Production readiness guide

Before production: a readiness checklist for small teams

Readiness is not a promise that incidents will never happen. It means the team knows what it is releasing, who may release it, how to detect a bad change, and how to recover with evidence rather than improvisation.

Release evidence

  • A reproducible build identifies source revision, dependencies, configuration assumptions, and the deployable artifact.
  • Automated checks cover the highest-risk behavior; their passing result is attached to the release.
  • Production-like validation covers migrations, permissions, external dependencies, and a minimal smoke test.
  • One named person authorizes the release and another path exists when that person is unavailable.
  • The release record states what changed, who approved it, when it ran, and which artifact was deployed.

Access and secrets

  • Human and workload identities are separate; shared administrator credentials are avoided.
  • Production permissions follow the necessary role and have a documented owner and removal process.
  • Secrets are supplied through an approved secret store or deployment environment, not committed to source or copied into documentation.
  • Rotation can occur without rebuilding unrelated systems; emergency access is controlled and reviewed.
  • A small team still records who can deploy, read production data, change DNS or infrastructure, and restore backups.

Rollback is a decision, not just a command

Document the signal and decision owner for rollback, the last known-good artifact, and the exact procedure. A database migration may be irreversible even if application code can be redeployed. Prefer backward-compatible changes or an explicitly tested forward-recovery plan where reversal is unsafe.

Run the rollback or recovery procedure in a suitable non-production environment. Record the result and the assumptions that were not tested. A written command that nobody has exercised is not strong recovery evidence.

Backups only help after restoration is demonstrated

  • Name the data and configuration included, the schedule, retention, storage location, encryption/access controls, and monitoring of failed backup jobs.
  • Define acceptable recovery point and recovery time as business decisions, then verify the setup can approach them.
  • Restore into an isolated location; verify integrity and application readability, not only that an archive exists.
  • Record the test date, operator, duration, result, gaps, and next test owner. Do not expose restored personal or production data unnecessarily.

Operational ownership handoff

  • Service owner, deployment approver, incident coordinator, infrastructure owner, and vendor contacts are named.
  • Dashboards and alerts have a human recipient and an agreed response window; monitoring alone is not 24/7 human response.
  • Runbooks cover common diagnosis, rollback, restoration, access recovery, and escalation.
  • Support hours, exclusions, maintenance tasks, and change authority are explicit. No uptime or round-the-clock response should be inferred unless contractually agreed.
  • A short handover session is completed and access is tested by the people who will actually operate the system.

Source and scope

GitHub Docs: Deployments and environments. GitHub documents environment protection, approvals, branch restrictions, and environment secrets. Availability varies by plan and repository visibility. These controls do not themselves guarantee backups, recovery, uptime, or human response.