Back up the right state
Application files, database, environment, private data and relevant service configuration may all be needed. A backup should have a checksum and a tested restore path.
Define failure before launch
Health checks, important pages, forms, authentication, payment boundaries and service processes need explicit pass criteria. Without them, teams debate whether a release is broken while users wait.
Make rollback atomic
Versioned releases and controlled switching reduce partial deployments. Rollback should restore source, built output and private-state expectations together.
Capture evidence
Record version, hashes, route results, service health and invariant checks. Evidence helps diagnose subtle failures and gives the next release a trustworthy baseline.
Apply this to a real system.
Bring the context and we will identify the useful first decision.
Discuss the related system