Runbook — Rollback

Rollback is for failed application releases where database compatibility is preserved. If the release included irreversible migrations, use the disaster recovery or restore process instead.

Preconditions

• The previous chart revision is known.
• The failed release has not executed irreversible migrations.
• Incident commander approves rollback.

Steps

1. Inspect Helm history:

   helm -n finnest-prod history finnest-power

2. Roll back to the last healthy revision:

   helm -n finnest-prod rollback finnest-power <revision> --wait --timeout 10m

3. Watch pods restart and stabilize:

   kubectl -n finnest-prod rollout status deploy --timeout=10m

Verification

• Public /healthz and /readyz return success.
• Error rate returns to baseline.
• New traces show the previous image tag and chart revision.
• A post-incident note records why the release failed and whether a data repair is required.