Blog
Longer walk-throughs of what the docs describe. RSS feed.
En españolSeries: Making litek8s fast
The pod that waits: how Pending works, and the deadline that never passed
What a Pending pod means, how the scheduler decides whether it fits, and the bug that let a Job with a deadline wait for ever.
kubectl get 2.4× faster: the copy hidden in json!
A profile of kubectl get's Table showed half the time going to copying values. The cause was one serde_json macro, and the fix was moving instead.
Giving an LLM agent a Kubernetes cluster without fear
litek8s mcp gives an agent diagnose_pod, logs, PromQL and GitOps tools, and refuses clusters that are not litek8s. The guardrails, and what they don't cover.
A home lab across NATs: one cluster from a desktop, a cloud VM and a friend's machine
Three machines behind different NATs, one litek8s cluster. How the WireGuard overlay finds direct paths, when it falls back to a relay, and what that costs.
We ran the official Kubernetes conformance suite on litek8s: what passes, what fails, and why
413 of 419 conformance specs, against 416 for k3s. The six failures, the first run that scored zero, and what each one says about Kubernetes.
Kubernetes in CI in ~2.5 s instead of ~45 s: litek8s against kind
A one-line change in a GitHub Actions workflow: a cluster Ready in ~2.5 s instead of ~45 s, Helm jobs 3.7x shorter. Measured, with where it does not pay off.
Metrics, logs and alerts without kube-prometheus-stack and Loki: 1,371 MiB to 1.8 MiB
On a small k3s node the usual stack added 1,371 MiB. litek8s's built-in metrics, logs and alerts added 1.8 MiB. What that buys, and what it leaves out.
October 2026 in litek8s: 21 releases, RBAC, NetworkPolicies and a UI for any cluster
What shipped in litek8s in October 2026: RBAC and NetworkPolicies enforced, a GitHub Action for CI, a terminal UI for any cluster, and litecd for GitOps.
One binary instead of Docker Desktop, and it is also Kubernetes
litek8s dev up serves the Docker Engine API: docker, compose and Testcontainers work unchanged, and kubectl and Helm use the same cluster. What is missing, too.
One LLM across three 4 GiB machines: you get memory, not speed
Qwen2.5 7B needs 4.4 GiB of weights. No node had that. Split over three with llama.cpp's RPC backend on litek8s, it ran at 7.5 tokens/s.
Replacing Argo CD's controller with litecd: 666 MiB to 100, and what it cost
How we swapped Argo CD's application controller for litecd on a production k3s cluster: the plan, the traps, the numbers, and what we gave up.
Small is not enough: measuring litek8s's speed, and a 13× faster list
We had numbers for memory and disk, none for speed. A load test for the API server, a list made 13× faster, and how CI now catches regressions.
The node that was too full: why Kubernetes evicted our database
One node at 97% memory, a database evicted twice, and a scheduler that saw room. Requests against usage, and how we moved pods without downtime.
Cloning a production cluster onto litek8s: what Kubernetes taught us
We copied a running production cluster onto litek8s without touching it. Four bugs, two missing pieces of Kubernetes, and what each says about how it works.
You don't need a space shuttle to ship your app
Why litek8s is so fast, what that speed costs, and why starting small on the real Kubernetes API lets you migrate later without rewriting anything.
Kubernetes on a Raspberry Pi and an old laptop with litek8s
Turn a Raspberry Pi, an old laptop and a cheap cloud VM into one Kubernetes cluster with litek8s: one binary per machine, installed over SSH from your laptop.
