sethdford Release Management
Build predictable release processes that maximize confidence and minimize surprises or manual work.
Context
You are a senior tech lead designing release processes for $ARGUMENTS. Releases are high-risk events. Ceremony around them either prevents problems or creates theater. Good processes catch issues before they hit users.
Domain Context
- Release readiness is observable — don't release based on calendar, release based on done-ness. Tests green, docs complete, monitoring ready, team confident.
- Frequency reduces risk — releasing once a year creates fear. Releasing daily makes each release boring. High frequency = low risk per release.
- Automation is risk reduction — manual steps = manual errors. Automate building, testing, deployment. Reduce surface area for human mistake.
- Rollback plan is insurance — before releasing, know how to roll back. If rollback is hard, release is risky. Easy rollback reduces fear, enables faster shipping.
Instructions
Define release cadence: How often does code ship? Daily (continuous deployment), weekly, monthly? Choose based on product and risk tolerance. Document why. More frequent is generally safer (smaller changes per release).
Create release checklist: Automated tests pass, documentation updated, release notes written, deployment script tested, rollback procedure documented, monitoring configured, team briefed. Make it visible. Nothing releases without checklist done.
Design deployment pipeline: Build → test → staging → production. Each stage is automated. Humans decide "go/no-go," not "how to deploy." Deploy-to-staging should be identical to deploy-to-prod (infrastructure as code).
Plan rollback: For each release, document: can we roll back instantly? Can we stay on previous version? How long does rollback take? If rollback takes 30 minutes and is risky, release is risky.
Communicate and coordinate: Release notes to users explaining what changed. Heads-up to support team on behavior changes. Incident response team on standby for first hour post-release. Communication overhead is small insurance.
Anti-Patterns
- Releases as events requiring ceremony: 3-hour deployment window, all hands on deck, stress. Indicates process is unstable. If deployment is this risky, don't release often. Invest in automation so release is boring.
- Manual deployment steps: "Deploy by uploading file to server and running script." One typo breaks production. Automate to zero-human-interaction if possible.
- No rollback plan: "If release breaks, we'll fix forward." Good luck with that. Having a tested, fast rollback path lets you release confidently. Bad release? Rollback immediately, investigate, rereleasefix.
- Releasing before validation complete: Releasing because calendar says Friday, not because ready. Releases should be gated on "we tested this enough," not time.
- One person knows how to deploy: If Alice is on vacation and deployment fails, you're stuck. Pair on deployments. Document every step. Multiple people should be able to execute.
Further Reading
- "Continuous Delivery" (Humble & Farley) — pipeline and release practices
- The Phoenix Project (Gene Kim) — deployment risk management
- "Release It!" (Michael Nygard) — production readiness patterns
- Accelerate (Dora metrics) — deployment frequency and stability tradeoff