SRE and platform engineer at Ænix, Tbilisi. Proud dog dad. 🐕
Maintainer of Cozystack (CNCF Sandbox), where I also lead the GPU/AI track. Patches in the Linux kernel and Git, printer boards in Armbian, and a long tail of controllers, charts, and MCP servers.
Writing here about infrastructure, reliability, Kubernetes, open source as practice, and herding LLMs.
🔗 awesome.lexfrei.dev · all projects | me.lex.la · contacts and PGP
English version of the Russian original published on Habr. From each according to his ability, to each according to his needs. Karl Marx, “Critique of the Gotha Programme” After the previous article about Klipper the most frequent objection was not “you are wrong”. The most frequent one was: “So what? Fork it, patch it, use it.” And it came from people who live exactly like that, and who, by the look of it, are not bothered at all that their work will drown among millions of forks and never be of use to anyone. ...
English version of the Russian original published on Habr. The man whose code has booted your virtual machines for the last 20 years or so (SeaBIOS has been the default BIOS of QEMU/KVM since about 2010) today decides single-handedly what goes into Klipper, the firmware behind a good half of the planet’s 3D printers, from home builds to factory machines. This is not a metaphor about spiritual leadership. It is literally a one-row table in the official documentation. ...
I have a Raspberry Pi 5 with a Samsung 990 EVO. Samsung released new firmware. Their official utility is x86-64 only. What now? When You Need This Samsung’s official method is booting from an ISO and running their utility. That’s not always possible: ARM systems — Raspberry Pi, Ampere, Apple Silicon, AWS Graviton. Samsung’s utility simply won’t run. Fleet updates — you have 50 servers and want to update firmware via Ansible, not run around with a USB stick. Headless servers — no physical access, SSH only. Rebooting into recovery isn’t an option. CI/CD automation — hardware provisioning before deployment, firmware as part of the pipeline. Remote management — server in a datacenter, iLO/IPMI available, but mounting an ISO and waiting is slow and inconvenient. In all these cases, you need a way to flash firmware from a running system via nvme-cli. ...
Welcome to my new tradition: monthly recaps! This month was mostly about tackling personal backlog and closing out accumulated issues. ML310 Migration Continuing my migration from MicroServer to HP ML310e. Ran into a wall of proprietary nonsense like “you need OUR specific fan for $150”. Turns out it’s easy to fool — just short pins 4-5-6 to ground and all errors disappear. Caught some Illegal OpCode errors with the HBA controller and NIC, but disabling Optional ROM in BIOS fixed it. Not a universal solution, but I boot from USB anyway, not disks or network, so it works for me. ...
People often ask me “how do I make my first PR?”, the answer is simple — just do it. But I want to write a series of notes about open source culture and how to make life convenient for everyone. Today we’ll look at a case I encountered yesterday, namely — multiple changes to one semi-abandoned repo. TL;DR When preparing a large refactor in an open source project: Break into small logical PRs — reduces cognitive load on reviewers Propose integration branch for coordination — allows incremental review and batch changes If in doubt, don’t hesitate to reach out to the team early — openness and communication matter Be ready for iterations — open source is about people This reduces load on reviewers and increases chances of successful merge. ...
VIP for Kubernetes API Alright, time to brag a bit: I became completely disillusioned with solutions like metallb, cilium, kube-vip, etc. for announcing kube API. Also, I don’t want to do this externally. I don’t want to manage it on hosts. I don’t want to manage peer lists. Therefore, vipalived. (Won the naming contest; “lube-vip” came in second). A bit of context: Cilium replaced my kube-proxy, metallb, ingress controller, and CNI. It’s cool and optimal — previously there were several network tools, now just one. But there’s a side effect: Cilium became critically dependent on the kube API. This makes it unsuitable for VIP of the kube API itself — a vicious cycle. ...
Welcome to my personal blog! Here I’ll be writing about SRE, DevOps, infrastructure, and other tech topics. What to Expect Technical deep dives Infrastructure best practices DevOps workflows and automation Real-world experience from production systems Multilingual Content This blog supports both English and Russian. Use the language switcher in the header to switch between languages. Stay tuned for more content!