Files
vps/runbooks/rollback.md
T
windyboy b0c01b2551 docs(runbooks): add runbook spec, template, index and 6 first-batch runbooks; light-enhance existing 10
- RUNBOOKS.md: repo-level spec (six-field model, naming, safety, maturity path)
- runbooks/_template.md + README.md: standard template and 16-entry routing index
- new: issue-to-merge, fix-ci, release, rollback, network-change, network-recovery
- light-enhance 10 existing runbooks with Purpose/Scope/Safety headers
- AGENTS.md: point step 3 at index/spec, add runbook execution rules
- docs/agent-runbook-guide.md: archive of Manus AI guide
2026-08-17 15:59:46 +08:00

1.8 KiB

Runbook: roll back a change

Purpose

Restore a known-good state after a change failed verification or caused a regression.

Scope

  • Applicable: Compose services, DNS records, firewall rules, and Ansible-managed config that have a rollback point.
  • Not applicable: destructive migrations without a backup (route to the owning runbook, e.g. vaultwarden-sqlite-to-postgres.md §Rollback).

Preconditions

  • A rollback point exists: backup, prior config, or known-good commit/playbook state.
  • The failing change and its symptoms are recorded.

Safety

  • Confirm the rollback target before acting; do not roll back unrelated changes.
  • Verify after rollback; a rollback that is not verified is not complete.

Procedure

Step 1 — Confirm the target

Action — record the change to revert (playbook run, config edit, DNS change) and its intended prior state.

Expected — the exact prior state is known and reachable.

Decision — prior state known → Step 2; unknown → STOP and escalate.

Step 2 — Revert

Action — apply the documented reversal for the change type:

  • Compose config: re-apply the prior config and reconcile (or git revert the change then reconcile).
  • DNS: restore the previous record value on PowerDNS.
  • Firewall: re-apply the prior allowlist via the gated playbook.
  • General repo change: git revert <sha>.

Expected — the prior configuration is in place.

Verification — re-run the relevant health report / endpoint check and confirm green.

Rollback — if the rollback itself fails, STOP; keep the original management session open and escalate.

Final Verification

  • Health report green after rollback.
  • The incident and rollback are recorded in the Linear vps issue.

References