Case Study

Booking.com + Ververica Platform

Building a platform to provide Security-as-a-Service.
Delivering streaming-in security related data for
processing
July 2023
 2
Contents

About booking.com 03

The Challenge 04

The Solution 05

The Results 07
 3
About
Part of Booking Holdings Inc. (NASDAQ: BKNG), Booking.com’s mission is to make it easier for
everyone to experience the world. By investing in the technology that helps take the friction out
of travel, Booking.com’s marketplace seamlessly connects millions of travelers with memorable
experiences every day. For more information, follow @bookingcom on social media or visit
globalnews.booking.com.

© Copyright Ververica 2023 www.ververica.com

 4
The Challenge
What wasn’t working?

With the increasing need to process security data streams, we needed to have an orchestration
tool that can manage a large number of Flink applications (as well as their dependencies) and

manage their stateful upgrade easily. Furthermore, the tool should enable multi-tenancy and
separation of duties. Orchestrating Flink applications themselves was the problem to solve.

How were you managing your jobs previously?

Before exploring Ververica (or other solutions), we were using helm charts and shell scripts to
manage our Flink jobs. It involved a lot of operational toil and manual management. In particular,
it did not work well for the stateful upgrades of Flink applications.

Can you elaborate on the need for a Security-as-a-Service and if there are any industry
need factors at play?

Booking.com provides an ecosystem used by both partners (accommodations, flights, rides

and attractions) and travelers. A lot of data needs to be processed to provide a secure end
to end experience for our users and to report cases of unsafe hotels and properties. Since we
have a lot of security teams focusing on different use-cases (within security), we need to enable
segregation of duties and access control. The former means the people in charge of maintaining
and supporting our infrastructure should not be able to see the contents of the data. The latter;
only the people specifically authorized for it should have access to the data.

© Copyright Ververica 2023 www.ververica.com

 5
The Solution
What were the key deciding factors in selecting Ververica?

For us, it is crucial to have the ability to switch to a model where Booking.com’s other security
teams would manage and leverage the control of the lifecycle for their own Flink applications,
allowing the Platform team to just take care of keeping the solution, which makes that possible,
running.

What else played a role in the selection to use Ververica Platform?

Built on a well-understood and widely used Kubernetes standard, Ververica Platform provides a
security-as-a-service platform for Booking.com to manage a large number of Flink applications
with an intuitive Web UI (multi-tenancy) without any limitations on Flink and Kubernetes
features. Ververica Platform is flexible enough to be implemented in the Booking.com ecosystem
with minimal friction and simple enough to use, providing value from the very start.

What does Ververica Platform provide for the needs of Booking.com?

Ververica Platform allowed us to solve the problem of lifecycle management of Flink applications
and presenting this as a service for users to leverage on. It extends Kubernetes APIs; we’re
building on well-understood and widely used industry standards. Its reduction of complexity
shortens the lead time to bring Flink application lifecycle management to a production-grade
solution. It, in turn, brings more stability and reliability. It also provides visibility for the owners to
better debug and maintain their applications themselves.

Any particular features that Ververica offered over other platform providers?

No other solution that we had a chance to study in depth worked on Kubernetes properly with
abstracted snapshot management and high availability using all the native features of Flink itself,
which seemed like the most forward-evolving solution.

What would you say were the instant short-term wins?

A user can log in and immediately see what Flink jobs there are in a namespace. Simple things
like an inventory for Flink jobs in a namespace instead of having to rely on helm, words, or shell
scripts.

A web UI also always makes life easier. Anything that shortens the work time to achieve an action
has a significant impact because we’re talking about avoiding downtime and faster velocity for
development as well as detecting and solving some issues.

© Copyright Ververica 2023 www.ververica.com

 6
Great! Anything from a long-term perspective?

The one we nailed is the scale-up of a number of applications. That is definitely something that
is well achieved, and we took advantage of it very fast. It was not initially thought to be possible
until we onboarded.

We are managing more jobs with fewer work hours from our team. That’s a net improvement of
the ratio. The hours required for the scale-up were completely detached from the number of job
applications executed. It was always the ultimate desire, but at the time, it wasn’t clear how that
would be in reality. The reality has proven to be above expectations; I would say perfect.

It was certainly more a midterm to a long-term desired state. We achieved it in the short term, so
it was a quick win.

What exceeded your expectations?

In terms of exceeding expectations, the Web UI itself was above expectations. We set out to build
an orchestration of Flink applications - declarative seamless, reducing downtime and friction. The
Web UI was just something we didn’t expect to have as much influence and users as it ended up
having - a huge win!

Another aspect is the ability to scale up without complications other than quotas and
Kubernetes. The amount of control and flexibility that it provided was exceptional because it
allows full flexibility while allowing you to use pretty much everything from Flink (that you would
want to be able to use) from an orchestration solution.

During implementation, we had some technical requirements to fit Ververica in our corporate
environment. We needed some tweaks to Ververica itself to fit it. Your engineering team
responded faster and better than we expected to accommodate this technical need.

© Copyright Ververica 2023 www.ververica.com

 7
The Results
Easily manage hundreds of Flink applications without increasing the complexity and manpower
inventory of Flink jobs at a glance (no relying on helm, shell scripts, etc.).

• Shorten the lead time to go to production: the deployment time and stateful upgrade time
reduced from hours to a few minutes.

• Fast response time on Flink or Ververica Platform questions reduced from potentially infinity
to a few minutes/hours.

How many jobs do you manage with VVP?

Right now we have about 250 Flink Applications managed by Ververica Platform. Before
Ververica, we could only handle between 10-20.

Ververica is allowing us to
build separate environments
for application lifecycles.

What changes did you see to your deployment times?

Deployment times went from a single job taking around 20 minutes to complete to only 1-4
minutes.

For example, in the past upgrading the code of an application would have taken anywhere
between ten minutes to half an hour and would require the intervention of both developers and
platform engineers. With Ververica the upgrade itself takes a couple of minutes and can be fully
performed by the developers. They have also been able to improve our monitoring and visibility
of what’s going on during and after the upgrade

Having support means we have SLAs for getting guidance from experts. In contrast, getting help
from the Flink community could go from a few hours, or even days, to infinity. With Ververica’s
support we can have Flink experts on the line in minutes, if needed.

© Copyright Ververica 2023 www.ververica.com

 8

About Ververica

Ververica’s mission is to power the core business of every company with cutting-edge
real-time stream processing technology. In order to do that, the team at Ververica
focuses on building the best technology available for stream processing, while at the
same time creating a global and open community around this technology.

We build and develop Ververica Platform, a stream processing platform that enables
every enterprise to power their real-time business and use a production-grade stream-
ing infrastructure while at the same time we actively contribute and participate in
the open source Apache Flink® community, the underlying technology framework of
Ververica Platform itself.

For more information:

[email protected]

@VervericaData

www.ververica.com