Skip to content

RB016: Database maintenance window

Applicability: Cloud SQL primary; RDS optional. Operator: data and platform owner. Verified: 2026-10-23.

Symptom

A provider maintenance event is scheduled or active, pending modifications exist, or database availability changes during maintenance.

Investigation

  1. Record provider event ID, database identity, engine/version, scheduled window, affected changes, expected interruption, backup status, and application load.
  2. Check connection headroom, active migrations/restores, range operations, worker backlog, and deployment activity. Avoid overlapping mutations.
  3. For AWS, inspect the configured RDS maintenance/backup windows and pending modifications. The deployment checks pending changes with environment-specific policy.
  4. The GCP Terraform module does not declare an equivalent maintenance-window contract. Report the provider event and this gap; do not claim Terraform controls the window.

Resolution

Pause conflicting release and administrative work. Use provider-supported rescheduling only with data-owner approval. During impact, keep new range admission bounded and watch database, proxy, portal, and workers. Afterward, verify instance state/version, connections, schema, /health/, worker backlog, and a read-only application path. Reconcile pending modifications through Terraform and the protected workflow.

Prevention

Track provider events, define service maintenance communications and change freeze windows, test client reconnection, and keep backup/restore evidence current. Add a Terraform-managed GCP window only through separate reviewed architecture work.

Escalate unexpected version changes, maintenance outside the agreed window, failed reconnection, pending destructive modifications, or missing backup evidence.