Envio’s Post

Cozystack v0.9.0 Release: Enhancing Stability for User Kubernetes Clusters The release of Cozystack v0.9.0 (https://lnkd.in/gEpqA3FN) is now available for download, installation, or updating your existing installation. Key Updates in Kubernetes Application: — Node Groups: Allow upgrading existing node groups. — Version Upgrade: Kubernetes and its surrounding tooling have been updated to v1.30.1, bringing the latest features and improvements. Other Component Updates: — KubeVirt: Updated to version v1.2.2, bringing enhanced virtualization capabilities. — Kamaji: Updated to version v1.0.0, providing better control plane management. — Cluster API: Updated with hardcoded component versions to ensure stability and compatibility. — Piraeus: Updated to version v2.5.1, enhancing storage management and performance. Download the latest version of Cozystack v0.9.0 to take advantage of these enhancements: https://lnkd.in/gEpqA3FN

🎉NEW FLOW RELEASE 🎉 In this new release of the flow application, I handled the error of init. Here is the link: https://lnkd.in/dYcdw5tu

k8gb v0.14.0 is out: https://lnkd.in/e5ycAVCr - among many other things in this significant release, the k8gb controller can now speak Istio 🎉 Next up - Gateway API. What do you want to see from k8gb?

I have released another version of the flow application in which I added the alert functionality with flags. Here is the link: https://lnkd.in/drAetwaB

✨ nf-schema 2.1.0 has just been released ✨ A lot has been improved and some new exciting features have been added: - 🆘 The help messages have been completely reworked! Simply enable the validation.help.enabled = true config option option and your pipeline will automatically emit a help message when the --help parameter is used! - 📰 Full support for nested parameters! - ⚙ A whole bunch of configuration options to customize some parts of the plugin See the full changelog: https://lnkd.in/eD2n_g8J

🎉 NEW FLOW RELEASE 🎉 I released another version of the flow application in which I added the init command to initialize the application. Here is the link: https://lnkd.in/d3NptQXU

Infinity v0.5.0 is released ! 1. Support for high availability architecture based on shared object storage (s3 compatible). 2. Support product quantization for dense vectors 3. Support 8 bit scalar quantization for dense vectors 4. Support binary dense vectors, and provide Hamming distance similarity. 5. Support search results caching and pagination. 6. Allow to specify comment when creating database and index. 7. Support functions including: regular expression, md5, substring, lower, upper, ltrim, rtrim, trim, char_posiition, sqrt, round, ceil, floor, IN, isnan, isinf, and isfinite. 8. Improve the performance of full-text phrase search by providing dynamic query pruning. 9. Allow threshold to be specified to limit search results when using BM25 / L2 / IP / Cosine. 10. Support Arm64 architecture. 11. Improve the performance of RAG analyzer. 12. Provide New IK analzyer for Chinese full text search 13. Integration into RAGFlow https://lnkd.in/gKp5uZze

Next big step in development of my real-time operating system? I want to use the fact that #RP2040 is affordable piece of hardware and has multiple cores working connected as equal. So the next feature will be SMP support. SMP is a no easy feature to support. A kernel, if designed well is accessed mostly in a serialized fashion. Syscalls can't preempt each other, because, well, you can't call a syscall from a thread if kernel is being executed currently. That changes with two or more cores as both of them can be executing syscalls concurrently. There's an intermediate step between unicore and SMP kernel - the stage at which you allow your interrupts to preempt kernel, do something useful and your kernel won't melt down doing so. As it turns out this is almost as hard as having full SMP support. The thing is that in real-time environment you don't want to delay interrupts for longer than absolutely necessary. So they need to be able to preempt whatever kernel is doing. One critical task kernels are usually doing is thread scheduling. This involves two actions: selecting the best thread to run next and then performing the thread switch. The former one can be quite costly in terms of time. Unfortunately, in my RTOS, basically the only action interrupts are eligible for in terms of talking to the kernel is changing of scheduling priorities. This is for handing the work down to the threads rather than processing it in interrupt contexts. As it turns out, this is the worst case scenario as you have one modifying action potentially being preempted by another modifying action. A race condition. If you guard the scheduler lookups by mutexes, then you get a deadlock (due to priority inversion). No bueno. If you don't lock it, then concurrent run will turn your kernel state into cabbage. You can disable interrupts while hunting for next thread but that will delay your interrupt handler by hundreds of cycles. It was clear that the solution must be lock-free. And due to the nuances of real-time environment interrupts, it must even be wait-free. We did not even start considering the second core and we are in the land of wait-free algorithms. One thing you'll learn early about wait-free algorithms is that they are hard and some things are not even possible. How to deal with it? We need changes to the scheduler be atomic and we need ability to abort write of outdated changes if ISR was involved. That smells like transactions! The more I am thinking about it, the more transactions look like the solution for SMP issues. And the prototype seems to be really cheap. If you have spare RP2040 laying around and are willing to play with it, here's repository of examples: https://lnkd.in/epaQ7JHa The version there does not support transactions yet, so it is possible your ISRs will cause kernel to run into weeds from time to time! Feel free to ramble that it doesn't build, it is still in stage of hot mess. #rtos #cortex #arm #cmrxrtos

Huggingface serving runtime with vLLM support will be introduced in the next KServe release. Runtime details - https://lnkd.in/gZnz4jTU This serving runtime provides out-of-box support to run text generation based Huggingface LLMs with ease on Kubernetees using KServe. With this change there is built-in support for vLLM (https://vllm.ai) enabling users to run vLLM supported architectures to gain up to 20x higher throughput than Huggingface Transformers, without requiring any model architecture changes. The following implements the newly introduced Text Generate REST API schema - https://lnkd.in/gnxj8dwq We are planning to add more tasks based support like embeddings to this runtime in the future. We like to get an early community feedbacks on this runtime. Dan Sun Johnu George Yuan Tang