What a hosting provider did today

Disk-full cleanups that delete error logs destroy the incident timeline.

Mid-incident, disk alarms fire. Someone deletes “old” logs — including the error log and slow log that held the only timeline.

Lost

  • Crash fingerprints
  • Alignment between app errors and DB waits
  • Evidence of storage-layer glitches

Controls

  1. Ship logs off-box continuously
  2. Treat log retention as part of backup policy
  3. Document sacred paths in provider/customer runbooks
  4. Prefer environments where roles during incidents are explicit

Disk-full is an emergency. Deleting the crime scene is not a resolution.