DevOps&SRE Library

Personal-AI-Router

NVIDIA Personal AI Router (PAIR) is a local inference router for a group of compatible computers on the same network. It discovers participating nodes, manages supported inference engines, and presents Ollama-compatible and OpenAI-compatible proxy endpoints to applications and agents. Independent requests can be routed to eligible nodes according to engine availability, model availability, and current workload.

https://github.com/NVIDIA/Personal-AI-Router

web-llm

WebLLM is a high-performance in-browser LLM inference engine that brings language model inference directly onto web browsers with hardware acceleration. Everything runs inside the browser with no server support and is accelerated with WebGPU.

https://github.com/mlc-ai/web-llm

Автоматизация платформы не отбирает у вас интересные задачи. Она забирает рутину. Deckhouse Platform берёт на себя обновление, масштабирование и поддержку инфраструктуры «из коробки». Освободившееся время остаётся вам — на то, что вам действительно нравится. Обсудите с инженерами Deckhouse, что можно автоматизировать в вашем стеке 👈

solo

Ship one musl-linked executable. At runtime, load the user's existing glibc-linked GPU driver. No container, no AppImage, and no second libc in the process.

https://github.com/pg83/solo

Inside Cilium CNI: solving mysterious Kubernetes pod setup timeouts

In high-performance networking, 35 seconds is a lifetime. This was the latency required to iterate through our connection tracking table of 7 million entries at a maximum speed of 200,000 entries per second. At our 16-million-entry peak, this sequential lookup could take up to 80 seconds, leading to Cilium CNI timeouts preventing new pods from starting on affected nodes.

https://www.adyen.com/knowledge-hub/inside-cilium-cni-solving-kubernetes-pod-setup-timeouts

tailcat

Tailcat is a remix of Tailscale open source pieces to act like netcat, but over Tailscale's data plane, without Tailscale's control plane. Tailscale's data plane (magicsock, internally) gives you point-to-point WireGuard®-encrypted tunnels between two machines with DERP as the NAT-hole-punching communication side channel and the ultimate relay-of-last-resort if NAT traversal fails. Instead of using the Tailscale control plane, all tailcat connection metadata is exchanged out of band, however you want.

https://github.com/tailscale/tailcat

When declaring an incident becomes everyone's favorite workaround

You see someone declare a Sev-2 and you wonder: wait, why is that even an incident? Nothing is down. Customers aren't affected. But a manager needed to get their team's problem to the top of another team's priority queue, and the incident process was a reliable way to make it happen.

https://greatcircle.com/blog/2026/08/11/declaring-incidents-for-side-effects

eBPF: рентгеновское зрение для production Сервис замедлился, соединения обрываются, а привычные показатели указывают только на симптом. Чтобы найти настоящую причину, иногда нужно увидеть, что происходит глубже — на уровне ядра Linux. 23 сентября в 20:00 на открытом уроке курса «DevOps практики и инструменты» познакомитесь с eBPF — технологией, которая помогает исследовать сетевые события, производительность и безопасность работающей системы. На демонстрации вы увидите, как Cilium Hubble показывает сетевые взаимодействия и помогает находить проблемы с трафиком. С помощью Tetragon разберёте обнаружение подозрительной активности на уровне ядра. Также рассмотрите диагностику узких мест без остановки сервисов. Преподаватель объяснит архитектуру eBPF простыми словами — как программы безопасно запускаются в ядре, какие данные можно получать и почему этот подход расширяет возможности традиционного мониторинга. Вы поймёте, для каких задач eBPF действительно полезен, где он дополняет существующие средства наблюдаемости и когда его внедрение будет избыточным. 👉 Зарегистрируйтесь: https://vk.cc/d1Nx9k Реклама. ООО «Отус онлайн-образование», ОГРН 1177746618576, erid: 2VtzqxT3ABE

HAProxy Data Plane API

Data Plane API is a sidecar process that runs next to HAProxy and provides API endpoints for managing HAProxy. It requires HAProxy version 1.9.0 or higher.

https://github.com/haproxytech/dataplaneapi

substrate

Agent Substrate delivers a performant, high density runtime environment for large scale agent deployments. At its core, Agent Substrate maps a larger set of "actors" (applications such as agents) onto a smaller set of ready "workers", relying on the fact that agent-like applications tend to be idle most of the time to achieve heavy multiplexing.

https://github.com/agent-substrate/substrate

ksolver

Kubernetes cluster cost optimizer. Connects to a live cluster (or a saved snapshot), collects every scheduling constraint, and uses CP-SAT constraint programming to find the cheapest node fleet that still satisfies all placement rules.

https://github.com/syslenslabs/ksolver

grpc-streams-checker

Synthetic monitoring for gRPC server-side streams — open each stream, verify frames actually arrive, and export Prometheus metrics.

https://github.com/youngpabl0/grpc-streams-checker

k8s-mechanic

k8s-mechanic is a Kubernetes controller that watches your cluster for failures, investigates them automatically, and opens pull requests on your GitOps repository with proposed fixes — all without leaving your cluster. When a Pod is crash-looping, a Deployment is degraded, or a Node goes NotReady, mechanic spawns an in-cluster OpenCode agent that inspects the live cluster, locates the relevant manifests in your GitOps repo, determines the root cause, and opens a PR.

https://github.com/lenaxia/k8s-mechanic

🤖 DevOps в эпоху AI трансформации: что работает уже сейчас! 17 сентября Сбер собирает DevOps-инженеров, SRE и платформенные команды на встречу о том, как AI меняет работу с инфраструктурой, релизами и безопасностью. Обсудим: 🔘Platform Engineering и IDP: как строить внутренние платформы без лишней сложности 🔘Agentic engineering: каких агентов уже можно доверить пайплайнам 🔘Security на скорости: как закрывать уязвимости, когда счет идет на часы Своим опытом поделятся практики из Авито, Сбера, Cloud․ru и Т-Банка. Встречаемся 17 сентября в 17:00 в Сбер.Среде (Москва, Земляной Вал 9А) или онлайн. Мест всего 150 — успей занять своё по ссылке!

Ephemeral Environments: Automating QA for Reliable Cloud-Native Systems

Ephemeral environments are short-lived, isolated, fully provisioned, on-demand test environments created automatically from a pull request and destroyed the moment their job is done. This guide walks through why traditional environments fail, what ephemeral environments actually are, the modern stack that powers them, and a concrete, code-level blueprint for building your first ephemeral testing pipeline.

https://medium.com/@shezakhanashraf/ephemeral-environments-automating-qa-for-reliable-cloud-native-systems-9c94035feaf6

An 8-minute outage from a dead NLB and a JVM that cached DNS forever

We drained a Network Load Balancer during a planned migration, and one internal service kept hammering the dead IPs for 8 minutes. The cause wasn't the failover. It was a JVM caching the DNS record forever. The fix was a 30-second TTL and a health-check tweak, not a smarter system.

https://dev.to/claire_nguyen/an-8-minute-outage-from-a-dead-nlb-and-a-jvm-that-cached-dns-forever-4dmj

Dagster on Kubernetes: When More Nodes Won't Save You

You've scaled up your Kubernetes cluster. EC2 instances are running. CPU and memory metrics look reasonable. And yet — the number of completed Dagster runs stubbornly refuses to climb. This article walks through the full diagnostic journey — from the obvious knobs to the hidden PostgreSQL bottleneck that probably nobody warns you about.

https://medium.com/@wojciech.szyba/dagster-on-kubernetes-when-more-nodes-wont-save-you-d591ba6669fd

OpenViking

OpenViking is an open-source context database for AI agents. It stores memories, resources, and skills as one virtual filesystem under the viking:// protocol, so an agent browses its own context with ls, tree, and find instead of querying a black-box vector store. Content is processed into three tiers — L0 abstract, L1 overview, L2 details — and loaded on demand. Every retrieval leaves a trajectory you can watch and debug.

https://github.com/volcengine/OpenViking

kern

A fast, rootless sandbox and virtual resource runtime for any workload, including untrusted and AI-generated code. Daemonless: a real, kernel-enforced container in ~3.5 ms from an OCI image, no daemon, one 1.52 MB static binary.

https://github.com/getkern/kern

picomq

PicoMQ is durable, real-time streams over HTTP, built on S3-compatible object storage.

https://github.com/picomq/picomq