Back up and restore the cluster

The platform backup, its configuration, targets, alarms and restore command are the same as on one node: see Backup and restore. A cluster adds the points below. A platform backup is the only way back from losing the cluster’s database, and the only copy of its certificate authority outside OpenBao.

Caution

Keep the age identity and the target’s location and credentials outside the cluster. Without them a restore is impossible.

On a cluster

• The leader’s backup reconciler writes the bundle. A bundle holds the database, OpenBao’s store and unseal key, the node credentials and the cluster CA.
• A directory target must be an off-cluster NFS export mounted at the same path on every control-plane node, mounted before the control plane starts, with targetOptions.requireMountpoint: true. A path on one node’s disk is not disaster recovery. An s3 target needs no mounts.
• stratumctl admin selfcheck reports backup-freshness, and readiness BackupsOff while backups are off.

Which restore

You lost	Do
One control-plane node	Nothing to restore. Add a node the normal way.
The database or the quorum: corruption, a bad migration, most nodes at once	Reinstall the first node, restore into it, and join the other nodes again.
The whole site	Reinstall every node from the ISO, then restore into the first.

The restore requires the same host names and topology as the lost cluster: the node records in the restored database must match the machines.

Restore

On the freshly installed first node, with its install complete:

sudo stratum-bootstrap restore \
  --bundle stratum-backup-<stamp>-<id>.tar.age \
  --identity backup-identity.txt \
  --yes

Install the first node and the joiners from a build at least as new as the one that wrote the bundle. The restore refuses an older image before touching anything (--force overrides).

Then:

1. Sign in with the original cluster’s admin password. If nobody knows it, sudo stratum-bootstrap recover admin-password --yes sets a new one-time password.
2. Join the other nodes as usual.
3. Run stratumctl admin selfcheck.

After a full-site restore, Ceph forms empty: tenant volume and image data on the lost site is gone. Volume records come back, and volumes whose data was lost are recreated empty or show as errors. Instances come back stopped. Back up volumes that matter with backup plans, which work on Ceph volumes.