the node is responsible for. For example, in a four node cluster, the data in this example is distributed as
follows:
Node Start range End range PartitionHash value

key
A -9223372036854775808 -4611686018427387903 johnny -6723372854036780875
B -4611686018427387904 -1 jim -2245462676723223822
C 0 4611686018427387903 suzy 1168604627387940318
D 4611686018427387904 9223372036854775807 carol 7723358927203680754
Virtual nodes
Overview of virtual nodes (vnodes).
Vnodes simplify many tasks in Cassandra:
• You no longer have to calculate and assign tokens to each node.
• Rebalancing a cluster is no longer necessary when adding or removing nodes. When a node joins the
cluster, it assumes responsibility for an even portion of data from the other nodes in the cluster. If a
node fails, the load is spread evenly across other nodes in the cluster.
• Rebuilding a dead node is faster because it involves every other node in the cluster.
• Improves the use of heterogeneous machines in a cluster. You can assign a proportional number of
vnodes to smaller and larger machines.
For more information, see the article Virtual nodes in Cassandra 1.2 and Enabling virtual nodes on an
existing production cluster.
To convert an existing cluster to vnodes, see Enabling virtual nodes on an existing production cluster.
17
How data is distributed across a cluster (using virtual nodes)

Vnodes use consistent hashing to distribute data without requiring new token generation and assignment.
Prior to version 1.2, you had to calculate and assign a single token to each node in a cluster. Each token
determined the node's position in the ring and its portion of data according to its hash value. Starting in
version 1.2, Cassandra allows many tokens per node. The new paradigm is called virtual nodes (vnodes).
Vnodes allow each node to own a large number of small partition ranges distributed throughout the cluster.
Vnodes also use consistent hashing to distribute data but using them doesn't require token generation and
assignment.
Figure 1: Virtual vs single-token architecture
The top portion of the graphic shows a cluster without vnodes. In this paradigm, each node is assigned
a single token that represents a location in the ring. Each node stores data determined by mapping the
partition key to a token value within a range from the previous node to its assigned value. Each node also
contains copies of each row from other nodes in the cluster. For example, range E replicates to nodes 5, 6,
and 1. Notice that a node owns exactly one contiguous partition range in the ring space.
The bottom portion of the graphic shows a ring with vnodes. Within a cluster, virtual nodes are randomly
selected and non-contiguous. The placement of a row is determined by the hash of the partition key within
many smaller partition ranges belonging to each node.
Data replication
Cassandra stores replicas on multiple nodes to ensure reliability and fault tolerance. A replication strategy
determines the nodes where replicas are placed.
Cassandra stores replicas on multiple nodes to ensure reliability and fault tolerance. A replication strategy
determines the nodes where replicas are placed. The total number of replicas across the cluster is referred
to as the replication factor. A replication factor of 1 means that there is only one copy of each row on one
node. A replication factor of 2 means two copies of each row, where each copy is on a different node. All
replicas are equally important; there is no primary or master replica. As a general rule, the replication factor
18
should not exceed the number of nodes in the cluster. However, you can increase the replication factor and
then add the desired number of nodes later.
Two replication strategies are available:
• SimpleStrategy: Use for a single data center only. If you ever intend more than one data center, use
the NetworkTopologyStrategy.
• NetworkTopologyStrategy: Highly recommended for most deployments because it is much easier
to expand to multiple data centers when required by future expansion.
SimpleStrategy
Use only for a single data center. SimpleStrategy places the first replica on a node determined by
the partitioner. Additional replicas are placed on the next nodes clockwise in the ring without considering
topology (rack or data center location).
NetworkTopologyStrategy
Use NetworkTopologyStrategy when you have (or plan to have) your cluster deployed across multiple
data centers. This strategy specify how many replicas you want in each data center.
NetworkTopologyStrategy places replicas in the same data center by walking the ring clockwise until
reaching the first node in another rack. NetworkTopologyStrategy attempts to place replicas on distinct
racks because nodes in the same rack (or similar physical grouping) often fail at the same time due to
power, cooling, or network issues.
When deciding how many replicas to configure in each data center, the two primary considerations are (1)
being able to satisfy reads locally, without incurring cross data-center latency, and (2) failure scenarios. The
two most common ways to configure multiple data center clusters are:
• Two replicas in each data center: This configuration tolerates the failure of a single node per replication
group and still allows local reads at a consistency level of ONE.
• Three replicas in each data center: This configuration tolerates either the failure of a one node per
replication group at a strong consistency level of LOCAL_QUORUM or multiple node failures per data
center using consistency level ONE.
Asymmetrical replication groupings are also possible. For example, you can have three replicas in one data
center to serve real-time application requests and use a single replica elsewhere for running analytics.
Choosing keyspace replication options

Options to set replication strategy.
To set the replication strategy for a keyspace, see CREATE KEYSPACE.
When you use NetworkToplogyStrategy, during creation of the keyspace, you use the data center
names defined for the snitch used by the cluster. To place replicas in the correct location, Cassandra
requires a keyspace definition that uses the snitch-configured data center names. For example, if the
cluster uses the PropertyFileSnitch, create the keyspace using the user-defined data center and rack
names in the cassandra-topologies.properties file. If the cluster uses the Ec2Snitch, create the
keyspace using EC2 data center and rack names. If the cluster uses the GoogleCloudSnitch, create the
keyspace using GoogleCloud data center and rack names.
Partitioners
A partitioner determines how data is distributed across the nodes in the cluster (including replicas).
A partitioner determines how data is distributed across the nodes in the cluster (including replicas).
Basically, a partitioner is a function for deriving a token representing a row from its partion key, typically by
hashing. Each row of data is then distributed across the cluster by the value of the token.
Both the Murmur3Partitioner and RandomPartitioner use tokens to help assign equal portions
of data to each node and evenly distribute data from all the tables throughout the ring or other grouping,
such as a keyspace. This is true even if the tables use different partition keys, such as usernames or
19
timestamps. Moreover, the read and write requests to the cluster are also evenly distributed and load
balancing is simplified because each part of the hash range receives an equal number of rows on average.
For more detailed information, see Consistent hashing.
Cassandra offers the following partitioners:
• Murmur3Partitioner (default): uniformly distributes data across the cluster based on MurmurHash
hash values.
• RandomPartitioner: uniformly distributes data across the cluster based on MD5 hash values.
• ByteOrderedPartitioner: keeps an ordered distribution of data lexically by key bytes
The Murmur3Partitioner is the default partitioning strategy for new Cassandra clusters and the right
choice for new clusters in almost all cases.
Set the partitioner in the cassandra.yaml file:
• Murmur3Partitioner: org.apache.cassandra.dht.Murmur3Partitioner
• RandomPartitioner: org.apache.cassandra.dht.RandomPartitioner
• ByteOrderedPartitioner: org.apache.cassandra.dht.ByteOrderedPartitioner
Note: If using virtual nodes (vnodes), you do not need to calculate the tokens. If not using vnodes,
you must calculate the tokens to assign to the initial_token parameter in the cassandra.yaml file.
See Generating tokens and use the method for the type of partitioner you are using.
The location of the cassandra.yaml file depends on the type of installation:
Package installations /etc/cassandra/cassandra.yaml

Tarball installations install_location/resources/cassandra/
conf/cassandra.yaml
Murmur3Partitioner
The Murmur3Partitioner provides faster hashing and improved performance than the previous default
partitioner.
The Murmur3Partitioner provides faster hashing and improved performance than the previous default
partitioner (RandomPartitioner). You can only use Murmur3Partitioner for new clusters; you cannot
change the partitioner in existing clusters. The Murmur3Partitioner uses the MurmurHash function.
This hashing function creates a 64-bit hash value of the partition key. The possible range of hash values is
63 63
from -2 to +2 -1.
When using the Murmur3Partitioner, you can page through all rows using the token function in a CQL
query.
RandomPartitioner
The default partitioner prior to Cassandra 1.2.
The RandomPartitioner was the default partitioner prior to Cassandra 1.2. It is included for backwards
compatibility. You can use it in later versions of Cassandra, even when using virtual nodes (vnodes).
However, if you don't use vnodes, you must calculate the tokens, as described in Generating tokens.
The RandomPartition distributes data evenly across the nodes using an MD5 hash value of the row
127
key. The possible range of hash values is from 0 to 2 -1.
When using the RandomPartitioner, you can page through all rows using the token function in a CQL
query.
20
ByteOrderedPartitioner
Cassandra provides this partitioner for ordered partitioning. It is included for backwards compatibility.
Cassandra provides the ByteOrderedPartitioner for ordered partitioning. It is included for backwards
compatibility. This partitioner orders rows lexically by key bytes. You calculate tokens by looking at the
actual values of your partition key data and using a hexadecimal representation of the leading character(s)
in a key. For example, if you wanted to partition rows alphabetically, you could assign an A token using its
hexadecimal representation of 41.
Using the ordered partitioner allows ordered scans by primary key. This means you can scan rows as
though you were moving a cursor through a traditional index. For example, if your application has user
names as the partition key, you can scan rows for users whose names fall between Jake and Joe. This
type of query is not possible using randomly partitioned partition keys because the keys are stored in the
order of their MD5 hash (not sequentially).
Although having the capability to do range scans on rows sounds like a desirable feature of ordered
partitioners, there are ways to achieve the same functionality using table indexes.
Using an ordered partitioner is not recommended for the following reasons:
Difficult load balancing
More administrative overhead is required to load balance the cluster. An ordered partitioner requires
administrators to manually calculate partition ranges based on their estimates of the partition key distribution.
In practice, this requires actively moving node tokens around to accommodate the actual distribution of data
once it is loaded.
Sequential writes can cause hot spots
If your application tends to write or update a sequential block of rows at a time, then the writes are not be
distributed across the cluster; they all go to one node. This is frequently a problem for applications dealing
with timestamped data.
Uneven load balancing for multiple tables
If your application has multiple tables, chances are that those tables have different row keys and different
distributions of data. An ordered partitioner that is balanced for one table may cause hot spots and uneven
distribution for another table in the same cluster.
Snitches
A snitch determines which data centers and racks nodes belong to.
A snitch determines which data centers and racks nodes belong to. They inform Cassandra about the
network topology so that requests are routed efficiently and allows Cassandra to distribute replicas by
grouping machines into data centers and racks. Specifically, the replication strategy places the replicas
based on the information provided by the new snitch. All nodes must return to the same rack and data
center. Cassandra does its best not to have more than one replica on the same rack (which is not
necessarily a physical location).
Note: If you change snitches, you may need to perform additional steps because the snitch affects
where replicas are placed. See Switching snitches.
Dynamic snitching
Monitors the performance of reads from the various replicas and chooses the best replica based on this
history.
By default, all snitches also use a dynamic snitch layer that monitors read latency and, when possible,
routes requests away from poorly-performing nodes. The dynamic snitch is enabled by default and is
recommended for use in most deployments. For information on how this works, see Dynamic snitching
in Cassandra: past, present, and future. Configure dynamic snitch thresholds for each node in the
cassandra.yaml configuration file.
21
For more information, see the properties listed under Failure detection and recovery.

conf/cassandra.yaml
SimpleSnitch
The SimpleSnitch is used only for single-data center deployments.
The SimpleSnitch (default) is used only for single-data center deployments. It does not recognize data
center or rack information and can be used only for single-data center deployments or single-zone in public
clouds. It treats strategy order as proximity, which can improve cache locality when disabling read repair.
Using a SimpleSnitch, you define the keyspace to use SimpleStrategy and specify a replication factor.
RackInferringSnitch
Determines the location of nodes by rack and data center corresponding to the IP addresses.
The RackInferringSnitch determines the proximity of nodes by rack and data center, which are assumed to
correspond to the 3rd and 2nd octet of the node's IP address, respectively. This snitch is best used as an
example for writing a custom snitch class (unless this happens to match your deployment conventions).
PropertyFileSnitch
Determines the location of nodes by rack and data center.
About this task

This snitch determines proximity is determined by rack and data center. It uses the network details located
in the cassandra-topology.properties file. When using this snitch, you can define your data center names
to be whatever you want. Make sure that the data center names correlate to the name of your data
centers in the keyspace definition. Every node in the cluster should be described in the cassandra-
topology.properties file, and this file should be exactly the same on every node in the cluster.
The location of the cassandra-topology.properties file depends on the type of installation:
Package installations /etc/cassandra/cassandra-

topology.properties
Tarball installations install_location/conf/cassandra-
topology.properties
Procedure
If you had non-uniform IPs and two physical data centers with two racks in each, and a third logical data
center for replicating analytics data, the cassandra-topology.properties file might look like this:
Note: Data center and rack names are case-sensitive.
# Data Center One
22
175.56.12.105=DC1:RAC1
175.50.13.200=DC1:RAC1
175.54.35.197=DC1:RAC1
120.53.24.101=DC1:RAC2
120.55.16.200=DC1:RAC2
120.57.102.103=DC1:RAC2
# Data Center Two
110.56.12.120=DC2:RAC1
110.50.13.201=DC2:RAC1
110.54.35.184=DC2:RAC1
50.33.23.120=DC2:RAC2
50.45.14.220=DC2:RAC2
50.17.10.203=DC2:RAC2
# Analytics Replication Group
172.106.12.120=DC3:RAC1
172.106.12.121=DC3:RAC1
172.106.12.122=DC3:RAC1
# default for unknown nodes default =DC3:RAC1
GossipingPropertyFileSnitch
Automatically updates all nodes using gossip when adding new nodes and is recommended for production.
This snitch is recommended for production. It uses rack and data center information for the local node
defined in the cassandra-rackdc.properties file and propagates this information to other nodes via
gossip.
The cassandra-rackdc.properties file defines the default data center and rack used by this snitch:
dc=DC1
rack=RAC1
To save bandwidth, add the prefer_local=true option. This option tells Cassandra to use the local IP
address when communication is not across different data centers.
To allow migration from the PropertyFileSnitch, the GossipingPropertyFileSnitch uses the cassandra-
topology.properties file when present.
The location of the cassandra-rackdc.properties file depends on the type of installation:

rackdc.properties
rackdc.properties
Ec2Snitch
Use the Ec2Snitch with Amazon EC2 in a single region.
Use the Ec2Snitch for simple cluster deployments on Amazon EC2 where all nodes in the cluster are within
a single region.
23
In EC2 deployments , the region name is treated as the data center name and availability zones are
treated as racks within a data center. For example, if a node is in the us-east-1 region, us-east is the data
center name and 1 is the rack location. (Racks are important for distributing replicas, but not for data center
naming.) Because private IPs are used, this snitch does not work across multiple regions.
If you are using only a single data center, you do not need to specify any properties.
If you need multiple data centers, set the dc_suffix options in the cassandra-rackdc.properties file. Any
other lines are ignored.
For example, for each node within the us-east region, specify the data center in its cassandra-
rackdc.properties file:
Note: Data center names are case-sensitive.
• node0
dc_suffix=_1_cassandra
• node1
• node2
• node3
• node4
dc_suffix=_1_analytics
• node5
dc_suffix=_1_search
This results in three data centers for the region:
us-east_1_cassandra
us-east_1_analytics
us-east_1_search
Note: The data center naming convention in this example is based on the workload. You can use
other conventions, such as DC1, DC2 or 100, 200.
Keyspace strategy options

When defining your keyspace strategy options, use the EC2 region name, such as us-east, as your data
center name.

rackdc.properties
rackdc.properties
Ec2MultiRegionSnitch
Use the Ec2MultiRegionSnitch for deployments on Amazon EC2 where the cluster spans multiple regions.
Use the Ec2MultiRegionSnitch for deployments on Amazon EC2 where the cluster spans multiple regions.
You must configure settings in both the cassandra.yaml file and the property file (cassandra-
rackdc.properties) used by the Ec2MultiRegionSnitch.
24
Configuring cassandra.yaml for cross-region communication

The Ec2MultiRegionSnitch uses public IP designated in the broadcast_address to allow cross-region
connectivity. Configure each node as follows:
1. In the cassandra.yaml, set the listen_address to the private IP address of the node, and the
broadcast_address to the public IP address of the node.
This allows Cassandra nodes in one EC2 region to bind to nodes in another region, thus enabling
multiple data center support. For intra-region traffic, Cassandra switches to the private IP after
establishing a connection.
2. Set the addresses of the seed nodes in the cassandra.yaml file to that of the public IP. Private IP are
not routable between networks. For example:
seeds: 50.34.16.33, 60.247.70.52
To find the public IP address, from each of the seed nodes in EC2:
$ curl http://instance-data/latest/meta-data/public-ipv4
Note: Do not make all nodes seeds, see Internode communications (gossip).
3. Be sure that the storage_port or ssl_storage_port is open on the public IP firewall.
Configuring the snitch for cross-region communication

In EC2 deployments, the region name is treated as the data center name and availability zones are treated
as racks within a data center. For example, if a node is in the us-east-1 region, us-east is the data center
name and 1 is the rack location. (Racks are important for distributing replicas, but not for data center
naming.)
For each node, specify its data center in the cassandra-rackdc.properties. The dc_suffix option defines the
data centers used by the snitch. Any other lines are ignored.
In the example below, there are two cassandra data centers and each data center is named for its
workload. The data center naming convention in this example is based on the workload. You can use other
conventions, such as DC1, DC2 or 100, 200. (Data center names are case-sensitive.)
Region: us-east Region: us-west

Node and data center: Node and data center:
• node0 • node0
dc_suffix=_1_cassandra dc_suffix=_1_cassandra
• node1 • node1
• node2 • node2
• node3 • node3
• node4 • node4
dc_suffix=_1_analytics dc_suffix=_1_analytics
• node5 • node5
dc_suffix=_1_search dc_suffix=_1_search
This results in four us-east data centers: This results in four us-west data centers:
us-east_1_cassandra us-west_1_cassandra
25

us-east_1_analytics us-west_1_analytics
us-east_1_search us-west_1_search
Keyspace strategy options

When defining your keyspace strategy options, use the EC2 region name, such as us-east, as your data
center name.

rackdc.properties
rackdc.properties

conf/cassandra.yaml
GoogleCloudSnitch
Use the GoogleCloudSnitch for Cassandra deployments on Google Cloud Platform across one or more
regions.
Use the GoogleCloudSnitch for Cassandra deployments on Google Cloud Platform across one or more
regions. The region is treated as a data center and the availability zones are treated as racks within the
data center. All communication occurs over private IP addresses within the same logical network.
The region name is treated as the data center name and zones are treated as racks within a data center.
For example, if a node is in the us-central1-a region, us-central1 is the data center name and a is the rack
location. (Racks are important for distributing replicas, but not for data center naming.) This snitch can
work across multiple regions without additional configuration.
If you are using only a single data center, you do not need to specify any properties.
If you need multiple data centers, set the dc_suffix options in the cassandra-rackdc.properties file. Any
other lines are ignored.
For example, for each node within the us-central1 region, specify the data center in its cassandra-
rackdc.properties file:
Note: Data center names are case-sensitive.
• node0
dc_suffix=_a_cassandra
• node1
• node2
• node3
26
• node4
dc_suffix=_a_analytics
• node5
dc_suffix=_a_search
CloudstackSnitch
Use the CloudstackSnitch for Apache Cloudstack environments.
Use the CloudstackSnitch for Apache Cloudstack environments. Because zone naming is free-form in
Apache Cloudstack, this snitch uses the widely-used <country> <location> <az> notation.
Client requests
Client read or write requests can be sent to any node in the cluster because all nodes in Cassandra are
peers.
Client read or write requests can be sent to any node in the cluster because all nodes in Cassandra are
peers. When a client connects to a node and issues a read or write request, that node serves as the
coordinator for that particular client operation.
The job of the coordinator is to act as a proxy between the client application and the nodes (or replicas)
that own the data being requested. The coordinator determines which nodes in the ring should get the
request based on the cluster configured partitioner and replica placement strategy.
27
Planning a cluster deployment

Vital information about successfully deploying a Cassandra cluster.
When planning a Cassandra cluster deployment, you should have a good idea of the initial volume of data
you plan to store and a good estimate of your typical application workload. The following topics provide
information for planning your cluster:
Selecting hardware for enterprise implementations

Choosing appropriate hardware depends on selecting the right balance of the following resources:
memory, CPU, disk type, number of nodes, and network.
Choosing appropriate hardware depends on selecting the right balance of the following resources:
memory, CPU, disk type, number of nodes, and network. Anti-patterns in Cassandra also contains
important information about hardware, particularly SAN storage, NAS devices, and NFS.
Caution: Do not use a machine suited for development for load testing or production. Failure may
result.
Memory
The more memory a Cassandra node has, the better read performance. More RAM also allows memory
tables (memtables) to hold more recently written data. Larger memtables lead to a fewer number of
SSTables being flushed to disk and fewer files to scan during a read. The ideal amount of RAM depends
on the anticipated size of your hot data.
For both dedicated hardware and virtual environments:
• Production: 16GB to 64GB; the minimum is 8GB.
• Development in non-loading testing environments: no less than 4GB.
• For setting Java heap space, see Tuning Java resources.
CPU
Insert-heavy workloads are CPU-bound in Cassandra before becoming memory-bound. (All writes go to
the commit log, but Cassandra is so efficient in writing that the CPU is the limiting factor.) Cassandra is
highly concurrent and uses as many CPU cores as available:
• Production environments:
• For dedicated hardware, 8-core CPU processors are the current price-performance sweet spot.
• For virtual environments, 4 to 8-core CPU processors.
• Development in non-loading testing environments:
• For dedicated hardware, 2-core CPU processors.
• For virtual environments, 2-core CPU processors.
Spinning disks versus Solid State Drives

SSDs are recommended for Cassandra. The NAND Flash chips that power SSDs provide extremely
low-latency response times for random reads while supplying ample sequential write performance for
compaction operations. In recent years, drive manufacturers have improved overall endurance, usually in
conjunction with spare (unexposed) capacity. Additionally, because PBW/DWPD ratings are probabilistic
estimates based on worst case scenarios, such as random write workloads, and Cassandra does only
large sequential writes, drives significantly exceed their endurance ratings. However, it is important to
28
plan for drive failures and have spares available. A large variety of SSDs are available on the market from
server vendors and third-party drive manufacturers.
For purchasing SSDs, the best recommendation is to make SSD endurance decisions not based on
workload, but on how difficult it is to change drives when they fail. Remember, your data is protected
because Cassandra replicates data across the cluster. Buying strategies include:
• If drives are quickly available, buy the cheapest drives that provide the performance you want.
• If it is more challenging to swap the drives, consider higher endurance models, possibly starting in the
mid range, and then choose replacements of higher or lower endurance based on the failure rates of
the initial model chosen.
• Always buy cheap SSDs and keep several spares online and unused in the servers until the initial
drives fail. This way you can replace the drives without touching the server.
DataStax customers that need help in determining the most cost-effective option for a given deployment
and workload, should contact their Solutions Engineer or Architect.
Disk space
Disk space depends on usage, so it's important to understand the mechanism. Cassandra writes data
to disk when appending data to the commit log for durability and when flushing memtable to SSTable
data files for persistent storage. The commit log has a different access pattern (read/writes ratio) than the
pattern for accessing data from SSTables. This is more important for spinning disks than for SSDs (solid
state drives).
SSTables are periodically compacted. Compaction improves performance by merging and rewriting
data and discarding old data. However, depending on the type of compaction strategy and size of
the compactions, during compaction disk utilization and data directory volume temporarily increases.
For this reason, you should leave an adequate amount of free disk space available on a node.
For large compactions, leave an adequate amount of free disk space available on a node: 50%
(worst case) for SizeTieredCompactionStrategy and DateTieredCompactionStrategy, and 10% for
LeveledCompactionStrategy. For more information about compaction, see:
• Compaction
• The Apache Cassandra storage engine
• Leveled Compaction in Apache Cassandra
• When to Use Leveled Compaction
• DateTieredCompactionStrategy: Compaction for Time Series Data and Getting Started with Time
Series Data Modeling
For information on calculating disk size, see Calculating usable disk capacity.
Recommendations:
Capacity per node
Most workloads work best with a capacity under 500GB to 1TB per node depending on I/O. Maximum
recommended capacity for Cassandra 1.2 and later is 3 to 5TB per node for uncompressed data. For
Cassandra 1.1, it is 500 to 800GB per node. Be sure to account for replication.
Capacity and I/O
When choosing disks, consider both capacity (how much data you plan to store) and I/O (the write/read
throughput rate). Some workloads are best served by using less expensive SATA disks and scaling disk
capacity and I/O by adding more nodes (with more RAM).
Number of disks - SATA
Ideally Cassandra needs at least two disks, one for the commit log and the other for the data directories.
At a minimum the commit log should be on its own partition.
Commit log disk - SATA
The disk need not be large, but it should be fast enough to receive all of your writes as appends (sequential
I/O).
29
Commit log disk - SSD

Unlike spinning disks, it's all right to store both commit logs and SSTables are on the same mount point.
Data disks
Use one or more disks and make sure they are large enough for the data volume and fast enough to both
satisfy reads that are not cached in memory and to keep up with compaction.
RAID on data disks
It is generally not necessary to use RAID for the following reasons:
• Data is replicated across the cluster based on the replication factor you've chosen.
• Starting in version 1.2, Cassandra includes a JBOD (Just a bunch of disks) feature to take care of disk
management. Because Cassandra properly reacts to a disk failure either by stopping the affected node
or by blacklisting the failed drive, you can deploy Cassandra nodes with large disk arrays without the
overhead of RAID 10. You can configure Cassandra to stop the affected node or blacklist the drive
according to your availability/consistency requirements. You can also recover a disk using JBOD.
RAID on the commit log disk
Generally RAID is not needed for the commit log disk. Replication adequately prevents data loss. If you
need extra redundancy, use RAID 1.
Extended file systems
DataStax recommends deploying Cassandra on XFS or ext4. On ext2 or ext3, the maximum file size is 2TB
even using a 64-bit kernel. On ext4 it is 16TB.
Because Cassandra can use almost half your disk space for a single file when using
SizeTieredCompactionStrategy, use XFS when using large disks, particularly if using a 32-bit kernel. XFS
file size limits are 16TB max on a 32-bit kernel, and essentially unlimited on 64-bit.
Number of nodes
Prior to version 1.2, the recommended size of disk space per node was 300 to 500GB. Improvement to
Cassandra 1.2, such as JBOD support, virtual nodes (vnodes), off-heap Bloom filters, and parallel leveled
compaction (SSD nodes only), allow you to use few machines with multiple terabytes of disk space.
Network
Since Cassandra is a distributed data store, it puts load on the network to handle read/write requests and
replication of data across nodes. Be sure that your network can handle traffic between nodes without
bottlenecks. You should bind your interfaces to separate Network Interface Cards (NIC). You can use
public or private depending on your requirements.
• Recommended bandwidth is 1000 Mbit/s (gigabit) or greater.
• Thrift/native protocols use the rpc_address.
• Cassandra's internal storage protocol uses the listen_address.
Cassandra efficiently routes requests to replicas that are geographically closest to the coordinator node
and chooses a replica in the same rack if possible; it always chooses replicas located in the same data
center over replicas in a remote data center.
Firewall
If using a firewall, make sure that nodes within a cluster can reach each other. See Configuring firewall port
access.
30
Planning an Amazon EC2 cluster

Important information for deploying a production Cassandra cluster on Amazon EC2.
Before planning an Amazon EC2 cluster, please see the User guide in the Amazon Elastic Compute Cloud
Documentation.
DataStax AMI deployments

The DataStax AMI is intended only for a single region and availability zone. For an EC2 cluster that spans
multiple regions and availability zones, see EC2 clusters spanning multiple regions and availability zones.
Use AMIs from trusted sources

Use only AMIs from a trusted source. Random AMI's pose a security risk and may perform slower than
expected due to the way the EC2 install is configured. The following are examples of trusted AMIs:
• Ubuntu Amazon EC2 AMI Locator
• Debian AmazonEC2Image
• CentOS-6 images on Amazon's EC2 Cloud
EC2 clusters spanning multiple regions and availability zones

When creating an EC2 cluster that spans multiple regions and availability zones, use OpsCenter to set up
your cluster. You can use any of the supported platforms. It is best practice to use the same platform on
all nodes. If your cluster was instantiated using the DataStax AMI, use Ubuntu for the additional nodes.
Configure the cluster as a multiple data center cluster using the Ec2MultiRegionSnitch. The following topics
describe OpsCenter provisioning:
• Provisioning a new cluster
• Adding an existing cluster
• Adding nodes to a cluster
Production Cassandra clusters on EC2

For production Cassandra clusters on EC2, use these guidelines for choosing the instance types:
• Development and light production: m3.large
• Moderate production: m3.xlarge
• SSD production with light data: c3.2xlarge
• Largest heavy production: m3.2xlarge (PV) or i2.2xlarge (HVM)
Note: The main difference between m1 and m3 instance types for use with Cassandra is that
m3 instance types have faster, smaller SSD drives and m1 instance types have slower, larger
rotational drives. Use m1 instance types when you have higher tolerance for latency SLAs and you
require smaller cluster sizes, or both. For more aggressive workloads use m3 instance types with
appropriately sized clusters.
EBS volumes are not recommended

EBS volumes are not recommended for Cassandra data storage volumes for the following reasons:
• EBS volumes contend directly for network throughput with standard packets. This contention means
that EBS throughput is likely to fail if you saturate a network link.
• EBS volumes have unreliable performance. I/O performance can be exceptionally slow, causing the
system to back load reads and writes until the entire cluster becomes unresponsive.
31
• Adding capacity by increasing the number of EBS volumes per host does not scale. You can easily
surpass the ability of the system to keep effective buffer caches and concurrently serve requests for all
of the data it is responsible for managing.
Note: Use only ephemeral instance-store devices for Cassandra data storage.
For more information and graphs related to ephemeral versus EBS performance, see the blog article
Systematic Look at EC2 I/O.
Disk Performance Optimization

To ensure high disk performance to mounted drives, it is recommended that you pre-warm your drives
by writing once to every drive location before production use. Depending on EC2 conditions, you can get
moderate to enormous increases in throughput. See Optimizing Disk Performance in the Amazon Elastic
Compute Cloud Documentation.
Storage recommendations for Cassandra 1.2 and later

Cassandra 1.2 and later supports JBOD (just a bunch of disks). JBOD excels at tolerating partial failures
in a disk array. Configure using the disk_failure_policy in the cassandra.yaml file. Addition information is
available in the Handling Disk Failures In Cassandra 1.2 blog and Recovering using JBOD.
Note: Cassandra JBOD support allows you to use standard disks. However, RAID0 may provide
better throughput because it splits every block to be on another device. This means that writes are
written in parallel fashion instead of written serially on disk.

conf/cassandra.yaml
Storage recommendations for Cassandra 1.1 and earlier

RAID 0 the ephemeral disks. Then put both the data directory and the commit log on that volume.
This has proved to be better in practice than putting the commit log on the root volume, which is also a
shared resource. For more data redundancy, consider deploying your Cassandra cluster across multiple
availability zones or using EBS volumes to store your Cassandra backup files.
Calculating usable disk capacity

Determining how much data your Cassandra nodes can hold.
About this task

To calculate how much data your Cassandra nodes can hold, calculate the usable disk capacity per node
and then multiply that by the number of nodes in your cluster. Remember that in a production cluster, you
will typically have your commit log and data directories on different disks.
Procedure
1. Start with the raw capacity of the physical disks:
raw_capacity = disk_size * number_of_data_disks
32
2. Calculate the formatted disk space as follows:
formatted_disk_space = (raw_capacity * 0.9)
During normal operations, Cassandra routinely requires disk capacity for compaction and repair
operations. For optimal performance and cluster health, DataStax recommends not filling your disks
to capacity, but running at 50% to 80% capacity depending on the compaction strategy and size of the
compactions.
3. Calculate the usuable disk space accounting for file system formatting overhead (roughly 10 percent):
usable_disk_space = formatted_disk_space * (0.5 to 0.8)
Calculating user data size

Accounting for storage overhead in determining user data size.
About this task

The size of your raw data may be larger or smaller once it is loaded into Cassandra due to storage
overhead. How much depends on how well it compresses and the characteristics of your data and tables.
The following calculations account for data persisted to disk, not for data stored in memory.
Procedure
• Determine column overhead:
regular_total_column_size = column_name_size + column_value_size + 15
counter - expiring_total_column_size = column_name_size +

column_value_size + 23
Every column in Cassandra incurs 15 bytes of overhead. Since each row in a table can have different
column names as well as differing numbers of columns, metadata is stored for each column. For
counter columns and expiring columns, you should add an additional 8 bytes (23 bytes total).
• Account for row overhead.
Every row in Cassandra incurs 23 bytes of overhead.
• Estimate primary key index size:
primary_key_index = number_of_rows * ( 32 + average_key_size )
Every table also maintains a partition index. This estimation is in bytes.

• Determine replication overhead:
replication_overhead = total_data_size * ( replication_factor - 1 )
The replication factor plays a role in how much disk capacity is used. For a replication factor of 1, there
is no overhead for replicas (as only one copy of data is stored in the cluster). If replication factor is
greater than 1, then your total data storage requirement will include replication overhead.
33
Anti-patterns in Cassandra
Implementation or design patterns that are ineffective and/or counterproductive in Cassandra production
installations. Correct patterns are suggested in most cases.
Implementation or design patterns that are ineffective and/or counterproductive in Cassandra production
installations. Correct patterns are suggested in most cases.
Storage area network

SAN storage is not recommended for on-premises deployments. DataStax does not support SAN-backed
deployments.
Note: Storage in clouds works very differently. Customers should contact DataStax for questions.
Although used frequently in Enterprise IT environments, SAN storage has proven to be a difficult and
expensive architecture to use with distributed databases for a variety of reasons, including:
• SAN ROI does not scale along with that of Cassandra, in terms of capital expenses and engineering
resources.
• In distributed architectures, SAN generally introduces a bottleneck and single point of failure because
Cassandra's IO frequently surpasses the array controller's ability to keep up.
• External storage, even when used with a high-speed network and SSD, adds latency for all operations.
• Heap pressure is increased because pending I/O operations take longer.
• When the SAN transport shares operations with internal and external Cassandra traffic, it can saturate
the network and lead to network availability problems.
Taken together these factors can create problems that are difficult to resolve in production. In particular,
new users deploying Cassandra with SAN must first develop adequate methods and allocate sufficient
engineering resources to identify these issues before they become a problem in production. For example,
methods are needed for all key scaling factors, such as operational rates and SAN fiber saturation.
Impact of Shared Storage on Cassandra details metrics on how severely performance can be affected.
Network attached storage

Storing SSTables on a network attached storage (NAS) device is not recommended. Using a NAS device
often results in network related bottlenecks resulting from high levels of I/O wait time on both reads and
writes. The causes of these bottlenecks include:
• Router latency.
• The Network Interface Card (NIC) in the node.
• The NIC in the NAS device.
If you are required to use NAS for your environment, please contact a technical resource from DataStax for
assistance.
Shared network file systems

Shared network file systems (NFS) have the same limitations as NAS. The temptation with NFS
implementations is to place all SSTables in a node into one NFS. Doing this deprecates one of
Cassandra's strongest features: No Single Point of Failure (SPOF). When all SSTables from all nodes are
stored onto a single NFS, the NFS becomes a SPOF. To best use Cassandra, avoid using NFS.
Excessive heap space size

DataStax recommends using the default heap space size for most use cases. Exceeding this size can
impair the Java virtual machine's (JVM) ability to perform fluid garbage collections (GC). The following
table shows a comparison of heap space performances reported by a Cassandra user:
34
Heap CPU utilization Queries per second Latency

40 GB 50% 750 1 second
8 GB 5% 8500 (not maxed out) 10 ms
For information on heap sizing, see Tuning Java resources.
Cassandra's rack feature

This information applies only to single-token architecture, not to virtual nodes.
Defining one rack for the entire cluster is the simplest and most common implementation. Multiple racks
should be avoided for the following reasons:
• Most users tend to ignore or forget rack requirements that racks should be organized in an alternating
order. This order allows the data to get distributed safely and appropriately.
• Many users are not using the rack information effectively. For example, setting up with as many racks
as nodes (or similar non-beneficial scenarios).
• Expanding a cluster when using racks can be tedious. The procedure typically involves several node
moves and must ensure that racks are distributing data correctly and evenly. When clusters need
immediate expansion, racks should be the last concern.
To use racks correctly:
• Use the same number of nodes in each rack.
• Use one rack and place the nodes in different racks in an alternating pattern. This allows you to still get
the benefits of Cassandra's rack feature, and allows for quick and fully functional cluster expansions.
Once the cluster is stable, you can swap nodes and make the appropriate moves to ensure that nodes
are placed in the ring in an alternating fashion with respect to the racks.
Also see About Replication in Cassandra in the Cassandra 1.1 documentation.
SELECT ... IN or index lookups

SELECT ... IN and index lookups (formerly secondary indexes) should be avoided except for specific
scenarios. See When not to use IN in SELECT and When not to use an index in Indexing in CQL for
Cassandra 2.0.
Using the Byte Ordered Partitioner

The Byte Ordered Partitioner (BOP) is not recommended.
Use virtual nodes (vnodes) and use either the Murmur3Partitioner (default) or the RandomPartitioner.
Vnodes allow each node to own a large number of small ranges distributed throughout the cluster. Using
vnodes saves you the effort of generating tokens and assigning tokens to your nodes. If not using vnodes,
these partitioners are recommended because all writes occur on the hash of the key and are therefore
spread out throughout the ring amongst tokens range. These partitioners ensure that your cluster evenly
distributes data by placing the key at the correct token using the key's hash value.
Reading before writing

Reads take time for every request, as they typically have multiple disk hits for uncached reads. In work
flows requiring reads before writes, this small amount of latency can affect overall throughput. All write I/
O in Cassandra is sequential so there is very little performance difference regardless of data size or key
distribution.
35
Load balancers
Cassandra was designed to avoid the need for load balancers. Putting load balancers between Cassandra
and Cassandra clients is harmful to performance, cost, availability, debugging, testing, and scaling. All
high-level clients, such as the Java and Python drivers for Cassandra, implement load balancing directly.
Insufficient testing
Be sure to test at scale and production loads. This the best way to ensure your system will function
properly when your application goes live. The information you gather from testing is the best indicator of
what throughput per node is needed for future expansion calculations.
To properly test, set up a small cluster with production loads. There will be a maximum throughput
associated with each node count before the cluster can no longer increase performance. Take the
maximum throughput at this cluster size and apply it linearly to a cluster size of a different size. Next
extrapolate (graph) your results to predict the correct cluster sizes for required throughputs for your
production cluster. This allows you to predict the correct cluster sizes for required throughputs in the future.
The Netflix case study shows an excellent example for testing.
Too many keyspaces or tables

Each Cassandra keyspace has a certain amount of overhead space that uses JVM memory. Each table
uses approximately 1MB of memory. For example, 3,500 tables would use about 3.5TB of JVM memory.
Using too many tables, or by extension, too many keyspaces will bloat the memory requirements. A good
rule of thumb is to keep the number of tables within a cluster to 1,000 at most, and aim for 500 or less.
Lack of familiarity with Linux

Linux has a great collection of tools. Become familiar with the Linux built-in tools. It will help you greatly
and ease operation and management costs in normal, routine functions. The essential list of tools and
techniques to learn are:
• Parallel SSH and Cluster SSH: The pssh and cssh tools allow SSH access to multiple nodes. This is
useful for inspections and cluster wide changes.
• Passwordless SSH: SSH authentication is carried out by using public and private keys. This allows SSH
connections to easily hop from node to node without password access. In cases where more security is
required, you can implement a bastion host and/or VPN.
• Useful common command-line tools include:
• dstat: Shows all system resources instantly. For example, you can compare disk usage in
combination with interrupts from your IDE controller, or compare the network bandwidth numbers
directly with the disk throughput (in the same interval).
• top: Provides an ongoing look at CPU processor activity in real time.
• System performance tools: Tools such as iostat, mpstat, iftop, sar, lsof, netstat, htop, vmstat, and
similar can collect and report a variety of metrics about the operation of the system.
• vmstat: Reports information about processes, memory, paging, block I/O, traps, and CPU activity.
• iftop: Shows a list of network connections. Connections are ordered by bandwidth usage, with the
pair of hosts responsible for the most traffic at the top of list. This tool makes it easier to identify the
hosts causing network congestion.
Running without the recommended settings

Be sure to use the recommended settings in the Cassandra documentation.
Also be sure to consult the Planning a Cassandra cluster deployment documentation, which discusses
hardware and other recommendations before making your final hardware purchases.
More anti-patterns
For more about anti-patterns, visit Matt Dennis` slideshare.
36
Installing DataStax Community

Various installation methods.
Installing DataStax Community on RHEL-based systems

Install using Yum repositories on RHEL, CentOS, and Oracle Linux.
About this task

Use these steps to install Cassandra using Yum repositories on RHEL, CentOS, and Oracle Linux.
Note: To install on SUSE, use the Cassandra binary tarball distribution.
For a complete list of supported platforms, see DataStax Community – Supported Platforms.
Before you begin

• Yum Package Management application installed.
• Root or sudo access to the install machine.
• Latest version of Oracle Java SE Runtime Environment (JRE) 8 (recommended) or OpenJDK 7.
• Python 2.6+ (needed if installing OpsCenter).
About this task

The packaged releases create a cassandra user. When starting Cassandra as a service, the service runs
as this user. The following utilities are included in a separate package: sstable2json, sstablelevelreset,
sstablemetadata, json2sstable, sstablerepairedset, sstablesplit, and token-generator.
Procedure
In a terminal window:
1. Check which version of Java is installed by running the following command:
$ java -version
It is recommended to use the latest version of Oracle Java 8 on all nodes. (Oracle Java 7 is also
supported.)
See Installing the JRE on RHEL-based systems.
2. Add the DataStax Community repository to the /etc/yum.repos.d/datastax.repo:
[datastax]
name = DataStax Repo for Apache Cassandra
baseurl = http://rpm.datastax.com/community
enabled = 1
gpgcheck = 0
3. Install the packages:
$ sudo yum install dsc21

$ sudo yum install cassandra21-tools ## Installs optional utilities.
The DataStax Community distribution of Cassandra is ready for configuration.
37
What to do next
• Initializing a multiple node cluster (single data center)
• Initializing a multiple node cluster (multiple data centers)
• Recommended production settings
• Installing OpsCenter
• Key components for configuring Cassandra
• Starting Cassandra as a service
• Package installation directories
Installing DataStax Community on Debian-based systems

Install using APT repositories on Debian and Ubuntu.
About this task

Use these steps to install Cassandra using APT repositories on Debian and Ubuntu Linux.
Before you begin

• Advanced Package Tool is installed.
• Root or sudo access to the install machine.
About this task

The packaged releases create a cassandra user. When starting Cassandra as a service, the service runs
as this user. The following utilities are included in a separate package: sstable2json, sstablelevelreset,
sstablemetadata, json2sstable, sstablerepairedset, sstablesplit, and token-generator.
Procedure
$ java -version
supported.)
2. Add the DataStax Community repository to the /etc/apt/sources.list.d/
cassandra.sources.list
$ echo "deb http://debian.datastax.com/community stable main" | sudo tee -

a /etc/apt/sources.list.d/cassandra.sources.list
3. Debian systems only:
a) In /etc/apt/sources.list, find the line that describes your source repository for Debian and
add contrib non-free to the end of the line. For example:
deb http://some.debian.mirror/debian/ $distro main contrib non-free
This allows installation of the Oracle JVM instead of the OpenJDK JVM.
38
b) Save and close the file when you are done adding/editing your sources.
4. Add the DataStax repository key to your aptitude trusted keys.
$ curl -L http://debian.datastax.com/debian/repo_key | sudo apt-key add -

5. Install the latest package:
$ sudo apt-get update

$ sudo apt-get install dsc21
$ sudo apt-get install cassandra-tools ## Optional utilities
This installs the DataStax Community distribution of Cassandra. .
6. Because the Debian packages start the Cassandra service automatically, you must stop the server and
clear the data:
Doing this removes the default cluster_name (Test Cluster) from the system table. All nodes must use
the same cluster name.
$ sudo service cassandra stop

$ sudo rm -rf /var/lib/cassandra/data/system/*
What to do next
• Starting Cassandra as a service
• Package installation directories
Installing DataStax Community on any Linux-based platform

Install on all Linux-based platforms using a binary tarball.
About this task

Use these steps to install Cassandra on all Linux-based platforms using a binary tarball.
About this task

Use this install for Mac OS X and platforms without package support, or if you do not have or want a root
installation.
Before you begin

• If you are using an older RHEL-based Linux distribution, such as CentOS-5, you may see the following
error: GLIBCXX_3.4.9 not found. You must replace the Snappy compression/decompression
library (snappy-java-1.0.5.jar) with the snappy-java-1.0.4.1.jar.
39
About this task

The binary tarball runs as a stand-alone process.
Procedure
$ java -version
supported.)
2. Download the DataStax Community:
$ curl -L http://downloads.datastax.com/community/dsc.tar.gz | tar xz

You can also download from Planet Cassandra.
3. Go to the install directory:
$ cd dsc-cassandra-2.1.x
4. Go to the install directory:
$ cd install_location/apache-cassandra-2.1.x
Cassandra is ready for configuration.
What to do next
• Tarball installation directories
• Starting Cassandra as a stand-alone process
Installing DataStax Community on Windows systems

About installing on Windows systems.
About this task

To install or uninstall DataStax Community on Windows systems, see Cassandra and DataStax Enterprise
Essentials.
40
Installing prior releases of DataStax Community

Steps for installing the same version as other nodes in your cluster.
About this task

How to install the same version as other nodes in your cluster.
Follow the install instructions in the relevant documentation and specify the specific version in the install
command.
Installing the packages on RHEL-based platforms

Examples:
$ sudo yum install dsc21-2.1.2.1 cassandra21-2.1.2-1

$ sudo yum install dsc21-2.1.2.1 cassandra21-2.1.2-1 cassandra21-tools-2.1.2-1
## Optional
Installing the packages on Debian-based platforms

Examples:
$ sudo apt-get install dsc21=2.1.2-1 cassandra=2.1.2

$ sudo apt-get install dsc21=2.1.2-1 cassandra=2.1.2 cassandra-tools=2.1.2 ##
Optional
Installing from the binary tarball

1. Download the tarball using the URL for the prior version.
Cassandra 2.1.0 example:
http://downloads.datastax.com/community/dsc-cassandra-2.1.0-bin.tar.gz
Cassandra 2.1.3 example:
http://downloads.datastax.com/community/dsc-cassandra-2.1.3-bin.tar.gz
2. Unpack the distribution. For example:
$ tar -xzvf dsc-cassandra-2.1.3-bin.tar.gz
The files are extracted into the dsc-cassandra-2.1.3 directory.
Uninstalling DataStax Community

Steps for uninstalling Cassandra by install type.
About this task

Select the uninstall method for your type of installation.
Uninstalling Debian- and RHEL-based packages

Use this method when you have installed DataStax Community using APT or Yum.
1. Stop the Cassandra and DataStax Agent services:
41

2. Make sure all services are stopped:
$ ps auwx | grep cassandra

$ ps auwx | grep datastax-agent ## If the DataStax Agent was installed.
3. If services are still running, use the PID to kill the service:
$ sudo kill cassandra_pid

$ sudo kill datastax_agent_pid ## If the DataStax Agent was installed.
4. Remove the library and log directories:
$ sudo rm -r /var/lib/cassandra
$ sudo rm -r /var/log/cassandra
5. Remove the installation directories:
RHEL-based packages:
$ sudo yum remove "cassandra-*" "datastax-*" ## datastax-x required if the

DataStax Agent was installed.
Debian-based packages:
$ sudo apt-get purge "cassandra-*" "datastax-*" ## datastax-x required if

the DataStax Agent was installed..
Uninstalling the binary tarball

Use this method when you have installed DataStax Community using the binary tarball.
1. Stop the node:

$ sudo kill <pid>
2. Stop the DataStax Agent if installed:
$ sudo kill datastax_agent_pid

3. Remove the installation directory.
Installing on cloud providers

Installation methods for the supported cloud providers.
Installing a Cassandra cluster on Amazon EC2

A step-by-step guide for installing the DataStax Community AMI (Amazon Machine Image).
About this task

The DataStax AMI allows you to set up a simple DataStax Community cluster using the Amazon Web
Services EC2 Management Console. Installing via the AMI allows you to quickly deploy a Cassandra
cluster within a single availability zone.
The AMI does the following:
• Installs the latest version of Cassandra with an Ubuntu 12.04 LTS (Precise Pangolin), image (Ubuntu
Cloud 20140227 release), Kernel 3.8+.
• Installs Oracle Java 7.
• Install metrics tools such as dstat, ethtool, make, gcc, and s3cmd.
42
• Uses RAID0 ephemeral disks for data storage and commit logs.
• Choice of PV (Para-virtualization) or HVM (Hardware-assisted Virtual Machine) instance types. See
Amazon documentation.
• Launches EBS-backed instances for faster start-up, not database storage.
• Uses the private interface for intra-cluster communication.
• Sets the seed nodes cluster-wide.
• Installs OpsCenter (by default).
Note: When creating an EC2 cluster that spans multiple regions and availability zones, use
OpsCenter to set up your cluster. See EC2 clusters spanning multiple regions and availability
zones.
Because Amazon changes the EC2 console intermittently, these instructions have been generalized. For
details on each step, see the User guide in the Amazon Elastic Compute Cloud Documentation.
To install a Cassandra cluster from the DataStax AMI, complete the following tasks:
Creating an EC2 security group

An EC2 Security Group acts as a firewall for designation which protocols and ports are open in your
cluster.
About this task

An EC2 Security Group acts as a firewall that allows you to choose which protocols and ports are open in
your cluster. You must specify a security group in the same region as your instances.
You can specify the protocols and ports either by a range of IP addresses or by security group. To protect
your cluster, you should define a security group. Be aware that specifying a Source IP of 0.0.0.0/0 allows
every IP address access by the specified protocol and port range.
Procedure
If you need more help, click an informational icon or a link to the Amazon EC2 User Guide.
1. Sign in to the AWS console.
2. From the Amazon EC2 console navigation bar, select the same region as where you will launch the
DataStax Community AMI.
Step 1 in Launch instances provides a list of the available regions.
3. Open the Security Groups page.

4. Create a security group with a name and description of your choice, then save it. It is recommended
that you include the region name in the description.
Note: Creating and saving the securing group allows you to create rules based on the group.
After the security group is saved it is available in the Source field drop-list.
5. Create rules for the security group using the following table:
43
Table 1: Ports
Port Type Protocol Source Description

number
Public ports
22 SSH TCP 0.0.0.0/0 SSH port
8888 Custom TCP 0.0.0.0/0 OpsCenter website. The opscenterd daemon listens
TCP on this port for HTTP requests coming directly from
Rule the browser.
Cassandra inter-node ports
1024 - Custom TCP Your Cassandra 1.2 or earlier only. Because JMX
65355 TCP security connects on port 7199, handshakes, and then
Rule group uses any port within the 1024+ range, use SSH to
execute commands remotely to connect to JMX
locally or use the DataStax OpsCenter.
7000 Custom TCP Your Cassandra inter-node cluster communication.
TCP security
Rule group
7001 Custom TCP Your Cassandra SSL inter-node cluster communication.
TCP security
Rule group
7199 Custom TCP Your Cassandra JMX monitoring port.
TCP security
Rule group
Cassandra client ports
9042 Custom TCP 0.0.0.0/0 Cassandra client port.
TCP
Rule
9160 Custom TCP 0.0.0.0/0 Cassandra client port (Thrift).
TCP
Rule
OpsCenter inter-node ports
61620 Custom TCP Your OpsCenter monitoring port. The opscenterd daemon
TCP security listens on this port for TCP traffic coming from the
Rule group agent.
61621 Custom TCP Your OpsCenter agent port. The agents listen on this port
TCP security for SSL traffic initiated by OpsCenter.
Rule group
The completed port rules should look similar to this:
44
Warning: The security configuration shown in this example opens up all externally accessible
ports to incoming traffic from any IP address (0.0.0.0/0). The risk of data loss is high. If you
desire a more secure configuration, see the Amazon EC2 help on security groups.
Creating a key pair

Amazon EC2 uses public–key cryptography to encrypt and decrypt login information.
About this task

Amazon EC2 uses public–key cryptography to encrypt and decrypt login information. Public–key
cryptography uses a public key to encrypt data and the recipient uses the private key to decrypt the data.
The public and private keys are known as a key pair.
Procedure
You must create a key pair for each region you use.
1. From the Amazon EC2 console navigation bar, select the same region as where you will launch the
DataStax Community AMI.
Step 1 in Launch instances provides a list of the available regions.
2. Create the key pair and save it to your home directory.

3. Set the permissions of your private key file so that only you can read it.
$ chmod 400 my-key-pair.pem
Launching the DataStax Community AMI

Launching your Cassandra Amazon Machine Images.
About this task

After creating the security group, you can launch your AMI instances.
Procedure
If you need more help, click an informational icon or a link to the Amazon EC2 User Guide.
45
1. Launch the AMI using the links in the following table:

Amazon EC2 offers a number of geographic regions for launching the AMI. Factors for choosing a
region include: reduce latency, cost, or regulatory requirements.
Region AMI
HVM instances (Hardware-assisted Virtual Machine - see Amazon documentation.)
us-east-1 ami-ada2b6c4
us-west-1 ami-3cf7c979
us-west-2 ami-1cff962c
eu-west-1 ami-7f33cd08
ap-southeast-1 ami-b47828e6
ap-southeast-2 ami-55d54d6f
ap-northeast-1 ami-714a3770
sa-east-1 ami-1dda7800
PV instances (Paravirtualization - see Amazon documentation.
us-east-1 ami-f9a2b690
us-west-1 ami-32f7c977
us-west-2 ami-16ff9626
eu-west-1 ami-8932ccfe
ap-southeast-1 ami-8c7828de
ap-southeast-2 ami-57d54d6d
ap-northeast-1 ami-6b4a376a
sa-east-1 ami-15da7808
2. In Step 2: Choose an Instance Type, choose the appropriate type.
The recommended instances are:
• Development and light production: m3.large
• Moderate production: m3.xlarge
• SSD production with light data: c3.2xlarge
• Largest heavy production: m3.2xlarge (PV) or i2.2xlarge (HVM)
• Micro, small, and medium types are not supported.
Note: The main difference between m1 and m3 instance types for use with Cassandra is that
m3 instance types have faster, smaller SSD drives and m1 instance types have slower, larger
rotational drives. Use m1 instance types when you have higher tolerance for latency SLAs and
you require smaller cluster sizes, or both. For more aggressive workloads use m3 instance types
with appropriately sized clusters.
When the instance is selected, its specifications are displayed:
46
Because Amazon updates instance types periodically, see the following docs to help you determine
your hardware and storage requirements:
• Planning a cluster deployment
• User guide in the Amazon Elastic Compute Cloud Documentation
• What is the story with AWS storage
• Get in the Ring with Cassandra and EC2
3. Click Next: Configure Instance Details and configure the instances to suit your requirements:
• Number of instances
• Network - Select Launch into EC2-Classic.
• Advanced Details - Open and add the following options (as text) to the User Data section.
Option Description
--clustername Required. The name of the cluster.
name
--totalnodes Required. The total number of nodes in the cluster.
#_nodes
--version Required. The version of the cluster. Use community to install the latest
community version of DataStax Community.
--opscenter [no] Optional. By default, DataStax OpsCenter is installed on the first instance.
Specify no to disable.
--reflector url Optional. Allows you to use your own reflector. Default: http://
reflector2.datastax.com/reflector2.php
For example: --clustername myDSCcluster --totalnodes 6 --version community
4. Click Next: Add Storage, and add volumes as needed.

The number of instance store devices available to the machine depends on the instance type. EBS
volumes are not recommended for database storage.
5. Click Next: Tag Instance and give a name to your DSE instance, such as workload-dsc.
Tags enable you to categorize your AWS resources in different ways, such as purpose, owner, or
environment.
6. Click Next: Configure Security Group and configure as follows:
a) Choose Select an existing security group.
b) Select the Security Group you created earlier.
c) Click Review and Launch.
7. On the Step 7: Review Instance Launch page, make any needed changes.
8. Click Launch and then in the Select an existing key pair or create a new key pair dialog, do one of
the following:
• Select an existing key pair from the Select a key pair drop list.
47
• If you need to create a new key pair, click Choose an existing key pair drop list and select Create
a new key pair. Then create the new key pair as described in Create key pair.
9. Click Launch Instances.
The AMI image configures your cluster and starts Cassandra services. The Launch Status page is
displayed.
10.Click View Instances.
Connecting to your DataStax Community EC2 instance

Connect to your Cassandra EC2 instances from a terminal or SSH client.
About this task

Once the cluster is launched, you can connect to it from a terminal or SSH client, such as PuTTY. Connect
as user ubuntu rather than as root.
Procedure
1. If necessary, from the EC2 Dashboard, click Running Instances.
You can connect to any node in the cluster. However, one node (Node0) runs OpsCenter and is the
Cassandra seed node.
2. To find which instance is Node0:

a) Select an instance.
b) Select the Description tab.
c) Scroll down the description information until you see AMI launch index.
d) Repeat until you find Node0.

3. To get the public DNS name of a node, select the node you want to connect to, and then click Connect.
4. In Connect To Your Instance, select A standalone SSH client.
5. Copy the Example command line and change the user from root to ubuntu, then paste it into your
SSH client.
48
The AMI image configures your cluster and starts the Cassandra services.
6. After you have logged into a node and the AMI has completed installing and setting up the nodes, the
status is displayed:
The URL for the OpsCenter is displayed when you connect to the node containing it; otherwise it is not
displayed.
7. If you installed OpsCenter, allow 60 to 90 seconds after the cluster has finished
initializing for OpsCenter to start. You can launch OpsCenter using the URL:
http://public_dns_of_first_instance:8888/
The Dashboard should show that the agents are connected.
8. If the agents have not automatically connected:

a) Click the Fix link located near the top left of the Dashboard.
b) When prompted for credentials for the agent nodes, use the username ubuntu and copy and paste
the entire contents from your private key (.pem).
49
The Dashboard shows the agents are connected.
Clearing the data for an AMI restart

Clearing the data for an Amazon Machine Image restart.
About this task

Use these steps to clear the data for an Amazon Machine Image (AMI) restart. The logs for AMIs are
stored in the same location as the data. To preserve the log files, you need to delete everything except the
log directory.
Procedure
To clear the data from the default directories:
After stopping the service, run the following command:
$ sudo rm -rf /var/lib/cassandra/commitlog

$ sudo rm -rf /var/lib/cassandra/data
$ sudo rm -rf /var/lib/cassandra/saved_caches
Expanding a Cassandra AMI cluster

Adding nodes to a Cassandra AMI cluster.
About this task

The best way to expand your EC2 implementations is to use OpsCenter:
• Provisioning a new cluster
• Adding an existing cluster
• Adding nodes to a cluster
Installing and deploying a Cassandra cluster using GoGrid

Steps for installing and deploying a developer (3-node) or production (5-node) Cassandra cluster using
GoGrid’s 1-Button Deploy.
About this task

You can use GoGrid’s 1-Button Deploy for installing and deploying a developer (3-node) or production (5-
node) Cassandra cluster. Additional introductory documentation is available from GoGrid at:
• GoGrid Cassandra Wiki
• Getting Started
The 1-Button Deploy of Cassandra does the following:
• Installs the latest version of Cassandra on X-Large SSD Cloud Servers running Debian 7.2.
• Installs OpsCenter.
• Installs Oracle JDK 7.
• Installs Python Driver.
• Enables the Firewall Service - All services are blocked except SSH (22) and ping for public traffic.
• Deploys Cassandra using virtual nodes (vnodes).
Procedure
1. Register with GoGrid.
2. Fill out the registration form and complete the account verification.
3. Access the management console with the login credentials you received in your email.
50
Your cluster automatically starts deploying. A green status indicator shows that a server is up and
running.
Hover over any item to view its details or right-click to display a context menu.
4. Login to one of the servers and validate that the servers are configured and communicating:
Note: You can login to any member of the cluster either with SSH, a third-party client (like
PuTTY), or through the GoGrid Console service.
a) To find your server credentials, right-click the server and select Passwords.
b) From your secure connection client, login to the server with the proper credentials. For example from
SSH:
$ ssh root@ip_address
c) Validate that the cluster is running:
$ nodestool status
Each node should be listed and it's status and state should be UN (Up Normal):
Datacenter: datacenter1
=======================
Status=Up/Down |/ State=Normal/Leaving/Joining/Moving
-- Address Load Tokens Owns (effective) Host ID
Rack
UN 10.110.94.2 71.46 KB 256 65.9%
3ed072d6-49cb-4713-bd55-ea4de25576a9 rack1
UN 10.110.94.5 40.91 KB 256 66.7%
d5d982bc-6e30-40a0-8fe7-e46d6622c1d5 rack1
UN 10.110.94.4 73.33 KB 256 67.4% f6c3bf08-
d9e5-43c8-85fa-5420db785052 rack1
What to do next
The following provides information about using and configuring Cassandra, OpsCenter, GoGrid, and the
Cassandra Query Language (CQL):
• About Apache Cassandra
51
• OpsCenter documentation
• GoGrid documentation
• CQL for Cassandra 2.x
Installing the Oracle JRE

Instructions for various platforms.
Installing Oracle JRE on RHEL-based Systems

You must configure your operating system to use the Oracle JRE 7 or 8, or OpenJDK 7.
About this task

Configure your operating system to use the latest version of Oracle Java SE Runtime Environment 8 or
OpenJDK 7.
Note: After installing the JRE, you may need to set JAVA_HOME to your profile:
For shell or bash: export JAVA_HOME=path_to_java_home
For csh (C shell): setenv JAVA_HOME=path_to_java_home
Procedure
1. Check which version of the JRE your system is using:
$ java -version
If Oracle Java is used, the results should look like:
java version "1.8.0_45"

Java(TM) SE Runtime Environment (build 1.8.0_45-b14)
Java HotSpot(TM) 64-Bit Server VM (build 25.45-b02, mixed mode)
2. If necessary, go to Oracle Java SE Downloads, accept the license agreement, and download the
installer for your distribution.
Note: If installing the Oracle JRE in a cloud environment, accept the license agreement,
download the installer to your local client, and then use scp (secure copy) to transfer the file to
your cloud machines.
3. From the directory where you downloaded the package, run the install:
$ sudo rpm -ivh jre-8uversion-linux-x64.rpm
The RPM installs the JRE into the /usr/java/ directory.

4. Use the alternatives command to add a symbolic link to the Oracle JRE installation so that your
system uses the Oracle JRE instead of the OpenJDK JRE:
$ sudo alternatives --install /usr/bin/java java /usr/java/

jre1.8.0_version/bin/java 200000
If you have any problems, set the PATH and JAVA_HOME variables:
export PATH="$PATH:/usr/java/latest/bin"
set JAVA_HOME=/usr/java/latest
52
5. Make sure your system is now using the correct JRE. For example:
$ java -version

6. If the OpenJDK JRE is still being used, use the alternatives command to switch it. For example:
$ sudo alternatives --config java
There are 2 programs which provide java.
Selection Command
------------------------------------------------------------
1 /usr/lib/jvm/jre-1.7.0-openjdk.x86_64/bin/java
*+ 2 /usr/java/jre1.8.0_45/bin/java
Enter to keep the current selection [+ ], or type selection number:
Installing Oracle JRE on Debian or Ubuntu Systems

Steps for installing Oracle Java on Debian-based systems.
About this task

Configure your operating system to use the latest version of Oracle Java SE Runtime Environment 8 or
OpenJDK 7.
Note: After installing the JRE, you may need to set JAVA_HOME to your profile:
For shell or bash: export JAVA_HOME=path_to_java_home
For csh (C shell): setenv JAVA_HOME=path_to_java_home
About this task

The Oracle Java Runtime Environment (JRE) has been removed from the official software repositories of
Ubuntu and only provides a binary (.bin) version. You can get the JRE from the Java SE Downloads.
Procedure
1. Check which version of the JRE your system is using:
$ java -version
If Oracle Java is used, the results should look like:

2. If necessary, go to Oracle Java SE Downloads, accept the license agreement, and download the
installer for your distribution.
Note: If installing the Oracle JRE in a cloud environment, accept the license agreement,
download the installer to your local client, and then use scp (secure copy) to transfer the file to
your cloud machines.
53
3. Make a directory for the JRE:
$ sudo mkdir -p /usr/lib/jvm
4. Unpack the tarball and install the JRE:
$ sudo tar zxvf jre-7u25-linux-x64.tar.gz -C /usr/lib/jvm
The JRE files are installed into a directory called /usr/lib/jvm/jre-7u_version.

5. Tell the system that there's a new Java version available:
$ sudo update-alternatives --install "/usr/bin/java" "java" "/usr/lib/jvm/

jre1.8.0_version/bin/java" 1
If updating from a previous version that was removed manually, execute the above command twice,
because you'll get an error message the first time.
6. Set the new JRE as the default:
$ sudo update-alternatives --set java /usr/lib/jvm/jre1.8.0_version/bin/

java
7. Make sure your system is now using the correct JRE. For example:
$ java -version

Recommended production settings

Recommendations for production environments.
Recommendations for production environments; adjust them accordingly for your implementation.
Optimizing SSDs
For the majority of Linux distributions, SSDs are not configured optimally by default. The following steps
ensures best practice settings for SSDs:
1. Ensure that the SysFS rotational flag is set to false (zero).
This overrides any detection by the operating system to ensure the drive is considered an SSD.
2. Repeat for any block devices created from SSD storage, such as mdarrays.
3. Set the IO scheduler to either deadline or noop:
• The noop scheduler is the right choice when the target block device is an array of SSDs behind a
high-end IO controller that performs IO optimization.
• The deadline scheduler optimizes requests to minimize IO latency. If in doubt, use the deadline
scheduler.
4. Set the read-ahead value for the block device to 8KB.
This setting tells the operating system not to read extra bytes, which can increase IO time and pollute
the cache with bytes that weren’t requested by the user.
For example, if the SSD is /dev/sda, in /etc/rc.local:
echo deadline > /sys/block/sda/queue/scheduler
54
#OR...
#echo noop > /sys/block/sda/queue/scheduler
echo 0 > /sys/class/block/sda/queue/rotational
echo 8 > /sys/class/block/sda/queue/read_ahead_kb
Disable zone_reclaim_mode on NUMA systems

The Linux kernel can be inconsistent in enabling/disabling zone_reclaim_mode. This can result in odd
performance problems.
To ensure that zone_reclaim_mode is disabled:
$ echo 0 > /proc/sys/vm/zone_reclaim_mode

For more information, see Peculiar Linux kernel performance problem on NUMA systems.
User resource limits

You can view the current limits using the ulimit -a command. Although limits can also be temporarily
set using this command, DataStax recommends making the changes permanent:
Packaged installs: Ensure that the following settings are included in the /etc/security/limits.d/
cassandra.conf file:
cassandra - memlock unlimited

cassandra - nofile 100000
cassandra - nproc 32768
cassandra - as unlimited
Tarball installs: Ensure that the following settings are included in the /etc/security/limits.conf file:
* - memlock unlimited
* - nofile 100000
* - nproc 32768
* - as unlimited
If you run Cassandra as root, some Linux distributions such as Ubuntu, require setting the limits for root
explicitly instead of using *:
root - memlock unlimited

root - nofile 100000
root - nproc 32768
root - as unlimited
For CentOS, RHEL, OEL systems, also set the nproc limits in /etc/security/limits.d/90-
nproc.conf:
* - nproc 32768
For all installations, add the following line to /etc/sysctl.conf:
vm.max_map_count = 131072
To make the changes take effect, reboot the server or run the following command:
$ sudo sysctl -p
To confirm the limits are applied to the Cassandra process, run the following command where pid is the
process ID of the currently running Cassandra process:
$ cat /proc/<pid>/limits
55
For more information, see Insufficient user resource limits errors.
Disable swap
You must disable swap entirely. Failure to do so can severely lower performance. Because Cassandra
has multiple replicas and transparent failover, it is preferable for a replica to be killed immediately when
memory is low rather than go into swap. This allows traffic to be immediately redirected to a functioning
replica instead of continuing to hit the replica that has high latency due to swapping. If your system has a
lot of DRAM, swapping still lowers performance significantly because the OS swaps out executable code
so that more DRAM is available for caching disks.
If you insist on using swap, you can set vm.swappiness=1. This allows the kernel swap out the absolute
least used parts.
$ sudo swapoff --all

To make this change permanent, remove all swap file entries from /etc/fstab.
For more information, see Nodes seem to freeze after some period of time.
Synchronize clocks
The clocks on all nodes should be synchronized. You can use NTP (Network Time Protocol) or other
methods.
This is required because columns are only overwritten if the timestamp in the new version of the column is
more recent than the existing column.
Optimum blockdev --setra settings for RAID

Typically, a readahead of 128 is recommended, especially on Amazon EC2 RAID0 devices.
Check to ensure setra is not set to 65536:
$ sudo blockdev --report /dev/<device>

To set setra:
$ sudo blockdev --setra 128 /dev/<device>
Java Virtual Machine

The latest 64-bit version of Java 8 is recommended or OpenJDK 7.
56
Initializing a cluster
Topics for deploying a cluster.
Initializing a multiple node cluster (single data center)

A deployment scenario for a Cassandra cluster with a single data center.
About this task

This topic contains information for deploying a Cassandra cluster with a single data center. If you're new to
Cassandra, and haven't set up a cluster, see Cassandra and DataStax Enterprise Essentials or 10 Minute
Cassandra Walkthrough.
Before you begin

Each node must be correctly configured before starting the cluster. You must determine or perform the
following before starting the cluster:
• A good understanding of how Cassandra works. Be sure to read at least Understanding the
architecture, Data replication, and Cassandra's rack feature.
• Install Cassandra on each node.
• Choose a name for the cluster.
• Get the IP address of each node.
• Determine which nodes will be seed nodes. Do not make all nodes seed nodes. Please read
Internode communications (gossip).
• Determine the snitch and replication strategy. The GossipingPropertyFileSnitch and
NetworkTopologyStrategy are recommended for production environments.
• If using multiple data centers, determine a naming convention for each data center and rack, for
example: DC1, DC2 or 100, 200 and RAC1, RAC2 or R101, R102. Choose the name carefully;
renaming a data center is not possible.
• Other possible configuration settings are described in cassandra.yaml configuration file and property
files such as cassandra-rackdc.properties.
About this task

This example describes installing a 6 node cluster spanning 2 racks in a single data center. Each node is
configured to use the GossipingPropertyFileSnitch and 256 virtual nodes (vnodes).
In Cassandra, the term data center is a grouping of nodes. Data center is synonymous with replication
group, that is, a grouping of nodes configured together for replication purposes.
Procedure
1. Suppose you install Cassandra on these nodes:
node0 110.82.155.0 (seed1)

node1 110.82.155.1
node2 110.82.155.2
node3 110.82.156.3 (seed2)
node4 110.82.156.4
node5 110.82.156.5
Note: It is a best practice to have more than one seed node per data center.
57
2. If you have a firewall running in your cluster, you must open certain ports for communication between
the nodes. See Configuring firewall port access.
3. If Cassandra is running, you must stop the server and clear the data:
Package installations:
a) Stop Cassandra:

b) Clear the data:

Tarball installations:
a) Stop Cassandra:

$ sudo kill pid
b) Clear the data:

4. Set the properties in the cassandra.yaml file for each node:
Note: After making any changes in the cassandra.yaml file, you must restart the node for the
changes to take effect.
Properties to set:
• num_tokens: recommended value: 256
• -seeds: internal IP address of each seed node
Seed nodes do not bootstrap, which is the process of a new node joining an existing cluster. For
new clusters, the bootstrap process on seed nodes is skipped.
• listen_address:
If not set, Cassandra asks the system for the local address, the one associated with its hostname.
In some cases Cassandra doesn't produce the correct address and you must specify the
listen_address.
• endpoint_snitch: name of snitch (See endpoint_snitch.) If you are changing snitches, see Switching
snitches.
• auto_bootstrap: false (Add this setting only when initializing a fresh cluster with no data.)
Note: If the nodes in the cluster are identical in terms of disk layout, shared libraries, and so on,
you can use the same copy of the cassandra.yaml file on all of them.
Example:
cluster_name: 'MyCassandraCluster'
num_tokens: 256
seed_provider:
- class_name: org.apache.cassandra.locator.SimpleSeedProvider
parameters:
- seeds: "110.82.155.0,110.82.155.3"
listen_address:
rpc_address: 0.0.0.0
endpoint_snitch: GossipingPropertyFileSnitch
58
5. In the cassandra-rackdc.properties file, assign the data center and rack names you determined
in the Prerequisites. For example:
# indicate the rack and dc for this node

dc=DC1
rack=RAC1
6. After you have installed and configured Cassandra on all nodes, start the seed nodes one at a time,
and then start the rest of the nodes.
Note: If the node has restarted because of automatic restart, you must first stop the node and
clear the data directories, as described above.
$ sudo service cassandra start

$ cd install_location
$ bin/cassandra
7. To check that the ring is up and running, run:
$ nodetool status
$ bin/nodetool status
Each node should be listed and it's status and state should be UN (Up Normal).

conf/cassandra.yaml
Initializing a multiple node cluster (multiple data centers)

A deployment scenario for a Cassandra cluster with multiple data centers.
About this task

This topic contains information for deploying a Cassandra cluster with multiple data centers. If you're new
to Cassandra, and haven't set up a cluster, see Cassandra and DataStax Enterprise Essentials or 10
Minute Cassandra Walkthrough.
59
This example describes installing a six node cluster spanning two data centers. Each node is configured to
use the GossipingPropertyFileSnitch (multiple rack aware) and 256 virtual nodes (vnodes).
In Cassandra, the term data center is a grouping of nodes. Data center is synonymous with replication
group, that is, a grouping of nodes configured together for replication purposes.
Before you begin

Each node must be correctly configured before starting the cluster. You must determine or perform the
following before starting the cluster:
• A good understanding of how Cassandra works. Be sure to read at least Understanding the
architecture, Data replication, and Cassandra's rack feature.
• Install Cassandra on each node.
• Choose a name for the cluster.
• Get the IP address of each node.
• Determine which nodes will be seed nodes. Do not make all nodes seed nodes. Please read
Internode communications (gossip).
• Determine the snitch and replication strategy. The GossipingPropertyFileSnitch and
NetworkTopologyStrategy are recommended for production environments.
• If using multiple data centers, determine a naming convention for each data center and rack, for
example: DC1, DC2 or 100, 200 and RAC1, RAC2 or R101, R102. Choose the name carefully;
renaming a data center is not possible.
• Other possible configuration settings are described in cassandra.yaml configuration file and property
files such as cassandra-rackdc.properties.
Procedure
1. Suppose you install Cassandra on these nodes:
node0 10.168.66.41 (seed1)

node1 10.176.43.66
node2 10.168.247.41
node3 10.176.170.59 (seed2)
node4 10.169.61.170
node5 10.169.30.138
Note: It is a best practice to have more than one seed node per data center.
2. If you have a firewall running in your cluster, you must open certain ports for communication between
the nodes. See Configuring firewall port access.
3. If Cassandra is running, you must stop the server and clear the data:
a) Stop Cassandra:

b) Clear the data:

a) Stop Cassandra:

$ sudo kill pid
60
b) Clear the data:

4. Set the properties in the cassandra.yaml file for each node:
Note: After making any changes in the cassandra.yaml file, you must restart the node for the
changes to take effect.
Properties to set:
• num_tokens: recommended value: 256
• -seeds: internal IP address of each seed node
Seed nodes do not bootstrap, which is the process of a new node joining an existing cluster. For
new clusters, the bootstrap process on seed nodes is skipped.
• listen_address:
If not set, Cassandra asks the system for the local address, the one associated with its hostname.
In some cases Cassandra doesn't produce the correct address and you must specify the
listen_address.
• endpoint_snitch: name of snitch (See endpoint_snitch.) If you are changing snitches, see Switching
snitches.
• auto_bootstrap: false (Add this setting only when initializing a fresh cluster with no data.)
Note: If the nodes in the cluster are identical in terms of disk layout, shared libraries, and so on,
you can use the same copy of the cassandra.yaml file on all of them.
Example:
cluster_name: 'MyCassandraCluster'
num_tokens: 256
seed_provider:
- class_name: org.apache.cassandra.locator.SimpleSeedProvider
parameters:
- seeds: "10.168.66.41,10.176.170.59"
listen_address:
Note: Include at least one node from each data center.

5. In the cassandra-rackdc.properties file, assign the data center and rack names you determined
in the Prerequisites. For example:
Nodes 0 to 2

dc=DC1
rack=RAC1
Nodes 3 to 5

dc=DC2
rack=RAC1
6. After you have installed and configured Cassandra on all nodes, start the seed nodes one at a time,
and then start the rest of the nodes.
Note: If the node has restarted because of automatic restart, you must first stop the node and
clear the data directories, as described above.
61

$ bin/cassandra
7. To check that the ring is up and running, run:
$ nodetool status
$ bin/nodetool status
Each node should be listed and it's status and state should be UN (Up Normal).

conf/cassandra.yaml
62
Security
Security
Topics for securing Cassandra.
Securing Cassandra
Cassandra provides various security features to the open source community.
Cassandra provides these security features to the open source community.
• Client-to-node encryption
Cassandra includes an optional, secure form of communication from a client machine to a database
cluster. Client to server SSL ensures data in flight is not compromised and is securely transferred back/
forth from client machines.
• Authentication based on internally controlled login accounts/passwords
Administrators can create users who can be authenticated to Cassandra database clusters using the
CREATE USER command. Internally, Cassandra manages user accounts and access to the database
cluster using passwords. User accounts may be altered and dropped using the Cassandra Query
Language (CQL).
• Object permission management
Once authenticated into a database cluster using either internal authentication, the next security issue
to be tackled is permission management. What can the user do inside the database? Authorization
capabilities for Cassandra use the familiar GRANT/REVOKE security paradigm to manage object
permissions.
SSL encryption
Topics for using SSL in Cassandra.
Client-to-node encryption
Client-to-node encryption protects data in flight from client machines to a database cluster using SSL
(Secure Sockets Layer).
About this task

Client-to-node encryption protects data in flight from client machines to a database cluster using SSL
(Secure Sockets Layer). It establishes a secure channel between the client and the coordinator node.

conf/cassandra.yaml
Before you begin

All nodes must have all the relevant SSL certificates on all nodes. See Preparing server certificates.
About this task

To enable client-to-node SSL, you must set the client_encryption_options in the cassandra.yaml file.
63
Security
Procedure
On each node under client_encryption_options:
• Enable encryption.
• Set the appropriate paths to your .keystore and .truststore files.
• Provide the required passwords. The passwords must match the passwords used when generating the
keystore and truststore.
• To enable client certificate authentication, set require_client_auth to true. (Available starting with
Cassandra 1.2.3.)
Example
client_encryption_options:
enabled: true
keystore: conf/.keystore ## The path to your .keystore file
keystore_password: <keystore password> ## The password you used when
generating the keystore.
truststore: conf/.truststore
truststore_password: <truststore password>
require_client_auth: <true or false>
Node-to-node encryption
Node-to-node encryption protects data transferred between nodes in a cluster, including gossip
communications, using SSL (Secure Sockets Layer).
About this task

Node-to-node encryption protects data transferred between nodes in a cluster, including gossip
communications, using SSL (Secure Sockets Layer).
Before you begin

All nodes must have all the relevant SSL certificates on all nodes. See Preparing server certificates.
About this task

To enable node-to-node SSL, you must set the server_encryption_options in the cassandra.yaml file.

conf/cassandra.yaml
Procedure
On each node under sever_encryption_options:
• Enable internode_encryption.
The available options are:
• all
• none
• dc: Cassandra encrypts the traffic between the data centers.
• rack: Cassandra encrypts the traffic between the racks.
• Set the appropriate paths to your .keystore and .truststore files.
64
Security
• Provide the required passwords. The passwords must match the passwords used when generating the
keystore and truststore.
• To enable client certificate authentication, set require_client_auth to true. (Available starting with
Cassandra 1.2.3.)
Example
server_encryption_options:
internode_encryption: <internode_option>
keystore: resources/dse/conf/.keystore
keystore_password: <keystore password>
truststore: resources/dse/conf/.truststore
truststore_password: <truststore password>
require_client_auth: <true or false>
Using cqlsh with SSL encryption

Using a cqlshrc file with SSL encryption.
About this task

Using a cqlshrc file means you don't have to override the SSL_CERTFILE environmental variables every
time.
Procedure
1. To run cqlsh with SSL encryption, create a .cassandra/cqlshrc file in your home or client program
directory.
Sample files are available in the following directories:
• Package installations: /etc/cassandra
• Tarball installations: install_location/conf
2. Start cqlsh with the --ssl option.
$ cqlsh --ssl ## Package installations

$ install_location/bin/cqlsh -ssl ## Tarball installations
Example
[authentication]
username = fred
password = !!bang!!$
[connection]
hostname = 127.0.0.1
port = 9042
[ssl]
certfile = ~/keys/cassandra.cert
validate = true ## Optional, true by default
userkey = ~/key.pem ## Provide when require_client_auth=true
usercert = ~/cert.pem ## Provide when require_client_auth=true
[certfiles] ## Optional section, overrides the default certfile in the

[ssl] section
192.168.1.3 = ~/keys/cassandra01.cert
192.168.1.4 = ~/keys/cassandra02.cert
65
Security
Note: When validate is enabled, the host in the certificate is compared to the host of the machine
that it is connected to. The SSL certificate must be provided either in the configuration file or as an
environment variable. The environment variables (SSL_CERTFILE and SSL_VALIDATE) override
any options set in this file.
Preparing server certificates

Steps to generate SSL certificates for client-to-node encryption or node-to-node encryption.
About this task

How to generate SSL certificates for client-to-node encryption or node-to-node encryption. If you generate
the certificates for one type of encryption, you do not need to generate them again for the other: the same
certificates are used for both. All nodes must have all the relevant SSL certificates on all nodes. A keystore
contains private keys. The truststore contains SSL certificates for each node and doesn't require signing by
a trusted and recognized public certification authority.
Procedure
1. Generate the private and public key pair for the nodes of the cluster.
A prompt for the new keystore and key password appears.
2. Leave key password the same as the keystore password.
3. Repeat steps 1 and 2 on each node using a different alias for each one.
keytool -genkey -keyalg RSA -alias <cassandra_node0> -keystore .keystore
4. Export the public part of the certificate to a separate file and copy these certificates to all other nodes.
keytool -export -alias cassandra -file cassandranode0.cer -

keystore .keystore
5. Add the certificate of each node to the truststore of each node, so nodes can verify the identity of other
nodes.
keytool -import -v -trustcacerts -alias <cassandra_node0> -file

<cassandra_node0>.cer -keystore .truststore
keytool -import -v -trustcacerts -alias <cassandra_node1> -file
<cassandra_node1>.cer -keystore .truststore
. . .
6. Distribute the .keystore and .truststore files to all Cassandra nodes.

7. Make sure .keystore is readable only to the Cassandra daemon and not by any user of the system.
Adding new trusted users

Add new users when client certificate authentication is enabled.
About this task

How to add new users when client certificate authentication is enabled.
Before you begin

The client certificate authentication must be enabled (require_client_auth=true).
Procedure
1. Generate the certificate as described in Client-to-node encryption.
66
Security
2. Import the user's certificate into every node's truststore using keytool:
keytool -import -v -trustcacerts -alias <username> -file <certificate

file> -keystore .truststore
Internal authentication
Topics for internal authentication.
Internal authentication
Internal authentication is based on Cassandra-controlled login accounts and passwords.
Like object permission management using internal authorization, internal authentication is based on
Cassandra-controlled login accounts and passwords. Internal authentication works for the following clients
when you provide a user name and password to start up the client:
• Astyanax
• cassandra-cli
• cqlsh
• DataStax drivers - produced and certified by DataStax to work with Cassandra.
• Hector
• pycassa
Internal authentication stores usernames and bcrypt-hashed passwords in the system_auth.credentials
table.
PasswordAuthenticator is an IAuthenticator implementation that you can use to configure Cassandra for
internal authentication out-of-the-box.
Configuring authentication
Steps for configuring authentication.
About this task

To configure Cassandra to use internal authentication, first make a change to the cassandra.yaml file and
increase the replication factor of the system_auth keyspace, as described in this procedure. Next, start up
Cassandra using the default user name and password (cassandra/cassandra), and start cqlsh using the
same credentials. Finally, use these CQL statements to set up user accounts to authorize users to access
the database objects:
• ALTER USER
• CREATE USER
• DROP USER
• LIST USERS
Note: To configure authorization, see Configuring internal authorization.

conf/cassandra.yaml
Procedure
1. Change the authenticator option in the cassandra.yaml file to PasswordAuthenticator.
67
Security
By default, the authenticator option is set to AllowAllAuthenticator.
authenticator: PasswordAuthenticator
2. Increase the replication factor for the system_auth keyspace to N (number of nodes).
If you use the default, 1, and the node with the lone replica goes down, you will not be able to log into
the cluster because the system_auth keyspace was not replicated.
3. Restart the Cassandra client.
The default superuser name and password that you use to start the client is stored in Cassandra.
<client startup string> -u cassandra -p cassandra
4. Start cqlsh using the superuser name and password.
./cqlsh -u cassandra -p cassandra
5. Create another superuser, not named cassandra. This step is optional but highly recommended.
6. Log in as that new superuser.
7. Change the cassandra user password to something long and incomprehensible, and then forget about
it. It won't be used again.
8. Take away the cassandra user's superuser status.
9. Use the CQL statements listed previously to set up user accounts and then grant permissions to access
the database objects.
Logging in using cqlsh

How to create a cqlshrc file to avoid having enter credentials every time you launch cqlsh.
About this task

Typically, after configuring authentication, you log into cqlsh using the -u and -p options to the cqlsh
command. To avoid having enter credentials every time you launch cqlsh, you can create a cqlshrc file
in the .cassandra directory, which is in your home directory. When present, this file passes default login
information to cqlsh.
Note: Sample cqlshrc files are available in:
• Package installations: /etc/cassandra
Procedure
1. Open a text editor and create a file that specifies a user name and password.
[authentication]
username = fred
password = !!bang!!$
2. Save the file in your home/.cassandra directory and name it cqlshrc.

3. Set permissions on the file.
To protect database login information, ensure that the file is secure from unauthorized access.
Internal authorization
Topics about internal authorization.
68
Security
Object permissions
Granting or revoking permissions to access Cassandra data.
Cassandra provides the familiar relational database GRANT/REVOKE paradigm to grant or revoke
permissions to access Cassandra data. A superuser grants initial permissions, and subsequently a user
may or may not be given the permission to grant/revoke permissions. Object permission management is
based on internal authorization.
Read access to these system tables is implicitly given to every authenticated user because the tables are
used by most Cassandra tools:
• system.schema_keyspace
• system.schema_columns
• system.schema_columnfamilies
• system.local
• system.peers
Configuring internal authorization

Steps for adding the CassandraAuthorizer.
About this task

CassandraAuthorizer is one of many possible IAuthorizer implementations, and the one that stores
permissions in the system_auth.permissions table to support all authorization-related CQL statements.
Configuration consists mainly of changing the authorizer option in the cassandra.yaml to use the
CassandraAuthorizer.
Note: To configure authentication, see Configuring authentication.

conf/cassandra.yaml
Procedure
1. In the cassandra.yaml file, comment out the default AllowAllAuthorizer and add the
CassandraAuthorizer.
authorizer: CassandraAuthorizer
You can use any authenticator except AllowAll.

2. Configure the replication factor for the system_auth keyspace to increase the replication factor to a
number greater than 1.
3. Adjust the validity period for permissions caching by setting the permissions_validity_in_ms option in
the cassandra.yaml file.
Alternatively, disable permission caching by setting this option to 0.
Results
CQL supports these authorization statements:
• GRANT
• LIST PERMISSIONS
• REVOKE
69
Security
Configuring firewall port access

Which ports to open when nodes are protected by a firewall.
If you have a firewall running on the nodes in your Cassandra cluster, you must open up the following ports
to allow communication between the nodes, including certain Cassandra ports. If this isn’t done, when
you start Cassandra on a node, the node acts as a standalone database server rather than joining the
database cluster.
Table 2: Public ports
Port Description
number
22 SSH port
8888 OpsCenter website. The opscenterd daemon listens
on this port for HTTP requests coming directly from
the browser.
Table 3: Cassandra inter-node ports
Port Description
number
7000 Cassandra inter-node cluster communication.
7001 Cassandra SSL inter-node cluster communication.
7199 Cassandra JMX monitoring port.
Table 4: Cassandra client ports
Port Description
number
9042 Cassandra client port.
9160 Cassandra client port (Thrift).
Table 5: Cassandra OpsCenter ports
Port Description
number
61620 OpsCenter monitoring port. The opscenterd daemon
listens on this port for TCP traffic coming from the
agent.
61621 OpsCenter agent port. The agents listen on this port
for SSL traffic initiated by OpsCenter.
70
Security
Enabling JMX authentication

The default settings for Cassandra make JMX accessible only from localhost. To enable remote JMX
connections, change the LOCAL_JMX setting in cassandra-env.sh.
About this task

The default settings for Cassandra make JMX accessible only from localhost. If you want to enable remote
JMX connections, change the LOCAL_JMX setting in cassandra-env.sh and enable authentication
and/or ssl. After you enable JMX authentication, ensure that tools that use JMX, such as nodetool and
DataStax OpsCenter, are configured to use authentication.
To use JMX authentication for OpsCenter, follow the steps in Modifying OpsCenter cluster connections.
Procedure
1. Open the cassandra-env.sh file for editing and update or add these lines:
JVM_OPTS="$JVM_OPTS -Dcom.sun.management.jmxremote.authenticate=true"
JVM_OPTS="$JVM_OPTS -Dcom.sun.management.jmxremote.password.file=/etc/
cassandra/jmxremote.password"
If the LOCAL_JMX setting is in your file:
LOCAL_JMX=no
2. Copy the jmxremote.password.template from /jre_install_location/lib/management/
to /etc/cassandra/ and rename it to jmxremote.password:
cp /<jre_install_dir/lib/management/jmxremote.password.template /etc/
cassandra/jmxremote.password
3. Change the ownership of jmxremote.password to the user you run cassandra with and change
permission to read only:
chown cassandra:cassandra /etc/cassandra/jmxremote.password
chmod 400 /etc/cassandra/jmxremote.password
4. Edit jmxremote.password and add the user and password for JMX-compliant utilities:
monitorRole QED
controlRole R&D
cassandra cassandrapassword
Note: This cassandra user and cassandra password is just an example. Specify the user and
password for your environment.
5. Add the cassandra user with read and write permission to /jre_install_location/lib/
management/jmxremote.access:
monitorRole readonly
cassandra readwrite
controlRole readwrite \
create javax.management.monitor.,javax.management.timer. \
unregister
6. Restart Cassandra.
7. Run nodetool with the cassandra user and password.
$ nodetool status -u cassandra -pw cassandra
Results
If you run nodetool without user and password, you see an error similar to:
root@VM1 cassandra]# nodetool status
71
Security
Exception in thread "main" java.lang.SecurityException: Authentication failed!

Credentials required
at
com.sun.jmx.remote.security.JMXPluggableAuthenticator.authenticationFailure(Unknown
Source)
at com.sun.jmx.remote.security.JMXPluggableAuthenticator.authenticate(Unknown
Source)
at sun.management.jmxremote.ConnectorBootstrap
$AccessFileCheckerAuthenticator.authenticate(Unknown Source)
at javax.management.remote.rmi.RMIServerImpl.doNewClient(Unknown Source)
at javax.management.remote.rmi.RMIServerImpl.newClient(Unknown Source)
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(Unknown Source)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(Unknown Source)
at java.lang.reflect.Method.invoke(Unknown Source)
at sun.rmi.server.UnicastServerRef.dispatch(Unknown Source)
at sun.rmi.transport.Transport$1.run(Unknown Source)
at sun.rmi.transport.Transport$1.run(Unknown Source)
at java.security.AccessController.doPrivileged(Native Method)
at sun.rmi.transport.Transport.serviceCall(Unknown Source)
at sun.rmi.transport.tcp.TCPTransport.handleMessages(Unknown Source)
at sun.rmi.transport.tcp.TCPTransport$ConnectionHandler.run0(Unknown Source)
at sun.rmi.transport.tcp.TCPTransport$ConnectionHandler.run(Unknown Source)
at java.util.concurrent.ThreadPoolExecutor.runWorker(Unknown Source)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(Unknown Source)
at java.lang.Thread.run(Unknown Source)
at sun.rmi.transport.StreamRemoteCall.exceptionReceivedFromServer(Unknown
Source)
at sun.rmi.transport.StreamRemoteCall.executeCall(Unknown Source)
at sun.rmi.server.UnicastRef.invoke(Unknown Source)
at javax.management.remote.rmi.RMIServerImpl_Stub.newClient(Unknown Source)
at javax.management.remote.rmi.RMIConnector.getConnection(Unknown Source)
at javax.management.remote.rmi.RMIConnector.connect(Unknown Source)
at javax.management.remote.JMXConnectorFactory.connect(Unknown Source)
at org.apache.cassandra.tools.NodeProbe.connect(NodeProbe.java:146)
at org.apache.cassandra.tools.NodeProbe.<init>(NodeProbe.java:116)
at org.apache.cassandra.tools.NodeCmd.main(NodeCmd.java:1099)
72
Database internals
Database internals
Topics about the Cassandra database.
Storage engine
A description about Cassandra's storage structure and engine.
Cassandra uses a storage structure similar to a Log-Structured Merge Tree, unlike a typical relational
database that uses a B-Tree. Cassandra avoids reading before writing. Read-before-write, especially in
a large distributed system, can produce stalls in read performance and other problems. For example, two
clients read at the same time, one overwrites the row to make update A, and then the other overwrites the
row to make update B, removing update A. Reading before writing also corrupts caches and increases
IO requirements. To avoid a read-before-write condition, the storage engine groups inserts/updates to be
made, and sequentially writes only the updated parts of a row in append mode. Cassandra never re-writes
or re-reads existing data, and never overwrites the rows in place.
A log-structured engine that avoids overwrites and uses sequential IO to update data is essential for writing
to hard disks (HDD) and solid-state disks (SSD). On HDD, writing randomly involves a higher number of
seek operations than sequential writing. The seek penalty incurred can be substantial. Using sequential
IO, and thereby avoiding write amplification and disk failure, Cassandra accommodates inexpensive,
consumer SSDs extremely well.
Separate table directories

Cassandra provides fine-grained control of table storage on disk.
Cassandra provides fine-grained control of table storage on disk, writing tables to disk using separate
directories for each table. From the installation directory, data files are stored using this directory and file
naming format on default tarball installations:
/data/data/ks1/cf1-5be396077b811e3a3ab9dc4b9ac088d/ks1-cf1-hc-1-Data.db
On packaged installations, the data files are stored in the same format, but in /var/lib/cassandra/
data by default. In this example, ks1 represents the keyspace name to distinguish the keyspace for
streaming or bulk loading data. A hexadecimal string, 5be396077b811e3a3ab9dc4b9ac088d in this
example, is appended to table names to represent unique table IDs.
Cassandra creates a subdirectory for each table, which allows you to symlink a table to a chosen physical
drive or data volume. This provides the capability to move very active tables to faster media, such as
SSD’s for better performance, and also divvy up tables across all attached storage devices for better I/O
balance at the storage layer.
Cassandra storage basics

Understanding how Casssandra stores data.
To manage and access data in Cassandra, it is important to understand how Casssandra stores data. The
hinted handoff feature and Cassandra conformance and non-conformance to the ACID (atomic, consistent,
isolated, durable) database properties are key concepts in this discussion. In Cassandra, consistency
refers to how up-to-date and synchronized a row of data is on all of its replicas.
Client utilities and application programming interfaces (APIs) for developing applications for data storage
and retrieval are available.
73
Database internals
The write path to compaction

Cassandra processes data at several stages on the write path.
Cassandra processes data at several stages on the write path, starting with the immediate logging of a
write and ending in compaction:
• Logging data in the commit log
• Writing data to the memtable
• Flushing data from the memtable
• Storing data on disk in SSTables
• Compaction
Logging writes and memtable storage

When a write occurs, Cassandra stores the data in a structure in memory, the memtable, and also
appends writes to the commit log on disk, providing configurable durability. The commit log receives every
write made to a Cassandra node, and these durable writes survive permanently even after power failure.
The memtable is a write-back cache of data partitions that Cassandra looks up by key. The memtable
stores writes until reaching a limit, and then is flushed.
Flushing data from the memtable

When memtable contents exceed a configurable threshold, the memtable data, which includes
indexes, is put in a queue to be flushed to disk. You can configure the length of the queue by changing
memtable_flush_queue_size in the cassandra.yaml. If the data to be flushed exceeds the queue size,
Cassandra blocks writes until the next flush succeeds. You can manually flush a table using the nodetool
flush command. Typically, before restarting nodes, flushing the memtable is recommended to reduce
commit log replay time. To flush the data, Cassandra sorts memtables by token and then writes the data to
disk sequentially.
Storing data on disk in SSTables

Data in the commit log is purged after its corresponding data in the memtable is flushed to an SSTable.
Memtables and SSTables are maintained per table. SSTables are immutable, not written to again after the
memtable is flushed. Consequently, a partition is typically stored across multiple SSTable files.
For each SSTable, Cassandra creates these structures:
74
Database internals
• Partition index
A list of partition keys and the start position of rows in the data file (on disk)
• Partition summary (in memory)
A sample of the partition index.
• Bloom filter
Compaction
Periodic compaction is essential to a healthy Cassandra database because Cassandra does not insert/
update in place. As inserts/updates occur, instead of overwriting the rows, Cassandra writes a new
timestamped version of the inserted or updated data in another SSTable. Cassandra manages the
accumulation of SSTables on disk using compaction.
Cassandra also does not delete in place because the SSTable is immutable. Instead, Cassandra marks
data to be deleted using a tombstone. Tombstones exist for a configured time period defined by the
gc_grace_seconds value set on the table.
During compaction, there is a temporary spike in disk space usage and disk I/O because the old and new
SSTables co-exist. This diagram depicts the compaction process:
Compaction merges the data in each SSTable by partition key, selecting the latest data for storage
based on its timestamp. Cassandra can merge the data performantly, without random IO, because rows
are sorted by partition key within each SSTable. After evicting tombstones and removing deleted data,
columns, and rows, the compaction process consolidates SSTables into a single file. The old SSTable
files are deleted as soon as any pending reads finish using the files. Disk space occupied by old SSTables
becomes available for reuse.
Cassandra 2.1 improves read performance after compaction by performing an incremental replacement
of compacted SSTables. Instead of waiting for the entire compaction to finish and then throwing away the
75
Database internals
old SSTable (and cache), Cassandra can read data directly from the new SSTable even before it finishes
writing.
As data is written to the new SSTable and reads are directed to it, the corresponding data in the old
SSTables is no longer accessed and is evicted from the page cache. Thus begins an incremental process
of caching the new SSTable, while directing reads away from the old one. The dramatic cache miss is
gone. Cassandra provides predictable high performance even under heavy load.
Starting compaction
You can configure these types of compaction to run periodically:
• SizeTieredCompactionStrategy
For write-intensive workloads
• LeveledCompactionStrategy
For read-intensive workloads
• DateTieredCompactionStrategy
For time series data and expiring (TTL) data
You can manually start compaction using the nodetool compact command.
For more information about compaction strategies, see When to Use Leveled Compaction and Leveled
Compaction in Apache Cassandra.

conf/cassandra.yaml
How Cassandra stores and distributes indexes

A brief description of how Cassandra stores and distributes indexes.
Internally, a Cassandra index is a data partition. In the example of a music service, the playlists table
includes an artist column and uses a compound partition key: id is the partition key and song_order is the
clustering column.
CREATE TABLE playlists (

id uuid,
song_order int,
. . .
artist text,
PRIMARY KEY (id, song_order ) );
As shown in the music service example, to filter the data based on the artist, create an index on artist.
Cassandra uses the index to pull out the records in question. An attempt to filter the data before creating
the index will fail because the operation would be very inefficient. A sequential scan across the entire
playlists dataset would be required. After creating the artist index, Cassandra can filter the data in the
playlists table by artist, such as Fu Manchu.
The partition is the unit of replication in Cassandra. In the music service example, partitions are distributed
by hashing the playlist id and using the ring to locate the nodes that store the distributed data. Cassandra
would generally store playlist information on different nodes, and to find all the songs by Fu Manchu,
Cassandra would have to visit different nodes. To avoid these problems, each node indexes its own data.
This technique, however, does not guarantee trouble-free indexing, so know when and when not to use an
index.
76
Database internals
About index updates

A brief description about index updates.
As with relational databases, keeping indexes up to date is not free, so unnecessary indexes should be
avoided. When a column is updated, the index is updated as well. If the old column value was still in the
memtable, which typically occurs when updating a small set of rows repeatedly, Cassandra removes the
corresponding obsolete index entry; otherwise, the old entry remains to be purged by compaction. If a read
sees a stale index entry before compaction purges it, the reader thread invalidates it.
The write path of an update

A brief description of the write path of an update.
Inserting a duplicate primary key is treated as an upsert. Eventually, the updates are streamed to disk
using sequential I/O and stored in a new SSTable. During an update, Cassandra time-stamps and
writes columns to disk using the write path. During the update, if multiple versions of the column exist
in the memtable, Cassandra flushes only the newer version of the column to disk, as described in the
Compaction section.
About deletes
How Cassandra deletes data and why deleted data can reappear.
The way Cassandra deletes data differs from the way a relational database deletes data. A relational
database might spend time scanning through data looking for expired data and throwing it away or an
administrator might have to partition expired data by month, for example, to clear it out faster. Data in a
Cassandra column can have an optional expiration date called TTL (time to live). Use CQL to set the TTL
in seconds for data. Cassandra marks TTL data with a tombstone after the requested amount of time has
expired. A tombstone exists for gc_grace_seconds. After data is marked with a tombstone, the data is
automatically removed during the normal compaction process.
Facts about deleted data to keep in mind are:
• Cassandra does not immediately remove data marked for deletion from disk. The deletion occurs during
compaction.
• If you use the size-tiered or date-tiered compaction strategy, you can drop data immediately by
manually starting the compaction process. Before doing so, understand the documented disadvantages
of the process.
• Deleted data can reappear if you do not run node repair routinely.
Why deleted data can reappear
Marking data with a tombstone signals Cassandra to retry sending a delete request to a replica that was
down at the time of delete. If the replica comes back up within the grace period of time, it eventually
receives the delete request. However, if a node is down longer than the grace period, the node can miss
the delete because the tombstone disappears after gc_grace_seconds. Cassandra always attempts to
replay missed updates when the node comes back up again. After a failure, it is a best practice to run node
repair to repair inconsistencies across all of the replicas when bringing a node back into the cluster. If the
node doesn't come back within gc_grace,_seconds, remove the node, wipe it, and bootstrap it again.
About hinted handoff writes

How hinted handoff works and how it optimizes the cluster.
Hinted handoff is a Cassandra feature that optimizes the cluster consistency process and anti-entropy
when a replica-owning node is not available, due to network issues or other problems, to accept a replica
from a successful write operation. Hinted handoff is not a process that guarantees successful write
77
Database internals
operations, except when a client application uses a consistency level of ANY. You enable or disable hinted
handoff in the cassandra.yaml file.
How hinted handoff works

During a write operation, when hinted handoff is enabled and consistency can be met, the coordinator
stores a hint about dead replicas in the local system.hints table under either of these conditions:
• A replica node for the row is known to be down ahead of time.
• A replica node does not respond to the write request.
When the cluster cannot meet the consistency level specified by the client, Cassandra does not store a
hint.
A hint indicates that a write needs to be replayed to one or more unavailable nodes. The hint consists of:
• The location of the replica that is down
• Version metadata
• The actual data being written
By default, hints are saved for three hours after a replica fails because if the replica is down
longer than that, it is likely permanently dead. You can configure this interval of time using the
max_hint_window_in_ms property in the cassandra.yaml file. If the node recovers after the save time has
elapsed, run a repair to re-replicate the data written during the down time.
After a node discovers from gossip that a node for which it holds hints has recovered, the node sends the
data row corresponding to each hint to the target. Additionally, the node checks every ten minutes for any
hints for writes that timed out during an outage too brief for the failure detector to notice through gossip.
For example, in a cluster of two nodes, A and B, having a replication factor (RF) of 1, each row is stored on
one node. Suppose node A is down while we write row K to it with consistency level of one. The write fails
because reads always reflect the most recent write when:
W + R > replication factor
where W is the number of nodes to block for writes and R is the number of nodes to block for reads.
Cassandra does not write a hint to B and call the write good because Cassandra cannot read the data at
any consistency level until A comes back up and B forwards the data to A.
In a cluster of three nodes, A (the coordinator), B, and C, each row is stored on two nodes in a keyspace
having a replication factor of 2. Suppose node C goes down. The client writes row K to node A. The
coordinator, replicates row K to node B, and writes the hint for downed node C to node A.
78
Database internals
Cassandra, configured with a consistency level of ONE, calls the write good because Cassandra can read
the data on node B. When node C comes back up, node A reacts to the hint by forwarding the data to node
C. For more information about how hinted handoff works, see "Modern hinted handoff" by Jonathan Ellis.
Extreme write availability

For applications that want Cassandra to accept writes even when all the normal replicas are down, when
not even consistency level ONE can be satisfied, Cassandra provides consistency level ANY. ANY
guarantees that the write is durable and will be readable after an appropriate replica target becomes
available and receives the hint replay.
Performance
By design, hinted handoff inherently forces Cassandra to continue performing the same number of writes
even when the cluster is operating at reduced capacity. Pushing your cluster to maximum capacity with no
allowance for failures is a bad idea.
Hinted handoff is designed to minimize the extra load on the cluster.
All hints for a given replica are stored under a single partition key, so replaying hints is a simple sequential
read with minimal performance impact.
If a replica node is overloaded or unavailable, and the failure detector has not yet marked it down, then
expect most or all writes to that node to fail after the timeout triggered by write_request_timeout_in_ms,
which defaults to 10 seconds.
If this happens on many nodes at once this could become substantial memory pressure on the coordinator.
So the coordinator tracks how many hints it is currently writing, and if this number gets too high it will
temporarily refuse writes (withOverloadedException) whose replicas include the misbehaving nodes.
79
Database internals
Removal of hints
When removing a node from the cluster by decommissioning the node or by using the nodetool
removenode command, Cassandra automatically removes hints targeting the node that no longer exists.
Cassandra also removes hints for dropped tables.
Scheduling repair weekly

At first glance, it seems that hinted handoff eliminates the need for repair, but this is not true because
hardware failure is inevitable and has the following ramifications:
• Loss of the historical data necessary to tell the rest of the cluster exactly what data is missing.
• Loss of hints-not-yet-replayed from requests that the failed node coordinated.

conf/cassandra.yaml
Reads
How reads work and factors affecting them.
About reads
How Cassandra combines results from the active memtable and potentially mutliple SSTables to satisfy a
read.
To satisfy a read, Cassandra must combine results from the active memtable and potentially mutliple
SSTables.
First, Cassandra checks the Bloom filter. Each SSTable has a Bloom filter associated with it that checks
the probability of having any data for the requested partition in the SSTable before doing any disk I/O.
If the Bloom filter does not rule out the SSTable, Cassandra checks the partition key cache and takes one
of these courses of action:
• If an index entry is found in the cache:
• Cassandra goes to the compression offset map to find the compressed block having the data.
• Fetches the compressed data on disk and returns the result set.
• If an index entry is not found in the cache:
• Cassandra searches the partition summary to determine the approximate location on disk of the
index entry.
• Next, to fetch the index entry, Cassandra hits the disk for the first time, performing a single seek and
a sequential read of columns (a range read) in the SSTable if the columns are contiguous.
• Cassandra goes to the compression offset map to find the compressed block having the data.
• Fetches the compressed data on disk and returns the result set.
80
Database internals
How off-heap components affect reads

Factors for increasing the data handling capacity per node.
To increase the data handling capacity per node, Cassandra keeps these components off-heap:
• Bloom filter
• Compression offsets map
• Partition summary
Of the components in memory, only the partition key cache is a fixed size. Other components grow as the
data set grows.
Bloom filter
The Bloom filter grows to approximately 1-2 GB per billion partitions. In the extreme case, you can have
one partition per row, so you can easily have billions of these entries on a single machine. The Bloom filter
is tunable if you want to trade memory for performance.
Partition summary
By default, the partition summary is a sample of the partition index. You configure sample frequency
by changing the index_interval property in the table definition, also if you want to trade memory for
performance.
Compression offsets
The compression offset map grows to 1-3 GB per terabyte compressed. The more you compress data, the
greater number of compressed blocks you have and the larger the compression offset table. Compression
is enabled by default even though going through the compression offset map consumes CPU resources.
Having compression enabled makes the page cache more effective, and typically, almost always pays off.
81
Database internals
Reading from a partition

A brief description about reading from a partition.
Within a partition, all rows are not equally expensive to query. The very beginning of the partition—the
first rows, clustered by your key definition—is slightly less expensive to query because there is no need to
consult the partition-level index.
How write patterns affect reads

A brief description about how write patterns affect reads.
The type of compaction strategy Cassandra performs on your data is configurable and can significantly
affect read performance. Using the SizeTieredCompactionStrategy or DateTieredCompactionStrategy
tends to cause data fragmentation when rows are frequently updated. The LeveledCompactionStrategy
(LCS) was designed to prevent fragmentation under this condition. For more information about LCS, see
the article, Leveled Compaction in Apache Cassandra.
How the row cache affects reads

A brief description and illustration about how the row cache affects reads.
Typical of any database, reads are fastest when the most in-demand data (or hot working set) fits into
memory. Although all modern storage systems rely on some form of caching to allow for fast access to
hot data, not all of them degrade gracefully when the cache capacity is exceeded and disk I/O is required.
Cassandra's read performance benefits from built-in caching, shown in the following diagram.
The red lines in the SSTables in this diagram are fragments of a row that Cassandra needs to combine to
give the user the requested results. Cassandra caches the merged value, not the raw row fragments. That
saves some CPU and disk I/O.
The row cache is not write-through. If a write comes in for the row, the cache for it is invalidated and is not
cached again until it is read.
For rows that are accessed frequently, Cassandra 2.1 has improved the built-in partition key cache and an
optional row cache.
About transactions and concurrency control

A brief description about transactions and concurrency control.
Cassandra does not use RDBMS ACID transactions with rollback or locking mechanisms, but instead
offers atomic, isolated, and durable transactions with eventual/tunable consistency that lets the user decide
how strong or eventual they want each transaction’s consistency to be.
82
Database internals
As a non-relational database, Cassandra does not support joins or foreign keys, and consequently does
not offer consistency in the ACID sense. For example, when moving money from account A to B the total
in the accounts does not change. Cassandra supports atomicity and isolation at the row-level, but trades
transactional isolation and atomicity for high availability and fast write performance. Cassandra writes are
durable.
Atomicity
Everything in a transaction succeeds or the entire transaction is rolled back.
In Cassandra, a write is atomic at the partition-level, meaning inserting or updating columns in a row is
treated as one write operation. A delete operation is also performed atomically. By default, all operations
in a batch are performed atomically. Cassandra uses a batch log to ensure all operations in a batch
are applied atomically. There is a performance penalty for batch atomicity when a batch spans multiple
partitions. If you do not want to incur this penalty, use the UNLOGGED option. Using UNLOGGED makes
the batch operation atomic only within a single partition.
For example, if using a write consistency level of QUORUM with a replication factor of 3, Cassandra will
replicate the write to all nodes in the cluster and wait for acknowledgement from two nodes. If the write fails
on one of the nodes but succeeds on the other, Cassandra reports a failure to replicate the write on that
node. However, the replicated write that succeeds on the other node is not automatically rolled back.
Cassandra uses timestamps to determine the most recent update to a column. Depending on the version
of the Native CQL Protocol, the timestamp is provided by either the client application or the server. The
latest timestamp always wins when requesting data, so if multiple client sessions update the same columns
in a row concurrently, the most recent update is the one seen by readers.
Important: Native CQL Protocol V3 supports client-side timestamps. Be sure to check your client's
documentation to ensure that it generates client-side timestamps and that this feature is activated.
Consistency
A transaction cannot leave the database in an inconsistent state. Cassandra offers different types of
consistency.
Cassandra offers two consistency types:
• Tunable consistency
Availability and consistency can be tuned, and can be strong in the CAP sense--data is made
consistent across all the nodes in a distributed database cluster.
• Linearizable consistency
In ACID terms, linearizable consistency is a serial (immediate) isolation level for lightweight
transactions.
In Cassandra, there are no locking or transactional dependencies when concurrently updating multiple
rows or tables. Tuning availability and consistency always gives you partition tolerance. A user can pick
and choose on a per operation basis how many nodes must receive a DML command or respond to a
SELECT query.
For in-depth information about this new consistency level, see the article, Lightweight transactions in
Cassandra.
To support linearizable consistency, a consistency level of SERIAL has been added to Cassandra.
Additions to CQL have been made to support lightweight transactions.
Isolation
Transactions cannot interfere with each other.
In early versions of Cassandra, it was possible to see partial updates in a row when one user was updating
the row while another user was reading that same row. For example, if one user was writing a row with two
83
Database internals
thousand columns, another user could potentially read that same row and see some of the columns, but
not all of them if the write was still in progress.
Full row-level isolation is in place, which means that writes to a row are isolated to the client performing
the write and are not visible to any other user until they are complete. Delete operations are performed in
isolation. All updates in a batch operation belonging to a given partition key are performed in isolation.
Durability
Completed transactions persist in the event of crashes or server failure.
Writes in Cassandra are durable. All writes to a replica node are recorded both in memory and in a commit
log on disk before they are acknowledged as a success. If a crash or server failure occurs before the
memtables are flushed to disk, the commit log is replayed on restart to recover any lost writes. In addition
to the local durability (data immediately written to disk), the replication of data on other nodes strengthens
durability.
You can manage the local durability to suit your needs for consistency using the commitlog_sync option in
the cassandra.yaml file. Set the option to either periodic or batch.

conf/cassandra.yaml
Lightweight transactions
A description about lightweight transactions and when to use them.
Lightweight transactions with linearizable consistency ensure transaction isolation level similar to the
serializable level offered by RDBMS’s. They are also known as compare and set transactions. You use
lightweight transactions instead of durable transactions with eventual/tunable consistency for situations the
require nodes in the distributed system to agree on changes to data. For example, two users attempting
to create a unique user account in the same cluster could overwrite each other’s work. Using a lightweight
transaction, the nodes can agree to create only one account.
Cassandra implements lightweight transactions by extending the Paxos consensus protocol, which is
based on a quorum-based algorithm. Using this protocol, a distributed system can agree on proposed data
additions/modifications without the need for a master database or two-phase commit.
You use extensions in CQL for lightweight transactions.
You can use an IF clause in a number of CQL statements, such as INSERT, for lightweight transactions.
For example, to ensure that an insert into a new accounts table is unique for a new customer, use the IF
NOT EXISTS clause:
INSERT INTO customer_account (customerID, customer_email)

VALUES (‘LauraS’, ‘[email protected]’)
IF NOT EXISTS;
DML modifications you make using UPDATE can also make use of the IF clause by comparing one or
more columns to various values:
UPDATE customer_account
SET customer_email=’[email protected]’
IF customerID=’LauraS’;
Cassandra 2.1.1 and later support non-equal conditions for lightweight transactions. You can use <, <=,
>, >=, != and IN operators in WHERE clauses to query lightweight tables. Behind the scenes, Cassandra
is making four round trips between a node proposing a lightweight transaction and any needed replicas
84
Database internals
in the cluster to ensure proper execution so performance is affected. Consequently, reserve lightweight
transactions for those situations where they are absolutely necessary; Cassandra’s normal eventual
consistency can be used for everything else.
A SERIAL consistency level allows reading the current (and possibly uncommitted) state of data without
proposing a new addition or update. If a SERIAL read finds an uncommitted transaction in progress,
Cassandra performs a read repair as part of the commit.
Data consistency
Topics about how up-to-date and synchronized a row of data is on all replicas.
About data consistency

An introduction to how Cassandra extends eventual consistency with tunable consistency.
Consistency refers to how up-to-date and synchronized a row of Cassandra data is on all of its replicas.
Cassandra extends the concept of eventual consistency by offering tunable consistency. Tunable
consistency means for any given read or write operation, the client application decides how consistent the
requested data must be.
Even at low consistency levels, Cassandra writes to all replicas of the partition key, even replicas in other
data centers. The consistency level determines only the number of replicas that need to acknowledge
the write success to the client application. Typically, a client specifies a consistency level that is less than
the replication factor specified by the keyspace. This practice ensures that the coordinating server node
reports the write successful even if some replicas are down or otherwise not responsive to the write.
The read consistency level specifies how many replicas must respond to a read request before returning
data to the client application. Cassandra checks the specified number of replicas for data to satisfy the
read request.
The CQL documentation contains a tutorial comparing consistency levels using cqlsh tracing.
About built-in consistency repair features

Built-in repair utilities to ensure that data remains consistent across replicas.
You can use these built-in repair utilities to ensure that data remains consistent across replicas.
• Read repair
• Hinted handoff
• Anti-entropy node repair
Configuring data consistency

Consistency levels in Cassandra can be configured to manage availability versus data accuracy.
Consistency levels in Cassandra can be configured to manage availability versus data accuracy. You
can configure consistency on a cluster, data center, or individual I/O operation basis. Consistency among
participating nodes can be set globally and also controlled on a per-operation basis (for example insert or
update) using Cassandra’s drivers and client libraries.
Write consistency levels

This table describes the write consistency levels in strongest-to-weakest order.
85
Database internals
Table 6: Write Consistency Levels
Level Description Usage

ALL A write must be written to the commit log Provides the highest consistency and the
and memtable on all replica nodes in the lowest availability of any other level.
cluster for that partition.
EACH_QUORUM Strong consistency. A write must be written Used in multiple data center clusters to
to the commit log and memtable on a strictly maintain consistency at the same
quorum of replica nodes in all data centers. level in each data center. For example,
choose this level if you want a read to
fail when a data center is down and the
QUORUM cannot be reached on that data
center.
QUORUM A write must be written to the commit log Provides strong consistency if you can
and memtable on a quorum of replica tolerate some level of failure.
nodes.
LOCAL_QUORUMStrong consistency. A write must be written Used in multiple data center clusters with
to the commit log and memtable on a a rack-aware replica placement strategy
quorum of replica nodes in the same data ( NetworkTopologyStrategy) and a
center as the coordinator node. Avoids properly configured snitch. Fails when
latency of inter-data center communication. using SimpleStrategy. Use to maintain
consistency locally (within the single data
center).
ONE A write must be written to the commit log Satisfies the needs of most users because
and memtable of at least one replica node. consistency requirements are not stringent.
TWO A write must be written to the commit log Similar to ONE.
and memtable of at least two replica nodes.
THREE A write must be written to the commit log Similar to TWO.
and memtable of at least three replica
nodes.
LOCAL_ONE A write must be sent to, and successfully In a multiple data center clusters, a
acknowledged by, at least one replica node consistency level of ONE is often desirable,
in the local datacenter. but cross-DC traffic is not. LOCAL_ONE
accomplishes this. For security and quality
reasons, you can use this consistency level
in an offline datacenter to prevent automatic
connection to online nodes in other data
centers if an offline node goes down.
ANY A write must be written to at least one node. Provides low latency and a guarantee that
If all replica nodes for the given partition key a write never fails. Delivers the lowest
are down, the write can still succeed after a consistency and highest availability.
hinted handoff has been written. If all replica
nodes are down at write time, an ANY write
is not readable until the replica nodes for
that partition have recovered.
SERIAL Achieves linearizable consistency for You cannot configure this level as a normal
lightweight transactions by preventing consistency level, configured at the driver
unconditional updates. level using the consistency level field.
You configure this level using the serial
86
Database internals

consistency field as part of the native
protocol operation. See failure scenarios.
LOCAL_SERIALSame as SERIAL but confined to the data Same as SERIAL. Used for disaster
center. A write must be written conditionally recovery. See failure scenarios.
to the commit log and memtable on a
quorum of replica nodes in the same data
center.
SERIAL and LOCAL_SERIAL write failure scenarios

If one of three nodes is down, the Paxos commit fails under the following conditions:
• CQL query-configured consistency level of ALL
• Driver-configured serial consistency level of SERIAL
• Replication factor of 3
A WriteTimeout with a WriteType of CAS occurs and further reads do not see the write. If the node goes
down in the middle of the operation instead of before the operation started, the write is committed, the
value is written to the live nodes, and a WriteTimeout with a WriteType of SIMPLE occurs.
Under the same conditions, if two of the nodes are down at the beginning of the operation, the Paxos
commit fails and nothing is committed. If the two nodes go down after the Paxos proposal is accepted,
the write is committed to the remaining live nodes and written there, but a WriteTimeout with WriteType
SIMPLE is returned.
Read consistency levels

This table describes read consistency levels in strongest-to-weakest order.
Table 7: Read Consistency Levels

ALL Returns the record after all replicas have Provides the highest consistency of all
responded. The read operation will fail if a levels and the lowest availability of all
replica does not respond. levels.
EACH_QUORUM Not supported for reads. Not supported for reads.
QUORUM Returns the record after a quorum of Ensures strong consistency if you can
replicas has responded from any data tolerate some level of failure.
center.
LOCAL_QUORUMReturns the record after a quorum of Used in multiple data center clusters with
replicas in the current data center as the a rack-aware replica placement strategy
coordinator node has reported. Avoids ( NetworkTopologyStrategy) and a
latency of inter-data center communication. properly configured snitch. Fails when using
SimpleStrategy.
ONE Returns a response from the closest replica, Provides the highest availability of all the
as determined by the snitch. By default, a levels if you can tolerate a comparatively
read repair runs in the background to make high probability of stale data being read.
the other replicas consistent. The replicas contacted for reads may not
always have the most recent write.
TWO Returns the most recent data from two of Similar to ONE.
the closest replicas.
87
Database internals

THREE Returns the most recent data from three of Similar to TWO.
the closest replicas.
LOCAL_ONE Returns a response from the closest replica Same usage as described in the table about
in the local data center. write consistency levels.
SERIAL Allows reading the current (and possibly To read the latest value of a column
uncommitted) state of data without after a user has invoked a lightweight
proposing a new addition or update. If transaction to write to the column, use
a SERIAL read finds an uncommitted SERIAL. Cassandra then checks the inflight
transaction in progress, it will commit the lightweight transaction for updates and, if
transaction as part of the read. Similar to found, returns the latest data.
QUORUM.
LOCAL_SERIALSame as SERIAL, but confined to the data Used to achieve linearizable consistency for
center. Similar to LOCAL_QUORUM. lightweight transactions.
About the QUORUM levels

The QUORUM level writes to the number of nodes that make up a quorum. A quorum is calculated, and then
rounded down to a whole number, as follows:
quorum = (sum_of_replication_factors / 2) + 1
The sum of all the replication_factor settings for each data center is the
sum_of_replication_factors.
sum_of_replication_factors = datacenter1_RF + datacenter2_RF + . . . +

datacentern_RF
Examples:
• Using a replication factor of 3, a quorum is 2 nodes. The cluster can tolerate 1 replica down.
• Using a replication factor of 6, a quorum is 4. The cluster can tolerate 2 replicas down.
• In a two data center cluster where each data center has a replication factor of 3, a quorum is 4 nodes.
The cluster can tolerate 2 replica nodes down.
• In a five data center cluster where two data centers have a replication factor of 3 and three data centers
have a replication factor of 2, a quorum is 6 nodes.
The more data centers, the higher number of replica nodes need to respond for a successful operation.
If consistency is a top priority, you can ensure that a read always reflects the most recent write by using the
following formula:
(nodes_written + nodes_read) > replication_factor
For example, if your application is using the QUORUM consistency level for both write and read operations
and you are using a replication factor of 3, then this ensures that 2 nodes are always written and 2 nodes
are always read. The combination of nodes written and read (4) being greater than the replication factor (3)
ensures strong read consistency.
Similar to QUORUM, the LOCAL_QUORUM level is calculated based on the replication factor of the same data
center as the coordinator node. That is, even if the cluster has more than one data center, the quorum is
calculated only with local replica nodes.
In EACH_QUORUM, every data center in the cluster must reach a quorum based on that data center's
replication factor in order for the read or write request to succeed. That is, for every data center in the
88
Database internals
cluster a quorum of replica nodes must respond to the coordinator node in order for the read or write
request to succeed.
Configuring client consistency levels

You can use a new cqlsh command, CONSISTENCY, to set the consistency level for queries
from the current cqlsh session. The WITH CONSISTENCY clause has been removed from CQL
commands. You set the consistency level programmatically (at the driver level). For example, call
QueryBuilder.insertInto with a setConsistencyLevel argument. The consistency level defaults
to ONE for all write and read operations.
Read requests
The three types of read requests that a coordinator node can send to a replica.
There are three types of read requests that a coordinator can send to a replica:
• A direct read request
• A digest request
• A background read repair request
The coordinator node contacts one replica node with a direct read request. Then the coordinator sends
a digest request to a number of replicas determined by the consistency level specified by the client. The
digest request checks the data in the replica node to make sure it is up to date. Then the coordinator
sends a digest request to all remaining replicas. If any replica nodes have out of date data, a background
read repair request is sent. Read repair requests ensure that the requested row is made consistent on all
replicas.
For a digest request the coordinator first contacts the replicas specified by the consistency level. The
coordinator sends these requests to the replicas that are currently responding the fastest. The nodes
contacted respond with a digest of the requested data; if multiple nodes are contacted, the rows from each
replica are compared in memory to see if they are consistent. If they are not, then the replica that has the
most recent data (based on the timestamp) is used by the coordinator to forward the result back to the
client.
To ensure that all replicas have the most recent version of frequently-read data, the coordinator also
contacts and compares the data from all the remaining replicas that own the row in the background. If the
replicas are inconsistent, the coordinator issues writes to the out-of-date replicas to update the row to the
most recent values. This process is known as read repair. Read repair can be configured per table for
non-QUORUM consistency levels (using read_repair_chance), and is enabled by default.
For illustrated examples of read requests, see the examples of read consistency levels.
Rapid read protection using speculative_retry

Rapid read protection allows Cassandra to still deliver read requests when the originally selected
replica nodes are either down or taking too long to respond. If the table has been configured with the
speculative_retry property, the coordinator node for the read request will retry the request with
another replica node if the original replica node exceeds a configurable timeout value to complete the read
request.
89
Database internals
Recovering from replica node failure with rapid read protection

R1 replica node failed
1
12 2
R2
11 3
Client 10 4
coodinator node
9 resends after 5
tim eout
R3
Coordinat or node 8 6
Chosen node 7
Examples of read consistency levels

Read request examples with different consistency levels.
The following diagrams show examples of read requests using these consistency levels:
• QUORUM in a single data center
• ONE in a single data center
• QUORUM in two data centers
• LOCAL_QUORUM in two data centers
• ONE in two data centers
• LOCAL_ONE in two data centers
Rapid read protection diagram shows how the speculative retry table property affects consistency.
A single data center cluster with a consistency level of QUORUM

In a single data center cluster with a replication factor of 3, and a read consistency level of QUORUM, 2
of the 3 replicas for the given row must respond to fulfill the read request. If the contacted replicas have
different versions of the row, the replica with the most recent version will return the requested data. In the
background, the third replica is checked for consistency with the first two, and if needed, a read repair is
initiated for the out-of-date replicas.
Single data center cluster with 3 replica

nodes and consistency set to QUORUM
R1
1
12 2
R2
11 3
Client 10 4
Chosen node
R3
Read response 8 6
Read repair 7
A single data center cluster with a consistency level of ONE

In a single data center cluster with a replication factor of 3, and a read consistency level of ONE, the
closest replica for the given row is contacted to fulfill the read request. In the background a read repair is
potentially initiated, based on the read_repair_chance setting of the table, for the other replicas.
90
Database internals
Single data center cluster with 3 replica nodes and consistency set to ONE
R1
1
12 2
R2
11 3
Client 10 4
Chosen node R3
Read response 8 6
7
Read repair
A two data center cluster with a consistency level of QUORUM

In a two data center cluster with a replication factor of 3, and a read consistency of QUORUM, 4 replicas for
the given row must respond to fulfill the read request. The 4 replicas can be from any data center. In the
background, the remaining replicas are checked for consistency with the first four, and if needed, a read
repair is initiated for the out-of-date replicas.
Mult iple dat a cent er clust er wit h 3 replica

nodes and consist ency set t o QUORUM
R1
1
12 2
R2
11 3
Dat a Cent er Alpha

Client 10 4
9 5
R3
8 6
7
1
12 2
R1
11 3
R2
10 Dat a Cent er Bet a 4
9 5
Coordinat or node R3
8 6
Chosen node
7
Read response
Read repair
A two data center cluster with a consistency level of LOCAL_QUORUM

In a multiple data center cluster with a replication factor of 3, and a read consistency of LOCAL_QUORUM,
2 replicas in the same data center as the coordinator node for the given row must respond to fulfill the
read request. In the background, the remaining replicas are checked for consistency with the first 2, and if
needed, a read repair is initiated for the out-of-date replicas.
91
Database internals

nodes and consist ency set t o LOCAL_QUORUM
R1
1
12 2
R2
11 3
Dat a Cent er Alpha

Client 10 4
9 5
R3
8 6
7
1
12 2
R1
11 3
R2
9 5
Coordinat or node R3
8 6
Chosen node
7
Read response
Read repair
A two data center cluster with a consistency level of ONE

In a multiple data center cluster with a replication factor of 3, and a read consistency of ONE, the
closest replica for the given row, regardless of data center, is contacted to fulfill the read request. In the
background a read repair is potentially initiated, based on the read_repair_chance setting of the table,
for the other replicas.
Multiple data center cluster with 3 replica nodes and consistency set to ONE
R1
1
12 2
R2
11 3
Data Center Alpha

Client 10 4
9 5
R3
8 6
7
1
12 2
R1
11 3
R3
10 Data Center Beta 4
Chosen node R2
Read response 8 6
7
Read repair
92
Database internals
A two data center cluster with a consistency level of LOCAL_ONE

In a multiple data center cluster with a replication factor of 3, and a read consistency of LOCAL_ONE, the
closest replica for the given row in the same data center as the coordinator node is contacted to fulfill the
read request. In the background a read repair is potentially initiated, based on the read_repair_chance
setting of the table, for the other replicas.
Multiple data center cluster with 3 replica nodes and consistency set to LOCAL_ONE
R1
1
12 2
R2
11 3
Data Center Alpha

Client 10 4
9 5
R3
8 6
7
1
12 2
R1
11 3
R3
10 Data Center Beta 4
Chosen node
R2
Read response 8 6
7
Read repair
Write requests
How write requests work.
The coordinator sends a write request to all replicas that own the row being written. As long as all replica
nodes are up and available, they will get the write regardless of the consistency level specified by the
client. The write consistency level determines how many replica nodes must respond with a success
acknowledgment in order for the write to be considered successful. Success means that the data was
written to the commit log and the memtable as described in About writes.
For example, in a single data center 10 node cluster with a replication factor of 3, an incoming write will
go to all 3 nodes that own the requested row. If the write consistency level specified by the client is ONE,
the first node to complete the write responds back to the coordinator, which then proxies the success
message back to the client. A consistency level of ONE means that it is possible that 2 of the 3 replicas
could miss the write if they happened to be down at the time the request was made. If a replica misses
a write, Cassandra will make the row consistent later using one of its built-in repair mechanisms: hinted
handoff, read repair, or anti-entropy node repair.
That node forwards the write to all replicas of that row. It responds back to the client once it receives a
write acknowledgment from the number of nodes specified by the consistency level.
93
Database internals
Single data center cluster with 3 replica nodes and consistency set to ONE
R1
1
12 2
R2
11 3
Client 10 4
9 5
Coordinat or node
R3
Chosen node 8 6
Writ e response 7
Multiple data center write requests

How write requests work when using multiple data centers.
In multiple data center deployments, Cassandra optimizes write performance by choosing one coordinator
node. The coordinator node contacted by the client application forwards the write request to each replica
node in each all the data centers.
If using a consistency level of LOCAL_ONE or LOCAL_QUORUM, only the nodes in the same data center
as the coordinator node must respond to the client request in order for the request to succeed. This way,
geographical latency does not impact client request response times.

nodes and consist ency set t o QUORUM
R1
1
12 2
R2
11 3
Dat a Cent er Alpha

Client 10 4
9 5
R3
8 6
7
1
12 2
R1
11 3
R2
9 5
R3
7
Nodes t hat m ake up a quorum
Writ e response
94
Configuration
Configuration
Configuration topics.
The cassandra.yaml configuration file

The cassandra.yaml file is the main configuration file for Cassandra.
The cassandra.yaml file is the main configuration file for Cassandra.
Important: After changing properties in the cassandra.yaml file, you must restart the node for
the changes to take effect. It is located in the following directories:
• Cassandra package installations: /etc/cassandra
• Cassandra tarball installations: install_location/conf
• DataStax Enterprise package installations: /etc/dse/cassandra
• DataStax Enterprise tarball installations: install_location/resources/cassandra/conf
The configuration properties are grouped into the following sections:
• Quick start
The minimal properties needed for configuring a cluster.
• Commonly used
Properties most frequently used when configuring Cassandra.
• Performance tuning
Tuning performance and system resource utilization, including commit log, compaction, memory, disk I/
O, CPU, reads, and writes.
• Advanced
Properties for advanced users or properties that are less commonly used.
• Security
Server and client security settings.
note
Note: Values with indicate default values that are defined internally, missing, commented out,
or implementation depends on other properties in the cassandra.yaml file. Additionally, some
commented out values may not match the actual default value; these values are recommended
when changing from the default.
Quick start properties

The minimal properties needed for configuring a cluster.
Related information: Initializing a multiple node cluster (single data center) and Initializing a multiple node
cluster (multiple data centers).
cluster_name
(Default: Test Cluster) The name of the cluster. This setting prevents nodes in one logical cluster from joining
another. All nodes in a cluster must have the same value.
listen_address
(Default: localhost) The IP address or hostname that Cassandra binds to for connecting to other Cassandra
nodes. Set this parameter or listen_interface, not both. You must change the default setting for multiple
nodes to communicate:
95
Configuration
• Generally set to empty. If the node is properly configured (host name, name resolution, and so on),
Cassandra uses InetAddress.getLocalHost() to get the local address from the system.
• For a single node cluster, you can use the default setting (localhost).
• If Cassandra can't find the correct address, you must specify the IP address or host name.
• Never specify 0.0.0.0; it is always wrong.
listen_interface
note
(Default: eth0) The interface that Cassandra binds to for connecting to other Cassandra nodes. Interfaces
must correspond to a single address, IP aliasing is not supported. See listen_address.
Default directories
If you have changed any of the default directories during installation, make sure you have root access and
set these properties:
commitlog_directory
The directory where the commit log is stored. Default locations:
• Package installations: /var/lib/cassandra/commitlog
• Tarball installations: install_location/data/commitlog
For optimal write performance, place the commit log be on a separate disk partition, or (ideally) a separate
physical device from the data file directories. Because the commit log is append only, an HDD for is
acceptable for this purpose.
data_file_directories
The directory location where table data (SSTables) is stored. Cassandra distributes data evenly across the
location, subject to the granularity of the configured compaction strategy. Default locations:
• Package installations: /var/lib/cassandra/data
As a production best practice, use RAID 0 and SSDs.
saved_caches_directory
The directory location where table key and row caches are stored. Default location:
• Package installations: /var/lib/cassandra/saved_caches
• Tarball installations: install_location/data/saved_caches
Commonly used properties

Properties most frequently used when configuring Cassandra.
Before starting a node for the first time, you should carefully evaluate your requirements.
Common initialization properties
Note: Be sure to set the properties in the Quick start section as well.
commit_failure_policy
(Default: stop) Policy for commit disk failures:
• die
Shut down gossip and Thrift and kill the JVM, so the node can be replaced.
• stop
Shut down gossip and Thrift, leaving the node effectively dead, but can be inspected using JMX.
• stop_commit
Shut down the commit log, letting writes collect but continuing to service reads (as in pre-2.0.5
Cassandra).
96
Configuration
• ignore
Ignore fatal errors and let the batches fail.
disk_failure_policy
(Default: stop) Sets how Cassandra responds to disk failure. Recommend settings are stop or best_effort.
• die
Shut down gossip and Thrift and kill the JVM for any file system errors or single SSTable errors, so the
node can be replaced.
• stop_paranoid
Shut down gossip and Thrift even for single SSTable errors.
• stop
Shut down gossip and Thrift, leaving the node effectively dead, but available for inspection using JMX.
• best_effort
Stop using the failed disk and respond to requests based on the remaining available SSTables. This
means you will see obsolete data at consistency level of ONE.
• ignore
Ignores fatal errors and lets the requests fail; all file system errors are logged but otherwise ignored.
Cassandra acts as in versions prior to 1.2.
Related information: Handling Disk Failures In Cassandra 1.2 blog and Recovering using JBOD.
endpoint_snitch
(Default: org.apache.cassandra.locator.SimpleSnitch) Set to a class that implements the
IEndpointSnitch. Cassandra uses snitches for locating nodes and routing requests.
• SimpleSnitch
Use for single-data center deployments or single-zone in public clouds. Does not recognize data center
or rack information. It treats strategy order as proximity, which can improve cache locality when disabling
read repair.
• GossipingPropertyFileSnitch
Recommended for production. The rack and data center for the local node are defined in the
cassandra-rackdc.properties file and propagated to other nodes via gossip. To allow migration
from the PropertyFileSnitch, it uses the cassandra-topology.properties file if it is present.
• PropertyFileSnitch
Determines proximity by rack and data center, which are explicitly configured in the cassandra-
topology.properties file.
• Ec2Snitch
For EC2 deployments in a single region. Loads region and availability zone information from the EC2
API. The region is treated as the data center and the availability zone as the rack. Uses only private IPs.
Subsequently it does not work across multiple regions.
• Ec2MultiRegionSnitch
Uses public IPs as the broadcast_address to allow cross-region connectivity. This means you must also
set seed addresses to the public IP and open the storage_port or ssl_storage_port on the public IP
firewall. For intra-region traffic, Cassandra switches to the private IP after establishing a connection.
• RackInferringSnitch:
Proximity is determined by rack and data center, which are assumed to correspond to the 3rd and 2nd
octet of each node's IP address, respectively. This snitch is best used as an example for writing a custom
snitch class (unless this happens to match your deployment conventions).
Related information: Snitches
97
Configuration
rpc_address
(Default: localhost) The listen address for client connections (Thrift RPC service and native transport).Valid
values are:
• unset:
Resolves the address using the hostname configuration of the node. If left unset, the hostname must
resolve to the IP address of this node using /etc/hostname, /etc/hosts, or DNS.
• 0.0.0.0:
Listens on all configured interfaces, but you must set the broadcast_rpc_address to a value other than
0.0.0.0.
• IP address
• hostname
Related information: Network
rpc_interface
note
(Default: eth1) The listen address for client connections. Interfaces must correspond to a single address,
IP aliasing is not supported. See rpc_address.
seed_provider
The addresses of hosts deemed contact points. Cassandra nodes use the -seeds list to find each other and
learn the topology of the ring.
• class_name (Default: org.apache.cassandra.locator.SimpleSeedProvider)
The class within Cassandra that handles the seed logic. It can be customized, but this is typically not
required.
• - seeds (Default: 127.0.0.1)
A comma-delimited list of IP addresses used by gossip for bootstrapping new nodes joining a cluster.
When running multiple nodes, you must change the list from the default value. In multiple data-center
clusters, the seed list should include at least one node from each data center (replication group). More
than a single seed node per data center is recommended for fault tolerance. Otherwise, gossip has to
communicate with another data center when bootstrapping a node. Making every node a seed node is not
recommended because of increased maintenance and reduced gossip performance. Gossip optimization
is not critical, but it is recommended to use a small seed list (approximately three nodes per data center).
Common compaction settings
compaction_throughput_mb_per_sec
(Default: 16) Throttles compaction to the specified total throughput across the entire system. The faster you
insert data, the faster you need to compact in order to keep the SSTable count down. The recommended
value is 16 to 32 times the rate of write throughput (in MB/second). Setting the value to 0 disables compaction
throttling.
Related information: Configuring compaction
Common memtable settings
memtable_total_space_in_mb
note
(Default: 1/4 of heap) Specifies the total memory used for all memtables on a node. This replaces the
per-table storage settings memtable_operations_in_millions and memtable_throughput_in_mb.
Related information: Tuning the Java heap
Common disk settings
concurrent_reads
98
Configuration
note
(Default: 32) For workloads with more data than can fit in memory, the bottleneck is reads fetching data
from disk. Setting to (16 × number_of_drives) allows operations to queue low enough in the stack so that
the OS and drives can reorder them. The default setting applies to both logical volume managed (LVM)
and RAID drives.
concurrent_writes
note
(Default: 32) Writes in Cassandra are rarely I/O bound, so the ideal number of concurrent writes depends
on the number of CPU cores in your system. The recommended value is 8 × number_of_cpu_cores.
concurrent_counter_writes
note
(Default: 32) Counter writes read the current values before incrementing and writing them back. The
recommended value is (16 × number_of_drives) .
Common automatic backup settings
incremental_backups
(Default: false) Backs up data updated since the last snapshot was taken. When enabled, Cassandra creates
a hard link to each SSTable flushed or streamed locally in a backups/ subdirectory of the keyspace data.
Removing these links is the operator's responsibility.
Related information: Enabling incremental backups
snapshot_before_compaction
(Default: false) Enable or disable taking a snapshot before each compaction. This option is useful to back
up data when there is a data format change. Be careful using this option because Cassandra does not clean
up older snapshots automatically.
Common fault detection setting
phi_convict_threshold
note
(Default: 8) Adjusts the sensitivity of the failure detector on an exponential scale. Generally this setting
never needs adjusting.
Related information: Failure detection and recovery
Performance tuning properties

Tuning performance and system resource utilization, including commit log, compaction, memory, disk I/O,
CPU, reads, and writes.
Commit log settings
commitlog_sync
(Default: periodic) The method that Cassandra uses to acknowledge writes in milliseconds:
• periodic: (Default: 10000 milliseconds [10 seconds])
Used with commitlog_sync_period_in_ms to control how often the commit log is synchronized to disk.
Periodic syncs are acknowledged immediately.
note
• batch: (Default: disabled)
Used with commitlog_sync_batch_window_in_ms (Default: 2 ms) to control how long Cassandra waits
for other writes before performing a sync. When using this method, writes are not acknowledged until
fsynced to disk.
Related information: Durability
commitlog_segment_size_in_mb
(Default: 32MB) Sets the size of the individual commitlog file segments. A commitlog segment may be
archived, deleted, or recycled after all its data has been flushed to SSTables. This amount of data can
99
Configuration
potentially include commitlog segments from every table in the system. The default size is usually suitable
for most commitlog archiving, but if you want a finer granularity, 8 or 16 MB is reasonable.
Related information: Commit log archive configuration
commitlog_total_space_in_mb
note
(Default: 32MB for 32-bit JVMs, 8192MB for 64-bit JVMs) Total space used for commitlogs. If the used
space goes above this value, Cassandra rounds up to the next nearest segment multiple and flushes
memtables to disk for the oldest commitlog segments, removing those log segments. This reduces the
amount of data to replay on start-up, and prevents infrequently-updated tables from indefinitely keeping
commitlog segments. A small total commitlog space tends to cause more flush activity on less-active tables.
Related information: Configuring memtable throughput
Compaction settings
compaction_preheat_key_cache
(Default: true) When set to true, cached row keys are tracked during compaction, and re-cached to their
new positions in the compacted SSTable. If you have extremely large key caches for tables, set the value
to false; see Global row and key caches properties.
concurrent_compactors
(Default: Smaller of number of disks or number of cores, with a minimum of 2 and a maximum of 8
note
per CPU core) Sets the number of concurrent compaction processes allowed to run simultaneously
on a node, not including validation compactions for anti-entropy repair. Simultaneous compactions
help preserve read performance in a mixed read-write workload by mitigating the tendency of small
SSTables to accumulate during a single long-running compaction. If your data directories are backed by
SSD, increase this value to the number of cores. If compaction running too slowly or too fast, adjust
compaction_throughput_mb_per_sec first.
in_memory_compaction_limit_in_mb
(Default: 64MB) Size limit for rows being compacted in memory. Larger rows spill to disk and use a
slower two-pass compaction process. When this occurs, a message is logged specifying the row key. The
recommended value is 5 to 10 percent of the available Java heap size.
preheat_kernel_page_cache
(Default: false) Enable or disable kernel page cache preheating from contents of the key cache after
compaction. When enabled it preheats only first page (4KB) of each row to optimize for sequential access.
It can be harmful for fat rows, see CASSANDRA-4937 for more details.
sstable_preemptive_open_interval_in_mb
(Default: 50MB) When compacting, the replacement opens SSTables before they are completely written
and uses in place of the prior SSTables for any range previously written. This setting helps to smoothly
transfer reads between the SSTables by reducing page cache churn and keeps hot rows hot.
Memtable settings
memtable_allocation_type
(Default: heap_buffers) Specify the way Cassandra allocates and manages memtable memory. See Off-
heap memtables in Cassandra 2.1. Options are:
• heap_buffers
On heap NIO (non-blocking I/O) buffers.
• offheap_buffers
Off heap (direct) NIO buffers.
• offheap_objects
Native memory, eliminating NIO buffer heap overhead.
100
Configuration
memtable_cleanup_threshold
note
(Default: 0.11 1/(memtable_flush_writers + 1)) Ratio of occupied non-flushing memtable size to total
permitted size for triggering a flush of the largest memtable. Larger values mean larger flushes and less
compaction, but also less concurrent flush activity, which can make it difficult to keep your disks saturated
under heavy write load.
file_cache_size_in_mb
(Default: Smaller of 1/4 heap or 512) Total memory to use for SSTable-reading buffers.
memtable_flush_queue_size
(Default: 4) The number of full memtables to allow pending flush (memtables waiting for a write thread). At
a minimum, set to the maximum number of indexes created on a single table.
Related information: Flushing data from the memtable
memtable_flush_writers
note
(Default: Smaller of number of disks or number of cores with a minimum of 2 and a maximum of 8) Sets
the number of memtable flush writer threads. These threads are blocked by disk I/O, and each one holds
a memtable in memory while blocked. If your data directories are backed by SSD, increase this setting to
the number of cores.
memtable_heap_space_in_mb
note
(Default: 1/4 heap) Total permitted memory to use for memtables. Triggers a flush based on
memtable_cleanup_threshold. Cassandra stops accepting writes when the limit is exceeded until a flush
completes. If unset, sets to default.
memtable_offheap_space_in_mb
note
(Default: 1/4 heap) See memtable_heap_space_in_mb.
Cache and index settings
column_index_size_in_kb
(Default: 64) Granularity of the index of rows within a partition. For huge rows, decrease this setting to
improve seek time. If you use key cache, be careful not to make this setting too large because key cache
will be overwhelmed. If you're unsure of the size of the rows, it's best to use the default setting.
index_summary_capacity_in_mb
note
(Default: 5% of the heap size [empty]) Fixed memory pool size in MB for SSTable index summaries. If
the memory usage of all index summaries exceeds this limit, any SSTables with low read rates shrink their
index summaries to meet this limit. This is a best-effort process. In extreme conditions, Cassandra may
need to use more than this amount of memory.
index_summary_resize_interval_in_minutes
(Default: 60 minutes) How frequently index summaries should be re-sampled. This is done periodically to
redistribute memory from the fixed-size pool to SSTables proportional their recent read rates. To disable,
set to -1. This leaves existing index summaries at their current sampling level.
reduce_cache_capacity_to
(Default: 0.6) Sets the size percentage to which maximum cache capacity is reduced when Java heap usage
reaches the threshold defined by reduce_cache_sizes_at.
reduce_cache_sizes_at
(Default: 0.85) When Java heap usage (after a full concurrent mark sweep (CMS) garbage collection)
exceeds this percentage, Cassandra reduces the cache capacity to the fraction of the current size as
specified by reduce_cache_capacity_to. To disable, set the value to 1.0.
Disks settings
stream_throughput_outbound_megabits_per_sec
note
(Default: 200 seconds) Throttles all outbound streaming file transfers on a node to the specified
throughput. Cassandra does mostly sequential I/O when streaming data during bootstrap or repair, which
can lead to saturating the network connection and degrading client (RPC) performance.
101
Configuration
inter_dc_stream_throughput_outbound_megabits_per_sec
note
(Default: unset) Throttles all streaming file transfer between the data centers. This setting allows throttles
streaming throughput betweens data centers in addition to throttling all network stream traffic as configured
with stream_throughput_outbound_megabits_per_sec.
trickle_fsync
(Default: false) When doing sequential writing, enabling this option tells fsync to force the operating system
to flush the dirty buffers at a set interval trickle_fsync_interval_in_kb. Enable this parameter to avoid sudden
dirty buffer flushing from impacting read latencies. Recommended to use on SSDs, but not on HDDs.
trickle_fsync_interval_in_kb
(Default: 10240). Sets the size of the fsync in kilobytes.
Advanced properties
Properties for advanced users or properties that are less commonly used.
Advanced initialization properties
auto_bootstrap
(Default: true) This setting has been removed from default configuration. It makes new (non-seed) nodes
automatically migrate the right data to themselves. When initializing a fresh cluster without data, add
auto_bootstrap: false.
batch_size_warn_threshold_in_kb
(Default: 5KB per batch) Log WARN on any batch size exceeding this value in kilobytes. Caution should be
taken on increasing the size of this threshold as it can lead to node instability.
broadcast_address
note
(Default: listen_address) The IP address a node tells other nodes in the cluster to contact it by. It allows
public and private address to be different. For example, use the broadcast_address parameter in topologies
where not all nodes have access to other nodes by their private IP addresses.
If your Cassandra cluster is deployed across multiple Amazon EC2 regions and you use the
Ec2MultiRegionSnitch, set the broadcast_address to public IP address of the node and the
listen_address to the private IP. See Ec2MultiRegionSnitch.
initial_token
(Default: disabled) Used in the single-node-per-token architecture, where a node owns exactly one
contiguous range in the ring space. Setting this property overrides num_tokens.
If you not using vnodes or have num_tokens set it to 1 or unspecified (#num_tokens), you should always
specify this parameter when setting up a production cluster for the first time and when adding capacity. For
more information, see this parameter in the Cassandra 1.1 Node and Cluster Configuration documentation.
This parameter can be used with num_tokens (vnodes ) in special cases such as Restoring from a snapshot.
num_tokens
note
(Default: 256) Defines the number of tokens randomly assigned to this node on the ring when using
virtual nodes (vnodes). The more tokens, relative to other nodes, the larger the proportion of data that
the node stores. Generally all nodes should have the same number of tokens assuming equal hardware
capability. The recommended value is 256. If unspecified (#num_tokens), Cassandra uses 1 (equivalent
to #num_tokens : 1) for legacy compatibility and uses the initial_token setting.
If not using vnodes, comment #num_tokens : 256 or set num_tokens : 1 and use initial_token. If
you already have an existing cluster with one token per node and wish to migrate to vnodes, see Enabling
virtual nodes on an existing production cluster.
102
Configuration
Note: If using DataStax Enterprise, the default setting of this property depends on the type of node
and type of install.
partitioner
(Default: org.apache.cassandra.dht.Murmur3Partitioner) Distributes rows (by partition key)
across all nodes in the cluster. Any IPartitioner may be used, including your own as long as it is in the
class path. For new clusters use the default partitioner.
Cassandra provides the following partitioners for backwards compatibility:
• RandomPartitioner
• ByteOrderedPartitioner
• OrderPreservingPartitioner (deprecated)
Related information: Partitioners
storage_port
(Default: 7000) The port for inter-node communication.
Advanced automatic backup setting
auto_snapshot
(Default: true) Enable or disable whether a snapshot is taken of the data before keyspace truncation or
dropping of tables. To prevent data loss, using the default setting is strongly advised. If you set to false,
you will lose data on truncation or drop.
Key caches and global row properties
When creating or modifying tables, you enable or disable the key cache (partition key cache) or row cache
for that table by setting the caching parameter. Other row and key cache tuning and configuration options
are set at the global (node) level. Cassandra uses these settings to automatically distribute memory for
each table on the node based on the overall workload and specific table usage. You can also configure the
save periods for these caches globally.
Related information: Configuring caches
key_cache_keys_to_save
note
(Default: disabled - all keys are saved) Number of keys from the key cache to save.
key_cache_save_period
(Default: 14400 seconds [4 hours]) Duration in seconds that keys are saved in cache. Caches are saved
to saved_caches_directory. Saved caches greatly improve cold-start speeds and has relatively little effect
on I/O.
key_cache_size_in_mb
(Default: empty) A global cache setting for tables. It is the maximum size of the key cache in memory. When
no value is set, the cache is set to the smaller of 5% of the available heap, or 100MB. To disable set to 0.
Related information: setcachecapacity.
row_cache_keys_to_save
note
(Default: disabled - all keys are saved) Number of keys from the row cache to save.
row_cache_size_in_mb
(Default: 0- disabled) Maximum size of the row cache in memory. Row cache can save more time than
key_cache_size_in_mb, but is space-intensive because it contains the entire row. Use the row cache only
for hot rows or static rows. If you reduce the size, you may not get you hottest keys loaded on start up.
row_cache_save_period
(Default: 0- disabled) Duration in seconds that rows are saved in cache. Caches are saved to
saved_caches_directory. This setting has limited use as described in row_cache_size_in_mb.
memory_allocator
103
Configuration
(Default: NativeAllocator) The off-heap memory allocator. In addition to caches, this property affects storage
engine meta data. Supported values:
• NativeAllocator
• JEMallocAllocator
Experiments show that jemalloc saves some memory compared to the native allocator because it is
more fragmentation resistant. To use, install jemalloc as a library and modify cassandra-env.sh
(instructions in file).
Counter caches properties
Counter cache helps to reduce counter locks' contention for hot counter cells. In case of RF = 1 a counter
cache hit will cause Cassandra to skip the read before write entirely. With RF > 1 a counter cache hit will
still help to reduce the duration of the lock hold, helping with hot counter cell updates, but will not allow
skipping the read entirely. Only the local (clock, count) tuple of a counter cell is kept in memory, not the
whole counter, so it's relatively cheap.
Note: Reducing the size counter cache may result in not getting the hottest keys loaded on start-
up.
counter_cache_size_in_mb
note
(Default value: empty) When no value is specified a minimum of 2.5% of Heap or 50MB. If you perform
counter deletes and rely on low gc_grace_seconds, you should disable the counter cache. To disable, set
to 0.
counter_cache_save_period
(Default: 7200 seconds [2 hours]) Duration after which Cassandra should save the counter cache (keys
only). Caches are saved to saved_caches_directory.
counter_cache_keys_to_save
note
(Default value: disabled) Number of keys from the counter cache to save. When disabled all keys are
saved.
Tombstone settings
When executing a scan, within or across a partition, tombstones must be kept in memory to allow returning
them to the coordinator. The coordinator uses them to ensure other replicas know about the deleted rows.
Workloads that generate numerous tombstones may cause performance problems and exhaust the server
heap. See Cassandra anti-patterns: Queues and queue-like datasets. Adjust these thresholds only if you
understand the impact and want to scan more tombstones. Additionally, you can adjust these thresholds at
runtime using the StorageServiceMBean.
Related information: Cassandra anti-patterns: Queues and queue-like datasets
tombstone_warn_threshold
(Default: 1000) The maximum number of tombstones a query can scan before warning.
tombstone_failure_threshold
(Default: 100000) The maximum number of tombstones a query can scan before aborting.
Network timeout settings
range_request_timeout_in_ms
(Default: 10000 milliseconds) The time that the coordinator waits for sequential or index scans to complete.
read_request_timeout_in_ms
(Default: 5000 milliseconds) The time that the coordinator waits for read operations to complete.
counter_write_request_timeout_in_ms
(Default: 5000 milliseconds) The time that the coordinator waits for counter writes to complete.
cas_contention_timeout_in_ms
104
Configuration
(Default: 1000 milliseconds) The time that the coordinator continues to retry a CAS (compare and set)
operation that contends with other proposals for the same row.
truncate_request_timeout_in_ms
(Default: 60000 milliseconds) The time that the coordinator waits for truncates (remove all data from a
table) to complete. The long default value allows for a snapshot to be taken before removing the data. If
auto_snapshot is disabled (not recommended), you can reduce this time.
write_request_timeout_in_ms
(Default: 2000 milliseconds) The time that the coordinator waits for write operations to complete.
Related information: About hinted handoff writes
request_timeout_in_ms
(Default: 10000 milliseconds) The default time for other, miscellaneous operations.
Inter-node settings
cross_node_timeout
(Default: false) Enable or disable operation timeout information exchange between nodes (to accurately
measure request timeouts). If disabled Cassandra assumes the request are forwarded to the replica instantly
by the coordinator, which means that under overload conditions extra time is required for processing already-
timed-out requests..
Caution: Before enabling this property make sure NTP (network time protocol) is installed and the
times are synchronized between the nodes.
internode_send_buff_size_in_bytes
note
(Default: N/A) Sets the sending socket buffer size in bytes for inter-node calls.
When setting this parameter and internode_recv_buff_size_in_bytes, the buffer size is limited by
net.core.wmem_max. When unset, buffer size is defined by net.ipv4.tcp_wmem. See man tcp and:
• /proc/sys/net/core/wmem_max
• /proc/sys/net/core/rmem_max
• /proc/sys/net/ipv4/tcp_wmem
• /proc/sys/net/ipv4/tcp_wmem
internode_recv_buff_size_in_bytes
note
(Default: N/A) Sets the receiving socket buffer size in bytes for inter-node calls.
internode_compression
(Default: all) Controls whether traffic between nodes is compressed. The valid values are:
• all
All traffic is compressed.
• dc
Traffic between data centers is compressed.
• none
No compression.
inter_dc_tcp_nodelay
(Default: false) Enable or disable tcp_nodelay for inter-data center communication. When disabled larger,
but fewer, network packets are sent. This reduces overhead from the TCP protocol itself. However, if cross
data-center responses are blocked, it will increase latency.
streaming_socket_timeout_in_ms
105
Configuration
note
(Default: 0 - never timeout streams) Enable or disable socket timeout for streaming operations. When
a timeout occurs during streaming, streaming is retried from the start of the current file. Avoid setting this
value too low, as it can result in a significant amount of data re-streaming.
Native transport (CQL Binary Protocol)
start_native_transport
(Default: true) Enable or disable the native transport server. Uses the same address as the rpc_address,
but the port is different from the rpc_port. See native_transport_port.
native_transport_port
(Default: 9042) Port on which the CQL native transport listens for clients.
native_transport_max_threads
note
(Default: 128) The maximum number of thread handling requests. Similar to rpc_max_threads and differs
as follows:
• Default is different (128 versus unlimited).
• No corresponding native_transport_min_threads.
• Idle threads are stopped after 30 seconds.
native_transport_max_frame_size_in_mb
(Default: 256MB) The maximum size of allowed frame. Frame (requests) larger than this are rejected as
invalid.
RPC (remote procedure call) settings
Settings for configuring and tuning client connections.
broadcast_rpc_address
note
(Default: unset) RPC address to broadcast to drivers and other Cassandra nodes. This cannot be set to
0.0.0.0. If blank, it is set to the value of the rpc_address or rpc_interface. If rpc_address or rpc_interfaceis
set to 0.0.0.0, this property must be set.
rpc_port
(Default: 9160) Thrift port for client connections.
start_rpc
(Default: true) Starts the Thrift RPC server.
rpc_keepalive
(Default: true) Enable or disable keepalive on client connections (RPC or native).
rpc_max_threads
note
(Default: unlimited) Regardless of your choice of RPC server (rpc_server_type), the number of maximum
requests in the RPC thread pool dictates how many concurrent requests are possible. However, if you are
using the parameter sync in the rpc_server_type, it also dictates the number of clients that can be connected.
For a large number of client connections, this could cause excessive memory usage for the thread stack.
Connection pooling on the client side is highly recommended. Setting a maximum thread pool size acts as a
safeguard against misbehaved clients. If the maximum is reached, Cassandra blocks additional connections
until a client disconnects.
rpc_min_threads
note
(Default: 16) Sets the minimum thread pool size for remote procedure calls.
rpc_recv_buff_size_in_bytes
note
(Default: N/A) Sets the receiving socket buffer size for remote procedure calls.
rpc_send_buff_size_in_bytes
note
(Default: N/A) Sets the sending socket buffer size in bytes for remote procedure calls.
rpc_server_type
106
Configuration
(Default: sync) Cassandra provides three options for the RPC server. On Windows, sync is about 30%
slower than hsha. On Linux, sync and hsha performance is about the same, but hsha uses less memory.
• sync: (Default One thread per Thrift connection.)
For a very large number of clients, memory is the limiting factor. On a 64-bit JVM, 180KB is the minimum
stack size per thread and corresponds to your use of virtual memory. Physical memory may be limited
depending on use of stack space.
• hsha:
Half synchronous, half asynchronous. All Thrift clients are handled asynchronously using a small number
of threads that does not vary with the number of clients and thus scales well to many clients. The RPC
requests are synchronous (one thread per active request).
Note: When selecting this option, you must change the default value (unlimited) of
rpc_max_threads.
• Your own RPC server
You must provide a fully-qualified class name of an o.a.c.t.TServerFactory that can create a
server instance.
Advanced fault detection settings
Settings to handle poorly performing or failing nodes.
dynamic_snitch_badness_threshold
(Default: 0.1) Sets the performance threshold for dynamically routing client requests away from a poorly
performing node. Specifically, it controls how much worse a poorly performing node has to be before the
dynamic snitch prefers other replicas over it. A value of 0.2 means Cassandra continues to prefer the static
snitch values until the node response time is 20% worse than the best performing node. Until the threshold
is reached, incoming requests are statically routed to the closest replica (as determined by the snitch). If
the value of this parameter is greater than zero and read_repair_chance is less than 1.0, cache capacity
is maximized across the nodes.
dynamic_snitch_reset_interval_in_ms
(Default: 600000 milliseconds) Time interval to reset all node scores, which allows a bad node to recover.
dynamic_snitch_update_interval_in_ms
(Default: 100 milliseconds) The time interval for how often the snitch calculates node scores. Because score
calculation is CPU intensive, be careful when reducing this interval.
hinted_handoff_enabled
(Default: true) Enable or disable hinted handoff. To enable per data center, add data center list. For
example: hinted_handoff_enabled: DC1,DC2. A hint indicates that the write needs to be replayed to
an unavailable node. Where Cassandra writes the hint depends on the version:
• Prior to 1.0
Writes to a live replica node.
• 1.0 and later
Writes to the coordinator node.
hinted_handoff_throttle_in_kb
(Default: 1024) Maximum throttle per delivery thread in kilobytes per second. This rate reduces proportionally
to the number of nodes in the cluster. For example, if there are two nodes in the cluster, each delivery thread
will use the maximum rate. If there are three, each node will throttle to half of the maximum, since the two
nodes are expected to deliver hints simultaneously.
max_hint_window_in_ms
107
Configuration
(Default: 10800000 milliseconds [3 hours]) Maximum amount of time that hints are generates hints for an
unresponsive node. After this interval, new hints are no longer generated until the node is back up and
responsive. If the node goes down again, a new interval begins. This setting can prevent a sudden demand
for resources when a node is brought back online and the rest of the cluster attempts to replay a large
volume of hinted writes.
Related information: Failure detection and recovery
max_hints_delivery_threads
(Default: 2) Number of threads with which to deliver hints. In multiple data-center deployments, consider
increasing this number because cross data-center handoff is generally slower.
batchlog_replay_throttle_in_kb
(Default: 1024KB per second) Total maximum throttle. Throttling is reduced proportionally to the number
of nodes in the cluster.
Request scheduler properties
Settings to handle incoming client requests according to a defined policy. If you need to use these
properties, your nodes are overloaded and dropping requests. It is recommended that you add more nodes
and not try to prioritize requests.
request_scheduler
(Default: org.apache.cassandra.scheduler.NoScheduler) Defines a scheduler to handle incoming
client requests according to a defined policy. This scheduler is useful for throttling client requests in single
clusters containing multiple keyspaces. This parameter is specifically for requests from the client and does
not affect inter-node communication. Valid values are:
• org.apache.cassandra.scheduler.NoScheduler
No scheduling takes place.
• org.apache.cassandra.scheduler.RoundRobinScheduler
Round robin of client requests to a node with a separate queue for each request_scheduler_id property.
• A Java class that implements the RequestScheduler interface.
request_scheduler_id
note
(Default: keyspace) An identifier on which to perform request scheduling. Currently the only valid value
is keyspace. See weights.
request_scheduler_options
(Default: disabled) Contains a list of properties that define configuration options for request_scheduler:
• throttle_limit
The number of in-flight requests per client. Requests beyond this limit are queued up until running
requests complete. Recommended value is ((concurrent_reads + concurrent_writes) × 2).
note
• default_weight: (Default: 1)
How many requests are handled during each turn of the RoundRobin.
• weights: (Default: Keyspace: 1)
Takes a list of keyspaces. It sets how many requests are handled during each turn of the RoundRobin,
based on the request_scheduler_id.
Thrift interface properties
Legacy API for older clients. CQL is a simpler and better API for Cassandra.
thrift_framed_transport_size_in_mb
(Default: 15) Frame size (maximum field length) for Thrift. The frame is the row or part of the row that the
application is inserting.
thrift_max_message_length_in_mb
108
Configuration
(Default: 16) The maximum length of a Thrift message in megabytes, including all fields and internal Thrift
overhead (1 byte of overhead for each frame). Message length is usually used in conjunction with batches.
A frame length greater than or equal to 24 accommodates a batch with four inserts, each of which is 24
bytes. The required message length is greater than or equal to 24+24+24+24+4 (number of frames).
Security properties
Server and client security settings.
authenticator
(Default: AllowAllAuthenticator) The authentication backend. It implements IAuthenticator for
identifying users. The available authenticators are:
• AllowAllAuthenticator:
Disables authentication; no checks are performed.
• PasswordAuthenticator
Authenticates users with user names and hashed passwords stored in the system_auth.credentials table.
If you use the default, 1, and the node with the lone replica goes down, you will not be able to log into
the cluster because the system_auth keyspace was not replicated.
Related information: Internal authentication
internode_authenticator
note
(Default: enabled) Internode authentication backend. It implements
org.apache.cassandra.auth.AllowAllInternodeAuthenticator to allows or disallow
connections from peer nodes.
authorizer
(Default: AllowAllAuthorizer) The authorization backend. It implements IAuthenticator to limit access
and provide permissions. The available authorizers are:
• AllowAllAuthorizer
Disables authorization; allows any action to any user.
• CassandraAuthorizer
Stores permissions in system_auth.permissions table. If you use the default, 1, and the node with the
lone replica goes down, you will not be able to log into the cluster because the system_auth keyspace
was not replicated.
Related information: Object permissions
permissions_validity_in_ms
(Default: 2000) How long permissions in cache remain valid. Depending on the authorizer, such as
CassandraAuthorizer, fetching permissions can be resource intensive. This setting disabled when set
to 0 or when AllowAllAuthorizer is set.
Related information: Object permissions
permissions_update_interval_in_ms
note
(Default: same value as permissions_validity_in_ms) Refresh interval for permissions cache (if enabled).
After this interval, cache entries become eligible for refresh. On next access, an async reload is scheduled
and the old value is returned until it completes. If permissions_validity_in_ms , then this property must
benon-zero.
server_encryption_options
Enable or disable inter-node encryption. You must also generate keys and provide the appropriate key and
trust store locations and passwords. No custom encryption options are currently enabled. The available
options are:
109
Configuration
• internode_encryption: (Default: none) Enable or disable encryption of inter-node communication

using the TLS_RSA_WITH_AES_128_CBC_SHA cipher suite for authentication, key exchange, and
encryption of data transfers. Use the DHE/ECDHE ciphers if running in (Federal Information Processing
Standard) FIPS 140 compliant mode. The available inter-node options are:
• all
Encrypt all inter-node communications.
• none
No encryption.
• dc
Encrypt the traffic between the data centers (server only).
• rack
Encrypt the traffic between the racks (server only).
• keystore: (Default: conf/.keystore)
The location of a Java keystore (JKS) suitable for use with Java Secure Socket Extension (JSSE), which
is the Java version of the Secure Sockets Layer (SSL), and Transport Layer Security (TLS) protocols.
The keystore contains the private key used to encrypt outgoing messages.
• keystore_password: (Default: cassandra)
Password for the keystore.
• truststore: (Default: conf/.truststore)
Location of the truststore containing the trusted certificate for authenticating remote servers.
• truststore_password: (Default: cassandra)
Password for the truststore.
The passwords used in these options must match the passwords used when generating the keystore and
truststore. For instructions on generating these files, see Creating a Keystore to Use with JSSE.
The advanced settings are:
• protocol: (Default: TLS)
• algorithm: (Default: SunX509)
• store_type: (Default: JKS)
• cipher_suites: (Default:
TLS_RSA_WITH_AES_128_CBC_SHA,TLS_RSA_WITH_AES_256_CBC_SHA)
• require_client_auth: (Default: false)
Enables or disables certificate authentication.
Related information: Node-to-node encryption
client_encryption_options
Enable or disable client-to-node encryption. You must also generate keys and provide the appropriate key
and trust store locations and passwords. No custom encryption options are currently enabled. The available
options are:
• enabled: (Default: false)
To enable, set to true.
• keystore: (Default: conf/.keystore)
The location of a Java keystore (JKS) suitable for use with Java Secure Socket Extension (JSSE), which
is the Java version of the Secure Sockets Layer (SSL), and Transport Layer Security (TLS) protocols.
The keystore contains the private key used to encrypt outgoing messages.
• keystore_password: (Default: cassandra)
110
Configuration
Password for the keystore. This must match the password used when generating the keystore and
truststore.
• require_client_auth: (Default: false)
Enables or disables certificate authentication. (Available starting with Cassandra 1.2.3.)
• truststore: (Default: conf/.truststore)
Set if require_client_auth is true.
• truststore_password: <truststore_password>
Set if require_client_auth is true.
The advanced settings are:
• protocol: (Default: TLS)
• algorithm: (Default: SunX509)
• store_type: (Default: JKS)
• cipher_suites: (Default:
TLS_RSA_WITH_AES_128_CBC_SHA,TLS_RSA_WITH_AES_256_CBC_SHA)
Related information: Client-to-node encryption
ssl_storage_port
(Default: 7001) The SSL port for encrypted communication. Unused unless enabled in encryption_options.
Configuring gossip settings

Using the cassandra.yaml file to configure gossip.
About this task

When a node first starts up, it looks at its cassandra.yaml configuration file to determine the name of the
Cassandra cluster it belongs to; which nodes (called seeds) to contact to obtain information about the other
nodes in the cluster; and other parameters for determining port and range information.

conf/cassandra.yaml
Procedure
In the cassandra.yaml file, set the following parameters:
Property Description
cluster_name Name of the cluster that this node is joining. Must
be the same for every node in the cluster.
listen_address The IP address or hostname that Cassandra binds
to for connecting to other Cassandra nodes.
(Optional) broadcast_address The IP address a node tells other nodes in the
cluster to contact it by. It allows public and private
address to be different. For example, use the
broadcast_address parameter in topologies
where not all nodes have access to other nodes
111
Configuration
Property Description
by their private IP addresses. The default is the
listen_address.
seed_provider A -seeds list is comma-delimited list of hosts
(IP addresses) that gossip uses to learn the
topology of the ring. Every node should have
the same list of seeds. In multiple data-center
clusters, the seed list should include at least one
node from each data center (replication group).
More than a single seed node per data center
is recommended for fault tolerance. Otherwise,
gossip has to communicate with another data
center when bootstrapping a node. Making every
node a seed node is not recommended because
of increased maintenance and reduced gossip
performance. Gossip optimization is not critical,
but it is recommended to use a small seed list
(approximately three nodes per data center).
storage_port The inter-node communication port (default is
7000). Must be the same for every node in the
cluster.
initial_token For legacy clusters. Used in the single-node-per-
token architecture, where a node owns exactly
one contiguous range in the ring space.
num_tokens For new clusters. Defines the number of tokens
randomly assigned to this node on the ring when
using virtual nodes (vnodes).
Configuring the heap dump directory

Analyzing the heap dump file can help troubleshoot memory problems.
About this task

Analyzing the heap dump file can help troubleshoot memory problems. Cassandra starts Java with the
option -XX:+HeapDumpOnOutOfMemoryError. Using this option triggers a heap dump in the event of
an out-of-memory condition. The heap dump file consists of references to objects that cause the heap to
overflow. By default, Cassandra puts the file a subdirectory of the working, root directory when running as
a service. If Cassandra does not have write permission to the root directory, the heap dump fails. If the root
directory is too small to accommodate the heap dump, the server crashes.
For a heap dump to succeed and to prevent crashes, configure a heap dump directory that meets these
requirements:
• Accessible to Cassandra for writing
• Large enough to accommodate a heap dump
The location of the cassandra-env.sh file depends on the type of installation:
Package installations /etc/cassandra/cassandra-env.sh

Tarball installations install_location/conf/cassandra-env.sh
Base the size of the directory on the value of the Java -mx option.
112
Configuration
Procedure
Set the location of the heap dump in the cassandra-env.sh file.
1. Open the cassandra-env.sh file for editing.
# set jvm HeapDumpPath with CASSANDRA_HEAPDUMP_DIR
2. Scroll down to the comment about the heap dump path:
3. On the line after the comment, set the CASSANDRA_HEAPDUMP_DIR to the path you want to use:

CASSANDRA_HEAPDUMP_DIR =<path>
4. Save the cassandra-env.sh file and restart.
Configuring virtual nodes

Topics about configuring virtual nodes.
Enabling virtual nodes on a new cluster

Steps and recommendations for enabling virtual nodes (vnodes) on a new cluster.
About this task

Generally when all nodes have equal hardware capability, they should have the same number of virtual
nodes (vnodes). If the hardware capabilities vary among the nodes in your cluster, assign a proportional
number of vnodes to the larger machines. For example, you could designate your older machines to use
128 vnodes and your new machines (that are twice as powerful) with 256 vnodes.

conf/cassandra.yaml
Procedure
Set the number of tokens on each node in your cluster with the num_tokens parameter in the
cassandra.yaml file.
The recommended value is 256. Do not set the initial_token parameter.
Enabling virtual nodes on an existing production cluster

Steps and recommendations for enabling virtual nodes (vnodes) on an existing production cluster.
About this task

You cannot directly convert a single-token nodes to a vnode. However, you can configure another data
center configured with vnodes already enabled and let Cassandra automatic mechanisms distribute the
existing data into the new nodes. This method has the least impact on performance.
Procedure
1. Add a new data center to the cluster.
113
Configuration
2. Once the new data center with vnodes enabled is up, switch your clients to use the new data center.
3. Run a full repair with nodetool repair.
This step ensures that after you move the client to the new data center that any previous writes are
added to the new data center and that nothing else, such as hints, is dropped when you remove the old
data center.
4. Update your schema to no longer reference the old data center.
5. Remove the old data center from the cluster.
See Decommissioning a data center.
Using multiple network interfaces

Steps for configuring Cassandra for multiple network interfaces or when using different regions in cloud
implementations.
About this task

How to configure Cassandra for multiple network interfaces or when using different regions in cloud
implementations.
You must configure settings in both the cassandra.yaml file and the property file (cassandra-
rackdc.properties or cassandra-topology.properties) used by the snitch.
Configuring cassandra.yaml for multiple networks or across regions in cloud implementations

In multiple networks or cross-region cloud scenarios, communication between data centers can only take
place using an external IP address. The external IP address is defined in the cassandra.yaml file using
the broadcast_address setting. Configure each node as follows:
1. In the cassandra.yaml, set the listen_address to the private IP address of the node, and the
broadcast_address to the public address of the node.
This allows Cassandra nodes to bind to nodes in another network or region, thus enabling multiple
data-center support. For intra- network or region traffic, Cassandra switches to the private IP after
establishing a connection.
2. Set the addresses of the seed nodes in the cassandra.yaml file to that of the public IP. Private IP are
not routable between networks. For example:
seeds: 50.34.16.33, 60.247.70.52
3. Be sure that the storage_port or ssl_storage_port is open on the public IP firewall.
Caution: Be sure to enable encryption and authentication when using public IP's. See Node-to-
node encryption. Another option is to use a custom VPN to have local, inter-region/datacenter IP's.
Configuring the snitch for multiple networks

External communication between the data centers can only happen when using the broadcast_address
(public IP).
The GossipingPropertyFileSnitch is recommended for production. The cassandra-
rackdc.properties file defines the data centers used by this snitch.
For each node in the network, specify its data center in cassandra-rackdc.properties file.
114
Configuration
Network A Network B
• node0 • node0
dc=DC_A_cassandra dc=DC_A_cassandra
rack=RAC1 rack=RAC1
• node1 • node1
dc=DC_A_cassandra dc=DC_A_cassandra
rack=RAC1 rack=RAC1
• node2 • node2
dc=DC_B_cassandra dc=DC_B_cassandra
rack=RAC1 rack=RAC1
• node3 • node3
dc=DC_B_cassandra dc=DC_B_cassandra
rack=RAC1 rack=RAC1
• node4 • node4
dc=DC_A_analytics dc=DC_A_analytics
rack=RAC1 rack=RAC1
• node5 • node5
dc=DC_A_search dc=DC_A_search
rack=RAC1 rack=RAC1
Configuring the snitch for cross-region communication in cloud implementations

Note: Be sure to use the appropriate snitch for your implementation. If your deploying on Amazon
EC2, see the instructions in Ec2MultiRegionSnitch.
In cloud deployments, the region name is treated as the data center name and availability zones are
treated as racks within a data center. For example, if a node is in the us-east-1 region, us-east is the data
center name and 1 is the rack location. (Racks are important for distributing replicas, but not for data center
naming.)
For each node, specify its data center in the cassandra-rackdc.properties. The dc_suffix option defines the
data centers used by the snitch. Any other lines are ignored.

• node0 • node0
• node1 • node1
• node2 • node2
115
Configuration

• node3 • node3
• node4 • node4
dc_suffix=_1_analytics dc_suffix=_1_analytics
• node5 • node5
dc_suffix=_1_search dc_suffix=_1_search
This results in four us-east data centers: This results in four us-west data centers:
us-east_1_analytics us-west_1_analytics
us-east_1_search us-west_1_search

conf/cassandra.yaml
Configuring logging
About Cassandra logging functionality using Simple Logging Facade for Java (SLF4J) with a logback
backend.
Cassandra provides logging functionality using Simple Logging Facade for Java (SLF4J) with a logback
backend. Logs are written to the system.log file in the Cassandra logging directory. You can configure
logging programmatically or manually. Manual ways to configure logging are:
• Run the nodetool setlogginglevel command
• Configure the logback-test.xml or logback.xml file installed with Cassandra
• Use the JConsole tool to configure logging through JMX.
Logback looks for logback-test.xml first, and then for logback.xml. The logback.xml location for
different types of installations is listed in the "File locations" section. For example, on tarball and source
installations, logback.xml is located in the install_location/conf directory.
The XML configuration files look something like this:
<configuration scan="true">
<jmxConfigurator />
<appender name="FILE"
class="ch.qos.logback.core.rolling.RollingFileAppender">
<file>${cassandra.logdir}/system.log</file>
<rollingPolicy
class="ch.qos.logback.core.rolling.FixedWindowRollingPolicy">
<fileNamePattern>${cassandra.logdir}/system.log.%i.zip</
fileNamePattern>
<minIndex>1</minIndex>
<maxIndex>20</maxIndex>
</rollingPolicy>
116
Configuration
<triggeringPolicy
class="ch.qos.logback.core.rolling.SizeBasedTriggeringPolicy">
<maxFileSize>20MB</maxFileSize>
</triggeringPolicy>
<encoder>
<pattern>%-5level [%thread] %date{ISO8601} %F:%L - %msg%n</pattern>

</encoder>
</appender>
<appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">

<encoder>
<pattern>%-5level %date{HH:mm:ss,SSS} %msg%n</pattern>
</encoder>
</appender>
<root level="INFO">
<appender-ref ref="FILE" />
<appender-ref ref="STDOUT" />
</root>
<logger name="com.thinkaurelius.thrift" level="ERROR"/>

</configuration>
The appender configurations specify where to print the log and its configuration. The first appender directs
logs to a file. The second appender directs logs to the console. You can change the following logging
functionality:
• Rolling policy
• The policy for rolling logs over to an archive
• Location and name of the log file
• Location and name of the archive
• Minimum and maximum file size to trigger rolling
• Format of the message
• The log level
Log levels
The valid values for setting the log level include ALL for logging information at all levels, TRACE through
ERROR, and OFF for no logging. TRACE creates the most verbose log, and ERROR, the least.
• ALL
• TRACE
• DEBUG
• INFO (Default)
• WARN
• ERROR
• OFF
Note: Increasing logging levels can generate heavy logging output on a moderately trafficked
cluster.
You can use the nodetool getlogginglevels command to see the current logging configuration.
$ nodetool getlogginglevels
Logger Name Log Level
ROOT INFO
117
Configuration
com.thinkaurelius.thrift ERROR
Migrating to logback from log4j

If you upgrade from a previous version of Cassandra that used log4j, you can convert log4j.properties files
to logback.xml using the logback PropertiesTranslator web-application.
Using log file rotation

The default policy rolls the system.log file after the size exceeds 20MB. Archives are compressed in zip
format. Logback names the log files system.log.1.zip, system.log.2.zip, and so on. For more
information, see logback documentation.
Commit log archive configuration

Cassandra provides commitlog archiving and point-in-time recovery.
About this task

Cassandra provides commitlog archiving and point-in-time recovery. You configure this feature in the
commitlog_archiving.properties configuration file, which is located in the following directories:
• Cassandra package installations: /etc/cassandra
• Cassandra tarball installations: install_location/conf
• DataStax Enterprise package installations: /etc/dse/cassandra
• DataStax Enterprise tarball installations: install_location/resources/cassandra/conf
The commands archive_command and restore_command expect only a single command with
arguments. The parameters must be entered verbatim. STDOUT and STDIN or multiple commands cannot
be executed. To workaround, you can script multiple commands and add a pointer to this file. To disable a
command, leave it blank.
Procedure
• Archive a commitlog segment:
Command archive_command=
Parameters %path Fully qualified path of the segment to
archive.
%name Name of the commit log.
Example archive_command=/bin/ln %path /
backup/%name
• Restore an archived commitlog:
Command restore_command=
Parameters %from Fully qualified path of the an archived
commitlog segment from the
restore_directories.
%t Name of live commit log directory.
Example restore_command=cp -f %from %to
• Set the restore directory location:
118
Configuration
Command restore_directories=
Format restore_directories=restore_directory_location
• Restore mutations created up to and including the specified timestamp:
Command restore_point_in_time=
Format <timestamp> (YYYY:MM:DD HH:MM:SS)
Example restore_point_in_time=2013:12:11
17:00:00
Restore stops when the first client-supplied timestamp is greater than the restore point timestamp.
Because the order in which Cassandra receives mutations does not strictly follow the timestamp order,
this can leave some mutations unrecovered.
Generating tokens
If not using virtual nodes (vnodes), you must calculate tokens for your cluster.
If not using virtual nodes (vnodes), you must calculate tokens for your cluster.
The following topics in the Cassandra 1.1 documentation provide conceptual information about tokens:
• Data Distribution in the Ring
• Replication Strategy
About calculating tokens for single or multiple data centers in Cassandra 1.2 and later
• Single data center deployments: calculate tokens by dividing the hash range by the number of nodes in
the cluster.
• Multiple data center deployments: calculate the tokens for each data center so that the hash range is
evenly divided for the nodes in each data center.
For more explanation, see be sure to read the conceptual information mentioned above.
The method used for calculating tokens depends on the type of partitioner:
Calculating tokens for the Murmur3Partitioner

Use this method for generating tokens when you are not using virtual nodes (vnodes) and using the
63
Murmur3Partitioner (default). This partitioner uses a maximum possible range of hash values from -2 to
63
+2 -1. To calculate tokens for this partitioner:
python -c 'print [str(((2**64 / number_of_tokens) * i) - 2**63) for i in

range(number_of_tokens)]'
For example, to generate tokens for 6 nodes:
python -c 'print [str(((2**64 / 6) * i) - 2**63) for i in range(6)]'
The command displays the token for each node:
[ '-9223372036854775808', '-6148914691236517206', '-3074457345618258604',

'-2', '3074457345618258600', '6148914691236517202' ]
119
Configuration
Calculating tokens for the RandomPartitioner

To calculate tokens when using the RandomPartitioner in Cassandra 1.2 clusters, use the Cassandra 1.1
Token Generating Tool.
Hadoop support
Cassandra support for integrating Hadoop with Cassandra.
Cassandra support for integrating Hadoop with Cassandra includes:
• MapReduce
• Apache Pig
You can use Cassandra 2.1 with Hadoop 2.x or 1.x with some restrictions.
• Isolate Cassandra and Hadoop nodes in separate data centers.
• Before starting the data centers of Cassandra/Hadoop nodes, disable virtual nodes (vnodes).
To disable virtual nodes:
1. In the cassandra.yaml file, set num_tokens to 1.
2. Uncomment the initial_token property and set it to 1 or to the value of a generated token for a multi-
node cluster.
3. Start the cluster for the first time.
Do not disable or enable vnodes on an existing cluster.
Setup and configuration, described in the Apache docs, involves overlaying a Hadoop cluster on
Cassandra nodes, configuring a separate server for the Hadoop NameNode/JobTracker, and installing
a Hadoop TaskTracker and Data Node on each Cassandra node. The nodes in the Cassandra data
center can draw from data in the HDFS Data Node as well as from Cassandra. The Job Tracker/Resource
Manager (JT/RM) receives MapReduce input from the client application. The JT/RM sends a MapReduce
job request to the Task Trackers/Node Managers (TT/NM) and optional clients, MapReduce and Pig. The
data is written to Cassandra and results sent back to the client.
120
Configuration
The Apache docs also cover how to get configuration and integration support.
Input and Output Formats

Hadoop jobs can receive data from CQL tables and indexes and you can load data into Cassandra from a
Hadoop job. Cassandra 2.1 supports the following formats for these tasks:
• CQL partition input format: ColumnFamilyInputFormat class.
• BulkOutputFormat class introduced in Cassandra 1.1
Cassandra 2.1.1 and later supports the CqlOutputFormat, which is the CQL-compatible version of the
BulkOutputFormat class. The CQLOutputFormat acts as a Hadoop-specific OutputFormat. Reduce tasks
can store keys (and corresponding bound variable values) as CQL rows (and respective columns) in a
given CQL table.
Cassandra 2.1.1 supports using the CQLOutputFormat with Apache Pig.
Running the wordcount example

Wordcount example JARs are located in the examples directory of the Cassandra source
code installation. There are CQL and legacy examples in the hadoop_cql3_word_count and
hadoop_word_count subdirectories, respectively. Follow instructions in the readme files.
Isolating Hadoop and Cassandra workloads

When you create a keyspace using CQL, Cassandra creates a virtual data center for a cluster, even a one-
node cluster, automatically. You assign nodes that run the same type of workload to the same data center.
The separate, virtual data centers for different types of nodes segregate workloads running Hadoop from
those running Cassandra. Segregating workloads ensures that only one type of workload is active per data
center. Separating nodes running a sequential data load, from nodes running any other type of workload,
such as Cassandra real-time OLTP queries is a best practice.
121
Configuration

conf/cassandra.yaml
122
Operations
Operations
Operation topics.
Monitoring Cassandra
Monitoring topics.
Monitoring a Cassandra cluster

Understanding the performance characteristics of a Cassandra cluster is critical to diagnosing issues and
planning capacity.
Understanding the performance characteristics of a Cassandra cluster is critical to diagnosing issues and
planning capacity.
Cassandra exposes a number of statistics and management operations via Java Management Extensions
(JMX). JMX is a Java technology that supplies tools for managing and monitoring Java applications
and services. Any statistic or operation that a Java application has exposed as an MBean can then be
monitored or manipulated using JMX.
During normal operation, Cassandra outputs information and statistics that you can monitor using JMX-
compliant tools, such as:
• The Cassandra nodetool utility
• DataStax OpsCenter management console
• JConsole
Using the same tools, you can perform certain administrative commands and operations such as flushing
caches or doing a node repair.
Monitoring using the nodetool utility

The nodetool utility is a command-line interface for monitoring Cassandra and performing routine database
operations. Included in the Cassandra distribution, nodetool and is typically run directly from an operational
Cassandra node.
The nodetool utility supports the most important JMX metrics and operations, and includes other useful
commands for Cassandra administration, such as the proxyhistogram command. This example shows the
output from nodetool proxyhistograms after running 4,500 insert statements and 45,000 select statements
on a three ccm node-cluster on a local computer.
$ nodetool proxyhistograms
proxy histograms
Percentile Read Latency Write Latency Range Latency
(micros) (micros) (micros)
50% 1502.50 375.00 446.00
75% 1714.75 420.00 498.00
95% 31210.25 507.00 800.20
98% 36365.00 577.36 948.40
99% 36365.00 740.60 1024.39
Min 616.00 230.00 311.00
Max 36365.00 55726.00 59247.00
For a summary of the ring and its current state of general health, use the status command. For example:
$ nodetool status
123
Operations
Note: Ownership information does not include topology; for complete

information, specify a keyspace
=======================
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
-- Address Load Tokens Owns Host ID
Rack
UN 127.0.0.1 47.66 KB 1 33.3% aaa1b7c1-6049-4a08-
ad3e-3697a0e30e10 rack1
UN 127.0.0.2 47.67 KB 1 33.3% 1848c369-4306-4874-
afdf-5c1e95b8732e rack1
UN 127.0.0.3 47.67 KB 1 33.3% 49578bf1-728f-438d-b1c1-
d8dd644b6f7f rack1
The nodetool utility provides commands for viewing detailed metrics for tables, server metrics, and
compaction statistics:
• nodetool cfstats displays statistics for each table and keyspace.
• nodetool cfhistograms provides statistics about a table, including read/write latency, row size,
column count, and number of SSTables.
• nodetool netstats provides statistics about network operations and connections.
• nodetool tpstats provides statistics about the number of active, pending, and completed tasks for
each stage of Cassandra operations by thread pool.
DataStax OpsCenter
DataStax OpsCenter is a graphical user interface for monitoring and administering all nodes in a
Cassandra cluster from one centralized console. DataStax OpsCenter is bundled with DataStax support
offerings. You can register for a free version for development or non-production use.
OpsCenter provides a graphical representation of performance trends in a summary view that is hard
to obtain with other monitoring tools. The GUI provides views for different time periods as well as
the capability to drill down on single data points. Both real-time and historical performance data for a
Cassandra or DataStax Enterprise cluster are available in OpsCenter. OpsCenter metrics are captured and
stored within Cassandra.
Within OpsCenter you can customize the performance metrics viewed to meet your monitoring needs.
Administrators can also perform routine node administration tasks from OpsCenter. Metrics within
OpsCenter are divided into three general categories: table metrics, cluster metrics, and OS metrics. For
many of the available metrics, you can view aggregated cluster-wide information or view information on a
per-node basis.
124
Operations
Monitoring using JConsole

JConsole is a JMX-compliant tool for monitoring Java applications such as Cassandra. It is included with
Sun JDK 5.0 and higher. JConsole consumes the JMX metrics and operations exposed by Cassandra and
displays them in a well-organized GUI. For each node monitored, JConsole provides these six separate tab
views:
• Overview
Displays overview information about the Java VM and monitored values.
• Memory
Displays information about memory use.
• Threads
Displays information about thread use.
• Classes
Displays information about class loading.
• VM Summary
Displays information about the Java Virtual Machine (VM).
• Mbeans
Displays information about MBeans.
The Overview and Memory tabs contain information that is very useful for Cassandra developers.
The Memory tab allows you to compare heap and non-heap memory usage, and provides a control to
immediately perform Java garbage collection.
For specific Cassandra metrics and operations, the most important area of JConsole is the MBeans tab.
This tab lists the following Cassandra MBeans:
• org.apache.cassandra.db
Includes caching, table metrics, and compaction.
• org.apache.cassandra.internal
Internal server operations such as gossip and hinted handoff.
• org.apache.cassandra.net
Inter-node communication including FailureDetector, MessagingService and StreamingService.
• org.apache.cassandra.request
Tasks related to read, write, and replication operations.
When you select an MBean in the tree, its MBeanInfo and MBean Descriptor are displayed on the right,
and any attributes, operations or notifications appear in the tree below it. For example, selecting and
expanding the org.apache.cassandra.db MBean to view available actions for a table results in a display like
the following:
125
Operations
If you choose to monitor Cassandra using JConsole, keep in mind that JConsole consumes a significant
amount of system resources. For this reason, DataStax recommends running JConsole on a remote
machine rather than on the same host as a Cassandra node.
The JConsole CompactionManagerMBean exposes compaction metrics that can indicate when you need
to add capacity to your cluster.
Compaction metrics
Monitoring compaction performance is an important aspect of knowing when to add capacity to your
cluster.
Monitoring compaction performance is an important aspect of knowing when to add capacity to your
cluster. The following attributes are exposed through CompactionManagerMBean:
Table 8: Compaction Metrics
Attribute Description
CompletedTasks Number of completed compactions since the last start of this Cassandra
instance
PendingTasks Number of estimated tasks remaining to perform
ColumnFamilyInProgress The table currently being compacted. This attribute is null if no compactions
are in progress.
BytesTotalInProgress Total number of data bytes (index and filter are not included) being
compacted. This attribute is null if no compactions are in progress.
BytesCompacted The progress of the current compaction. This attribute is null if no compactions
are in progress.
Thread pool and read/write latency statistics

Increases in pending tasks on thread pool statistics can indicate when to add additional capacity.
Cassandra maintains distinct thread pools for different stages of execution. Each of the thread pools
provide statistics on the number of tasks that are active, pending, and completed. Trends on these pools
for increases in the pending tasks column indicate when to add additional capacity. After a baseline is
established, configure alarms for any increases above normal in the pending tasks column. Use nodetool
tpstats on the command line to view the thread pool details shown in the following table.
126
Operations
Thread Pool Description

AE_SERVICE_STAGE Shows anti-entropy tasks.
CONSISTENCY- Handles the background consistency checks if they were triggered from the
MANAGER client's consistency level.
FLUSH-SORTER-POOL Sorts flushes that have been submitted.
FLUSH-WRITER-POOL Writes the sorted flushes.
GOSSIP_STAGE Activity of the Gossip protocol on the ring.
LB-OPERATIONS The number of load balancing operations.
LB-TARGET Used by nodes leaving the ring.
MEMTABLE-POST- Memtable flushes that are waiting to be written to the commit log.
FLUSHER
MESSAGE- Streaming operations. Usually triggered by bootstrapping or decommissioning
STREAMING-POOL nodes.
MIGRATION_STAGE Tasks resulting from the call of system_* methods in the API that have
modified the schema.
MISC_STAGE
MUTATION_STAGE API calls that are modifying data.
READ_STAGE API calls that have read data.
RESPONSE_STAGE Response tasks from other nodes to message streaming from this node.
STREAM_STAGE Stream tasks from this node.
Read/Write latency metrics

Cassandra tracks latency (averages and totals) of read, write, and slicing operations at the server level
through StorageProxyMBean.
Table statistics
Compaction metrics provide a number of statistics that are important for monitoring performance trends.
For individual tables, ColumnFamilyStoreMBean provides the same general latency attributes as
StorageProxyMBean. Unlike StorageProxyMBean, ColumnFamilyStoreMBean has a number of other
statistics that are important to monitor for performance trends. The most important of these are:
Attribute Description
MemtableDataSize The total size consumed by this table's data (not including metadata).
MemtableColumnsCount Returns the total number of columns present in the memtable (across
all keys).
MemtableSwitchCount How many times the memtable has been flushed out.
RecentReadLatencyMicros The average read latency since the last call to this bean.
RecentWriterLatencyMicros The average write latency since the last call to this bean.
LiveSSTableCount The number of live SSTables for this table.
127
Operations
The recent read latency and write latency counters are important in making sure operations are happening
in a consistent manner. If these counters start to increase after a period of staying flat, you probably need
to add capacity to the cluster.
You can set a threshold and monitor LiveSSTableCount to ensure that the number of SSTables for a given
table does not become too great.
Tuning Bloom filters

Cassandra uses Bloom filters to determine whether an SSTable has data for a particular row.
Cassandra uses Bloom filters to determine whether an SSTable has data for a particular row. Bloom filters
are unused for range scans, but are used for index scans. Bloom filters are probabilistic sets that allow you
to trade memory for accuracy. This means that higher Bloom filter attribute settings bloom_filter_fp_chance
use less memory, but will result in more disk I/O if the SSTables are highly fragmented. Bloom filter
settings range from 0 to 1.0 (disabled). The default value of bloom_filter_fp_chance depends on the
compaction strategy. The LeveledCompactionStrategy uses a higher default value (0.1) than the
SizeTieredCompactionStrategy or DateTieredCompactionStrategy, which have a default of 0.01. Memory
savings are nonlinear; going from 0.01 to 0.1 saves about one third of the memory. SSTables using LCS
contain a relatively smaller ranges of keys than those using STCS, which facilitates efficient exclusion of
the SSTables even without a bloom filter; however, adding a small bloom filter helps when there are many
levels in LCS.
The settings you choose depend the type of workload. For example, to run an analytics application that
heavily scans a particular table, you would want to inhibit the Bloom filter on the table by setting it high.
To view the observed Bloom filters false positive rate and the number of SSTables consulted per read use
cfstats in the nodetool utility.
Bloom filters are stored off-heap so you don't need include it when determining the -Xmx settings (the
maximum memory size that the heap can reach for the JVM).
To change the bloom filter property on a table, use CQL. For example:
ALTER TABLE addamsFamily WITH bloom_filter_fp_chance = 0.1;
After updating the value of bloom_filter_fp_chance on a table, Bloom filters need to be regenerated in one
of these ways:
• Initiate compaction
• Upgrade SSTables
You do not have to restart Cassandra after regenerating SSTables.
Data caching
Data caching topics.
Configuring data caches

Cassandra includes integrated caching and distributes cache data around the cluster. The integrated
architecture facilitates troubleshooting and the cold start problem.
Cassandra includes integrated caching and distributes cache data around the cluster. When a node
goes down, the client can read from another cached replica of the data. The integrated architecture also
facilitates troubleshooting because there is no separate caching tier, and cached data matches what’s
in the database exactly. The integrated cache solves the cold start problem by saving the cache to disk
periodically. Cassandra reads contents back into the cache and distributes the data when it restarts. The
cluster does not start with a cold cache.
128
Operations
In Cassandra 2.1, the saved key cache files include the ID of the table in the file name. A saved key cache
file name for the users table in the mykeyspace keyspace in a Cassandra 2.1 looks something like this:
mykeyspace-users.users_name_idx-19bd7f80352c11e4aa6a57448213f97f-KeyCache-
b.db2046071785672832311.tmp
You can configure partial or full caching of each partition by setting the rows_per_partition table option.
Previously, the caching mechanism put the entire partition in memory. If the partition was larger than the
cache size, Cassandra never read the data from the cache. Now, you can specify the number of rows to
cache per partition to increase cache hits. You configure the cache using the CQL caching property.
About the partition key cache

The partition key cache is a cache of the partition index for a Cassandra table. Using the key cache instead
of relying on the OS page cache decreases seek times. However, enabling just the key cache results in
disk (or OS page cache) activity to actually read the requested data rows.
About the row cache

You can configure the number of rows to cache in a partition. To cache rows, if the row key is not already
in the cache, Cassandra reads the first portion of the partition, and puts the data in the cache. If the newly
cached data does not include all cells configured by user, Cassandra performs another read.
Typically, you enable either the partition key or row cache for a table, except archive tables, which are
infrequently read. Disable caching entirely for archive tables.
Enabling and configuring caching

Using CQL to enable or disable caching.
About this task

Use CQL to enable or disable caching by configuring the caching table property. Set parameters in the
cassandra.yaml file to configure global caching properties:
• Partition key cache size
• Row cache size
• How often Cassandra saves partition key caches to disk
• How often Cassandra saves row caches to disk
Set the caching property using CREATE TABLE or ALTER TABLE. For example, configuring the
row_cache_size_in_mb determines how much space in memory Cassandra allocates to store rows from
the most frequently read partitions of the table.

conf/cassandra.yaml
Procedure
Set the table caching property that configures the partition key cache and the row cache.
CREATE TABLE users (

userid text PRIMARY KEY,
first_name text,
last_name text,
)
WITH caching = { 'keys' : 'NONE', 'rows_per_partition' : '120' };
129
Operations
How caching works

Brief overview of how caching works.
When both row cache and partition key cache are configured, the row cache returns results whenever
possible. In the event of a row cache miss, the partition key cache might still provide a hit that makes
the disk seek much more efficient. This diagram depicts two read operations on a table with both caches
already populated.
One read operation hits the row cache, returning the requested row without a disk seek. The other read
operation requests a row that is not present in the row cache but is present in the partition key cache. After
accessing the row in the SSTable, the system returns the data and populates the row cache with this read
operation.
Tips for efficient cache use

Various tips for efficient cache use.
"Tuning the row cache in Cassandra 2.1" describes best practices of using the built-in caching
mechanisms and designing an effective data model. Some tips for efficient cache use are:
• Store lower-demand data or data with extremely long partitions in a table with minimal or no caching.
• Deploy a large number of Cassandra nodes under a relatively light load per node.
• Logically separate heavily-read data into discrete tables.
When you query a table, turn on tracing to check that the table actually gets data from the cache rather
than from disk. The first time you read data from a partition, the trace shows this line below the query
becasue the cache has not been populated yet:
Row cache miss [ReadStage:41]
In subsequent queries for the same partition, look for a line in the trace that looks something like this:
Row cache hit [ReadStage:55]
This output means the data was found in the cache and no disk read occurred. Updates invalidate the
cache. If you query rows in the cache plus uncached rows, request more rows than the global limit allows,
130
Operations
or the query does not grab the beginning of the partition, the trace might include a line that looks something
like this:
Ignoring row cache as cached value could not satisfy query [ReadStage:89]
This output indicates that an insufficient cache caused a disk read. Requesting rows not at the beginning
of the partition is a likely cause. Try removing constraints that might cause the query to skip the beginning
of the partition, or place a limit on the query to prevent results from overflowing the cache. To ensure that
the query hits the cache, try increasing the cache size limit, or restructure the table to position frequently
accessed rows at the head of the partition.
Monitoring and adjusting caching

Use nodetool to make changes to cache options and then monitor the effects of each change.
Make changes to cache options in small, incremental adjustments, then monitor the effects of each change
using DataStax Opscenter or the nodetool utility. The output of the nodetool info command shows the
following row cache and key cache metrics, which are configured in the cassandra.yaml file:
• Cache size in bytes
• Capacity in bytes
• Number of hits
• Number of requests
• Recent hit rate
• Duration in seconds after which Cassandra saves the key cache.
For example, on start-up, the information from nodetool info might look something like this:
ID : 387d15ba-7103-491b-9327-1a691dbb504a
Gossip active : true
Thrift active : true
Native Transport active: true
Load : 65.87 KB
Generation No : 1400189757
Uptime (seconds) : 148760
Heap Memory (MB) : 392.82 / 1996.81
Data Center : datacenter1
Rack : rack1
Exceptions : 0
Key Cache : entries 10, size 728 (bytes), capacity 103809024 (bytes),
93 hits, 102 requests, 0.912 recent hit rate, 14400 save period in seconds
Row Cache : entries 0, size 0 (bytes), capacity 0 (bytes), 0 hits, 0
requests, NaN recent hit rate, 0 save period in seconds
Counter Cache : entries 0, size 0 (bytes), capacity 51380224 (bytes), 0
hits, 0 requests, NaN recent hit rate, 7200 save period in seconds
Token : -9223372036854775808
In the event of high memory consumption, consider tuning data caches.


conf/cassandra.yaml
131
Operations
Configuring memtable throughput

Configuring memtable throughput to improve write performance.
Configuring memtable throughput can improve write performance. Cassandra flushes memtables to disk,
creating SSTables when the commit log space threshold has been exceeded. Configure the commit log
space threshold per node in the cassandra.yaml. How you tune memtable thresholds depends on your
data and write load. Increase memtable throughput under either of these conditions:
• The write load includes a high volume of updates on a smaller set of data.
• A steady stream of continuous writes occurs. This action leads to more efficient compaction.
Allocating memory for memtables reduces the memory available for caching and other internal Cassandra
structures, so tune carefully and in small increments.

conf/cassandra.yaml
Configuring compaction
Steps for configuring compaction. The compaction process merges keys, combines columns, evicts
tombstones, consolidates SSTables, and creates a new index in the merged SSTable.
About this task

As discussed in the Compaction topic, the compaction process merges keys, combines columns, evicts
tombstones, consolidates SSTables, and creates a new index in the merged SSTable.
In the cassandra.yaml file, you configure these global compaction parameters:
• snapshot_before_compaction
• in_memory_compaction_limit_in_mb
• compaction_preheat_key_cache
• concurrent_compactors
• compaction_throughput_mb_per_sec
The compaction_throughput_mb_per_sec parameter is designed for use with large partitions because
compaction is throttled to the specified total throughput across the entire system.
Cassandra provides a start-up option for testing compaction strategies without affecting the production
workload.
Using CQL, you configure a compaction strategy:
• SizeTieredCompactionStrategy (STCS): The default compaction strategy. This strategy triggers
a minor compaction when there are a number of similar sized SSTables on disk as configured by the
table subproperty, min_threshold. A minor compaction does not involve all the tables in a keyspace.
Also see STCS compaction subproperties.
• DateTieredCompactionStrategy (DTCS): Available in Cassandra 2.0.11 and 2.1.1 and later.
This strategy is particularly useful for time series data. DateTieredCompactionStrategy stores data
written within a certain period of time in the same SSTable. For example, Cassandra can store your last
hour of data in one SSTable time window, and the next 4 hours of data in another time window, and so
on. Compactions are triggered when the min_threshold (4 by default) for SSTables in those windows
is reached. The most common queries for time series workloads retrieve the last hour/day/month of
data. Cassandra can limit SSTables returned to those having the relevant data. Also, Cassandra can
132
Operations
store data that has been set to expire using TTL in an SSTable with other data scheduled to expire at
approximately the same time. Cassandra can then drop the SSTable without doing any compaction.
Also see DTCS compaction subproperties and DateTieredCompactionStrategy: Compaction for Time
Series Data.
Note: When using DTCS disabling read repair is recommended. Use full repair as necessary.
• LeveledCompactionStrategy (LCS): The leveled compaction strategy creates SSTables of a
fixed, relatively small size (160 MB by default) that are grouped into levels. Within each level, SSTables
are guaranteed to be non-overlapping. Each level (L0, L1, L2 and so on) is 10 times as large as the
previous. Disk I/O is more uniform and predictable on higher than on lower levels as SSTables are
continuously being compacted into progressively larger levels. At each level, row keys are merged
into non-overlapping SSTables. This can improve performance for reads, because Cassandra can
determine which SSTables in each level to check for the existence of row key data. This compaction
strategy is modeled after Google's leveldb implementation. Also see LCS compaction subproperties.
To configure the compaction strategy property and CQL compaction subproperties, such as the maximum
number of SSTables to compact and minimum SSTable size, use CREATE TABLE or ALTER TABLE.

conf/cassandra.yaml
Procedure
1. Update a table to set the compaction strategy using the ALTER TABLE statement.
ALTER TABLE users WITH

compaction = { 'class' : 'LeveledCompactionStrategy' }
2. Change the compaction strategy property to SizeTieredCompactionStrategy and specify the minimum
number of SSTables to trigger a compaction using the CQL min_threshold attribute.
ALTER TABLE users

WITH compaction =
{'class' : 'SizeTieredCompactionStrategy', 'min_threshold' : 6 }
Results
You can monitor the results of your configuration using compaction metrics, see Compaction metrics.
Compression
Compression maximizes the storage capacity of Cassandra nodes by reducing the volume of data on disk
and disk I/O, particularly for read-dominated workloads.
Compression maximizes the storage capacity of Cassandra nodes by reducing the volume of data on disk
and disk I/O, particularly for read-dominated workloads. Cassandra quickly finds the location of rows in the
SSTable index and decompresses the relevant row chunks.
Write performance is not negatively impacted by compression in Cassandra as it is in traditional
databases. In traditional relational databases, writes require overwrites to existing data files on disk. The
database has to locate the relevant pages on disk, decompress them, overwrite the relevant data, and
finally recompress. In a relational database, compression is an expensive operation in terms of CPU cycles
and disk I/O. Because Cassandra SSTable data files are immutable (they are not written to again after
they have been flushed to disk), there is no recompression cycle necessary in order to process writes.
133
Operations
SSTables are compressed only once when they are written to disk. Writes on compressed tables can show
up to a 10 percent performance improvement.
When to compress data

Compression is best suited for tables that have many rows and each row has the same columns, or at
least as many columns, as other rows.
Compression is best suited for tables that have many rows and each row has the same columns, or at
least as many columns, as other rows. For example, a table containing user data such as username, email,
and state, is a good candidate for compression. The greater the similarity of the data across rows, the
greater the compression ratio and gain in read performance.
A table that has rows of different sets of columns is not well-suited for compression.
Don't confuse table compression with compact storage of columns, which is used for backward
compatibility of old applications with CQL.
Depending on the data characteristics of the table, compressing its data can result in:
• 2x-4x reduction in data size
• 25-35% performance improvement on reads
• 5-10% performance improvement on writes
After configuring compression on an existing table, subsequently created SSTables are compressed.
Existing SSTables on disk are not compressed immediately. Cassandra compresses existing SSTables
when the normal Cassandra compaction process occurs. Force existing SSTables to be rewritten and
compressed by using nodetool upgradesstables (Cassandra 1.0.4 or later) or nodetool scrub.
Configuring compression
Steps for configuring compression.
About this task

You configure a table property and subproperties to manage compression. The CQL table properties
documentation describes the types of compression options that are available. Compression is enabled by
default.
Procedure
1. Disable compression, using CQL to set the compression parameters to an empty string.
CREATE TABLE DogTypes (

block_id uuid,
species text,
alias text,
population varint,
PRIMARY KEY (block_id)
)
WITH compression = { 'sstable_compression' : '' };
2. Enable compression on an existing table, using ALTER TABLE to set the compression algorithm
sstable_compression to LZ4Compressor (Cassandra 1.2.2 and later), SnappyCompressor, or
DeflateCompressor.
CREATE TABLE DogTypes (

block_id uuid,
species text,
alias text,
population varint,
PRIMARY KEY (block_id)
)
134
Operations
WITH compression = { 'sstable_compression' :

'LZ4Compressor' };
3. Change compression on an existing table, using ALTER TABLE and setting the compression algorithm
sstable_compression to DeflateCompressor.
ALTER TABLE CatTypes

WITH compression = { 'sstable_compression' : 'DeflateCompressor',
'chunk_length_kb' : 64 }
You tune data compression on a per-table basis using CQL to alter a table.
Testing compaction and compression

Enabling write survey mode.
About this task

Write survey mode is a Cassandra startup option for testing new compaction and compression strategies.
In write survey mode, you can test out new compaction and compression strategies on that node and
benchmark the write performance differences, without affecting the production cluster.
Write survey mode adds a node to a database cluster. The node accepts all write traffic as if it were part of
the normal Cassandra cluster, but the node does not officially join the ring.
Also use write survey mode to try out a new Cassandra version. The nodes you add in write survey mode
to a cluster must be of the same major release version as other nodes in the cluster. The write survey
mode relies on the streaming subsystem that transfers data between nodes in bulk and differs from one
major release to another.
If you want to see how read performance is affected by modifications, stop the node, bring it up as a
standalone machine, and then benchmark read operations on the node.
Procedure
Enable write survey mode by starting a Cassandra node using the write_survey option.
$ bin/cassandra – Dcassandra.write_survey=true
This example shows how to start a tarball installation of Cassandra.
Tuning Java resources

Consider tuning Java resources in the event of a performance degradation or high memory consumption.
Consider tuning Java resources in the event of a performance degradation or high memory consumption.
There are two files that control environment settings for Cassandra:
• conf/cassandra-env.sh
Java Virtual Machine (JVM) configuration settings
• bin/cassandra-in.sh
Sets up Cassandra environment variables such as CLASSPATH and JAVA_HOME.
Heap sizing options

If you decide to change the Java heap sizing, both MAX_HEAP_SIZE and HEAP_NEWSIZE should should
be set together in conf/cassandra-env.sh.
135
Operations
• MAX_HEAP_SIZE
Sets the maximum heap size for the JVM. The same value is also used for the minimum heap size. This
allows the heap to be locked in memory at process start to keep it from being swapped out by the OS.
• HEAP_NEWSIZE
The size of the young generation. The larger this is, the longer GC pause times will be. The shorter it is,
the more expensive GC will be (usually). A good guideline is 100 MB per CPU core.
Tuning the Java heap

Because Cassandra is a database, it spends significant time interacting with the operating system's
I/O infrastructure through the JVM, so a well-tuned Java heap size is important. Cassandra's default
configuration opens the JVM with a heap size that is based on the total amount of system memory:
System Memory Heap Size

Less than 2GB 1/2 of system memory
2GB to 4GB 1GB
Greater than 4GB 1/4 system memory, but not more than 8GB
Many users new to Cassandra are tempted to turn up Java heap size too high, which consumes the
majority of the underlying system's RAM. In most cases, increasing the Java heap size is actually
detrimental for these reasons:
• In most cases, the capability of Java to gracefully handle garbage collection above 8GB quickly
diminishes.
• Modern operating systems maintain the OS page cache for frequently accessed data and are very good
at keeping this data in memory, but can be prevented from doing its job by an elevated Java heap size.
If you have more than 2GB of system memory, which is typical, keep the size of the Java heap relatively
small to allow more memory for the page cache.
Some Solr users have reported that increasing the stack size improves performance under Tomcat. To
increase the stack size, uncomment and modify the default setting in the cassandra-env.sh file. Also,
decreasing the memtable space to make room for Solr caches might improve performance. Modify the
memtable space using the memtable_total_space_in_mb property in the cassandra.yaml file.
Because MapReduce runs outside the JVM, changes to the JVM do not affect Analytics/Hadoop
operations directly.
How Cassandra uses memory

Using a java-based system like Cassandra, you can typically allocate about 8GB of memory on the heap
before garbage collection pause time starts to become a problem. Modern machines have much more
memory than that and Cassandra can make use of additional memory as page cache when files on disk
are accessed. Allocating more than 8GB of memory on the heap poses a problem due to the amount of
Cassandra metadata about data on disk. The Cassandra metadata resides in memory and is proportional
to total data. Some of the components grow proportionally to the size of total memory.
In Cassandra 1.2 and later, the Bloom filter and compression offset map that store this metadata reside off-
heap, greatly increasing the capacity per node of data that Cassandra can handle efficiently. The partition
summary also resides off-heap.
About the off-heap row cache

Cassandra can store cached rows in native memory, outside the Java heap. This results in both a smaller
per-row memory footprint and reduced JVM heap requirements, which helps keep the heap size in the
sweet spot for JVM garbage collection performance.
136
Operations
Tuning Java garbage collection

Cassandra's GCInspector class logs information about garbage collection whenever a garbage collection
takes longer than 200ms. Garbage collections that occur frequently and take a moderate length of time
to complete (such as ConcurrentMarkSweep taking a few seconds), indicate that there is a lot of garbage
collection pressure on the JVM. Remedies include adding nodes, lowering cache sizes, or adjusting the
JVM options regarding garbage collection.
JMX options
Cassandra exposes a number of statistics and management operations via Java Management Extensions
(JMX). Java Management Extensions (JMX) is a Java technology that supplies tools for managing and
monitoring Java applications and services. Any statistic or operation that a Java application has exposed
as an MBean can then be monitored or manipulated using JMX. JConsole, the nodetool utility, and
DataStax OpsCenter are examples of JMX-compliant management tools.
By default, you can modify the following properties in the conf/cassandra-env.sh file to configure JMX to
listen on port 7199 without authentication.
• com.sun.management.jmxremote.port
The port on which Cassandra listens from JMX connections.
• com.sun.management.jmxremote.ssl
Enable/disable SSL for JMX.
• com.sun.management.jmxremote.authenticate
Enable/disable remote authentication for JMX.
• -Djava.rmi.server.hostname
Sets the interface hostname or IP that JMX should use to connect. Uncomment and set if you are
having trouble connecting.

conf/cassandra.yaml
Purging gossip state on a node

Correcting a problem in the gossip state.
About this task

Gossip information is persisted locally by each node to use immediately on node restart without having to
wait for gossip communications.
Procedure
In the unlikely event you need to correct a problem in the gossip state:
1. Using MX4J or JConsole, connect to the node's JMX port and then use the JMX method
Gossiper.unsafeAssassinateEndpoints(ip_address) to assassinate the problem node.
This takes a few seconds to complete so wait for confirmation that the node is deleted.
2. If the JMX method above doesn't solve the problem, stop your client application from sending writes to
the cluster.
3. Take the entire cluster offline:
137
Operations
a) Drain each node.
$ nodetool options drain

b) Stop each node:
• Package installations:

• Tarball installations:

4. Clear the data from the peers directory:
$ sudo rm -r /var/lib/cassandra/data/system/peers/*
5. Clear the gossip state when the node starts:
• For tarball installations, you can use a command line option or edit the cassandra-env.sh. To use
the command line:
$ install_location/bin/cassandra -Dcassandra.load_ring_state=false
• For package installations or if you are not using the command line option above, add the following
line to the cassandra-env.sh file:
JVM_OPTS="$JVM_OPTS -Dcassandra.load_ring_state=false"
• Package installations: /usr/share/cassandra/cassandra-env.sh

• Tarball installations: install_location/conf/cassandra-env.sh
6. Bring the cluster online one node at a time, starting with the seed nodes.

$ bin/cassandra
What to do next
Remove the line you added in the cassandra-env.sh file.
Repairing nodes
Node repair makes data on a replica consistent with data on other nodes.
Node repair makes data on a replica consistent with data on other nodes. Anti-entropy node repairs are
important for every Cassandra cluster. Frequent data deletions and a node going down are common
causes of inconsistency. You use the nodetool repair command to repair a node. There are several types
of node repair:
• Sequential - One node is repaired after another, and done in full, all SSTables are repaired (default).
• Incremental - Persists already repaired data, which allows the repair process to stay performant and
lightweight as datasets grow providing repairs are run frequently.
• Partitioner range - Repairs only the first range returned by the partitioner for a node. This repair type
operates on each node in the cluster in succession without duplicating work
You can combine repair options, such as parallel and incremental repair. This combination does an
incremental repair to all nodes at the same time. You can restrict repairs to local or other data centers or to
138
Operations
nodes between a certain token range. You can specify which nodes have the good data for replacing the
outdated data.
Incremental repair
The repair process involves computing a Merkle tree for each range of data on that node. The Merkle tree
is a binary tree of hashes used by Cassandra for calculating the differences in datasets between nodes in
a cluster. Each node involved in the repair has to construct its Merkle tree from all the SSTables it stores,
making the calculation resource intensive.
To reduce the expense of constructing trees, Cassandra 2.1 introduces incremental repair. An incremental
repair makes already repaired data persistent, and only calculates a Merkle tree for unrepaired SSTables.
Reducing the size of the Merkle tree improves the performance of the incremental repair process,
assuming repairs are run frequently.
Incremental repairs begin with the repair leader sending out a prepare message to its peers. Each node
builds a Merkle tree from the unrepaired SSTables. After the leader receives a Merkle tree from each node,
the leader compares the trees and issues streaming requests. Finally, the leader issues an anticompaction
command.
For more information about incremental repairs, see the "More efficient repairs" article.
Anticompaction
Anticompaction in Cassandra 2.1 is the process of segregating repaired and unrepaired ranges into
separate SSTables unless the SSTable fits entirely within the repaired range. If the SSTable fits within the
repaired range, Cassandra just updates the SSTable metadata.
Anticompaction occurs after an incremental repair. Cassandra performs anticompaction only on the
SSTables that have a range of unrepaired data. If all node ranges are repaired, anticompaction does not
139
Operations
need to rewrite any data. During anticompaction, size/date-tiered compaction and leveled compaction
handle the segregation of the data differently.
• Size-tiered and date-tiered compaction splits repaired and unrepaired into separate pools for separate
compactions. A major compaction generates two SSTables, one containing repaired data and one
containing unrepaired data.
• Leveled compaction performs size-tiered compaction on unrepaired data. After repair completes,
Cassandra moves data from the set of unrepaired SSTables to L0.
Sequential versus parallel repair

By default, Cassandra 2.1 does a sequential repair of all SSTables, constructing the Merkle trees for
one node after the other. Cassandra takes a snapshot of each replica immediately and then sequentially
repairs each replica from the snapshots. For example, if you have RF=3 and A, B and C represents three
replicas, this command takes a snapshot of each replica immediately and then sequentially repairs each
replica from the snapshots (A<->B, A<->C, B<->C).
A parallel repair does the repair of nodes A, B, and C all at once. During this type of repair, the dynamic
snitch maintains performance for your application using a replica in the snapshot that is not undergoing
repair.
Snapshots are hardlinks to existing SSTables. Snapshots are immutable and require almost no disk space.
Repair requires intensive disk I/O because validation compaction occurs during Merkle tree construction.
For any given replica set, only one replica at a time performs the validation compaction.
Partitioner range repair

Using the nodetool repair -pr (–partitioner-range) option repairs only the first range returned by the
partitioner for a node. Other replicas for that range still have to perform the Merkle tree calculation, causing
a validation compaction. Because all the replicas are compacting at the same time, all the nodes may be
slow to respond for that portion of the data. This type of repair operates on each node in the cluster in
succession without duplicating work.
Merkle trees do not have infinite resolution and Cassandra makes a tradeoff between the size and space.
Currently, Cassandra uses a fixed depth of 15 for the tree (32K leaf nodes). For a node containing a million
partitions with one damaged partition, about 30 partitions are streamed, which is the number that fall into
each of the leaves of the tree. The problem gets worse when more partitions exist per node, and results
in significant disk space usage and needless compaction. Generally, partitioner range repairs are not
recommended.
Repairing a subrange
To mitigate overstreaming, you can use subrange repair, but generally subrange repairs are not
recommended. A subrange repair does a portion of the data belonging to the node. Because the Merkle
tree precision is fixed, this effectively increases the overall precision.
To use subrange repair:
1. Use the Java describe_splits call to ask for a split containing 32K partitions.
2. Iterate throughout the entire range incrementally or in parallel. This completely eliminates the
overstreaming behavior and wasted disk usage overhead.
3. Pass the tokens you received for the split to the nodetool repair -st (–start-token) and -et (–end-
token) options.
4. Pass the -local (–in-local-dc) option to nodetool to repair only within the local data center. This reduces
the cross data-center transfer load.
Repair guidelines
Run repair in these situations:
140
Operations
• Daily if you run incremental repairs, weekly if you run full repairs.
Note: Even if deletions never occur, schedule regular repairs. Setting a column to null is a
delete.
• During node recovery. For example, when bringing a node back into the cluster after a failure.
• On nodes containing data that is not read frequently.
• To update data on a node that has been down.
• To recover missing data or corrupted SSTables. A non-incremental is required.
• To minimize impact, do not invoke more than one repair at a time.
Guidelines for running routine node repair include:
• Schedule regular repair operations for low-usage hours.
• A full repair is recommended to eliminate the need for anticompaction.
• Migrating to incremental repairs is recommended if you use leveled compaction.
• The hard requirement for routine repair frequency is the value of gc_grace_seconds. Run a repair
operation at least once on each node within this time period. Following this important guideline ensures
that deletes are properly handled in the cluster.
• In systems that seldom delete or overwrite data, you can raise the value of gc_grace_seconds with
minimal impact to disk space. A higher value schedules wider intervals between repair operations.
• To mitigate heavy disk usage, configure nodetool compaction throttling options
(setcompactionthroughput and setcompactionthreshold) before running a repair.
Migrating to incremental repairs

Migrating to incremental repairs by using the sstablerepairedset utility is recommended only under the
following conditions:
• You are doing an incremental repair for the first time.
• You are using the leveled compaction strategy.
Full, sequential repairs are the default because until the first incremental repair, Cassandra does not
know the repaired state of SSTables. After an incremental repair, anticompaction marks SSTables as
repaired or not. If you use the leveled compaction strategy and perform an incremental repair for the first
time, Cassandra performs size-tiering on all SSTables because the repair/unrepaired status is unknown.
This operation can take a long time. To save time, migrate to incremental repair one node at a time. The
migration procedure, covered in the next section, uses utilities in the tools/bin directory of installations
other than RHEL and Debian:
• sstablemetadata for checking the repaired or unrepaired status of an SSTable
• sstablerepairedset for manually marking an SSTable as repaired
The syntax of these commands is:
sstablemetadata <sstable filenames>
sstablerepairedset [--is-repaired | --is-unrepaired] [-f <sstable-list> |

<sstables>]
In Cassandra 2.1.1, sstablerepairedset can take as arguments a list of SSTables on the command
line or a file SSTables with a "-f" flag.
Note: In RHEL and Debian installations, you must install the tools packages.
This example shows how to use sstablerepairedset to clear the repaired state of an SSTable, rendering the
SSTable unrepaired.
1. Stop the node.
141
Operations
2. Run this command:
sstablerepairedset --is-unrepaired -f list_of_sstable_names.txt

3. Restart the node.
All data is changed to an unrepaired state.
Procedure for migrating to incremental repairs

To migrate to incremental repair, one node at a time:
1. Disable compaction on the node using nodetool disableautocompaction.
2. Run the default full, sequential repair.
3. Stop the node.
4. Use the tool sstablerepairedset to mark all the SSTables that were created before you disabled
compaction.
5. Restart cassandra
SSTables remain in a repaired state after running a full, but not a partition range, repair if you make no
changes to the SSTables.
Adding or removing nodes, data centers, or clusters

Topics for adding or removing nodes, data centers, or clusters.
Adding nodes to an existing cluster

Steps to add nodes when using virtual nodes.
About this task

Virtual nodes (vnodes) greatly simplify adding nodes to an existing cluster:
• Calculating tokens and assigning them to each node is no longer required.
• Rebalancing a cluster is no longer necessary because a node joining the cluster assumes responsibility
for an even portion of the data.
For a detailed explanation about how vnodes work, see Virtual nodes.
If you are using racks, you can safely bootstrap two nodes at a time when both nodes are on the same
rack.
Note: If you do not use vnodes, use OpsCenter to add capacity or see Adding or replacing single-
token nodes.
Procedure
Be sure to install the same version of Cassandra as is installed on the other nodes in the cluster. See
Installing prior releases.
1. Install Cassandra on the new nodes, but do not start Cassandra.
If you used the Debian install, Cassandra starts automatically and you must stop the node and clear the
data.
2. Set the following properties in the cassandra.yaml and, depending on the snitch, the cassandra-
topology.properties or cassandra-rackdc.properties configuration files:
• auto_bootstrap - If this option has been set to false, you must set it to true. This option is not listed in
the default cassandra.yaml configuration file and defaults to true.
• cluster_name - The name of the cluster the new node is joining.
142
Operations
• listen_address/broadcast_address - Can usually be left blank. Otherwise, use IP address or host

name that other Cassandra nodes use to connect to the new node.
• endpoint_snitch - The snitch Cassandra uses for locating nodes and routing requests.
• num_tokens - The number of vnodes to assign to the node. If the hardware capabilities vary among
the nodes in your cluster, you can assign a proportional number of vnodes to the larger machines.
• seed_provider - The -seeds list in this setting determines which nodes the new node should contact
to learn about the cluster and establish the gossip process.
Note: Seed nodes cannot bootstrap. Make sure the new node is not listed in the -seeds list.
Do not make all nodes seed nodes. Please read Internode communications (gossip).
• Change any other non-default settings you have made to your existing cluster in the
cassandra.yaml file and cassandra-topology.properties or cassandra-
rackdc.properties files. Use the diff command to find and merge (by head) any differences
between existing and new nodes.

topology.properties
topology.properties

rackdc.properties
rackdc.properties

conf/cassandra.yaml
3. Use nodetool status to verify that the node is fully bootstrapped and all other nodes are up (UN) and not
in any other state.
4. After all new nodes are running, run nodetool cleanup on each of the previously existing nodes to
remove the keys that no longer belong to those nodes. Wait for cleanup to complete on one node
before running nodetool cleanup on the next node.
Cleanup can be safely postponed for low-usage hours.
Note: Consistency issues might result when more than one node is added at a time. To assess
the risks to your environment, see JIRA issues CASSANDRA-2434 and CASSANDRA-7069.
Although adding multiple nodes at the same time is not a best practice, you can reduce data
consistency risks by following these steps:
1. Before you add the nodes, turn off consistent range movements with the -
Dconsistent.rangemovement=false property.
Package installations
Add the following option to the /usr/share/cassandra/cassandra-env.sh file:
JVM_OPTS="$JVM_OPTS -Dconsistent.rangemovement=false
Tarball installations
143
Operations
Start Cassandra with this option:
$ bin/cassandra -Dconsistent.rangemovement=false
2. Allow two minutes between node initializations.

3. After the nodes are added, turn consistent range movement back on.
Adding a data center to a cluster

Steps for adding a data center to an existing cluster.
About this task

Steps for adding a data center to an existing cluster.

conf/cassandra.yaml
Procedure
1. Ensure that you are using NetworkTopologyStrategy for all of your keyspaces.
2. For each node, set the following properties in the cassandra.yaml file:
a) Add (or edit) auto_bootstrap: false.
By default, this setting is true and not listed in the cassandra.yaml file. Setting this parameter to
false prevents the new nodes from attempting to get all the data from the other nodes in the data
center. When you run nodetool rebuild in the last step, each node is properly mapped.
b) Set other properties, such as -seeds and endpoint_snitch, to match the cluster settings.
For more guidance, see Initializing a multiple node cluster (multiple data centers).
c) If you want to enable vnodes, set num_tokens.
The recommended value is 256. Do not set the initial_token parameter.
3. Update the relevant property file for snitch used on all servers to include the new nodes. You do not
need to restart.
• GossipingPropertyFileSnitch: cassandra-rackdc.properties
• PropertyFileSnitch: cassandra-topology.properties
4. Ensure that your clients are configured correctly for the new cluster:
• If your client uses the DataStax Java, C#, or Python driver, set the load-balancing policy to
DCAwareRoundRobinPolicy (Java, C#, Python).
• If you are using another client such as Hector, make sure it does not auto-detect the new nodes so
that they aren't contacted by the client until explicitly directed. For example if you are using Hector,
use sethostConfig.setAutoDiscoverHosts(false);. If you are using Astyanax, use
ConnectionPoolConfigurationImpl.setLocalDatacenter("<data center name">) to
ensure you are connecting to the specified data center.
144
Operations
• If you are using Astyanax 2.x, with integration with the DataStax Java Driver 2.0,
you can set the load-balancing policy to DCAwareRoundRobinPolicy by calling
JavaDriverConfigBuilder.withLoadBalancingPolicy().
AstyanaxContext<Keyspace> context = new AstyanaxContext.Builder()

...
.withConnectionPoolConfiguration(new JavaDriverConfigBuilder()
.withLoadBalancingPolicy(new TokenAwarePolicy(new
DCAwareRoundRobinPolicy()))
.build())
...
5. If using a QUORUM consistency level for reads or writes, check the LOCAL_QUORUM or
EACH_QUORUM consistency level to see if the level meets your requirements for multiple data
centers.
6. Start Cassandra on the new nodes.
7. After all nodes are running in the cluster:
a) Change the keyspace properties to specify the desired replication factor for the new data center.
For example, set strategy options to DC1:2, DC2:2.
For more information, see ALTER KEYSPACE.
b) Run nodetool rebuild specifying the existing data center on all nodes in the new data center:
nodetool rebuild -- name_of_existing_data_center
Otherwise, requests to the new data center with LOCAL_ONE or ONE consistency levels may fail if
the existing data centers are not completely in-sync.
You can run rebuild on one or more nodes at the same time. The choices depends on whether your
cluster can handle the extra IO and network pressure of running on multiple nodes. Running on one
node at a time has the least impact on the existing cluster.
Attention: If you don't specify the existing data center in the command line, the new nodes
will appear to rebuild successfully, but will not contain any data.
8. Change to true or remove auto_bootstrap: false in the cassandra.yaml file.
Returns this parameter to its normal setting so the nodes can get all the data from the other nodes in
the data center if restarted.
Replacing a dead node or dead seed node

Steps to replace a node that has died for some reason, such as hardware failure.
About this task

Steps to replace a node that has died for some reason, such as hardware failure. Prepare and start the
replacement node, then attach it to the cluster. After the replacement node is running in the cluster, remove
the dead node.
Replacing a dead seed node

1. Promote an existing node to a seed node by adding its IP address to -seeds list and remove (demote)
the IP address of the dead seed node from the cassandra.yaml file for each node in the cluster.
2. Replace the dead node, as described in the next section.
145
Operations
Replacing a dead node
Procedure
1. Confirm that the node is dead using nodetool status:
The nodetool command shows a down status for the dead node (DN):
2. Note the Address of the dead node; it is used in step 5.

3. Install Cassandra on the new node, but do not start Cassandra.
If you used the Debian/Ubuntu install, Cassandra starts automatically and you must and stop the node
and clear the data.
4. Set the following properties in the cassandra.yaml and, depending on the snitch, the cassandra-
topology.properties or cassandra-rackdc.properties configuration files:
• auto_bootstrap - If this option has been set to false, you must set it to true. This option is not listed in
the default cassandra.yaml configuration file and defaults to true.
• cluster_name - The name of the cluster the new node is joining.
• listen_address/broadcast_address - Can usually be left blank. Otherwise, use IP address or host
name that other Cassandra nodes use to connect to the new node.
• endpoint_snitch - The snitch Cassandra uses for locating nodes and routing requests.
• num_tokens - The number of vnodes to assign to the node. If the hardware capabilities vary among
the nodes in your cluster, you can assign a proportional number of vnodes to the larger machines.
• seed_provider - The -seeds list in this setting determines which nodes the new node should contact
to learn about the cluster and establish the gossip process.
Note: Seed nodes cannot bootstrap. Make sure the new node is not listed in the -seeds list.
Do not make all nodes seed nodes. Please read Internode communications (gossip).
• Change any other non-default settings you have made to your existing cluster in the
cassandra.yaml file and cassandra-topology.properties or cassandra-
rackdc.properties files. Use the diff command to find and merge (by head) any differences
between existing and new nodes.
5. Start the replacement node with the replace_address option:
• Package installations: Add the following option to /usr/share/cassandra/cassandra-env.sh
file:
JVM_OPTS="$JVM_OPTS -Dcassandra.replace_address=address_of_dead_node
• Tarball installations: Start Cassandra with this option:
$ sudo bin/cassandra -Dcassandra.replace_address=address_of_dead_node

6. If using a packaged install, after the new node finishes bootstrapping, remove the option you added in
step 5.
What to do next
• Remove the old node's IP address from the cassandra-topology.properties or cassandra-
rackdc.properties file.
146
Operations
Caution: Wait at least 72 hours to ensure that old node information is removed from gossip. If
removed from the property file too soon, problems may result.
• Remove the node.
Caution: Wait at least 72 hours to ensure that old node information is removed from gossip. If
removed from the property file too soon, problems may result.

topology.properties
topology.properties

rackdc.properties
rackdc.properties

conf/cassandra.yaml
Replacing a running node

Steps to replace a node with a new node, such as when updating to newer hardware or performing
proactive maintenance.
About this task

Steps to replace a node with a new node, such as when updating to newer hardware or performing
proactive maintenance.
You must prepare and start the replacement node, integrate it into the cluster, and then decommision the
old node.
Note: To change the IP address of a node, simply change the IP of node and then restart
Cassandra. If you change the IP address of a seed node, you must update the -seeds parameter in
the seed_provider for each node in the cassandra.yaml file.

conf/cassandra.yaml
Procedure
1. Prepare and start the replacement node, as described in Adding nodes to an existing cluster.
147
Operations
Note: If not using vnodes, see Adding or replacing single-token nodes.

2. Confirm that the replacement node is alive:
• Run nodetool ring if not using vnodes.
• Run nodetool status if using vnodes.
The status should show:
• nodetool ring: Up
• nodetool status: UN
3. Note the Host ID of the node; it is used in the next step.
4. Using the Host ID of the original node, decommission the original node from the cluster using the
nodetool decommission command.
Moving a node from one rack to another

A common task is moving a node from one rack to another. For example, when using
GossipPropertyFileSnitch, a common error is mistakenly placing a node in the wrong rack. To correct the
error, use one of the following procedures.
• The preferred method is to decommission the node and re-add it to the correct rack and data center.
• This method takes longer to complete than the alternative method. Data is moved that the
decommissioned node doesn't need anymore. Then the node gets new data while bootstrapping.
The alternative method does both operations simultaneously.
• An alternative method is to update the node's topology and restart the node. Once the node is up, run a
full repair on the cluster.
Note: This method is not preferred because until the repair is completed. the node might blindly
handle requests for data the node doesn't yet have.
Decommissioning a data center

Steps to properly remove a data center so no information is lost.
About this task

Steps to properly remove a data center so no information is lost.
Procedure
1. Make sure no clients are still writing to any nodes in the data center.
2. Run a full repair with nodetool repair.
This ensures that all data is propagated from the data center being decommissioned.
3. Change all keyspaces so they no longer reference the data center being removed.
4. Run nodetool decommission on every node in the data center being removed.
Removing a node
Reduce the size of a data center.
About this task

Use these instructions when you want to remove nodes to reduce the size of your cluster, not for replacing
a dead node.
Attention: If you are not using virtual nodes (vnodes), you must rebalance the cluster.
148
Operations
Procedure
1. Check whether the node is up or down using nodetool status:
The nodetool command shows the status of the node (UN=up, DN=down):
2. If the node is up, run nodetool decommission.

This assigns the ranges that the node was responsible for to other nodes and replicates the data
appropriately.
Use nodetool netstats to monitor the progress.
3. If the node is down:
• If the cluster uses vnodes, remove the node using the nodetool removenode command.
• If the cluster does not use vnodes, before running the nodetool removenode command, adjust your
tokens to evenly distribute the data across the remaining nodes to avoid creating a hot spot.
4. If the node does not stop streaming data to other nodes because gossip has stale state data for the
node, assassinate the node.
Switching snitches
Steps for switching snitches.
About this task

Because snitches determine how Cassandra distributes replicas, the procedure to switch snitches depends
on whether or not the topology of the cluster will change:
• If data has not been inserted into the cluster, there is no change in the network topology. This means
that you only need to set the snitch; no other steps are necessary.
• If data has been inserted into the cluster, it's possible that the topology has changed and you will need
to perform additional steps.
A change in topology means that there is a change in the data centers and/or racks where the nodes are
placed. Topology changes may occur when the replicas are placed in different places by the new snitch.
Specifically, the replication strategy places the replicas based on the information provided by the new
snitch. The following examples demonstrate the differences:
• No topology change
Suppose you have 5 nodes using the SimpleSnitch in a single data center and you change to 5 nodes
in 1 data center and 1 rack using a network snitch such as the GossipingPropertyFileSnitch.
• Topology change
in 2 data centers using the PropertyFileSnitch.
Note: If splitting from one data center to two, you need to change the schema for the keyspace
that are splitting. Additionally, the data center names must change accordingly.
• Topology change
in 1 data center and 2 racks using the RackInferringSnitch.
Procedure
1. Create a properties file with data center and rack information.
149
Operations
• cassandra-rackdc.properties
GossipingPropertyFileSnitch, Ec2Snitch, and Ec2MultiRegionSnitch only.
• cassandra-topology.properties
All other network snitches.
2. Copy the cassandra-rackdc.properties or cassandra-topology.properties file to the Cassandra
configuration directory on all the cluster's nodes. They won't be used until the new snitch is enabled.

topology.properties
topology.properties

rackdc.properties
rackdc.properties

conf/cassandra.yaml
3. Change the snitch for each node in the cluster in the node's cassandra.yaml file. For example:
4. If the topology has not changed, you can restart each node one at a time.
Any change in the cassandra.yaml file requires a node restart.
5. If the topology of the network has changed:
a) Shut down all the nodes, then restart them.
b) Run a sequential repair and nodetool cleanup on each node.
Edge cases for transitioning or migrating a cluster

Unusual migration scenarios without interruption of service.
About this task

The information in this topic is intended for the following types of scenarios (without any interruption of
service):
• Transition a cluster on EC2 to a cluster on Amazon virtual private cloud (VPC).
• Migrate from a cluster when the network separates the current cluster from the future location.
• Migrate from an early Cassandra cluster to a recent major version.
Procedure
The following method ensures that if something goes wrong with the new cluster, you still have the existing
cluster until you no longer need it.
150
Operations
1. Set up and configure the new cluster as described in Provisioning a new cluster.
Note: If you're not using vnodes, be sure to configure the token ranges in the new nodes to
match the ranges in the old cluster.
2. Set up the schema for the new cluster using CQL.
3. Configure your client to write to both clusters.
Depending on how the writes are done, code changes may be needed. Be sure to use identical
consistency levels.
4. Ensure that the data is flowing to the new nodes so you won't have any gaps when you copy the
snapshots to the new cluster in step 6.
5. Snapshot the old EC2 cluster.
6. Copy the data files from your keyspaces to the nodes.
•If not using vnodes and the if the node ratio is 1:1, it's simpler and more efficient to simply copy the
data files to their matching nodes.
• If the clusters are different sizes or if you are using vnodes, use the Cassandra bulk loader
(sstableloader) (sstableloader).
7. You can either switch to the new cluster all at once or perform an incremental migration.
For example, to perform an incremental migration, you can set your client to designate a percentage of
the reads that go to the new cluster. This allows you to test the new cluster before decommissioning the
old cluster.
8. Decommission the old cluster:
a) Remove the cluster from the OpsCenter.
b) Remove the nodes.
Adding or replacing single-token nodes

Steps for adding or replacing nodes in single-token architecture clusters.
About this task

This topic applies only to clusters using single-token architecture, not vnodes.
About adding Capacity to an Existing Cluster

Cassandra allows you to add capacity to a cluster by introducing new nodes to the cluster in stages and by
adding an entire data center. When a new node joins an existing cluster, it needs to know:
• Its position in the ring and the range of data it is responsible for, which is assigned by the initial_token
and the partitioner.
• The seed nodes to contact for learning about the cluster and establish the gossip process.
• The name of the cluster it is joining and how the node should be addressed within the cluster.
• Any other non-default settings made to cassandra.yaml on your existing cluster.
When you add one or more nodes to a cluster, you must calculate the tokens for the new nodes. Use one
of the following approaches:
Add capacity by doubling the cluster size
Adding capacity by doubling (or tripling or quadrupling) the number of nodes is less complicated when
assigning tokens. Existing nodes can keep their existing token assignments, and new nodes are assigned
tokens that bisect (or trisect) the existing token ranges. For example, when you generate tokens for six
nodes, three of the generated token values will be the same as if you generated for three nodes. To clarify,
you first obtain the token values that are already in use, and then assign the newly calculated token values
to the newly added nodes.
Recalculate new tokens for all nodes and move nodes around the ring
151
Operations
When increases capacity by a non-uniform number of nodes, you must recalculate tokens for the entire
cluster, and then use nodetool move to assign the new tokens to the existing nodes. After all nodes are
restarted with their new token assignments, run a nodetool cleanup to remove unused keys on all nodes.
These operations are resource intensive and should be done during low-usage times.
Add one node at a time and leave the initial_token property empty
When the initial_token is empty, Cassandra splits the token range of the heaviest loaded node and places
the new node into the ring at that position. This approach is unlikely to result in a perfectly balanced ring,
but will alleviate hot spots.
Adding Nodes to a Cluster

1. Install Cassandra on the new nodes, but do not start them.
2. Calculate the tokens for the nodes based on the expansion strategy you are using the Token
Generating Tool.You can skip this step if you want the new nodes to automatically pick a token range
when joining the cluster.
3. Set the cassandra.yaml for the new nodes.
4. Set the initial_token according to your token calculations (or leave it unset if you want the new nodes to
automatically pick a token range when joining the cluster).
5. Start Cassandra on each new node. Allow two minutes between node initializations. You can monitor
the startup and data streaming process using nodetool netstats.
6. After the new nodes are fully bootstrapped, assign the new initial_token property value to the nodes that
required new tokens, and then run nodetool move new_token, one node at a time.
7. After all nodes have their new tokens assigned, run nodetool cleanup one node at a time for each node.
Wait for cleanup to complete before doing the next node. This step removes the keys that no longer
belong to the previously existing nodes.
Note: Cleanup may be safely postponed for low-usage hours.
Adding a Data Center to a Cluster

Before starting this procedure, please read the guidelines in Adding Capacity to an Existing Cluster above.
1. Ensure that you are using NetworkTopologyStrategy for all of your keyspaces.
2. For each new node, edit the configuration properties in the cassandra.yaml file:
• Set auto_bootstrap to False.
• Set the initial_token. Be sure to offset the tokens in the new data center, see Generating
tokens.
• Set the cluster name.
• Set any other non-default settings.
• Set the seed lists. Every node in the cluster must have the same list of seeds and include at least
one node from each data center. Typically one to three seeds are used per data center.
3. Update either the cassandra-topology.properties (PropertyFileSnitch) or cassandra-
rackdc.properties (GossipingPropertyFileSnitch) on all servers to include the new nodes. You do
not need to restart.
The location of the conf directory depends on the type of installation:
• Package installations: /etc/cassandra/conf
4. Ensure that your client does not auto-detect the new nodes so that they aren't contacted by the client
until explicitly directed. For example in Hector, set
hostConfig.setAutoDiscoverHosts(false);
152
Operations
5. If using a QUORUM consistency level for reads or writes, check the LOCAL_QUORUM or
EACH_QUORUM consistency level to make sure that the level meets the requirements for multiple data
centers.
6. Start the new nodes.
7. After all nodes are running in the cluster:
a. Change the replication factor for your keyspace for the expanded cluster.
b. Run nodetool rebuild on each node in the new data center.
Replacing a Dead Node

1. Confirm that the node is dead using the nodetool ring command on any live node in the cluster.
The nodetool ring command shows a Down status for the token value of the dead node:
2. Install Cassandra on the replacement node.

3. Remove any preexisting Cassandra data on the replacement node:
$ sudo rm -rf /var/lib/cassandra/*

4. Set auto_bootstrap: true. (If auto_bootstrap is not in the cassandra.yaml file, it automatically
defaults to true.)
5. Set the initial_token in the cassandra.yaml file to the value of the dead node's token -1. Using
the value from the above graphic, this is 28356863910078205288614550619314017621-1:
initial_token: 28356863910078205288614550619314017620
6. Configure any non-default settings in the node's cassandra.yaml to match your existing cluster.
7. Start the new node.
8. After the new node has finished bootstrapping, check that it is marked up using the nodetool ring
command.
9. Run nodetool repair on each keyspace to ensure the node is fully consistent. For example:
$ nodetool repair -h 10.46.123.12 keyspace_name

10.Remove the dead node.

conf/cassandra.yaml
153
Backing up and restoring data

Cassandra backs up data by taking a snapshot of all on-disk data files (SSTable files) stored in the data
directory.
About snapshots
A brief description of how Cassandra backs up data.
Cassandra backs up data by taking a snapshot of all on-disk data files (SSTable files) stored in the data
directory. You can take a snapshot of all keyspaces, a single keyspace, or a single table while the system
is online.
Using a parallel ssh tool (such as pssh), you can snapshot an entire cluster. This provides an eventually
consistent backup. Although no one node is guaranteed to be consistent with its replica nodes at the time
a snapshot is taken, a restored snapshot resumes consistency using Cassandra's built-in consistency
mechanisms.
After a system-wide snapshot is performed, you can enable incremental backups on each node to backup
data that has changed since the last snapshot: each time an SSTable is flushed, a hard link is copied into a
/backups subdirectory of the data directory (provided JNA is enabled).
Note: If JNA is enabled, snapshots are performed by hard links. If not enabled, I/O activity
increases as the files are copied from one location to another, which significantly reduces efficiency.
Taking a snapshot
Steps for taking a global snapshot or per node.
About this task

Snapshots are taken per node using the nodetool snapshot command. To take a global snapshot, run the
nodetool snapshot command using a parallel ssh utility, such as pssh.
A snapshot first flushes all in-memory writes to disk, then makes a hard link of the SSTable files for each
keyspace. You must have enough free disk space on the node to accommodate making snapshots of your
data files. A single snapshot requires little disk space. However, snapshots can cause your disk usage to
grow more quickly over time because a snapshot prevents old obsolete data files from being deleted. After
the snapshot is complete, you can move the backup files to another location if needed, or you can leave
them in place.
Note: Cassandra can only restore data from a snapshot when the table schema exists. It is
recommended that you also backup the schema.
Procedure
Run the nodetool snapshot command, specifying the hostname, JMX port, and keyspace. For example:
$ nodetool -h localhost -p 7199 snapshot mykeyspace
Results
The snapshot is created in data_directory_location/keyspace_name/table_name-UUID/
snapshots/snapshot_name directory. Each snapshot directory contains numerous .db files that
contain the data at the time of the snapshot.
154
For example:
/var/lib/cassandra/data/mykeyspace/users-081a1500136111e482d09318a3b15cc2/
snapshots/1406227071618/mykeyspace-users-ka-1-Data.db
install_location/data/data/mykeyspace/
users-081a1500136111e482d09318a3b15cc2/snapshots/1406227071618/mykeyspace-
users-ka-1-Data.db
Deleting snapshot files

Steps to delete snapshot files.
About this task

When taking a snapshot, previous snapshot files are not automatically deleted. You should remove old
snapshots that are no longer needed.
The nodetool clearsnapshot command removes all existing snapshot files from the snapshot directory of
each keyspace. You should make it part of your back-up process to clear old snapshots before taking a
new one.
Procedure
To delete all snapshots for a node, run the nodetool clearsnapshot command. For example:
$ nodetool -h localhost -p 7199 clearsnapshot
To delete snapshots on all nodes at once, run the nodetool clearsnapshot command using a parallel
ssh utility.
Enabling incremental backups

Steps to enable incremental backups. When incremental backups are enabled, Cassandra hard-links each
flushed SSTable to a backups directory under the keyspace data directory.
About this task

When incremental backups are enabled (disabled by default), Cassandra hard-links each flushed SSTable
to a backups directory under the keyspace data directory. This allows storing backups offsite without
transferring entire snapshots. Also, incremental backups combine with snapshots to provide a dependable,
up-to-date backup mechanism.
As with snapshots, Cassandra does not automatically clear incremental backup files. DataStax
recommends setting up a process to clear incremental backup hard-links each time a new snapshot is
created.

conf/cassandra.yaml
155
Procedure
Edit the cassandra.yaml configuration file on each node in the cluster and change the value of
incremental_backups to true.
Restoring from a snapshot

Methods for restoring from a snapshot.
About this task

Restoring a keyspace from a snapshot requires all snapshot files for the table, and if using incremental
backups, any incremental backup files created after the snapshot was taken.
Generally, before restoring a snapshot, you should truncate the table. If the backup occurs before the
delete and you restore the backup after the delete without first truncating, you do not get back the original
data (row). Until compaction, the tombstone is in a different SSTable than the original row, so restoring
the SSTable containing the original row does not remove the tombstone and the data still appears to be
deleted.
Cassandra can only restore data from a snapshot when the table schema exists. If you have not backed up
the schema, you can do the either of the following:
• Method 1
1. Restore the snapshot, as described below.
2. Recreate the schema.
• Method 2
1. Recreate the schema.
2. Restore the snapshot, as described below.
3. Run nodetool refresh.
Procedure
You can restore a snapshot in several ways:
• Use the sstableloader tool.
• Copy the snapshot SSTable directory (see Taking a snapshot) to the
data/keyspace/table_name-UUID directory and then call the JMX method
loadNewSSTables() in the column family MBean for each column family through JConsole. You can
use nodetool refresh instead of the loadNewSSTables() call.
The location of the data directory depends on the type of installation:
The tokens for the cluster you are restoring must match exactly the tokens of the backed-up cluster
at the time of the snapshot. Furthermore, the snapshot must be copied to the correct node with
matching tokens matching. If the tokens do not match, or the number of nodes do not match, use the
sstableloader procedure.
• Use the Node Restart Method described below.
156
Node restart method

Steps for restoring a snapshot. This method requires shutting down and starting nodes.
About this task

If restoring a single node, you must first shutdown the node. If restoring an entire cluster, you must shut
down all nodes, restore the snapshot data, and then start all nodes again.
Note: Restoring from snapshots and incremental backups temporarily causes intensive CPU and I/
O activity on the node being restored.
The location of the commitlog directory depends on the type of installation:
Package installations /var/lib/cassandra/commitlog

Tarball installations install_location/data/commitlog
Procedure
1. Shut down the node.
2. Clear all files in the commitlog directory.
This prevents the commitlog replay from putting data back, which would defeat the purpose of restoring
data to a particular point in time.
3. Delete all *.db files in data_directory_location/keyspace_name/keyspace_name-
keyspace_name directory, but DO NOT delete the /snapshots and /backups subdirectories.
where data_directory_location is:
4. Locate the most recent snapshot folder in this directory:
data_directory_location/keyspace_name/table_name-UUID/
snapshots/snapshot_name
5. Copy its contents into this directory:
data_directory_location/keyspace_name/table_name-UUID directory.
6. If using incremental backups, copy all contents of this directory:
data_directory_location/keyspace_name/table_name-UUID/backups
7. Paste it into this directory:
data_directory_location/keyspace_name/table_name-UUID
8. Restart the node.
Restarting causes a temporary burst of I/O activity and consumes a large amount of CPU resources.
9. Run nodetool repair.
Restoring a snapshot into a new cluster

Steps for restoring a snapshot by recovering the cluster into another newly created cluster.
About this task

Suppose you want to copy a snapshot of SSTable data files from a three node Cassandra cluster with
vnodes enabled (256 tokens) and recover it on another newly created three node cluster (256 tokens). The
token ranges will not match so you need to specify the tokens for the new cluster that were used in the old
cluster.
157
Note: This procedure assumes you are familiar with restoring a snapshot and configuring and
initializing a cluster. If not, see Initializing a cluster.

conf/cassandra.yaml
Procedure
To recover the snapshot on the new cluster:
1. From the old cluster, retrieve the list of tokens associated with each node's IP:
$ nodetool ring | grep ip_address_of_node | awk '{print $NF ","}' | xargs

2. In the cassandra.yaml file for each node in the new cluster, add the list of tokens you obtained in the
previous step to the initial_token parameter using the same num_tokens setting as in the old cluster.
3. Make any other necessary changes in the cassandra.yaml and property files so that the new nodes
match the old cluster settings.
4. Clear the system table data from each new node:

This allows the new nodes to use the initial tokens defined in the cassandra.yaml when they restart.
5. Restore the SSTable files snapshotted from the old cluster onto the new cluster using the same
directories. Otherwise the new cluster does not have data to read in when you restart the nodes.
6. Start each node using the specified list of token ranges in cassandra.yaml:
initial_token: -9211270970129494930, -9138351317258731895,

-8980763462514965928, ...
This allows Cassandra to read the SSTable snapshot from the old cluster.
Recovering from a single disk failure using JBOD

Recovering from a single disk failure in a disk array using JBOD.
About this task

How to recover from a single disk failure in a disk array using JBOD (just a bunch of disks).
Node can restart

1. Stop Cassandra and shut down the node.
2. Replace the failed disk.
3. Start the node and Cassandra.
4. Run nodetool repair on the node.
Node cannot restart

If the node cannot restart, it is possible the system directory is corrupted. If the node cannot restart after
completing these steps, see Replacing a dead node or dead seed node.
If using the node uses vnodes:
158

3. On a healthy node run the following command:
$ nodetool ring | grep ip_address_of_node | awk ' {print $NF ","}' | xargs
4. On the node with the new disk, add the list of tokens from the previous step (separated by commas),
under initial_token in the cassandra.yaml file.
5. Clear each system directory for every functioning drive:
Assuming disk1 has failed and the data_file_directories setting in the cassandra.yaml for each drive
is:
-/mnt1/cassandra/data
Run the following commands:
$ rm -fr /mnt2/cassandra/data/system
7. Run nodetool repair.
8. After the node is fully integrated into the cluster, it is recommended to return to normal vnode settings:
• num_tokens: number_of_tokens
• #initial_token
If the node uses assigned tokens (single-token architecture):
3. Clear each system directory for every functioning drive:
Assuming disk1 has failed and the data_file_directories setting in the cassandra.yaml for each drive
is:
Run the following commands:
5. Run nodetool repair on the node.

conf/cassandra.yaml
159
Cassandra tools
Cassandra tools
Topics for Cassandra tools.
The nodetool utility

A command line interface for managing a cluster.
The nodetool utility is a command line interface for managing a cluster.
Command format
• Package installations: nodetool [(-h <host> | --host <host>)] [(-p <port> | --port
<port>)] [(-pwf <passwordFilePath> | --password-file <passwordFilePath>)]
[(-u <username> | --username <username>)] [(-pw <password> | --password
<password>)] <command> [<args>]
• Tarball installations: Execute the command from install_location/bin
• If a username and password for RMI authentication are set explicitly in the cassandra-env.sh file for
the host, then you must specify credentials.
The repair and rebuild commands can affect multiple nodes in the cluster.
Most nodetool commands operate on a single node in the cluster if -h is not used to identify one or
more other nodes. If the node from which you issue the command is the intended target, you do not
need the -h option to identify the target; otherwise, for remote invocation, identify the target node, or
nodes, using -h.
Getting nodetool help

nodetool help
Provides a listing of nodetool commands.
nodetool help command name
Provides help on a specific command. For example
$ nodetool help upgradesstables
nodetool cfhistograms
Provides statistics about a table that could be used to plot a frequency function.
Provides statistics about a table that could be used to plot a frequency function.
Synopsis
$ nodetool <options> cfhistograms -- <keyspace> <table>

• options are:
• ( -h | --host ) <host name> | <ip address>
• ( -p | --port ) <port number>
• ( -pw | --password ) <password >
• ( -u | --username ) <user name>
• ( -pwf <passwordFilePath | --password-file <passwordFilePath> )
• -- separates an option from an argument that could be mistaken for a option.
• keyspace is the name of a keyspace.
160
Cassandra tools
• table is the name of a table.
Synopsis Legend
• Angle brackets (< >) mean not literal, a variable
• Italics mean optional
• The pipe (|) symbol means OR or AND/OR
• Ellipsis (...) means repeatable
• Orange ( and ) means not literal, indicates scope
Description
The nodetool cfhistograms command provides statistics about a table, including read/write latency,
partition size, column count, and number of SSTables. The report covers all operations since the last time
you ran nodetool cfhistograms in this session. The use of the metrics-core library in Cassandra 2.1 makes
the output more informative and easier to understand.
Example
For example, to get statistics about the libout table in the libdata keyspace on Linux, use this command:
$ install_location/bin/nodetool cfhistograms libdata libout
Output is:
libdata/libout histograms
Percentile SSTables Write Latency Read Latency Partition Size
Cell Count
(micros) (micros) (bytes)
50% 0.00 39.50 36.00 1597

42
75% 0.00 49.00 55.00 1597
42
95% 0.00 95.00 82.00 8239
258
98% 0.00 126.84 110.42 17084
446
99% 0.00 155.13 123.71 24601
770
Min 0.00 3.00 3.00 1110
36
Max 0.00 50772.00 314.00 126934
3973
The output shows the percentile rank of read and write latency values, the partition size, and the cell count
for the table.
nodetool cfstats
Provides statistics about tables.
Provides statistics about tables.
Synopsis
$ nodetool <options> cfstats -i -- (<keyspace>.<table> ... ) -H

• options are:
161
Cassandra tools

• i ignores the following tables, providing only information about other Cassandra tables
• keyspace.table is one or more keyspace and table names in dot notation.
• H converts bytes to a human readable form: kilobytes (KB), megabytes (MB), gigabytes (GB), or
terabytes (TB). (Cassandra 2.1.1)
Synopsis Legend
Description
The nodetool cfstats command provides statistics about one or more tables. You use dot notation
to specify one or more keyspace and table names. If you do not specify a keyspace and table, Cassandra
provides statistics about all tables. If you use the -i option, Cassandra provides statistics about all tables
except the given ones. The use of the metrics-core library in Cassandra 2.1 makes the output more
informative and easier to understand.
This table describes the nodetool cfstats output.
Table 11: nodetool cfstats output
Name of Example Brief Related information

statistic value description
Keyspace libdata Name of the Keyspace and table
keyspace
Read count 11207 Number of
requests to
read tables
in the libdata
keyspace
since startup
Read latency 0.047. . . Latency OpsCenter alert metrics
ms reading the
tables in
the libdata
keyspace
Write count 17598 Number of Same as above
requests
to update
tables in
the libdata
keyspace
since startup
162
Cassandra tools

Write latency 0.053. . . Latency Same as above
ms writing tables
in the libdata
keyspace
Pending tasks 0 Tasks in the OpsCenter pending task metrics
queue for
reads, writes,
and cluster
operations of
tables in the
keyspace
Table libout Name of the
Cassandra
table
SSTable count 3 Number of How to use the SSTable counts metric and OpsCenter alert
SSTables metrics
containing
data from the
table
Space used 9592399 Space used Advanced system alert metrics
(live), bytes: by the table
(depends
on operating
system)
Space used 9592399 Same as Same as above
(total), bytes: above
Space used 0 Same
by snapshots occupied by
(total), bytes: backup data
SSTable 0.367. . . Fraction Types of compression option)
compression of data-
ratio representation
size resulting
from
compression
Memtable cell 1022550 Number of Cassandra memtable structure in memory
count cells (storage
engine rows
x columns)
of data in the
memtable
Memtable data 32028148 Size of the Same as above
size, bytes memtable
data
Memtable switch 3 Number of How memtables are measured article
count times a full
memtable
163
Cassandra tools

was
swapped
for an
empty one
(Increases
each
time the
memtable
for a table
is flushed to
disk)
Local read count 11207 Number of OpsCenter alert documentation
local read
requests for
the libout
table since
startup
Local read 0.048 ms Round Factors that affect read latency
latency trip time in
milliseconds
to complete
a request
to read the
libout table
Local write 17598 Number OpsCenter alert documentation
count of local
requests to
update the
libout the
table since
startup
Local write 0.054 ms Round Factors that affect write latency
latency trip time in
milliseconds
to complete
an update
to the libout
table
Pending tasks 0 Number of OpsCenter pending task metrics documentation
read, write,
and cluster
operations
that are
pending
Bloom filter false 0 Number Tuning bloom filters
positives of false
positives,
which occur
when the
bloom filter
164
Cassandra tools

said the row
existed, but
it actually
did not exist
in absolute
numbers
Bloom filter false 0.00000 Fraction of Same as above
ratio all bloom
filter checks
resulting in a
false positive
Bloom filter 11688 Size of bytes Same as above
space used, of the bloom
bytes filter data
Compacted 1110 Lower size Used to calculate what the approximate row cache size
partition limit for the should be. Multiply the reported row cache size, which is
minimum bytes partition the number of rows in the cache, by the compacted row
being mean size for every table and sum them.
compacted in
memory
Compacted 126934 Upper size Configurable in the cassandra.yaml in_memory_compaction
partition limit for _limit_in_mb
maximum bytes compacted
table rows.
Compacted 2730 The average
partition mean size of
bytes compacted
table rows
Average 0.0 Average
live cells per of cells
slice (last five scanned
minutes) by single
key queries
during the
last five
minutes
Average 0.0 Average of
tombstones per tombstones
slice (last five scanned
minutes) by single
key queries
during the
last five
minutes
Examples
This example shows an excerpt of the output of the command after flushing a table of library data to disk.
165
Cassandra tools
$ nodetool cfstats libdata.libout

Keyspace: libdata
Read Count: 11207
Read Latency: 0.047931114482020164 ms.
Write Count: 17598
Write Latency: 0.053502954881236506 ms.
Pending Flushes: 0
Table: libout
SSTable count: 3
Space used (live), bytes: 9088955
Space used (total), bytes: 9088955
Space used by snapshots (total), bytes: 0
SSTable Compression Ratio: 0.36751363892150946
Memtable cell count: 0
Memtable data size, bytes: 0
Memtable switch count: 3
Local read count: 11207
Local read latency: 0.048 ms
Local write count: 17598
Local write latency: 0.054 ms
Pending flushes: 0
Bloom filter false positives: 0
Bloom filter false ratio: 0.00000
Bloom filter space used, bytes: 11688
Compacted partition minimum bytes: 1110
Compacted partition maximum bytes: 126934
Compacted partition mean bytes: 2730
Average live cells per slice (last five minutes): 0.0
Average tombstones per slice (last five minutes): 0.0
Using the human-readable option
Using the human-readable -H option provides output in easier-to-read units than bytes. For example:
$ nodetool cfstats demodb.nhanes -H

Keyspace: demodb
Read Count: 0
Read Latency: NaN ms.
Write Count: 20050
Write Latency: 0.08548014962593516 ms.
Pending Flushes: 0
Table: nhanes
SSTable count: 1
Space used (live): 13.75 MB
Space used (total): 13.75 MB
Space used by snapshots (total): 0 bytes
SSTable Compression Ratio: 0.3064650643762481
Memtable cell count: 0
Memtable data size: 0 bytes
Memtable switch count: 1
Local read count: 0
Local read latency: NaN ms
Local write count: 20050
Local write latency: 0.085 ms
Pending flushes: 0
Bloom filter false positives: 0
Bloom filter false ratio: 0.00000
Bloom filter space used: 23.73 KB
Compacted partition minimum bytes: 1.87 KB
Compacted partition maximum bytes: 2.69 KB
Compacted partition mean bytes: 2.26 KB
Average live cells per slice (last five minutes): 0.0
Maximum live cells per slice (last five minutes): 0.0
Average tombstones per slice (last five minutes): 0.0
166
Cassandra tools
Maximum tombstones per slice (last five minutes): 0.0
----------------

conf/cassandra.yaml
nodetool cleanup
Cleans up keyspaces and partition keys no longer belonging to a node.
Cleans up keyspaces and partition keys no longer belonging to a node.
Synopsis
$ nodetool <options> cleanup -- <keyspace> (<table> ...)

• options are:
• keyspace is a keyspace name.
• table is one or more table names, separated by a space.
Synopsis Legend
Description
Use this command to remove unwanted data after adding a new node to the cluster. Cassandra does
not automatically remove data from nodes that lose part of their partition range to a newly added node.
Run nodetool cleanup on the source node and on neighboring nodes that shared the same subrange
after the new node is up and running. Failure to run this command after adding a node causes Cassandra
to include the old data to rebalance the load on that node. Running the nodetool cleanup command
causes a temporary increase in disk space usage proportional to the size of your largest SSTable. Disk I/O
occurs when running this command.
Running this command affects nodes that use a counter column in a table. Cassandra assigns a new
counter ID to the node.
Optionally, this command takes a list of table names. If you do not specify a keyspace, this command
cleans all keyspaces no longer belonging to a node.
167
Cassandra tools
nodetool clearsnapshot
Removes one or more snapshots.
Removes one or more snapshots.
Synopsis
$ nodetool <options> clearsnapshot -t <snapshot> -- ( <keyspace> ... )

• options are:
• -t means the following file contains the snapshot.
• snapshot is the name of the snapshot.
• keyspace is one or more keyspace names, separated by a space.
Synopsis Legend
Description
Deletes snapshots in one or more keyspaces. To remove all snapshots, omit the snapshot name.
Caution: Removing all snapshots or all specific keyspace snapshots also removes OpsCenter
backups.
nodetool compact
Forces a major compaction on one or more tables.
Forces a major compaction on one or more tables.
Synopsis
$ nodetool <options> compact <keyspace> ( <table> ... )

• options are:
• -- separates an option and argument that could be mistaken for a option.
168
Cassandra tools
Synopsis Legend
Description
This command starts the compaction process on tables that use the SizeTieredCompactionStrategy
and DateTieredCompactionStrategy. You can specify a keyspace for compaction. If you do not specify
a keyspace, the nodetool command uses the current keyspace. You can specify one or more tables for
compaction. If you do not specify any tables, compaction of all tables in the keyspace occurs. This is called
a major compaction. If you do specify one or more tables, compaction of the specified tables occurs. This
is called a minor compaction. A major compaction combines each of the pools of repaired and unrepaired
SSTables into one repaired and one unreparied SSTable. During compaction, there is a temporary spike in
disk space usage and disk I/O because the old and new SSTables co-exist. A major compaction can cause
considerable disk I/O.
nodetool compactionhistory
Provides the history of compaction operations.
Provides the history of compaction operations.
Synopsis
$ nodetool <options> compactionhistory

options are:
Synopsis Legend
Example
The actual output of compaction history is seven columns wide. The first three columns show the id,
keyspace name, and table name of the compacted SSTable.
$ nodetool compactionhistory
Compaction History:
id keyspace_name
columnfamily_name
d06f7080-07a5-11e4-9b36-abc3a0ec9088 system
schema_columnfamilies
169
Cassandra tools
d198ae40-07a5-11e4-9b36-abc3a0ec9088 libdata users

0381bc30-07b0-11e4-9b36-abc3a0ec9088 Keyspace1 Standard1
74eb69b0-0621-11e4-9b36-abc3a0ec9088 system local
e35dd980-07ae-11e4-9b36-abc3a0ec9088 system
compactions_in_progress
8d5cf160-07ae-11e4-9b36-abc3a0ec9088 system
ba376020-07af-11e4-9b36-abc3a0ec9088 Keyspace1 Standard1
d18cc760-07a5-11e4-9b36-abc3a0ec9088 libdata libout
64009bf0-07a4-11e4-9b36-abc3a0ec9088 libdata libout
d04700f0-07a5-11e4-9b36-abc3a0ec9088 system sstable_activity
c2a97370-07a9-11e4-9b36-abc3a0ec9088 libdata users
cb928a80-07ae-11e4-9b36-abc3a0ec9088 Keyspace1 Standard1
cd8d1540-079e-11e4-9b36-abc3a0ec9088 system schema_columns
62ced2b0-07a4-11e4-9b36-abc3a0ec9088 system schema_keyspaces
d19cccf0-07a5-11e4-9b36-abc3a0ec9088 system
640bbf80-07a4-11e4-9b36-abc3a0ec9088 libdata users
6cd54e60-07ae-11e4-9b36-abc3a0ec9088 Keyspace1 Standard1
c29241f0-07a9-11e4-9b36-abc3a0ec9088 libdata libout
c2a30ad0-07a9-11e4-9b36-abc3a0ec9088 system
e3a6d920-079d-11e4-9b36-abc3a0ec9088 system schema_keyspaces
62c55cd0-07a4-11e4-9b36-abc3a0ec9088 system
62b07540-07a4-11e4-9b36-abc3a0ec9088 system schema_columns
cdd038c0-079e-11e4-9b36-abc3a0ec9088 system schema_keyspaces
b797af00-07af-11e4-9b36-abc3a0ec9088 Keyspace1 Standard1
8c918b10-07ae-11e4-9b36-abc3a0ec9088 Keyspace1 Standard1
377d73f0-07ae-11e4-9b36-abc3a0ec9088 system
62b9c410-07a4-11e4-9b36-abc3a0ec9088 system local
d0566a40-07a5-11e4-9b36-abc3a0ec9088 system schema_columns
ba637930-07af-11e4-9b36-abc3a0ec9088 system
cdbc1480-079e-11e4-9b36-abc3a0ec9088 system
e3456f80-07ae-11e4-9b36-abc3a0ec9088 Keyspace1 Standard1
d086f020-07a5-11e4-9b36-abc3a0ec9088 system schema_keyspaces
d06118a0-07a5-11e4-9b36-abc3a0ec9088 system local
cdaafd80-079e-11e4-9b36-abc3a0ec9088 system local
640fde30-07a4-11e4-9b36-abc3a0ec9088 system
37638350-07ae-11e4-9b36-abc3a0ec9088 Keyspace1 Standard1
The four columns to the right of the table name show the timestamp, size of the SSTtable before and after
compaction, and the number of partitions merged. The notation means {tables:rows}. For example: {1:3,
3:1} means 3 rows were taken from one SSTable (1:3) and 1 row taken from 3 SSTables (3:1) to make the
one SSTable in that compaction operation.
. . . compacted_at bytes_in bytes_out rows_merged

. . . 1404936947592 8096 7211 {1:3, 3:1}
. . . 1404936949540 144 144 {1:1}
. . . 1404941328243 1305838191 1305838191 {1:4647111}
. . . 1404770149323 5864 5701 {4:1}
. . . 1404940844824 573 148 {1:1, 2:2}
. . . 1404940700534 576 155 {1:1, 2:2}
. . . 1404941205282 766331398 766331398 {1:2727158}
. . . 1404936949462 8901649 8901649 {1:9315}
. . . 1404936336175 8900821 8900821 {1:9315}
. . . 1404936947327 223 108 {1:3, 2:1}
. . . 1404938642471 144 144 {1:1}
170
Cassandra tools
. . . 1404940804904 383020422 383020422 {1:1363062}

. . . 1404933936276 4889 4177 {1:4}
. . . 1404936334171 441 281 {1:3, 2:1}
. . . 1404936949567 379 79 {2:2}
. . . 1404936336248 144 144 {1:1}
. . . 1404940645958 307520780 307520780 {1:1094380}
. . . 1404938642319 8901649 8901649 {1:9315}
. . . 1404938642429 416 165 {1:3, 2:1}
. . . 1404933543858 692 281 {1:3, 2:1}
. . . 1404936334109 7760 7186 {1:3, 2:1}
. . . 1404936333972 4860 4724 {1:2, 2:1}
. . . 1404933936715 441 281 {1:3, 2:1}
. . . 1404941200880 1269180898 1003196133 {1:2623528,
2:946565}
. . . 1404940699201 297639696 297639696 {1:1059216}
. . . 1404940556463 592 148 {1:2, 2:2}
. . . 1404936334033 5760 5680 {2:1}
. . . 1404936947428 8413 5316 {1:2, 3:1}
. . . 1404941205571 429 42 {2:2}
. . . 1404933936584 7994 6789 {1:4}
. . . 1404940844664 306699417 306699417 {1:1091457}
. . . 1404936947746 601 281 {1:3, 3:1}
. . . 1404936947498 5840 5680 {3:1}
. . . 1404933936472 5861 5680 {3:1}
. . . 1404936336275 378 80 {2:2}
. . . 1404940556293 302170540 281000000 {1:924660, 2:75340}
nodetool compactionstats
Provide statistics about a compaction.
Provide statistics about a compaction.
Synopsis
$ nodetool <options> compactionstats -H

• options are:
• data center is the name of an arbitrarily chosen data center from which to select sources for streaming.
Synopsis Legend
171
Cassandra tools
Description
The total column shows the total number of uncompressed bytes of SSTables being compacted. The
system log lists the names of the SSTables compacted.
Example
$ bin/nodetool compactionstats
pending tasks: 5
compaction type keyspace table completed
total unit progress
Compaction Keyspace1 Standard1 282310680
302170540 bytes 93.43%
Compaction Keyspace1 Standard1 58457931
307520780 bytes 19.01%
Active compaction remaining time : 0h00m16s
nodetool decommission
Deactivates a node by streaming its data to another node.
Deactivates a node by streaming its data to another node.
Synopsis
$ nodetool <options> decommission

options are:
Synopsis Legend
Description
Causes a live node to decommission itself, streaming its data to the next node on the ring. Use netstats to
monitor the progress, as described on http://wiki.apache.org/cassandra/NodeProbe#Decommission and
http://wiki.apache.org/cassandra/Operations#Removing_nodes_entirely.
nodetool describecluster
Provide the name, snitch, partitioner and schema version of a cluster
Provide the name, snitch, partitioner and schema version of a cluster
Synopsis
$ nodetool <options> describecluster -- <data center>
172
Cassandra tools
• options are:
• data center is the name of an arbitrarily chosen data center from which to select sources for streaming.
Synopsis Legend
Description
Describe cluster is typically used to validate the schema after upgrading. If a schema disagreement occurs,
check for and resolve schema disagreements.
Example
$ bin/nodetool describecluster
Cluster Information:
Name: Test Cluster
Snitch: org.apache.cassandra.locator.DynamicEndpointSnitch
Partitioner: org.apache.cassandra.dht.Murmur3Partitioner
Schema versions:
65e78f0e-e81e-30d8-a631-a65dff93bf82: [127.0.0.1]
If a schema disagreement occurs, the last line of the output includes information about unreachable nodes.
Name: Production Cluster
Schema versions:
UNREACHABLE: 1176b7ac-8993-395d-85fd-41b89ef49fbb:
[10.202.205.203]
nodetool describering
Provides the partition ranges of a keyspace.
Provides the partition ranges of a keyspace.
Synopsis
$ nodetool <options> describering -- <keyspace>

• options are:
173
Cassandra tools

Synopsis Legend
Example
This example shows the sample output of the command on a three-node cluster.
$ nodetool describering demo_keyspace
Schema Version:1b04bd14-0324-3fc8-8bcb-9256d1e15f82
TokenRange:
TokenRange(start_token:3074457345618258602,
end_token:-9223372036854775808,
endpoints:[127.0.0.1, 127.0.0.2, 127.0.0.3],
rpc_endpoints:[127.0.0.1, 127.0.0.2, 127.0.0.3],
endpoint_details:[EndpointDetails(host:127.0.0.1,
datacenter:datacenter1, rack:rack1),
EndpointDetails(host:127.0.0.2, datacenter:datacenter1,
rack:rack1),
rack:rack1)])
TokenRange(start_token:-3074457345618258603,
end_token:3074457345618258602,
endpoints:[127.0.0.3, 127.0.0.1, 127.0.0.2],
rpc_endpoints:[127.0.0.3, 127.0.0.1, 127.0.0.2],
rack:rack1),
rack:rack1)])
TokenRange(start_token:-9223372036854775808,
end_token:-3074457345618258603,
endpoints:[127.0.0.2, 127.0.0.3, 127.0.0.1],
rpc_endpoints:[127.0.0.2, 127.0.0.3, 127.0.0.1],
rack:rack1),
rack:rack1)])
If a schema disagreement occurs, the last line of the output includes information about unreachable nodes.
174
Cassandra tools
Name: Production Cluster
Schema versions:
UNREACHABLE: 1176b7ac-8993-395d-85fd-41b89ef49fbb:
[10.202.205.203]
nodetool disableautocompaction
Disables autocompaction for a keyspace and one or more tables.
Disables autocompaction for a keyspace and one or more tables.
Synopsis
$ nodetool <options> disableautocompaction -- <keyspace> ( <table> ... )

• options are:
Synopsis Legend
Description
The keyspace can be followed by one or more tables.
nodetool disablebackup
Disables incremental backup.
Disables incremental backup.
Synopsis
$ nodetool <options> disablebackup

options are:
175
Cassandra tools
Synopsis Legend
nodetool disablebinary
Disables the native transport.
Disables the native transport.
Synopsis
$ nodetool <options> disablebinary

options are:
Synopsis Legend
Description
Disables the binary protocol, also known as the native transport.
nodetool disablegossip
Disables the gossip protocol.
Disables the gossip protocol.
Synopsis
$ nodetool <options> disablegossip

options are:
176
Cassandra tools
Synopsis Legend
Description
This command effectively marks the node as being down.
nodetool disablehandoff
Disables storing of future hints on the current node.
Disables storing of future hints on the current node.
Synopsis
$ nodetool <options> disablehandoff

options are:
Synopsis Legend
nodetool disablethrift
Disables the Thrift server.
Disables the Thrift server.
Synopsis
$ nodetool <options> disablethrift

options are:
Synopsis Legend
177
Cassandra tools

Description
nodetool disablethrift will disable thrift on a node preventing the node from acting as a coordinator.
The node can still be a replica for a different coordinator and data read at consistency level ONE could
be stale. To cause a node to ignore read requests from other coordinators, nodetool disablegossip
would also need to be run. However, if both commands are run, the node will not perform repairs, and
the node will continue to store stale data. If the goal is to repair the node, set the read operations to a
consistency level of QUORUM or higher while you run repair. An alternative approach is to delete the
node's data and restart the Cassandra process.
Examples
$ nodetool -u cassandra -pw cassandra disablethrift 192.168.100.1
C:\ nodetool -u cassandra -pw cassandra disablethrift 192.168.100.1
nodetool drain
Drains the node.
Drains the node.
Synopsis
$ nodetool <options> drain

options are:
Synopsis Legend
Description
Flushes all memtables from the node to SSTables on disk. Cassandra stops listening for connections from
the client and other nodes. You need to restart Cassandra after running nodetool drain. You typically
use this command before upgrading a node to a new version of Cassandra. To simply flush memtables to
disk, use nodetool flush.
178
Cassandra tools
nodetool enableautocompaction
Enables autocompaction for a keyspace and one or more tables.
Enables autocompaction for a keyspace and one or more tables.
Synopsis
$ nodetool <options> enableautocompaction -- <keyspace> ( <table> ... )

• options are:
• table is the name of one or more keyspaces, separated by a space.
Synopsis Legend
Description
The keyspace can be followed by one or more tables. Enables compaction for the named keyspace or the
current keyspace, and one or more named tables, or all tables.
nodetool enablebackup
Enables incremental backup.
Enables incremental backup.
Synopsis
$ nodetool <options> enablebackup

options are:
Synopsis Legend
179
Cassandra tools
nodetool enablebinary
Re-enables native transport.
Re-enables native transport.
Synopsis
$ nodetool <options> enablebinary

options are:
Synopsis Legend
Description
Re-enables the binary protocol, also known as native transport.
nodetool enablegossip
Re-enables gossip.
Re-enables gossip.
Synopsis
$ nodetool <options> enablegossip

options are:
Synopsis Legend
180
Cassandra tools
nodetool enablehandoff
Re-enables the storing of future hints on the current node.
Re-enables the storing of future hints on the current node.
Synopsis
$ nodetool <options> enablehandoff

• options are:
• <dc-name>,<dc-name> means enable hinted handoff only for these data centers
Synopsis Legend
nodetool enablethrift
Re-enables the Thrift server.
Re-enables the Thrift server.
Synopsis
$ nodetool <options> enablethrift

options are:
Synopsis Legend
nodetool flush
Flushes one or more tables from the memtable.
Flushes one or more tables from the memtable.
181
Cassandra tools
Synopsis
$ nodetool <options> flush -- <keyspace> ( <table> ... )

• options are:
• table is the name of one or more tables, separated by a space.
Synopsis Legend
Description
You can specify a keyspace followed by one or more tables that you want to flush from the memtable to
SSTables on disk.
nodetool getcompactionthreshold
Provides the minimum and maximum compaction thresholds in megabytes for a table.
Provides the minimum and maximum compaction thresholds in megabytes for a table.
Synopsis
$ nodetool <options> getcompactionthreshold -- <keyspace> <table>

• options are:
• table is the name of a table.
Synopsis Legend
182
Cassandra tools
nodetool getendpoints
Provides the IP addresses or names of replicas that own the partition key.
Provides the IP addresses or names of replicas that own the partition key.
Synopsis
$ nodetool <options> getendpoints -- <keyspace> <table> key

• options are:
• table is a table name.
• key is the partition key of the end points you want to get.
Synopsis Legend
Example
For example, which nodes own partition key_1, key_2, and key_3?
Note: The partitioner returns a token for the key. Cassandra will return an endpoint whether or not
data exists on the identified node for that token.
$ bin/nodetool -h 127.0.0.1 -p 7100 getendpoints myks mytable key_1
127.0.0.2
127.0.0.2
127.0.0.1
nodetool getlogginglevels
Get the runtime logging levels.
Get the runtime logging levels.
183
Cassandra tools
Synopsis
$ nodetool <options> getlogginglevels

options are:
• ( -pw | --password ) <password>
Synopsis Legend
nodetool getsstables
Provides the SSTables that own the partition key.
Provides the SSTables that own the partition key.
Synopsis
$ nodetool <options> getsstables -- <keyspace> <table> key

• options are:
• key is the partition key of the SSTables.
Synopsis Legend
nodetool getstreamthroughput
Provides the megabytes per second throughput limit for streaming in the system.
Provides the megabytes per second throughput limit for streaming in the system.
184
Cassandra tools
Synopsis
$ nodetool <options> getstreamthroughput

options are:
Synopsis Legend
nodetool gossipinfo
Provides the gossip information for the cluster.
Provides the gossip information for the cluster.
Synopsis
$ nodetool <options> gossipinfo

options are:
Synopsis Legend
nodetool help
Provides nodetool command help.
Provides nodetool command help.
Synopsis
$ nodetool help <command>
Synopsis Legend
185
Cassandra tools

Description
Using this command without the The help command provides a synopsis and brief description of each
nodetool command.
Examples
Using nodetool help lists all commands and usage information. For example, nodetool help netstats
provides the following information.
NAME
nodetool netstats - Print network information on provided host
(connecting node by default)
SYNOPSIS
nodetool [(-h <host> | --host <host>)] [(-p <port> | --port <port>)]
[(-pw <password> | --password <password>)]
[(-u <username> | --username <username>)] netstats
OPTIONS
-h <host>, --host <host>
Node hostname or ip address
-p <port>, --port <port>

Remote jmx agent port number
-pw <password>, --password <password>

Remote jmx agent password
-u <username>, --username <username>

Remote jmx agent username
nodetool info
Provides node information, such as load and uptime.
Provides node information, such as load and uptime.
Synopsis
$ nodetool <options> info ( -T | --tokens )

• options are:
• -T or --tokens means provide all token information.
Synopsis Legend
186
Cassandra tools

Description
Provides node information including the token and on disk storage (load) information, times started
(generation), uptime in seconds, and heap memory usage.
nodetool invalidatekeycache
Resets the global key cache parameter to the default, which saves all keys.
Resets the global key cache parameter to the default, which saves all keys.
Synopsis
$ nodetool <options> invalidatekeycache

options are:
Synopsis Legend
Description
By default the key_cache_keys_to_save is disabled in the cassandra.yaml. This command resets the
parameter to the default.

conf/cassandra.yaml
nodetool invalidaterowcache
Resets the global key cache parameter, row_cache_keys_to_save, to the default (not set), which saves all
keys.
Resets the global key cache parameter, row_cache_keys_to_save, to the default (not set), which saves all
keys.
Synopsis
$ nodetool <options> invalidaterowcache
187
Cassandra tools
options are:
Synopsis Legend
nodetool join
Causes the node to join the ring.
Causes the node to join the ring.
Synopsis
$ nodetool <options> join

options are:
Synopsis Legend
Description
Causes the node to join the ring, assuming the node was initially not started in the ring using the -
Djoin_ring=false cassandra utility option. The joining node should be properly configured with the desired
options for seed list, initial token, and auto-bootstrapping.
nodetool listsnapshots
Lists snapshot names, size on disk, and true size.
Lists snapshot names, size on disk, and true size.
Synopsis
nodetool <options> listsnapshots
188
Cassandra tools
options are:
Synopsis Legend
Description
Available in Cassandra 2.1 and later.
Example
Snapshot Details:
Snapshot Name Keyspace Column Family True Size Size on Disk
1387304478196 Keyspace1 Standard1 0 bytes 308.66 MB

Keyspace1 Standard1 0 bytes 308.66 MB
nodetool move
Moves the node on the token ring to a new token.
Moves the node on the token ring to a new token.
Synopsis
$ nodetool <options> move -- <new token>

• options are:
63 63
• new token is a number in the range -2 to +2 -1.
Synopsis Legend
189
Cassandra tools
Description
Escape negative tokens using \\ . For example: move \\-123 . This command essentially combines
decommission and bootstrapoperations.
nodetool netstats
Provides network information about the host.
Provides network information about the host.
Synopsis
$ nodetool <options> netstats -H

• options are:
Synopsis Legend
Description
The default host is the connected host if the user does not include a host name or IP address in the
command. The output includes the following information:
• JVM settings
• Mode
• Read repair statistics
• Attempted
The number of successfully completed read repair operations
• Mismatch (blocking)
The number of read repair operations since server restart that blocked a query.
• Mismatch (background)
The number of read repair operations since server restart performed in the background.
• Pool name
Information about client read and write requests by thread pool.
• Active, pending, and completed number of commands and responses
190
Cassandra tools
Example
Get the network information for a node 10.171.147.128:
$ nodetool -h 10.171.147.128 netstats

The output is:
Mode: NORMAL
Not sending any streams.
Read Repair Statistics:
Attempted: 0
Mismatch (Blocking): 0
Mismatch (Background): 0
Pool Name Active Pending Completed
Commands n/a 0 1156
Responses n/a 0 2750
nodetool pausehandoff
Pauses the hints delivery process
Pauses the hints delivery process
Synopsis
$ nodetool <options> pausehandoff

options are:
Synopsis Legend
nodetool proxyhistograms
Provides a histogram of network statistics.
Provides a histogram of network statistics.
Synopsis
$ nodetool <options> proxyhistograms

options are:
191
Cassandra tools
Synopsis Legend
Description
The output of this command shows the full request latency recorded by the coordinator. The output
includes the percentile rank of read and write latency values for inter-node communication. Typically, you
use the command to see if requests encounter a slow node.
Examples
This example shows the output from nodetool proxyhistograms after running 4,500 insert statements and
45,000 select statements on a three ccm node-cluster on a local computer.
$ nodetool proxyhistograms
proxy histograms
Percentile Read Latency Write Latency Range Latency
(micros) (micros) (micros)
50% 1502.50 375.00 446.00
75% 1714.75 420.00 498.00
95% 31210.25 507.00 800.20
98% 36365.00 577.36 948.40
99% 36365.00 740.60 1024.39
Min 616.00 230.00 311.00
Max 36365.00 55726.00 59247.00
nodetool rangekeysample
Provides the sampled keys held across all keyspaces.
Provides the sampled keys held across all keyspaces.
Synopsis
$ nodetool <options> rangekeysample

options are:
Synopsis Legend
192
Cassandra tools
nodetool rebuild
Rebuilds data by streaming from other nodes.
Rebuilds data by streaming from other nodes.
Synopsis
$ nodetool <options> rebuild -- <src-dc-name>

• options are:
• src-dc-name is the name of the data center from which to select sources for streaming. You can pick
any data center.
Synopsis Legend
Description
This command operates on multiple nodes in a cluster. Similar to bootstrap. Rebuild (like bootstrap) only
streams data from a single source replica per range. Use this command to bring up a new data center in an
existing cluster.
nodetool rebuild_index
Performs a full rebuild of the index for a table
Performs a full rebuild of the index for a table
Synopsis
$ nodetool <options> rebuild_index -- ( <keyspace>.<table>.<indexName> ... )

• options are:
• indexName is an optional list of index names separated by a space.
193
Cassandra tools
The keyspace and table name followed by a list of index names. For example: Standard3.IdxName
Standard3.IdxName1
Synopsis Legend
Description
Fully rebuilds one or more indexes for a table.
nodetool refresh
Loads newly placed SSTables onto the system without a restart.
Loads newly placed SSTables onto the system without a restart.
Synopsis
$ nodetool <options> refresh -- <keyspace> <table>

• options are:
Synopsis Legend
nodetool reloadtriggers
Reloads trigger classes.
Reloads trigger classes.
Synopsis
$ nodetool <options> reloadtriggers

options are:
194
Cassandra tools

Synopsis Legend
Description
Available in Cassandra 2.1 and later.
nodetool removenode
Provides the status of current node removal, forces completion of pending removal, or removes the
identified node.
Provides the status of current node removal, forces completion of pending removal, or removes the
identified node.
Synopsis
$ nodetool <options> removenode -- <status> | <force> | <ID>

• options are:
• status provides status information.
• force forces completion of the pending removal.
• ID is the host ID, in UUID format.
Synopsis Legend
Description
This command removes a node, shows the status of a removal operation, or forces the completion
of a pending removal. When the node is down and nodetool decommission cannot be used, use
nodetool removenode. Run this command only on nodes that are down. If the cluster does not use
vnodes, before running the nodetool removenode command, adjust the tokens.
195
Cassandra tools
Examples
Determine the UUID of the node to remove by running nodetool status. Use the UUID of the node that
is down to remove the node.
$ nodetool status
Datacenter: DC1
===============
Status=Up/Down
Rack
UN 192.168.2.101 112.82 KB 256 31.7% 420129fc-0d84-42b0-
be41-ef7dd3a8ad06 RAC1
DN 192.168.2.103 91.11 KB 256 33.9% d0844a21-3698-4883-
ab66-9e2fd5150edd RAC1
UN 192.168.2.102 124.42 KB 256 32.6% 8d5ed9f4-7764-4dbd-
bad8-43fddce94b7c RAC1
$ nodetool removenode d0844a21-3698-4883-ab66-9e2fd5150edd
View the status of the operation to remove the node:
$ nodetool removenode status
RemovalStatus: No token removals in process.
Confirm that the node has been removed.
$ nodetool removenode status
Datacenter: DC1
===============
Status=Up/Down
Rack
UN 192.168.2.101 112.82 KB 256 37.7% 420129fc-0d84-42b0-
be41-ef7dd3a8ad06 RAC1
UN 192.168.2.102 124.42 KB 256 38.3% 8d5ed9f4-7764-4dbd-
bad8-43fddce94b7c RAC1
nodetool repair
Repairs one or more tables.
Repairs one or more tables.
Synopsis
$ nodetool <options> repair

( -dc <dc_name> ) | ( --in-dc <dc_name> )
( -et <end_token>) | ( --end-token <end_token> )
( -hosts <host_name host_name . . . > ) | ( --in-hosts <host_name
host_name . . .> )
( -inc | --incremental )
( -local | --in-local-dc )
( -par | --parallel )
( -pr | --partitioner-range )
196
Cassandra tools
( -st <start_token> ) | ( --start-token <start_token> )

-- <keyspace> ( <table> ... )
• options are:
• -dc, or --in-dc, followed by dc_name, or means restrict repair to nodes in the named data center, which
must be the local data center.
• -dcpar, or --dc-parallel, means repair data centers in parallel.
• -et, or --end-token, followed by the UUID of a token means stop repairing a range of nodes after
repairing this token. Use -hosts to specify neighbor nodes.
• -hosts <host_name host_name . . . >, or --in-hosts <host_name host_name . . .>, means repair specific
hosts.
• -inc, or --incremental means do an incremental repair.
• -local, or --in-local-dc, means repair nodes in the same data center only.
• -par, or --parallel, means do a parallel repair.
• -pr, or --partitioner-range, means repair only the first range returned by the partitioner.
• -st, or --start-token, followed by the UUID of a token means start repairing a range of nodes at this
token.
• keyspace is the keyspace name. The default is all.
• table is one or more table names, separated by a space. The default is all.
Synopsis Legend
Description
Performing an anti-entropy node repair on a regular basis is important, especially when frequently deleting
data. The nodetool repair command repairs one or more nodes in a cluster, and includes an option to
restrict repair to a set of nodes. Anti-entropy node repair performs the following tasks:
• Ensures that all data on a replica is consistent.
• Repairs inconsistencies on a node that has been down.
By default, Cassandra 2.1 does a full, sequential repair.
Using options
You can use options to do these other types of repair:
• Incremental
• Parallel
Use the -hosts option to list the good nodes to use for repairing the bad nodes. Use -h to name the bad
nodes.
Use the -inc option for an incremental repair. An incremental repair eliminates the need for constant Merkle
tree construction by persisting already repaired data and calculating only the Merkle trees for SSTables
197
Cassandra tools
that have not been repaired. The repair process is likely more performant than the other types of repair
even as datasets grow, assuming you run repairs frequently. Before doing an incremental repair for the first
time, perform migration steps first if necessary.
Use the -par option for a parallel repair. Unlike sequential repair, parallel repair constructs the Merkle
tables for all nodes at the same time. Therfore, no snapshots are required (or generated). Use a parallel
repair to complete the repair quickly or when you have operational downtime that allows the resources to
be completely consumed during the repair.
Performing partitioner range repairs by using the -pr option is generally not recommended.
Example
All nodetool repair arguments are optional. The following examples show the following types of repair:
• An incremental, parallel repair of all keyspaces on the current node
• A partitioner range repair of the bad partition on current node using the good partitions on 10.2.2.20 or
10.2.2.21
• A start-point-to-end-point repair of all nodes between two nodes on the ring
$ nodetool repair -par -inc

$ nodetool repair -pr -hosts 10.2.2.20 10.2.2.21
$ nodetool -st a9fa31c7-f3c0-44d1-b8e7-a26228867840c -et f5bb146c-
db51-475ca44f-9facf2f1ad6e
To restrict the repair to the local data center, use the -dc option followed by the name of the data center.
Issue the command from a node in the data center you want to repair. Issuing the command from a data
center other than the named one returns an error. Do not use -pr with this option to repair only a local data
center.
$ nodetool repair -dc DC1
[2014-07-24 21:59:55,326] Nothing to repair for keyspace 'system'

[2014-07-24 21:59:55,617] Starting repair command #2, repairing 490 ranges
for keyspace system_traces (seq=true, full=true)
[2014-07-24 22:23:14,299] Repair session 323b9490-137e-11e4-88e3-
c972e09793ca
for range (820981369067266915,822627736366088177] finished
[2014-07-24 22:23:14,320] Repair session 38496a61-137e-11e4-88e3-
c972e09793ca
for range (2506042417712465541,2515941262699962473] finished
. . .
An inspection of the system.log shows repair taking place only on IP addresses in DC1.
. . .
INFO [AntiEntropyStage:1] 2014-07-24 22:23:10,708 RepairSession.java:171
- [repair #16499ef0-1381-11e4-88e3-c972e09793ca] Received merkle tree
for sessions from /192.168.2.101
INFO [RepairJobTask:1] 2014-07-24 22:23:10,740 RepairJob.java:145
- [repair #16499ef0-1381-11e4-88e3-c972e09793ca] requesting merkle trees
for events (to [/192.168.2.103, /192.168.2.101])
. . .
nodetool resetlocalschema
Reset the node's local schema and resynchronizes.
Reset the node's local schema and resynchronizes.
198
Cassandra tools
Synopsis
$ nodetool <options> resetlocalschema

options are:
Synopsis Legend
Description
Normally, this command is used to rectify schema disagreements on different nodes. It can be useful if
table schema changes have generated too many tombstones, on the order of 100,000s.
nodetool resetlocalschema drops the schema information of the local node and resynchronizes the
schema from another node. To drop the schema, the tool truncates all the system schema tables. The
node will temporarily lose metadata about the tables on the node, but will rewrite the information from
another node. If the node is experiencing problems with too many tombstones, the truncation of the tables
will eliminate the tombstones.
This command is useful when you have one node that is out of sync with the cluster. The system schema
tables must have another node from which to fetch the tables. It is not useful when all or many of your
nodes are in an incorrect state. If there is only one node in the cluster (replication factor of 1) – it does
not perform the operation, because another node from which to fetch the tables does not exist. Run the
command on the node experiencing difficulty.
nodetool resumehandoff
Resume hints delivery process.
Resume hints delivery process.
Synopsis
$ nodetool <options> resumehandoff

options are:
Synopsis Legend
199
Cassandra tools

nodetool ring
Provides node status and information about the ring.
Provides node status and information about the ring.
Synopsis
$ nodetool <options> ring ( -r | --resolve-ip ) -- <keyspace>

• options are:
• -r, or --resolve-ip, means to provide node names instead of IP addresses.
Synopsis Legend
Description
Displays node status and information about the ring as determined by the node being queried. This
information can give you an idea of the load balance and if any nodes are down. If your cluster is not
properly configured, different nodes may show a different ring. Check that the node appears the same way
in the ring.If you use virtual nodes (vnodes), use nodetool status for succinct output.
• Address
The node's URL.
• DC (data center)
The data center containing the node.
• Rack
The rack or, in the case of Amazon EC2, the availability zone of the node.
• Status - Up or Down
Indicates whether the node is functioning or not.
• State - N (normal), L (leaving), J (joining), M (moving)
The state of the node in relation to the cluster.
• Load - updates every 90 seconds
The amount of file system data under the cassandra data directory after excluding all content in the
snapshots subdirectories. Because all SSTable data files are included, any data that is not cleaned up,
such as TTL-expired cell or tombstoned data) is counted.
200
Cassandra tools
• Token
The end of the token range up to and including the value listed. For an explanation of token ranges, see
Data Distribution in the Ring .
• Owns
The percentage of the data owned by the node per data center times the replication factor. For
example, a node can own 33% of the ring, but show100% if the replication factor is 3.
• Host ID
The network ID of the node.
nodetool scrub
Rebuild SSTables for one or more Cassandra tables.
Rebuild SSTables for one or more Cassandra tables.
Synopsis
$ nodetool <options> scrub <keyspace> -- ( -ns | --no-snapshot ) ( -s | --

skip-corrupted ) ( <table> ... )
• options are:
• -ns, or --no-snapshot, triggers a snapshot of the scrubbed table first assuming snapshots are not
disabled (the default).
• - s, or --skip-corrupted skips corrupted partitions even when scrubbing counter tables. (default false)
Synopsis Legend
Description
Rebuilds SSTables on a node for the named tables and snapshots data files before rebuilding as a safety
measure. If possible use upgradesstables. While scrub rebuilds SSTables, it also discards data that it
deems broken and creates a snapshot, which you have to remove manually. If the -ns option is specified,
snapshot creation is disabled. If scrub can't validate the column value against the column definition's data
type, it logs the partition key and skips to the next partition. Skipping corrupted partitions in tables having
counter columns results in under-counting. By default the scrub operation stops if you attempt to skip such
a partition. To force the scrub to skip the partition and continue scrubbing, re-run nodetool scrub using
the --skip-corrupted option.
201
Cassandra tools
nodetool setcachecapacity
Set global key and row cache capacities in megabytes.
Set global key and row cache capacities in megabytes.
Synopsis
$ nodetool <options> setcachecapacity -- <key-cache-capacity> <row-cache-

capacity>
• options are:
• key-cache-capacity is the maximum size in MB of the key cache in memory.
• row-cache-capacity corresponds to the maximum size in MB of the row cache in memory.
• counter-cache-capacity corresponds to the maximum size in MB of the counter cache in memory.
Synopsis Legend
Description
The key-cache-capacity argument corresponds to the key_cache_size_in_mb parameter in the
cassandra.yaml. Each key cache hit saves one seek and each row cache hit saves a minimum of two
seeks. Devoting some memory to the key cache is usually a good tradeoff considering the positive effect
on the response time. The default value is empty, which means a minimum of five percent of the heap in
MB or 100 MB.
The row-cache-capacity argument corresponds to the row_cache_size_in_mb parameter in the
cassandra.yaml. By default, row caching is zero (disabled).
The counter-cache-capacity argument corresponds to the counter_cache_size_in_mb in the
cassandra.yaml. By default, counter caching is a minimum of 2.5% of Heap or 50MB.

conf/cassandra.yaml
nodetool setcachekeystosave
Sets the number of keys saved by each cache for faster post-restart warmup.
Sets the number of keys saved by each cache for faster post-restart warmup.
202
Cassandra tools
Synopsis
$ nodetool <options> setcachekeystosave -- <key-cache-keys-to-save> <row-

cache-keys-to-save>
• options are:
• key-cache-keys-to-save is the number of keys from the key cache to save to the saved caches
directory.
• row-cache-keys-to-save is the number of keys from the row cache to save to the saved caches
directory.
Synopsis Legend
Description
This command saves the specified number of key and row caches to the saved caches directory,
which you specify in the cassandra.yaml. The key-cache-keys-to-save argument corresponds to the
key_cache_keys_to_save in the cassandra.yaml, which is disabled by default, meaning all keys will
be saved. The row-cache-keys-to-save argument corresponds to the row_cache_keys_to_save in the
cassandra.yaml, which is disabled by default.

conf/cassandra.yaml
nodetool setcompactionthreshold
Sets minimum and maximum compaction thresholds for a table.
Sets minimum and maximum compaction thresholds for a table.
Synopsis
$ nodetool <options> setcompactionthreshold -- <keyspace> <table>

<minthreshold> <maxthreshold>
• options are:
203
Cassandra tools

• minthreshold sets the minimum number of SSTables to trigger a minor compaction when using
SizeTieredCompactionStrategy or DateTieredCompactionStrategy.
• maxthreshold sets the maximum number of SSTables to allow in a minor compaction when using
SizeTieredCompactionStrategy or DateTieredCompactionStrategy.
Synopsis Legend
Description
This parameter controls how many SSTables of a similar size must be present before a minor compaction
is scheduled. The max_threshold table property sets an upper bound on the number of SSTables that
may be compacted in a single minor compaction, as described in http://wiki.apache.org/cassandra/
MemtableSSTable.
When using LeveledCompactionStrategy, maxthreshold sets the MAX_COMPACTING_L0, which limits
the number of L0 SSTables that are compacted concurrently to avoid wasting memory or running out of
memory when compacting highly overlapping SSTables.
nodetool setcompactionthroughput
Sets the throughput capacity for compaction in the system, or disables throttling.
Sets the throughput capacity for compaction in the system, or disables throttling.
Synopsis
$ nodetool <options> setcompactionthroughput -- <value_in_mb>

• options are:
• value_in_mb is the throughput capacity in MB per second for compaction.
Synopsis Legend
204
Cassandra tools
Description
Set value_in_mb to 0 to disable throttling.
nodetool sethintedhandoffthrottlekb
Sets hinted handoff throttle in kb/sec per delivery thread. (Cassandra 2.1.1 and later)
Sets hinted handoff throttle in kb/sec per delivery thread. (Cassandra 2.1.1 and later)
Synopsis
$ nodetool <options> sethintedhandoffthrottlekb <value_in_kb/sec>

• options are:
• value_in_kb/sec is the throttle time.
Synopsis Legend
Description
When a node detects that a node for which it is holding hints has recovered, it begins sending the hints to
that node. This setting specifies the maximum sleep interval per delivery thread in kilobytes per second
after delivering each hint. The interval shrinks proportionally to the number of nodes in the cluster. For
example, if there are two nodes in the cluster, each delivery thread uses the maximum interval; if there are
three nodes, each node throttles to half of the maximum interval, because the two nodes are expected to
deliver hints simultaneously.
Example
nodetool sethintedhandoffthrottlekb 2048
nodetool setlogginglevel
Set the log level for a service.
Set the log level for a service.
Synopsis
$ nodetool <options> setlogginglevel -- < class_qualifier > < level >

options are:
205
Cassandra tools

• class_qualifier is the logger class qualifier, a fully qualified domain name, such as
org.apache.cassandra.service.StorageProxy.
• level is the logging level, for example DEBUG.
Synopsis Legend
Description
You can use this command to set logging levels for services instead of modifying the logback-text.xml file.
The following values are valid for the logger class qualifier:
• org.apache.cassandra
• org.apache.cassandra.db
• org.apache.cassandra.service.StorageProxy
The possible log levels are:
• ALL
• TRACE
• DEBUG
• INFO
• WARN
• ERROR
• OFF
If both class qualifier and level arguments to the command are empty or null, the command resets logging
to the initial configuration.
Example
This command sets the StorageProxy service to debug level.
$ nodetool setlogginglevel org.apache.cassandra.service.StorageProxy DEBUG
nodetool setstreamthroughput
Sets the throughput capacity in MB for streaming in the system, or disable throttling.
Sets the throughput capacity in MB for streaming in the system, or disable throttling.
Synopsis
$ nodetool <options> setstreamthroughput -- <value_in_mb>

• options are:
206
Cassandra tools

• value_in_mb is the throughput capacity in MB per second for streaming.
Synopsis Legend
Description
Set value_in_MB to 0 to disable throttling.
nodetool settraceprobability
Sets the probability for tracing a request.
Sets the probability for tracing a request.
Synopsis
$ nodetool <options> settraceprobability -- <value>

• options are:
• value is a probability between 0 and 1.
Synopsis Legend
Description
Probabilistic tracing is useful to determine the cause of intermittent query performance problems by
identifying which queries are responsible. This option traces some or all statements sent to a cluster.
Tracing a request usually requires at least 10 rows to be inserted.
A probability of 1.0 will trace everything whereas lesser amounts (for example, 0.10) only sample a certain
percentage of statements. Care should be taken on large and active systems, as system-wide tracing
will have a performance impact. Unless you are under very light load, tracing all requests (probability
1.0) will probably overwhelm your system. Start with a small fraction, for example, 0.001 and increase
only if necessary. The trace information is stored in a system_traces keyspace that holds two tables –
sessions and events, which can be easily queried to answer questions, such as what the most time-
207
Cassandra tools
consuming query has been since a trace was started. Query the parameters map and thread column in the
system_traces.sessions and events tables for probabilistic tracing information.
nodetool snapshot
Take a snapshot of one or more keyspaces, or of a table, to backup data.
Take a snapshot of one or more keyspaces, or of a table, to backup data.
Synopsis
$ nodetool <options> snapshot

( -cf <table> | --column-family <table> )
(-kc <ktlist> | --kc.list <ktlist> | -kt <ktlist> | --kt-list <ktlist>)
( -t <tag> | --tag <tag> )
-- ( <keyspace> ) | ( <keyspace> ... )
• options are:
• -cf, or --column-family, followed by the name of the table to be backed up.
• -kc, --kc.list, -kt, or --kt-list, followed by a list of keyspace.table names to be back up, ktlist.
• -t or --tag, followed by the snapshot name.
• keyspace is a single keyspace name that is required when using the -cf option
• keyspace_list is one or more optional keyspace names, separated by a space.
Synopsis Legend
Description
Use this command to backup and restore using a snapshot. See the examples below for various options.
Cassandra flushes the node before taking a snapshot, takes the snapshot, and stores the data in the
snapshots directory of each keyspace in the data directory. If you do not specify the name of a snapshot
directory using the -t option, Cassandra names the directory using the timestamp of the snapshot, for
example 1391460334889. Follow the procedure for taking a snapshot before upgrading Cassandra. When
upgrading, backup all keyspaces. For more information about snapshots, see Apache documentation.
Example: All keyspaces

Take a snapshot of all keyspaces on the node. On Linux, in the Cassandra bin directory, for example:
$ bin/nodetool snapshot
The following message appears:
Requested creating snapshot(s) for [all keyspaces] with snapshot name

[1391464041163]
208
Cassandra tools
Snapshot directory: 1391464041163
Because you did not specify a snapshot name, Cassandra names snapshot directories using the
timestamp of the snapshot. If the keyspace contains no data, empty directories are not created.
Example: Single keyspace snapshot

Assuming you created the keyspace and tables in the music service example, take a snapshot of the music
keyspace and name the snapshot 2014.06.24.
$ bin/nodetool snapshot -t 2014.06.24 music

Requested creating snapshot(s) for [music] with snapshot name [2014.06.24]

Snapshot directory: 2014.06.24
Assuming the music keyspace contains two tables, songs and playlists, taking a snapshot of the keyspace
creates multiple snapshot directories named 2014.06.24. A number of .db files containing the data are
located in these directories. For example, from the installation directory:
$ cd data/data/music/playlists-bf8118508cfd11e3972273ded3cb6170/
snapshots/1404936753154
$ ls
music-playlists-ka-1-CompressionInfo.db music-playlists-ka-1-Index.db
music-playlists-ka-1-TOC.txt
music-playlists-ka-1-Data.db music-playlists-ka-1-Statistics.db
music-playlists-ka-1-Filter.db music-playlists-ka-1-Summary.db
$ cd data/data/music/songs-b8e385a08cfd11e3972273ded3cb6170/2014.06.24/
snapshots/1404936753154
music-songs-ka-1-CompressionInfo.db music-songs-ka-1-Index.db music-songs-

ka-1-TOC.txt
music-songs-ka-1-Data.db music-songs-ka-1-Statistics.db
music-songs-ka-1-Filter.db music-songs-ka-1-Summary.db
Example: Multiple keyspaces snapshot

Assuming you created a keyspace named mykeyspace in addition to the music keyspace, take a snapshot
of both keyspaces.
$ bin/nodetool snapshot mykeyspace music

Requested creating snapshot(s) for [mykeyspace, music] with snapshot name

[1391460334889]
Example: Single table snapshot

Take a snapshot of only the playlists table in the music keyspace.
209
Cassandra tools
$ bin/nodetool snapshot -cf playlists music
Requested creating snapshot(s) for [music] with snapshot name

[1391461910600]
Cassandra creates the snapshot directory named 1391461910600 that contains the backup data of
playlists table in data/data/music/playlists-bf8118508cfd11e3972273ded3cb6170/
snapshots, for example.
Example: List of different keyspace.tables snapshot

Take a snapshot of several tables in different keyspaces, such as the playlists table in the music keyspace
and the users table in the test keyspace. The keyspace.table list should be comma-delimited with no
spaces.
$ bin/nodetool snapshot -kt music.playlists,test.users
Requested creating snapshot(s) for [music.playlists,test.users] with

snapshot name [1431045288401]
nodetool status
Provide information about the cluster, such as the state, load, and IDs.
Provide information about the cluster, such as the state, load, and IDs.
Synopsis
$ nodetool <options> status ( -r | --resolve-ip ) -- <keyspace>

• options are:
• -r, or --resolve-ip, means to provide node names instead of IP addresses.
Synopsis Legend
Description
The status command provides the following information:
• Status - U (up) or D (down)
Indicates whether the node is functioning or not.
210
Cassandra tools
• State - N (normal), L (leaving), J (joining), M (moving)

The state of the node in relation to the cluster.
• Address
The node's URL.
• Load - updates every 90 seconds
The amount of file system data under the cassandra data directory after excluding all content in the
snapshots subdirectories. Because all SSTable data files are included, any data that is not cleaned up,
such as TTL-expired cell or tombstoned data) is counted.
• Tokens
The number of tokens set for the node.
• Owns
The percentage of the data owned by the node per data center times the replication factor. For
example, a node can own 33% of the ring, but show 100% if the replication factor is 3.
Attention: If your cluster uses keyspaces having different replication strategies or replication
factors, specify a keyspace when you run nodetool status to get meaningful ownship
information.
• Host ID
The network ID of the node.
• Rack
The rack or, in the case of Amazon EC2, the availability zone of the node.
Example
This example shows the output from running nodetool status.
$ nodetool status mykeyspace
=======================
Status=Up/Down
-- Address Load Tokens Owns Host ID
Rack
UN 127.0.0.1 47.66 KB 1 33.3% aaa1b7c1-6049-4a08-ad3e-3697a0e30e10
rack1
UN 127.0.0.2 47.67 KB 1 33.3% 1848c369-4306-4874-afdf-5c1e95b8732e
rack1
UN 127.0.0.3 47.67 KB 1 33.3% 49578bf1-728f-438d-b1c1-d8dd644b6f7f
rack1
nodetool statusbackup
Provide the status of backup
Synopsis
$ nodetool <options> statusbackup

options are:
211
Cassandra tools
Synopsis Legend
In the synopsis section of each statement, formatting has the following meaning:
• Uppercase means literal
• Lowercase means not literal
A semicolon that terminates CQL statements is not included in the synopsis.
Description
Provides the status of backup.
nodetool statusbinary
Provide the status of native transport.
Provide the status of native transport.
Synopsis
$ nodetool <options> statusbinary

options are:
Synopsis Legend
Description
Provides the status of the binary protocol, also known as the native transport.
nodetool statusgossip
Provide the status of gossip.
Synopsis
$ nodetool <options> statusgossip

options are:
212
Cassandra tools

Synopsis Legend
Description
Provides the status of gossip.
nodetool statushandoff
Provides the status of hinted handoff.
Synopsis
$ nodetool <options> statushandoff

options are:
Synopsis Legend
In the synopsis section of each statement, formatting has the following meaning:
• Uppercase means literal
• Lowercase means not literal
A semicolon that terminates CQL statements is not included in the synopsis.
Description
Provides the status of hinted handoff.
nodetool statusthrift
Provide the status of the Thrift server.
Provide the status of the Thrift server.
Synopsis
$ nodetool <options> statusthrift

options are:
213
Cassandra tools

Synopsis Legend
nodetool stop
Stops the compaction process.
Stops the compaction process.
Synopsis
$ nodetool <options> stop -- <compaction_type>

• options are:
• A compaction type: COMPACTION, VALIDATION, CLEANUP, SCRUB, INDEX_BUILD
Synopsis Legend
Description
Stops an operation from continuing to run. This command is typically used to stop a compaction that has a
negative impact on the performance of a node. After the compaction stops, Cassandra continues with the
remaining operations in the queue. Eventually, Cassandra restarts the compaction.
nodetool stopdaemon
Stops the cassandra daemon.
Stops the cassandra daemon.
Synopsis
$ nodetool <options> stopdaemon
214
Cassandra tools
options are:
Synopsis Legend
nodetool tpstats
Provides usage statistics of thread pools.
Provides usage statistics of thread pools.
Synopsis
$ nodetool <options> tpstats

options are:
Synopsis Legend
Description
Cassandra is based on a Staged Event Driven Architecture (SEDA). Different tasks are separated into
stages that are connected by a messaging service. Stages have a queue and thread pool. Some stages
skip the messaging service and queue tasks immediately on a different stage if it exists on the same node.
The queues can back up if executing at the next stage is too busy. Having a queue get backed up can
cause performance bottlenecks. nodetool tpstats provides statistics about the number of active,
pending, and completed tasks for each stage of Cassandra operations by thread pool. A high number of
pending tasks for any pool can indicate performance problems, as described in http://wiki.apache.org/
cassandra/Operations#Monitoring.
Run the nodetool tpstatscommand on a local node for get thread pool statistics.
This table describes key indicators:
215
Cassandra tools
Table 12: nodetool tpstats output
Name of Task Related information

statistic
AntiEntropyStage Repair consistency Nodetool repair
CacheCleanupExecutor
Clears the cache
CommitlogArchiver
Archives commitlog
CompactionExecutor
Runs compaction
CounterMutationStage
Local counter changes Will back up if the write rate
exceeds the mutation rate. A
high pending count will be seen
if consistency level is set to
ONE and there is a high counter
increment workload.
GossipStage Handle gossip rounds every second Out of sync schemas can
cause issues. nodetool
resetlocalschema may need
to be used.
HintedHandoff Send missed mutations to other nodes Usually symptom of a problem
elsewhere. Use nodetool
disablehandoff and run
repair.
InternalResponseStage
Respond to non-client initiated messages, including
bootstrapping and schema checking
MemtableFlushWriter
Writes memtable contents to disk Will back up if the queue
is overrunning the disk I/O
capabilities. Sorting can also
cause issues if the queue has a
high load associated with a small
number of flushes. Cause can
be huge rows with large column
names or inserting too many
values into a CQL collection. For
disk issues, add nodes or tune
configuration.
MemtablePostFlush
Operations after flushing the memtable Discard commit log files and flush
secondary indexes.
MemtableReclaimMemory
Makes unused memory available
MigrationStage Make schema changes
MiscStage Miscellaneous operations Snapshotting, replicating data
after node remove completed.
MutationStage Local writes A high number of pending
write requests indicates a
problem handling them. Adding
a node, tuning hardware and
configuration, or updating data
models will improve handling.
216
Cassandra tools
Name of Task Related information

statistic
PendingRangeCalculator
Calculate pending ranges per bootstraps and Developer notes
departed nodes
ReadRepairStageA digest query and update of replicas of a key Fast providing good connectivity
between replicas exists. If
pending grows too large, attempt
to lower the rate for high-read
tables by altering the table to use
a smaller read_repair_chance
value, like 0.11.
ReadStage Local reads Performing a local read. Also
includes deserializing data from
row cache. Pending values can
cause increased read latency.
Generally resolved by adding
nodes or tuning the system.
RequestResponseStage
Handle responses from other nodes
ValidationExecutorValidates schema
Table 13: Droppable Messages
Message Type Stage Notes

BINARY n/a This is deprecated and no longer
has any use
_TRACE n/a (special) Used for recording traces
(nodetool settraceprobability)
Has a special executor (1 thread,
1000 queue depth) that throws
away messages on insertion
instead of within the execute
MUTATION MutationStage If a write message is
processed after its timeout
(write_request_timeout_in_ms) it
either sent a failure to the client or
it met its requested consistency
level and will relay on hinted
handoff and read repairs to do the
mutation if it succeeded.
COUNTER_MUTATION MutationStage If a write message is
processed after its timeout
(write_request_timeout_in_ms) it
either sent a failure to the client or
it met its requested consistency
level and will relay on hinted
handoff and read repairs to do the
mutation if it succeeded.
READ_REPAIR MutationStage Times out after
write_request_timeout_in_ms
217
Cassandra tools
READ ReadStage Times out after

read_request_timeout_in_ms. No
point in servicing reads after that
point since it would of returned
error to client
RANGE_SLICE ReadStage Times out after
range_request_timeout_in_ms.
PAGED_RANGE ReadStage Times out after
request_timeout_in_ms.
REQUEST_RESPONSE RequestResponseStage Times out after
request_timeout_in_ms.
Response was completed and
sent back but not before the
timeout
Example
Run the command every two seconds.
$ nodetool -h labcluster tpstats

Example output is:
Pool Name Active Pending Completed Blocked All

time blocked
CounterMutationStage 0 0 0 0
0
ReadStage 0 0 103 0
0
RequestResponseStage 0 0 0 0
0
MutationStage 0 0 13234794 0
0
ReadRepairStage 0 0 0 0
0
GossipStage 0 0 0 0
0
CacheCleanupExecutor 0 0 0 0
0
AntiEntropyStage 0 0 0 0
0
MigrationStage 0 0 11 0
0
ValidationExecutor 0 0 0 0
0
CommitLogArchiver 0 0 0 0
0
MiscStage 0 0 0 0
0
MemtableFlushWriter 0 0 126 0
0
MemtableReclaimMemory 0 0 126 0
0
PendingRangeCalculator 0 0 1 0
0
MemtablePostFlush 0 0 1468 0
0
CompactionExecutor 0 0 254 0
0
218
Cassandra tools
InternalResponseStage 0 0 1 0
0
HintedHandoff 0 0 0
Message type Dropped

RANGE_SLICE 0
READ_REPAIR 0
PAGED_RANGE 0
BINARY 0
READ 0
MUTATION 180
_TRACE 0
REQUEST_RESPONSE 0
COUNTER_MUTATION 0
nodetool truncatehints
Truncates all hints on the local node, or truncates hints for the one or more endpoints.
Truncates all hints on the local node, or truncates hints for the one or more endpoints.
Synopsis
$ nodetool <options> truncatehints -- ( <endpoint> ... )

• options are:
• endpoint is one or more endpoint IP addresses or host names which hints are to be deleted.
Synopsis Legend
nodetool upgradesstables
Rewrites older SSTables to the current version of Cassandra.
Rewrites older SSTables to the current version of Cassandra.
Synopsis
$ nodetool <options> upgradesstables

( -a | --include-all-sstables )
-- <keyspace> ( <table> ... )
• options are:
219
Cassandra tools

• -a or --include-all-sstables includes all SSTables, even those with the current settings. See Examples
below.
• keyspace is the keyspace name.
Synopsis Legend
Description
Rewrites SSTables on a node that are incompatible with the current version. Use this command when
upgrading your server or changing compression options.
Examples
$ upgradesstables
Reads only SSTables created by old major versions of Cassandra and re-writes them to the current version
one at a time.
$ upgradesstables -a
Reads all existing SSTables and re-writes them to the current Cassandra version one at a time.
$ upgradesstables -a keyspace table
Reads specific SSTables and re-writes them to the current Cassandra version one at a time.
nodetool version
Provides the version number of Cassandra running on the specified node.
Provides the version number of Cassandra running on the specified node.
Synopsis
$ nodetool <options> version

options are:
Synopsis Legend
220
Cassandra tools
Cassandra bulk loader (sstableloader)

Provides the ability to bulk load external data into a cluster, load existing SSTables into another cluster with
a different number of nodes or replication strategy, and restore snapshots.
About this task

The Cassandra bulk loader, also called the sstableloader, provides the ability to:
• Bulk load external data into a cluster.
• Load existing SSTables into another cluster with a different number of nodes or replication strategy.
• Restore snapshots.
The sstableloader streams a set of SSTable data files to a live cluster; it does not simply copy the set
of SSTables to every node, but transfers the relevant part of the data to each node, conforming to the
replication strategy of the cluster. The table into which the data is loaded does not need to be empty.
If tables are repaired in a different cluster, after being loaded, the tables will be unrepaired.
Prerequisites
Because sstableloader uses Cassandra gossip, make sure of the following:
• The cassandra.yaml configuration file is in the classpath and properly configured.
• At least one node in the cluster is configured as seed.
• In the cassandra.yaml file, the following properties are properly configured for the cluster that you
are importing into:
• cluster_name
• listen_address
• storage_port
• rpc_address
• rpc_port
When using sstableloader to load external data, you must first generate SSTables.
If using DataStax Enterprise, you can use Sqoop to migrate external data to Cassandra.
Generating SSTables
SSTableWriter is the API to create raw Cassandra data files locally for bulk load into your cluster. The
Cassandra source code includes the CQLSSTableWriter implementation for creating SSTable files from
external data without needing to understand the details of how those map to the underlying storage engine.
Import the org.apache.cassandra.io.sstable.CQLSSTableWriter class, and define the schema
for the data you want to import, a writer for the schema, and a prepared insert statement, as shown in
Cassandra 2.0.1, 2.0.2, and a quick peek at 2.0.3.
Using sstableloader
Before loading the data, you must define the schema of the tables with CQL or Thrift.
To get the best throughput from SSTable loading, you can use multiple instances of sstableloader to
stream across multiple machines. No hard limit exists on the number of SSTables that sstableloader can
run at the same time, so you can add additional loaders until you see no further improvement.
If you use sstableloader on the same machine as the Cassandra node, you can't use the same network
interface as the Cassandra node. However, you can use the JMX StorageService > bulkload() call from
that node. This method takes the absolute path to the directory where the SSTables are located, and
loads them just as sstableloader does. However, because the node is both source and destination for the
streaming, it increases the load on that node. This means that you should load data from machines that are
not Cassandra nodes when loading into a live cluster.
221
Cassandra tools
Usage:
$ sstableloader [options] path_to_keyspace

$ cd install_location/bin
$ sstableloader [options] path_to_keyspace
The sstableloader bulk loads the SSTables found in the keyspace directory to the configured target cluster,
where the parent directories of the directory path are used as the target keyspace/table name.
1. Go to the location of the SSTables:
$ cd /var/lib/cassandra/data/Keyspace1/Standard1/
2. To view the contents of the keyspace:
$ ls
Keyspace1-Standard1-jb-60-CRC.db
Keyspace1-Standard1-jb-60-Data.db
...
Keyspace1-Standard1-jb-60-TOC.txt
3. To bulk load the files, specify the path to Keyspace1/Standard1/ in the target cluster:
$ sstableloader -d 110.82.155.1 /var/lib/cassandra/data/Keyspace1/Standard1/

## package installation
$ install_location/bin/sstableloader -d 110.82.155.1 /var/lib/cassandra/

data/Keyspace1/Standard1/ ## tarball installation
This bulk loads all files.
Table 14: sstableloader options
Short option Long option Description

-alg --ssl-alg Client SSL algorithm (default: SunX509).
<ALGORITHM>
-ciphers --ssl-ciphers <CIPHER- Client SSL. Comma-separated list of encryption suites.
SUITES>
-cph --connections-per-host Number of concurrent connections-per-host.
<connectionsPerHost>
-d --nodes <initial_hosts> Required. Connect to a list of (comma separated) hosts for
initial cluster information.
-f --conf-path Path to the cassandra.yaml path for streaming
<path_to_config_file> throughput and client/server SSL.
-h --help Display help.
-i --ignore <NODES> Do not stream to this comma separated list of nodes.
-ks --keystore Client SSL. Full path to the keystore.
<KEYSTORE>
222
Cassandra tools
Short option Long option Description

-kspw --keystore-password Client SSL. Password for the keystore.
<KEYSTORE-
PASSWORD>
--no-progress Do not display progress.
-p --port <rpc port> RPC port (default: 9160 [Thift]).
-prtcl --ssl-protocol Client SSL. Connections protocol to use (default: TLS).
<PROTOCOL>
-pw --password Password for Cassandra authentication.
<password>
-st --store-type <STORE- Client SSL. Type of store.
TYPE>
-t --throttle <throttle> Throttle speed in Mbits (default: unlimited).
-tf --transport-factory Fully-qualified ITransportFactory class name for
<transport factory> creating a connection to Cassandra.
-ts --truststore Client SSL. Full path to truststore.
<TRUSTSTORE>
-tspw --truststore-password Client SSL. Password of the truststore.
<TRUSTSTORE-
PASSWORD>
-u --username User name for Cassandra authentication.
<username>
-v --verbose Verbose output.
The following cassandra.yaml options can be over-ridden from the command line:
Option in cassandra.yaml Command line example

stream_throughput_outbound_megabits_per_sec --throttle 300
server_encryption_options --ssl-protocol none
client_encryption_options --keystore-password MyPassword

conf/cassandra.yaml
The cassandra utility

Cassandra start-up parameters can be run from the command line (in Tarball installations) or specified in
the cassandra-env.sh file (Package or Tarball installations).
About this task

Cassandra start-up parameters can be run from the command line (Tarball installations) or specified in the
cassandra-env.sh file (Package or Tarball installations).
223
Cassandra tools
Note: You can also use the cassandra-env.sh file to pass additional options, such as maximum
and minimum heap size, to the Java virtual machine rather than setting them in the environment.

conf/cassandra.yaml

Usage
Add the following to the cassandra-env.sh file:
JVM_OPTS="$JVM_OPTS -D[PARAMETER]"

For Tarball installations, you can run this tool from the command line:
$ cassandra [OPTIONS]
Examples:
• cassandra-env.sh: JVM_OPTS="$JVM_OPTS -Dcassandra.load_ring_state=false"
• Command line: bin/cassandra -Dcassandra.load_ring_state=false
The Example section contains more examples.
Command line only options
Option Description
-f Start the cassandra process in foreground. The default is to start as background
process.
-h Help.
-p filename Log the process ID in the named file. Useful for stopping Cassandra by killing its PID.
-v Print the version and exit.
Start-up parameters
The -D option specifies the start-up parameters in both the command line and cassandra-env.sh file.
cassandra.auto_bootstrap=false
Facilitates setting auto_bootstrap to false on initial set-up of the cluster. The next time you start the cluster,
you do not need to change the cassandra.yaml file on each node to revert to true.
cassandra.available_processors=number_of_processors
224
Cassandra tools
In a multi-instance deployment, multiple Cassandra instances will independently assume that all CPU
processors are available to it. This setting allows you to specify a smaller set of processors.
cassandra.boot_without_jna=true
If JNA fails to initialize, Cassandra fails to boot. Use this command to boot Cassandra without JNA.
cassandra.config=directory
The directory location of the cassandra.yaml file. The default location depends on the type of installation.
cassandra.initial_token=token
Use when virtual nodes (vnodes) are not used. Sets the initial partitioner token for a node the first time the
node is started. (Default: disabled)
Note: Vnodes are highly recommended as they automatically select tokens.
cassandra.join_ring=true|false
Set to false to start Cassandra on a node but not have the node join the cluster. (Default: true) You can use
nodetool join and a JMX call to join the ring afterwards.
cassandra.load_ring_state=true|false
Set to false to clear all gossip state for the node on restart. (Default: true)
cassandra.metricsReporterConfigFile=file
Enable pluggable metrics reporter. See Pluggable metrics reporting in Cassandra 2.0.2.
cassandra.native_transport_port=port
Set the port on which the CQL native transport listens for clients. (Default: 9042)
cassandra.partitioner=partitioner
Set the partitioner. (Default: org.apache.cassandra.dht.Murmur3Partitioner)
cassandra.replace_address=listen_address or broadcast_address of dead node
To replace a node that has died, restart a new node in its place specifying the listen_address or
broadcast_address that the new node is assuming. The new node must not have any data in its data
directory, that is, it must be in the same state as before bootstrapping.
Note: The broadcast_address defaults to the listen_address except when using the
Ec2MultiRegionSnitch.
cassandra.replayList=table
Allow restoring specific tables from an archived commit log.
cassandra.ring_delay_ms=ms
Defines the amount of time a node waits to hear from other nodes before formally joining the ring. (Default:
1000ms)
cassandra.rpc_port=port
Set the port for the Thrift RPC service, which is used for client connections. (Default: 9160).
cassandra.ssl_storage_port=port
Set the SSL port for encrypted communication. (Default: 7001)
cassandra.start_native_transport=true|false
Enable or disable the native transport server. See start_native_transport in cassandra.yaml. (Default:
true)
cassandra.start_rpc=true/false
Enable or disable the Thrift RPC server. (Default: true)
cassandra.storage_port=port
Set the port for inter-node communication. (Default: 7000)
cassandra.triggers_dir=directory
Set the default location for the trigger JARs. (Default: conf/triggers)
225
Cassandra tools
cassandra.write_survey=true
For testing new compaction and compression strategies. It allows you to experiment with different strategies
and benchmark write performance differences without affecting the production workload. See Testing
compaction and compression.
consistent.rangemovement=true
True makes Cassandra 2.1 bootstrapping behavior effective. False makes Cassandra 2.0 behavior effective.
Example
Clear gossip state when starting a node:
• Command line: bin/cassandra -Dcassandra.load_ring_state=false
• cassandra-env.sh: JVM_OPTS="$JVM_OPTS -Dcassandra.load_ring_state=false"
Start Cassandra on a node and do not join the cluster:
• Command line: bin/cassandra -Dcassandra.join_ring=false
• cassandra-env.sh: JVM_OPTS="$JVM_OPTS -Dcassandra.join_ring=false"
Replacing a dead node:
• Command line: bin/cassandra -Dcassandra.replace_address=10.91.176.160
• cassandra-env.sh: JVM_OPTS="$JVM_OPTS -Dcassandra.replace_address=10.91.176.160"
The cassandra-stress tool

A Java-based stress testing utility for basic benchmarking and load testing a Cassandra cluster.
About this task

The cassandra-stress tool is a Java-based stress testing utility for basic benchmarking and load testing a
Cassandra cluster.
The choices you make when data modeling your application can make a big difference in how your
application performs. Creating the best data model requires significant load testing and multiple iterations.
The cassandra-stress tool helps you in this endeavor by populating your cluster and supporting stress
testing of arbitrary CQL tables and arbitrary queries on tables. Use the cassandra-stress tool to:
• Quickly determine how a schema performs.
• Understand how your database scales.
• Optimize your data model and settings.
• Determine production capacity.
Note: You can also use this tool on Cassandra 2.0 clusters.
The cassandra-stress tool also supports a YAML-based profile for defining specific schema with potential
compaction strategies, cache settings, and types. Sample files are located in:
• Package installations: /usr/share/doc/cassandra/examples
• Tarball installations: install_location/tools
For a complete description on using these sample files, see Improved Cassandra 2.1 Stress Tool:
Benchmark Any Schema – Part 1.
The cassandra-stress tool creates a keyspace called keyspace1 and within that, tables named standard1,
counter1, depending on what type of table is being tested. These are automatically created the first time
you run the stress test and are reused on subsequent runs unless you drop the keyspace using CQL. You
cannot change the names; they are hard-coded.
Usage:
226
Cassandra tools
• Package installations: cassandra-stress command [options]

• Tarball installations: install_location/tools/bin/cassandra-stress command
[options]
On tarball installations, you can use these commands and options with or without the cassandra-stress
daemon running.
Command Description
read Multiple concurrent reads. The cluster must first be populated by a write test.
write Multiple concurrent writes against the cluster.
mixed Interleave basic commands with configurable ratio and distribution. The cluster must first
be populated by a write test.
Multiple concurrent updates of counters.
counter_write
counter_readMultiple concurrent reads of counters. The cluster must first be populated by a
counter_write test.
user Interleave user provided queries with configurable ratio and distribution.
help Display help for a command or option.
Display help for an option: cassandra-stress help [options] For example:
cassandra-stress help -schema
print Inspect the output of a distribution definition.

legacy Legacy support mode.
Important: Additional sub options are available for each option in the following table. Format:
$ cassandra-stress help option

For an example, see View schema help.
Option Description
-pop Population distribution and intra-partition visit order.
Usage
$ -pop seq=? [no-wrap] [read-lookback=DIST(?)] [contents=?]
or
-pop [dist=DIST(?)] [contents=?]
-insert Insert specific options relating to various methods for batching and splitting partition updates.
Usage
$ -insert [revisit=DIST(?)] [visits=DIST(?)] partitions=DIST(?)
[batchtype=?] select-ratio=DIST(?)
-col Column details, such as size and count distribution, data generator, names, and comparator.
Usage
$ -col [n=DIST(?)] [slice] [super=?] [comparator=?] [timestamp=?]
[size=DIST(?)]
-rate Thread count, rate limit, or automatic mode (default is auto).

Usage
$ -rate threads=? [limit=?]
or
227
Cassandra tools
Option Description
$ -rate [threads>=?] [threads<=?] [auto]
-mode Thrift or CQL with options.

Usage
$ -mode thrift cql2 [prepared]
-errors How to handle errors when encountered during stress.

Usage
$ -errors [retries=?] [ignore]
-sample Specify the number of samples to collect for measuring latency.

Usage
$ -sample [history=?] [live=?] [report=?]
-schema Replication settings, compression, compaction, and so on.

Usage
$ -schema [replication(?)] [keyspace=?] [compaction(?)]
[compression=?]
-node Nodes to connect to.

Usage
$ -node [whitelist] [file=?] []
-log Where to log progress and the interval to use.

Usage
$ -log [level=?] [no-summary] [file=?] [interval=?]
- Custom transport factories.

transport
Usage
$ -transport [factory=?] [truststore=?] [truststore-password=?]
[ssl-protocol=?] [ssl-alg=?] [store-type=?] [ssl-ciphers=?]
-port Specify port for connecting Cassandra nodes.

Usage
$ -port [native=?] [thrift=?] [jmx=?]
-sendto Specify stress server to send this command to.

Usage
$ -sendToDaemon <host>
View schema help
$ cassandra-stress help -schema
replication([strategy=?][factor=?][<option 1..N>=?]): Define

the replication strategy and any parameters
strategy=? (default=org.apache.cassandra.locator.SimpleStrategy) The
replication strategy to use
factor=? (default=1) The
number of replicas
keyspace=? (default=keyspace1) The
keyspace name to use
228
Cassandra tools
compaction([strategy=?][<option 1..N>=?]): Define

the compaction strategy and any parameters
strategy=? The
compaction strategy to use
compression=?
Specify the compression to use for SSTable, default:no compression
Populate the database

Generally it is easier to let cassandra-stress create the basic schema and then modify it in CQL:
#Load one row with default schema

$ cassandra-stress write n=1 cl=one -mode native cql3 -log file=~/
create_schema.log
#Modify schema in CQL

$ cqlsh
#Run a real write workload

$ cassandra-stress write n=1000000 cl=one -mode native cql3 -schema
keyspace="keyspace1" -log file=~/load_1M_rows.log
Running a mixed workload

When running a mixed workload, you must escape parentheses, greater-than and less-than signs, and
other such things. This example invokes a workload that is one-quarter writes and three-quarters reads.
$ cassandra-stress mixed ratio$write=1,read=3$ n=100000 cl=ONE -pop

dist=UNIFORM$1..1000000$ -schema keyspace="keyspace1" -mode native cql3 -
rate threads\>=16 threads\<=256 -log file=~/mixed_autorate_50r50w_1M.log
Notice the following in this example:
1. The ratio requires backslash-escaped parenthesis.
2. The value of n is different than in write phase. During the write phase, n records are written. However in
the read phase, if n is too large, it is inconvenient to read all the records for simple testing. Generally, n
does not need be large when validating the persistent storage systems of a cluster.
The -pop dist=UNIFORM$1..1000000$ portion says that of the n=100,000 operations, select the
keys uniformly distributed between 1 and 1,000,000. Use this when you want to specify more data per
node than what fits in DRAM.
3. In the rate section, the greater-than and less-than signs are escaped. If not escaped, the shell will
attempt to use them for IO redirection. Specifically, the shell will try to read from a non-existent file
called =256 and create a file called =16. The rate section tells cassandra-stress to automatically
attempt different numbers of client threads and not test less that 16 or more than 256 client threads.
Standard mixed read/write workload keyspace for a single node
CREATE KEYSPACE "keyspace1" WITH replication = {

'class': 'SimpleStrategy',
'replication_factor': '1'
};
USE "keyspace1";
CREATE TABLE "standard1" (
key blob,
"C0" blob,
"C1" blob,
"C2" blob,
229
Cassandra tools
"C3" blob,
"C4" blob,
PRIMARY KEY (key)
) WITH
bloom_filter_fp_chance=0.010000 AND
caching='KEYS_ONLY' AND
comment='' AND
dclocal_read_repair_chance=0.000000 AND
gc_grace_seconds=864000 AND
index_interval=128 AND
read_repair_chance=0.100000 AND
replicate_on_write='true' AND
default_time_to_live=0 AND
speculative_retry='99.0PERCENTILE' AND
memtable_flush_period_in_ms=0 AND
compaction={'class': 'SizeTieredCompactionStrategy'} AND
compression={'sstable_compression': 'LZ4Compressor'};
Splitting up a load over multiple cassandra-stress instances on different nodes

This example is useful for loading into large clusters, where a single cassandra-stress load generator node
cannot saturate the cluster. In this example, $NODES is a variable whose value is a comma delimited list of
IP addresses such as 10.0.0.1,10.0.0.2, and so on.
#On Node1
keyspace="keyspace1" -pop seq=1..1000000 -log file=~/node1_load.log -node
$NODES
#On Node2
keyspace="keyspace1" -pop seq=1000001..2000000 -log file=~/node2_load.log -
node $NODES
Note the keyspace is defined and the -key option tells each instance which range of keys to populate.
Using the Daemon Mode

The daemon in larger-scale testing can prevent potential skews in the test results by keeping the JVM
warm.
The daemon is only available in tarball installations. Run the daemon from:
install_location/tools/bin/cassandra-stressd start|stop|status [-h

<host>]
During stress testing, you can keep the daemon running and send it commands through it using the --
send-to option.
Example
• Insert 1,000,000 rows to given host:
/tools/bin/cassandra-stress -d 192.168.1.101
When the number of rows is not specified, one million rows are inserted.
• Read 1,000,000 rows from given host:
tools/bin/cassandra-stress -d 192.168.1.101 -o read
230
Cassandra tools
• Insert 10,000,000 rows across two nodes:
/tools/bin/cassandra-stress -d 192.168.1.101,192.168.1.102 -n 10000000

• Insert 10,000,000 rows across two nodes using the daemon mode:
/tools/bin/cassandra-stress -d 192.168.1.101,192.168.1.102 -n 10000000 --

send-to 54.0.0.1
Interpreting the output of cassandra-stress

About the output from the running tests.
Each line reports data for the interval between the last elapsed time and current elapsed time.
Created keyspaces. Sleeping 1s for propagation.

Warming up WRITE with 50000 iterations...
INFO 23:57:05 Using data-center name 'datacenter1' for
DCAwareRoundRobinPolicy (if this is incorrect, please provide the correct
datacenter name with DCAwareRoundRobinPolicy constructor)
Connected to cluster: Test Cluster
Datacenter: datacenter1; Host: localhost/127.0.0.1; Rack: rack1
INFO 23:57:05 New Cassandra host localhost/127.0.0.1:9042 added
Sleeping 2s...
WARNING: uncertainty mode (err<) results in uneven workload between thread
runs, so should be used for high level analysis only
Running with 4 threadCount
Running WRITE with 4 threads until stderr of mean < 0.02
total ops , adj row/s, op/s, pk/s, row/s, mean, med, .95,
.99, .999, max, time, stderr, gc: #, max ms, sum ms, sdv
ms, mb
2552 , 2553, 2553, 2553, 2553, 1.5, 1.4, 2.5,
6.0, 12.6, 18.0, 1.0, 0.00000, 0, 0, 0,
0, 0
5173 , 2634, 2613, 2613, 2613, 1.5, 1.5, 1.8,
2.6, 8.6, 9.2, 2.0, 0.00000, 0, 0, 0,
0, 0
...
Results:
op rate : 3954
partition rate : 3954
row rate : 3954
latency mean : 1.0
latency median : 0.8
latency 95th percentile : 1.5
latency 99th percentile : 1.8
latency 99.9th percentile : 2.2
latency max : 73.6
total gc count : 25
total gc mb : 1826
total gc time (s) : 1
avg gc time(ms) : 37
stdev gc time(ms) : 10
Total operation time : 00:00:59
Sleeping for 15s
Running with 4 threadCount
231
Cassandra tools
Table 15: Output of cassandra-stress
Data Description
total ops Running total number of operations during the run.
op/s Number of operations per second performed during the run.
pk/s Number of partition operations per second performed during the
run.
row/s Number of row operations per second performed during the run.
mean Average latency in milliseconds for each operation during that run.
med Median latency in milliseconds for each operation during that run.
.95 95% of the time the latency was less than the number displayed in
the column.
.99 99% of the time the latency was less than the number displayed in
the column.
.999 99.9% of the time the latency was less than the number displayed
in the column.
max Maximum latency in milliseconds.
time Total operation time.
stderr Standard error of the mean. It is a measure of confidence in the
average throughput number; the smaller the number, the more
accurate the measure of the cluster's performance.
gc: # Number of garbage collections.
max ms Longest garbage collection in milliseconds.
sum ms Total of garbage collection in milliseconds.
sdv ms Standard deviation in milliseconds.
mb Size of the garbage collection in megabytes.
The sstablescrub utility

An offline version of nodetool scrub. It attempts to remove the corrupted parts while preserving non-
corrupted data.
About this task

The sstablescrub utility is an offline version of nodetool scrub. It attempts to remove the corrupted
parts while preserving non-corrupted data. Because sstablescrub runs offline, it can correct errors that
nodetool scrub cannot.
If scrubbing results in dropping rows, new SSTables become unrepaired. However, if no bad rows are
detected, the SSTable keeps its original repairedAt field, which denotes the time of the repair.
Procedure
1. Before using sstablescrub, try rebuilding the tables using nodetool scrub.
If nodetool scrub does not fix the problem, use this utility.
2. Shut down the node.
232
Cassandra tools
3. Run the utility:

• Package installations: sstablescrub [options] <keyspace> <table>
• Tarball installations: install_location/bin/sstablescrub [options] <keyspace>
<table>
Table 16: Options
Flag Option Description

--debug Display stack traces.
-m --manifest-check Only check and repair the leveled manifest, without actually scrubbing
the SSTables.
-s --skip-corrupted Skip corrupt rows in counter tables.
The sstablesplit utility

Use this tool to split SSTables files into multiple SSTables of a maximum designated size.
About this task

Use this tool to split SSTables files into multiple SSTables of a maximum designated size. For example,
if SizeTieredCompactionStrategy was used for a major compaction and results in a excessively large
SSTable, it's a good idea to split the table because won't get compacted again until the next huge
compaction.
Cassandra must be stopped to use this tool:


$ sudo kill pid
Usage:
• Package installations: sstablessplit [options] <filename> [<filename>]*
• Tarball installations: install_location/tools/bin/sstablessplit [options]
<filename> [<filename>]*
Example:
$ sstablesplit -s 40 /var/lib/cassandra/data/Keyspace1/Standard1/*
Table 17: Options

--no-snapshot Do not snapshot the SSTables before splitting.
233
Cassandra tools

-s --size <size> Maximum size in MB for the output SSTables (default: 50).
The sstablekeys utility

The sstablekeys utility dumps table keys.
About this task

The sstablekeys utility dumps table keys. To list the keys in an SSTable, find the name of the SSTable file.
The file is located in the data directory and has a .db extension. The location of the data directory, listed in
the "Install locations" section, depends on the type of installation. After finding the name of the file, use the
name as an argument to the sstablekeys command.
$ bin/sstablekeys <sstable_name>
Procedure
1. Create the playlists table in the music keyspace as shown in Data modeling.
2. Insert the row of data about ZZ Top in playlists:
INSERT INTO music.playlists (id, song_order, song_id, title, artist,

album)
VALUES (62c36092-82a1-3a00-93d1-46196ee77204,
1,
a3e64f8f-bd44-4f28-b8d9-6938726e34d4,
'La Grange',
'ZZ Top',
'Tres Hombres');
3. Flush the data to disk.
$ nodetool flush music playlists

4. Look at keys in the SSTable data. For example, use sstablekeys followed by the path to the data. Use
the path to data for your Cassandra installation:
$ sstablekeys <path to data>/data/data/music/

playlists-8b9f4cc0229211e4b02073ded3cb6170/music-playlists-ka-1-Data.db
The output appears, for example:
62c3609282a13a0093d146196ee77204
The sstableupgrade tool

Upgrade the SSTables in the specified table or snapshot to match the currently installed version of
Cassandra.
About this task

This tool rewrites the SSTables in the specified table or snapshot to match the currently installed version of
Cassandra.
If restoring with sstableloader, you must upgrade your snapshots before restoring for any snapshot taken in
a major version older than the major version that Cassandra is currently running.
234
Cassandra tools
Usage:
• Package installations: sstableupgrade [options] <keyspace> <cf> [snapshot]
• Tarball installations: install_location/bin/sstableupgrade [options] <keyspace>
<cf> [snapshot]
The snapshot option only upgrades the specified snapshot.
Table 18: Options

235
References
References
Reference topics.
Starting and stopping Cassandra

Topics for starting and stopping Cassandra.
Starting Cassandra as a service

Start the Cassandra Java server process for packaged installations.
About this task

Start the Cassandra Java server process for packaged installations.
Startup scripts are provided in the /etc/init.d directory. The service runs as the cassandra user.
Procedure
You must have root or sudo permissions to start Cassandra as a service.
On initial start-up, each node must be started one at a time, starting with your seed nodes:

On Enterprise Linux systems, the Cassandra service runs as a java process.
Starting Cassandra as a stand-alone process

Start the Cassandra Java server process for tarball installations.
About this task

Start the Cassandra Java server process for tarball installations.
Procedure
On initial start-up, each node must be started one at a time, starting with your seed nodes.
• To start Cassandra in the background:
$ bin/cassandra
• To start Cassandra in the foreground:
$ bin/cassandra -f
Stopping Cassandra as a service

Stopping the Cassandra Java server process on packaged installations.
About this task

Stopping the Cassandra Java server process on packaged installations.
236
References
Procedure
You must have root or sudo permissions to stop the Cassandra service:
Stopping Cassandra as a stand-alone process

Stop the Cassandra Java server process on tarball installations.
About this task

Stop the Cassandra Java server process on tarball installations.
Procedure
Find the Cassandra Java process ID (PID), and then kill the process using its PID number:

$ sudo kill pid
Clearing the data as a service

Remove all data from a package installation. Special instructions for AMI restart.
About this task

Remove all data from a package installation.
Procedure
To clear the data from the default directories:
After stopping the service, run the following command:
$ sudo rm -rf /var/lib/cassandra/*

Note: If you are clearing data from an AMI installation for restart, you need to preserve the log
files.
Clearing the data as a stand-alone process

Remove all data from a tarball installation.
About this task

Remove all data from a tarball installation.
Procedure
To clear all data from the default directories, including the commitlog and saved_caches:
1. After stopping the process, run the following command from the install directory:
$ sudo rm -rf data/*
2. To clear the only the data directory:
$ sudo rm -rf data/data/*
237
References
Install locations
Install location topics.
Tarball installation directories

Configuration files directory locations.
The configuration files are located in the following directories:
Configuration Files Locations

cassandra.yaml install_location/conf
cassandra-topology.properties install_location/conf
cassandra-rackdc.properties install_location/conf
cassandra-env.sh install_location/conf
cassandra.in.sh install_location/bin
The binary tarball releases install into the installation directory.
Directories Description
data Files for commitlog, data, and saved_caches
(unless set in cassandra.yaml)
bin Utilities and start scripts
conf Configuration files and environment settings
interface Thrift and Avro client APIs
javadoc Cassandra Java API documentation
lib JAR and license files
tools Cassandra tools and sample cassandra.yaml
files for stress testing.
For DataStax Enterprise installs, see the documentation for your DataStax Enterprise version.
Package installation directories

Configuration files directory locations.
The configuration files are located in the following directories:
Configuration Files Locations

cassandra.yaml /etc/cassandra
cassandra-topology.properties /etc/cassandra
cassandra-rackdc.properties /etc/cassandra
cassandra-env.sh /etc/cassandra
cassandra.in.sh /usr/share/cassandra
The packaged releases install into these directories:
238
References
Directories Description
/var/lib/ Data directories
cassandra
/var/log/ Log directory
cassandra
/var/run/ Runtime files
cassandra
/usr/share/ Environment settings
cassandra
/usr/share/ JAR files
cassandra/lib
/usr/bin Optional utilities, such as sstablelevelreset, sstablerepairedset, and sstablesplit
/usr/bin Binary files
/usr/sbin
/etc/cassandra Configuration files
/etc/init.d Service startup script
/etc/security/ Cassandra user limits
limits.d
/etc/default
/usr/share/ Sample cassandra.yaml files for stress testing.
doc/cassandra/
examples
For DataStax Enterprise installs, see the documentation for your DataStax Enterprise version.
Cassandra include file

Set environment variables (cassandra.in.sh).
To set environment variables (Linux only), Cassandra can use an include file, cassandra.in.sh. This
file is located in:
• Tarball installations: install_location/bin/cassandra.in.sh
• Package installations: /usr/share/cassandra/cassandra.in.sh
Cassandra-CLI utility (deprecated)

Deprecated configuration attributes. Will be removed in Cassandra 3.0.
Important: The CLI utility is deprecated and will be removed in Cassandra 3.0. For ease of use
and performance, switch from Thrift and CLI to CQL and cqlsh.
Keyspace attributes
Cassandra stores storage configuration attributes in the system keyspace. You can set storage engine
configuration attributes on a per-keyspace or per-table basis on the command line using the Cassandra-
CLI utility. A keyspace must have a user-defined name, a replica placement strategy, and options that
specify the number of copies per data center or node.
239
References
name
Required. The name for the keyspace.
placement_strategy
Required. Determines how Cassandra distributes replicas for a keyspace among nodes in the ring. Values
are:
• SimpleStrategy or org.apache.cassandra.locator.SimpleStrategy
• NetworkTopologyStrategy or
org.apache.cassandra.locator.NetworkTopologyStrategy
NetworkTopologyStrategy requires a snitch to be able to determine rack and data center locations of a node.
For more information about replication placement strategy, see Data replication.
strategy_options
Specifies configuration options for the chosen replication strategy class. The replication factor option is the
total number of replicas across the cluster. A replication factor of 1 means that there is only one copy of
each row on one node. A replication factor of 2 means there are two copies of each row, where each copy
is on a different node. All replicas are equally important; there is no primary or master replica. As a general
rule, the replication factor should not exceed the number of nodes in the cluster. However, you can increase
the replication factor and then add the desired number of nodes.
When the replication factor exceeds the number of nodes, writes are rejected, but reads are served as long
as the desired consistency level can be met.
For more information about configuring the replication placement strategy for a cluster and data centers,
see Choosing keyspace replication options.
durable_writes
(Default: true) When set to false, data written to the keyspace bypasses the commit log. Be careful using
this option because you risk losing data.
Table attributes
Attributes per table.
The following attributes can be declared per table.
bloom_filter_fp_chance
See CQL properties in CQL for Cassandra 2.x.
bucket_high
See CQL Compaction Subproperties in CQL for Cassandra 2.x.
bucket_low
caching
chunk_length_kb
See CQLCompression Subproperties in CQL for Cassandra 2.x.
column_metadata
(Default: N/A - container attribute) Column metadata defines these attributes of a column:
• name: Binds a validation_class and (optionally) an index to a column.
• validation_class: Type used to check the column value.
• index_name: Name of the index.
• index_type: Type of index. Currently the only supported value is KEYS.
240
References
Setting a value for the name option is required. The validation_class is set to the default_validation_class
of the table if you do not set the validation_class option explicitly. The value of index_type must be set to
create an index for a column. The value of index_name is not valid unless index_type is also set.
Setting and updating column metadata with the Cassandra-CLI utility requires a slightly different command
syntax than other attributes; note the brackets and curly braces in this example:
[default@demo ] UPDATE COLUMN FAMILY users WITH comparator =UTF8Type

AND column_metadata =[{column_name: full_name, validation_class: UTF8Type,
index_type: KEYS }];
column_type
(Default: Standard) The standard type of table contains regular columns.
comment
compaction_strategy
See compaction in CQL properties in CQL for Cassandra 2.x.
compaction_strategy_options
(Default: N/A - container attribute) Sets attributes related to the chosen compaction-strategy. Attributes are:
• bucket_high
• bucket_low
• max_compaction_threshold
• min_compaction_threshold
• min_sstable_size
• sstable_size_in_mb
• tombstone_compaction_interval
• tombstone_threshold
comparator
(Default: BytesType) Defines the data types used to validate and sort column names. There are several
built-in column comparators available. The comparator cannot be changed after you create a table.
compression_options
(Default: N/A - container attribute) Sets the compression algorithm and sub-properties for the table. Choices
are:
• sstable_compression
• chunk_length_kb
• crc_check_chance
crc_check_chance
See CQLCompression Subproperties in CQL for Cassandra 2.x.
default_time_to_live
default_validation_class
(Default: N/A) Defines the data type used to validate column values. There are several built-in column
validators available.
gc_grace
index_interval
key_validation_class
241
References
(Default: N/A) Defines the data type used to validate row key values. There are several built-in key validators
available, however CounterColumnType (distributed counters) cannot be used as a row key validator.
max_compaction_threshold
See max_threshold in CQL Compaction Subproperties in CQL for Cassandra 2.x.
min_compaction_threshold
See min_threshold in CQL Compaction Subproperties in CQL for Cassandra 2.x.
max_index_interval
min_index_interval
memtable_flush_after_mins
Deprecated as of Cassandra 1.0, but can still be declared for backward compatibility. Use
commitlog_total_space_in_mb.
memtable_flush_period_in_ms
memtable_operations_in_millions
memtable_throughput_in_mb
min_sstable_size
name
(Default: N/A) Required. The user-defined name of the table.
read_repair_chance
speculative_retry
sstable_size_in_mb
sstable_compression
See compression in CQL properties in CQL for Cassandra 2.x.
tombstone_compaction_interval
tombstone_threshold
242
Moving data to or from other databases
Moving data to or from other databases

Solutions for migrating from other databases.
Cassandra offers several solutions for migrating from other databases:
• The COPY command, which mirrors what the PostgreSQL RDBMS uses for file/export import.
• The Cassandra bulk loader provides the ability to bulk load external data into a cluster.
About the COPY command

You can use COPY in Cassandra’s CQL shell to load flat file data into Cassandra (nearly all RDBMS’s
have unload utilities that allow table data to be written to OS files) as well as data to be written out to OS
files.
ETL Tools
If you need more sophistication applied to a data movement situation (more than just extract-load), then
you can use any number of extract-transform-load (ETL) solutions that now support Cassandra. These
tools provide excellent transformation routines that allow you to manipulate source data in literally any way
you need and then load it into a Cassandra target. They also supply many other features such as visual,
point-and-click interfaces, scheduling engines, and more.
Many ETL vendors who support Cassandra supply community editions of their products that are free
and able to solve many different use cases. Enterprise editions are also available that supply many other
compelling features that serious enterprise data users need.
You can freely download and try ETL tools from Jaspersoft, Pentaho, and Talend that all work with
community Cassandra.
243
Troubleshooting
Troubleshooting
Peculiar Linux kernel performance problem on NUMA systems

Problems due to zone_reclaim_mode.
Problems due to zone_reclaim_mode.
The Linux kernel can be inconsistent in enabling/disabling zone_reclaim_mode. This can result in odd
performance problems:
• Random huge CPU spikes resulting in large increases in latency and throughput.
• Programs hanging indefinitely apparently doing nothing.
• Symptoms appearing and disappearing suddenly.
• After a reboot, the symptoms generally do not show again for some time.
To ensure that zone_reclaim_mode is disabled:
$ echo 0 > /proc/sys/vm/zone_reclaim_mode
Reads are getting slower while writes are still fast

The cluster's IO capacity is not enough to handle the write load it is receiving.
The cluster's IO capacity is not enough to handle the write load it is receiving.
Check the SSTable counts in cfstats. If the count is continually growing, the cluster's IO capacity is not
enough to handle the write load it is receiving. Reads have slowed down because the data is fragmented
across many SSTables and compaction is continually running trying to reduce them. Adding more IO
capacity, either via more machines in the cluster, or faster drives such as SSDs, will be necessary to solve
this.
If the SSTable count is relatively low (32 or less) then the amount of file cache available per machine
compared to the amount of data per machine needs to be considered, as well as the application's read
pattern. The amount of file cache can be formulated as (TotalMemory – JVMHeapSize) and if the
amount of data is greater and the read pattern is approximately random, an equal ratio of reads to the
cache:data ratio will need to seek the disk. With spinning media, this is a slow operation. You may be
able to mitigate many of the seeks by using a key cache of 100%, and a small amount of row cache
(10000-20000) if you have some hot rows and they are not extremely large.
Nodes seem to freeze after some period of time

Some portion of the JVM is being swapped out by the operating system (OS).
Some portion of the JVM is being swapped out by the operating system (OS).
Check your system.log for messages from the GCInspector. If the GCInspector is indicating that either
the ParNew or ConcurrentMarkSweep collectors took longer than 15 seconds, there is a high probability
that some portion of the JVM is being swapped out by the OS.
One way this might happen is if the mmap DiskAccessMode is used without JNA support. The address
space will be exhausted by mmap, and the OS will decide to swap out some portion of the JVM that
isn't in use, but eventually the JVM will try to GC this space. Adding the JNA libraries will solve this
(they cannot be shipped with Cassandra due to carrying a GPL license, but are freely available) or the
244
Troubleshooting
DiskAccessMode can be switched to mmap_index_only, which as the name implies will only mmap the
indices, using much less address space.
DataStax strongly recommends that you disable swap entirely (sudo swapoff --all). Because
Cassandra has multiple replicas and transparent failover, it is preferable for a replica to be killed
immediately when memory is low rather than go into swap. This allows traffic to be immediately redirected
to a functioning replica instead of continuing to hit the replica that has high latency due to swapping. If
your system has a lot of DRAM, swapping still lowers performance significantly because the OS swaps
out executable code so that more DRAM is available for caching disks. To make this change permanent,
remove all swap file entries from /etc/fstab.
If you insist on using swap, you can set vm.swappiness=1. This allows the kernel swap out the absolute
least used parts.
If the GCInspector isn't reporting very long GC times, but is reporting moderate times frequently
(ConcurrentMarkSweep taking a few seconds very often) then it is likely that the JVM is experiencing
extreme GC pressure and will eventually OOM. See the section below on OOM errors.
Nodes are dying with OOM errors

Nodes are dying with OutOfMemory exceptions.
Nodes are dying with OutOfMemory exceptions.
Check for these typical causes:
Row cache is too large, or is caching large rows
Row cache is generally a high-end optimization. Try disabling it and see if the OOM problems continue.
The memtable sizes are too large for the amount of heap allocated to the JVM
You can expect N + 2 memtables resident in memory, where N is the number of tables. Adding another
1GB on top of that for Cassandra itself is a good estimate of total heap usage.
If none of these seem to apply to your situation, try loading the heap dump in MAT and see which class is
consuming the bulk of the heap for clues.
Nodetool or JMX connections failing on remote nodes

Nodetool commands can be run locally but not on other nodes in the cluster.
Nodetool commands can be run locally but not on other nodes in the cluster.
If you can run nodetool commands locally but not on other nodes in the ring, you may have a common JMX
connection problem that is resolved by adding an entry like the following in install_location/conf/
cassandra-env.sh on each node:
JVM_OPTS = "$JVM_OPTS -Djava.rmi.server.hostname=<public name>"
If you still cannot run nodetool commands remotely after making this configuration change, do a full
evaluation of your firewall and network security. The nodetool utility communicates through JMX on port
7199.
Handling schema disagreements

Check for and resolve schema disagreements.
Check for and resolve schema disagreements.
245
Troubleshooting
About this task

In the event that a schema disagreement occurs, check for and resolve schema disagreements as follows:
Procedure
1. Run the nodetool describecluster command.
If any node is UNREACHABLE, you see output something like this:
Schema versions:
UNREACHABLE: 1176b7ac-8993-395d-85fd-41b89ef49fbb: [10.202.205.203]
9b861925-1a19-057c-ff70-779273e95aa6: [10.80.207.102]
8613985e-c49e-b8f7-57ae-6439e879bb2a: [10.116.138.23]
2. Restart unreachable nodes.
3. Repeat steps 1 and 2 until nodetool describecluster shows that all nodes have the same
schema version number#only one schema version appears in the output.
View of ring differs between some nodes

Indicates that the ring is in a bad state.
Indicates that the ring is in a bad state.
This situation can happen when not using virtual nodes (vnodes) and there are token conflicts (for
instance, when bootstrapping two nodes simultaneously with automatic token selection.) Unfortunately, the
only way to resolve this is to do a full cluster restart. A rolling restart is insufficient since gossip from nodes
with the bad state will repopulate it on newly booted nodes.
Java reports an error saying there are too many open files
Java may not have open enough file descriptors.
Java may not have open enough file descriptors.
Cassandra generally needs more than the default (1024) amount of file descriptors. To increase the
number of file descriptors, change the security limits on your Cassandra nodes as described in the
Recommended Settings section of Insufficient user resource limits errors.
Another, much less likely possibility, is a file descriptor leak in Cassandra. Run lsof -n | grep java
to check that the number of file descriptors opened by Java is reasonable and reports the error if the
number is greater than a few thousand.
Insufficient user resource limits errors

Insufficient resource limits may result in a number of errors in Cassandra and OpsCenter.
Insufficient resource limits may result in a number of errors in Cassandra and OpsCenter.
Cassandra errors
Insufficient as (address space) or memlock setting
246
Troubleshooting
ERROR [SSTableBatchOpen:1 ] 2012-07-25 15:46:02,913

AbstractCassandraDaemon.java (line 139)
Fatal exception in thread Thread [SSTableBatchOpen:1,5,main ]
java.io.IOError: java.io.IOException: Map failed at ...
Insufficient memlock settings
WARN [main ] 2011-06-15 09:58:56,861 CLibrary.java (line 118) Unable to lock
JVM memory (ENOMEM).
This can result in part of the JVM being swapped out, especially with mmapped
I/O enabled.
Increase RLIMIT_MEMLOCK or run Cassandra as root.
Insufficient nofiles setting
WARN 05:13:43,644 Transport error occurred during acceptance of message.
org.apache.thrift.transport.TTransportException: java.net.SocketException:
Too many open files ...
Insufficient nofiles setting
ERROR [MutationStage:11 ] 2012-04-30 09:46:08,102 AbstractCassandraDaemon.java
(line 139)
Fatal exception in thread Thread [MutationStage:11,5,main ]
java.lang.OutOfMemoryError: unable to create new native thread
Recommended settings
You can view the current limits using the ulimit -a command. Although limits can also be temporarily
set using this command, DataStax recommends making the changes permanent:
Packaged installs: Ensure that the following settings are included in the /etc/security/limits.d/
cassandra.conf file:
cassandra - memlock unlimited

cassandra - nofile 100000
cassandra - nproc 32768
cassandra - as unlimited
Tarball installs: Ensure that the following settings are included in the /etc/security/limits.conf file:
* - memlock unlimited
* - nofile 100000
* - nproc 32768
* - as unlimited
If you run Cassandra as root, some Linux distributions such as Ubuntu, require setting the limits for root
explicitly instead of using *:
root - memlock unlimited

root - nofile 100000
root - nproc 32768
root - as unlimited
For CentOS, RHEL, OEL systems, also set the nproc limits in /etc/security/limits.d/90-
nproc.conf:
* - nproc 32768
For all installations, add the following line to /etc/sysctl.conf:
vm.max_map_count = 131072
247
Troubleshooting
To make the changes take effect, reboot the server or run the following command:
$ sudo sysctl -p
To confirm the limits are applied to the Cassandra process, run the following command where pid is the
process ID of the currently running Cassandra process:
$ cat /proc/<pid>/limits
OpsCenter errors
See the OpsCenter Troubleshooting documentation.
Cannot initialize class org.xerial.snappy.Snappy

An error may occur when Snappy compression/decompression is enabled although its library is available
from the classpath.
An error may occur when Snappy compression/decompression is enabled although its library is available
from the classpath.
java.util.concurrent.ExecutionException: java.lang.NoClassDefFoundError:
Could not initialize class org.xerial.snappy.Snappy
...
Caused by: java.lang.NoClassDefFoundError: Could not initialize class
org.xerial.snappy.Snappy
at
org.apache.cassandra.io.compress.SnappyCompressor.initialCompressedBufferLength
(SnappyCompressor.java:39)
The native library snappy-1.0.4.1-libsnappyjava.so for Snappy compression is included in the
snappy-java-1.0.4.1.jar file. When the JVM initializes the JAR, the library is added to the default
temp directory. If the default temp directory is mounted with a noexec option, it results in the above
exception.
One solution is to specify a different temp directory that has already been mounted without the noexec
option, as follows:
• If you use the DSE/Cassandra command $_BIN/dse cassandra or $_BIN/cassandra, simply
append the command line:
• DSE: bin/dse cassandra -t -Dorg.xerial.snappy.tempdir=/path/to/newtmp
• Cassandra: bin/cassandra -Dorg.xerial.snappy.tempdir=/path/to/newtmp
• If starting from a package using service dse start or service cassandra start, add a system environment
variable JVM_OPTS with the value:
JVM_OPTS=-Dorg.xerial.snappy.tempdir=/path/to/newtmp
The default cassandra-env.sh looks for the variable and appends to it when starting the JVM.
Firewall idle connection timeout causing nodes to lose communication

during low traffic times
Steps to configure the default idle connection timeout.
About this task

During low traffic intervals, a firewall configured with an idle connection timeout can close connections to
local nodes and nodes in other data centers. The default idle connection timeout is usually 60 minutes and
configurable by the network administrator.
248
Troubleshooting
Procedure
To prevent connections between nodes from timing out, set the TCP keep alive variables:
1. Get a list of available kernel variables:
$ sysctl -A | grep net.ipv4

The following variables should exist:
• net.ipv4.tcp_keepalive_time
Time of connection inactivity after which the first keep alive request is sent.
• net.ipv4.tcp_keepalive_probes
Number of keep alive requests retransmitted before the connection is considered broken.
• net.ipv4.tcp_keepalive_intvl
Time interval between keep alive probes.
2. To change these settings:
$ sudo sysctl -w net.ipv4.tcp_keepalive_time=60

net.ipv4.tcp_keepalive_probes=3 net.ipv4.tcp_keepalive_intvl=10
This sample command changes TCP keepalive timeout to 60 seconds with 3 probes, 10 seconds gap
between each. This setting detects dead TCP connections after 90 seconds (60 + 10 + 10 + 10). There
is no need to be concerned about the additional traffic as it's negligible and permanently leaving these
settings shouldn't be an issue.
249
DataStax Community release notes

Release notes for DataStax Community.
The latest Cassandra version is 2.1.5. The CHANGES.txt file installed with Cassandra covers changes in
detail. An important changes to note is that incremental replacement of compacted SSTables has been
disabled for this release. This was done to prevent problems until the entire compaction chain of events is
stabilized.
Key features of Cassandra 2.1 were described earlier in this document. Noteworthy changes for
Cassandra 2.1 include:
• Cassandra 2.1 does not support pre-Cassandra 2.0 SSTables.
To upgrade to Cassandra 2.1 from a previous release that stored data in Cassandra 1.2.x SSTables,
start the node on Cassandra 2.0 and use the sstableupgrade tool after upgrading. Upgrade
SSTables even if you do not perform a rolling upgrade. Resolve schema disagreements if any exist,
and restart each node. For more upgrade information, see "Upgrading Cassandra" and NEWS.txt.
• The shuffle utility for migrating to virtual nodes (vnodes) and the nodetool taketoken command
have been removed. To migrate to vnodes, bootstrap a new data center.
• Cassandra 2.1 bundles and enables JNA. If JNA fails to initialize, you can disable JNA by using the -
Dcassandra.boot_without_jna=true option to start Cassandra.
• Cassandra rejects USING TIMESTAMP or USING TTL in the command to update a counter column,
and now generates an error message when you attempt such an operation.
• Configurable properties have been added to manage counter writes.
• A configurable counter cache reduces lock contention and helps with concurrency.
• In Cassandra 2.1, the CQL table property index_interval is replaced by min_index_interval and
max_index_interval. The max_index_interval is 2048 by default. The default would be reached only
when SSTables are infrequently-read and the index summary memory pool is full. When upgrading
from earlier releases, Cassandra uses the old index_interval value for the min_index_interval.
• CASSANDRA-6504 has been backported to Cassandra 2.0.5 so you can perform a rolling upgrade of a
database having counters to Cassandra 2.1.
• Default data and log locations have changed for tarball installations and source checkouts. By default,
the data file directory, commitlog directory, and saved caches directory are in $CASSANDRA_HOME/
data/data, $CASSANDRA_HOME/data/commitlog, and $CASSANDRA_HOME/data/
saved_caches, respectively. The log directory now defaults to $CASSANDRA_HOME/logs. If not set,
$CASSANDRA_HOME, defaults to the top-level directory of the installation. Deb and RPM packages
continue to use /var/lib/cassandra and /var/log/cassandra by default.
• Cassandra 2.1 maintains data consistency during bootstrapping. As you bootstrap a new node,
Cassandra streams the data for the new node from an existing node that is free from range movement.
If data inconsistency issues are present in the cluster, the improvement to bootstrapping handles these
issues. Data inconsistency commonly occurs after frequent data deletions and a node going down.
• To inhibit the new Cassandra 2.1 bootstrapping behavior, and make Cassandra 2.0 behavior effective,
start the node using the -Dcassandra.consistent.rangemovement=false property:
• Package installations: Add the following option to /usr/share/cassandra/cassandra-env.sh
file:
JVM_OPTS="$JVM_OPTS -Dcassandra.consistent.rangemovement=false
• Tarball installations: Start Cassandra with this option:
$ bin/cassandra -Dcassandra.consistent.rangemovement=false
To replace a dead node, you also need to specify the address of the node from which Cassandra
streams the data.
250
For a complete list of fixes and new features, see the Apache Cassandra 2.1.0 CHANGES.txt. You can
view all version changes by branch or tag in the branch drop-down list:
251
Tips for using DataStax documentation
Tips for using DataStax documentation
Navigating the documents

To navigate, use the table of contents or search in the left navigation bar. Additional controls are:
Hide or display the left navigation.

Go back or forward through the topics as listed in
the table of contents.
Toggle highlighting of search terms.
Print page.
See doc tweets and provide feedback.
Grab to adjust the size of the navigation pane.
Appears on headings for bookmarking. Right-click

the ¶ to get the link.
Toggles the legend for CQL statements and
nodetool options.
Other resources
You can find more information and help at:
• Documentation home page
• Datasheets
• Webinars
• Whitepapers
• Developer blogs
• Support
252

Cassandra 2.1 Used by DataStax Enterprise 4.7

Uploaded by

Copyright:

Available Formats

Cassandra 2.1 Used by DataStax Enterprise 4.7

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cassandra 2.1 Used by DataStax Enterprise 4.7

Uploaded by

Copyright:

Available Formats

Apache Cassandra™ 2.

© 2015 DataStax. All rights reserved.

About Apache Cassandra............................................................................................... 9

What's new in Cassandra 2.1....................................................................................... 11

Understanding the architecture................................................................................... 13

Planning a cluster deployment.................................................................................... 28

Installing on Amazon EC2......................................................................................................42

Configuring data consistency................................................................................................. 85

Backing up and restoring data.................................................................................. 154

Recovering using JBOD.................................................................................................................. 158

Cassandra tools........................................................................................................... 160

Moving data to/from other databases....................................................................... 243

Nodetool or JMX connections failing on remote nodes.................................................................. 245

Using the docs.............................................................................................................252

About Apache Cassandra

How does Cassandra work?

What's new in Cassandra 2.1

Understanding the architecture

Key components for configuring Cassandra

• Package installations: /var/lib/cassandra

Internode communications (gossip)

Failure detection and recovery

Data distribution and replication

name age car gender

Cassandra assigns a hash value to each partition key:

Partition key Murmur3 hash value

Partition key Murmur3 hash value

D Data Center Alpha B

Node Start range End range PartitionHash value

How data is distributed across a cluster (using virtual nodes)

Figure 1: Virtual vs single-token architecture

Choosing keyspace replication options

Package installations /etc/cassandra/cassandra.yaml

Package installations /etc/cassandra/cassandra.yaml

About this task

Package installations /etc/cassandra/cassandra-

# Data Center One

# Data Center Two

# Analytics Replication Group

# default for unknown nodes default =DC3:RAC1

Package installations /etc/cassandra/cassandra-

Keyspace strategy options

Package installations /etc/cassandra/cassandra-

Configuring cassandra.yaml for cross-region communication

seeds: 50.34.16.33, 60.247.70.52

Configuring the snitch for cross-region communication

Region: us-east Region: us-west

Region: us-east Region: us-west

Keyspace strategy options

Package installations /etc/cassandra/cassandra-

The location of the cassandra.yaml file depends on the type of installation:

Package installations /etc/cassandra/cassandra.yaml

Planning a cluster deployment

Selecting hardware for enterprise implementations

Spinning disks versus Solid State Drives

Commit log disk - SSD

Planning an Amazon EC2 cluster

DataStax AMI deployments

Use AMIs from trusted sources