Ops · Alex Morgan · 11.09.2026 · 890 views

Why we dropped Kubernetes

cluster.png

We had four people and a cluster that needed a fifth. Helm charts drifted. The on-call was YAML. Traffic never justified three availability zones.

Compose on one machine, backups to object storage, a load balancer in front. Deploys are a script. When we outgrow that we can rent Kubernetes again — as a choice, not a default.

Tags: ops, k8s, cost

Related in Ops: One box, two processes, a healthcheck

Comments

Maya Chen 11.09.2026 11:02

We did the same after the second etcd incident.

Alex Morgan 11.09.2026 11:40

etcd was the tell. Once you debug consensus you are running a second product.

Riley Cole 11.09.2026 16:18

Moderation note: keep the thread on the ops choice, not vendor bashing.