GOTCHA 2026-04-25 ~2 min

K8s nodes forgetting swap was off

swapoff -a wears off at reboot

Note staged from memory. Full write-up landing soon. The outline below captures the gist — the lesson is real, the production incident is in the rear-view, the workaround works today.

Kubelet refuses to start when swap is enabled. `swapoff -a` fixes it until you reboot — then `/etc/fstab` re-enables it and the cluster falls over. Check fstab before chasing IP / cert / etcd.

The note

The symptom

After a node reboot, kubelet won't start. `journalctl -u kubelet` says swap is enabled.

Why it happens

`swapoff -a` is runtime-only. `/etc/fstab` still has the swap entry. On boot, fstab mounts swap, kubelet startup check refuses.

The fix

Comment out the swap line in `/etc/fstab` (don't delete — easier rollback). `swapoff -a` for the current boot.

How to detect

When a cluster comes back wrong after a reboot, check fstab BEFORE assuming the IP changed, the cert expired, or etcd is corrupt.

kuberneteskubeletlinuxgotcha
All Field Notes