Uninstall

helm uninstall miroir -n miroir-system

By default this removes the miroir workloads but keeps every volume: the MiroirVolume and MiroirSnapshot objects, the DRBD resources, and the backing devices on the nodes are all left in place, and the CRDs (installed from the chart’s crds/ directory) survive the uninstall too. Mounted volumes keep serving I/O without the agents; a later reinstall re-adopts everything as it was.

Destroying the data

There is no undo

The armed hook deletes every MiroirVolume and MiroirSnapshot, and the agents then erase the DRBD resources and backing devices on the nodes. It includes volumes whose PV reclaimPolicy is Retain.

To delete every volume as part of the uninstall, arm the pre-delete hook first. Helm bakes hooks into the release at install/upgrade time, so the confirmation must be set before running helm uninstall:

helm upgrade miroir oci://ghcr.io/home-operations/charts/miroir \
  -n miroir-system --reuse-values \
  --version <installed-chart-version> \
  --set uninstall.confirmation=yes-really-destroy-data
helm uninstall miroir -n miroir-system

Pin --version to the chart version already installed (helm list -n miroir-system shows it) so this step arms the hook without also upgrading the release to a newer chart on its way out.

The hook Job deletes every MiroirVolume and MiroirSnapshot and waits while each node’s agent tears down its DRBD resources and backing devices through the finalizers. If a teardown cannot finish (a node is down), the job blocks: kubectl get miroirvolumes shows what is stuck, and the agent log on the affected node (kubectl logs -n miroir-system -l app.kubernetes.io/component=agent) shows the failing call to clean up manually.