2026-09-13  ·  10 min

You can checkpoint a container. You cannot restore one.

Kubernetes can capture a running container’s memory through a kubelet endpoint, and has no API to put it back. Memory, open files and child processes all survive the round trip; an open TCP connection stops the checkpoint being taken at all.

2026-09-11  ·  12 min

Kubernetes controllers don’t react to events. They react to state.

The four parts under every Kubernetes operator: one watches the cluster, one caches it, a queue picks what’s next, workers do the work. Deduplication only starts once you fall behind, the ten-hour resync never contacts the API server, and one missing call drops an object out of the loop for good.