k8s Cluster Maintenance

Connecting to the worker nodes

1. Access AWS EKS and go to Compute section.
2. Click on a node, then click through to the EC2 instance.
3. Connect using SSM Session Manager (the only option).

Worker nodes disk filling up

• The worker nodes pull images over time that they often no longer need (e.g. app upgrades).
• These images can fill up the disk space.

• To clean them:

  sudo nerdctl image prune -a
  sudo nerdctl system prune
  
  SRC_IMAGE="ghcr.io/containerd/pause"
  echo "Using pause image: $SRC_IMAGE"
  sudo ctr -n k8s.io images tag "$SRC_IMAGE" localhost/kubernetes/pause:latest
  sudo systemctl restart containerd
  sudo systemctl restart kubelet
  

Cleaning unused images manually

• Once you have an open session to the machine, it's as simple as:

  nerdctl image rm xxx