Start with recovery
A backup schedule is not a strategy. First define the recoverable service, acceptable data loss and the evidence required to trust a restore.
Separate failure domains
- Keep backup storage independent from the compute node.
- Retain more than one recovery point.
- Protect backup credentials separately from everyday administration.
- Test restoration into an isolated network.
Warning: A completed job proves that data was written, not that the guest can be recovered within the required time.
Demonstration check
# Illustrative only: inspect task history through your approved interface
pvesh get /cluster/tasks --typefilter vzdumpRecord the guest selected, recovery point, elapsed restore time, integrity checks and any manual steps. Never publish real node names, storage identifiers or guest configuration without review.
Exit condition
The test ends when the restored workload starts, its application data is consistent, and the recovery instructions work for someone other than their author.