deployment-safety

name	deployment-safety
description	Pre-deployment checklists, rollback strategies, and post-deploy verification. Use this skill when preparing to deploy code, reviewing deployment processes, or setting up CI/CD pipelines.
alwaysApply	false

Deployment Safety

You are a senior DevOps engineer reviewing deployments. Apply these checklists and strategies to ensure safe, reliable releases.

Pre-Deployment Checklist

Run through before every production deployment:

Code Readiness

• All tests passing (unit, integration, e2e)
• Code reviewed and approved
• No unresolved merge conflicts
• Feature flags in place for risky changes
• Database migrations tested on staging
• API backward compatibility verified (no breaking changes without versioning)
• Dependencies updated and locked (package-lock.json, go.sum, etc.)

Infrastructure Readiness

• Staging deployment successful and verified
• Resource limits configured (CPU, memory, replicas)
• Health check endpoints responding
• Monitoring and alerting configured for new features
• Log collection working for new components
• Secrets and environment variables configured in production
• SSL/TLS certificates valid and not expiring soon

Rollback Plan

• Previous version tagged and accessible
• Rollback procedure documented and tested
• Database rollback plan if migrations are involved
• Feature flags that can disable new features quickly
• Communication plan if rollback is needed

Timing

• Not deploying on Friday afternoon (unless critical)
• Not deploying during peak traffic hours
• Team available to monitor post-deploy
• No conflicting deployments from other teams

Deployment Strategies

Rolling Update (Default)

Old: [v1] [v1] [v1] [v1]
     [v2] [v1] [v1] [v1]  ← replace one at a time
     [v2] [v2] [v1] [v1]
     [v2] [v2] [v2] [v1]
     [v2] [v2] [v2] [v2]  ← done

Use when: Standard releases, stateless services Risk: Mixed versions serve traffic during rollout

Blue-Green

Blue  (current): [v1] [v1] [v1] ← all traffic
Green (new):     [v2] [v2] [v2] ← ready, no traffic

Switch: Blue → standby, Green → active

Use when: Zero-downtime required, easy rollback needed Risk: Requires 2x infrastructure temporarily

Canary

[v1] [v1] [v1] [v1] [v1]  ← 100% traffic
[v2] [v1] [v1] [v1] [v1]  ← 20% to v2, monitor
[v2] [v2] [v1] [v1] [v1]  ← 40% to v2, monitor
[v2] [v2] [v2] [v2] [v2]  ← 100% after validation

Use when: High-risk changes, gradual confidence building Risk: Slower rollout, users may see inconsistent behavior

Feature Flags

v2 deployed to all instances with flag OFF
Flag ON for internal team → test
Flag ON for 5% of users → canary
Flag ON for 100% → full release

Use when: Decoupling deploy from release, A/B testing Risk: Flag complexity, flag cleanup debt

Post-Deployment Verification

Immediate (First 5 minutes)

• Health check endpoints returning 200
• No spike in error rates (4xx, 5xx)
• Response times within normal range
• Logs show successful startup
• No crash loops or OOM kills

Short-term (First 30 minutes)

• Key business metrics stable (orders, sign-ups, API calls)
• No increase in support tickets
• Memory/CPU usage stable (no leaks)
• Database connections stable
• Queue depth not growing unexpectedly

Long-term (First 24 hours)

• No slow degradation patterns
• Scheduled jobs completing successfully
• No edge case errors accumulating
• Resource usage trending normally

Database Migration Safety

DO

• Add new columns as nullable or with defaults
• Create new tables before referencing them in code
• Add indexes concurrently (CREATE INDEX CONCURRENTLY in PostgreSQL)
• Test rollback of every migration on staging
• Run migrations before deploying new code (expand-then-contract)

DON'T

• Drop columns or tables in the same deploy that removes the code using them
• Add NOT NULL constraints without a default value on existing columns
• Run long-running migrations during peak traffic
• Combine schema changes with large data migrations

Expand-Contract Pattern

Deploy 1: Add new column (nullable)      ← expand
Deploy 2: Code writes to both old + new  ← dual-write
Deploy 3: Backfill old data to new column ← migrate
Deploy 4: Code reads from new column     ← switch
Deploy 5: Drop old column                ← contract

Rollback Procedures

Application Rollback

# Docker/K8s
kubectl rollout undo deployment/<name>
# or
kubectl set image deployment/<name> <container>=<previous-image>

# Git-based (Heroku, Render, etc.)
git revert HEAD && git push

# Blue-Green
# Switch load balancer back to blue environment

Database Rollback

# If using migration tool
migrate down 1

# If manual
# Run the DOWN migration SQL script
# Verify data integrity

When NOT to Roll Back

• Data has been written in new format (would lose data)
• External systems already received new-format data
• Rollback would cause more disruption than the bug → Instead: fix forward with a hotfix