As a part of the journey to transition Netflix's compute infrastructure to be more Kubernetes-native, we have leaned into incorporating components from the Kubernetes ecosystem into our container platform Titus. One example of this is our use of Kueue, a cloud-native job queueing system for batch workloads, which has largely replaced the custom queuing and scheduling logic in our homegrown managed batch solution Compute Managed Batch (CMB).

Server-side apply matters because Kubernetes objects are shared state: it moves field ownership into the API server, so apply-style tools can surface conflicts instead of hiding them as silent overwrites.

I maintain an app that builds on top of kubernetes port forwarding, so i track KEP-4006 because the streaming protocol underneath keeps changing. I wrote about this back in april 2024 around Kubernetes 1.30, and six releases later it's changed enough to be worth another look.

You've got ten Claude Code agents running. Each is working on a different issue. Agent 3 just created a PR that conflicts with Agent 7's changes. Agent 5 has been stuck on "Analyzing codebase…" for 45 minutes. Welcome to multi-agent chaos. Turns out, scaling AI coding agents from 1 to 10+ is a completely different problem from running a single agent in a terminal.

Cilium runs in the kernel-level networking path of millions of Kubernetes pods. If our supply chain were compromised, the blast radius would not be small. Hardening the project against that scenario is something we work on continuously, and we wanted to write down what we actually do, in detail. Most of what follows isn't Cilium-specific: any open source project running CI/CD on GitHub Actions can apply these patterns.

A short debugging story about a two-year-old credentials file, an EKS cluster, and the AWS credential chain quietly doing exactly what it was designed to do.

As you'll know if you've ever tried to build a reasonably sized microservices application locally, you can run out of system resources quickly. So, in order to free up some of that valuable memory, I figured I'd offload things to a Kubernetes cluster running on some relatively cheap hardware, and in this post I'll describe the steps I took so that you can do the same.

This guide breaks down the ultimate architectural blueprint to solve cluster sprawl and deployment friction for good. By unifying the centralized governance of Rancher, the resource-saving isolation of virtual control planes (vCluster), and the advanced multi-stage pipeline automation of Kargo GitOps, this write-up touches on how to build a highly scalable, zero-touch infrastructure environment.

In this article, we will use Kubernetes Service Accounts and the TokenReview API to authenticate requests between two in-cluster services, then improve the setup with audience-bound projected Service Account tokens.

DCGM exporter tells you a GPU is hot. It won't tell you whose job is frying it. l9gpu closes the loop. One agent per node emits vendor-neutral OTLP with workload attribution baked in — Kubernetes pod, namespace, deployment; Slurm job, user, partition.

A native iOS and Android app for monitoring and managing Argo CD deployments from your phone.

Ballast is a Kubernetes operator that automatically right-sizes workload resource requests and limits based on real operational history. It is a more active alternative to Fairwinds Goldilocks: rather than suggesting changes, it applies them — at admission time and on running pods via in-place resize (Kubernetes 1.35+).

Cordium is a free and open source, self-hosted, identity-based sandbox platform built on Kubernetes and Octelium. Cordium is a general-purpose platform that provides isolated, reproducible isolated sandboxes for developers, AI agents, and automated workloads.

Most load-testing tools, however, were designed for a simpler world. They assume requests are independent, stateless, and interchangeable. But an AI agent doesn't work that way. This gap was the motivation behind OpenTelemetry (OTel) Trace Replay, a new capability in Inference Perf, the Kubernetes SIG tool for GenAI inference benchmarking.

The first is the KubeBuddy Headlamp plugin, which brings the full KubeBuddy check catalog (excluding the Cloud provider best practices) into Headlamp without touching your cluster configuration. The second is the Headlamp Theme Builder, a Headlamp plugin and browser-based design tool that turns your brand colours into a ready-to-install Headlamp theme plugin in minutes.

I tried a bunch on a separate EKS cluster, like HAProxy and Traefik, but eventually I landed on using Envoy Gateway since to me it felt like the most approachable and stable solution for our use case. Since I started the migration I made a respectable number of mistakes. These are my notes.

In this post, we'll walk through how we redesigned our Kubernetes-based PostgreSQL clusters for failover safety, how we balanced durability against latency, and what we learned while validating this approach through benchmarking and failure testing.

This is the problem helmsniff solves. It is a small, fast command line tool. It reads rendered Kubernetes manifests and writes a simple report. One row per Kubernetes object. One column per security check.

A few weeks ago I finally sat down and built it. This article is about what I built, what broke, and why I think Sveltos is one of the most underrated tools in the Kubernetes ecosystem right now.