name: Deploy description: Ship applications reliably with CI/CD, rollback strategies, and zero-downtime deployment patterns. metadata: {"clawdbot":{"emoji":"🚀","os":["linux","darwin","win32"]}}

Deployment Rules

Pre-Deploy Checklist

• Tests passing in CI — never deploy with failing tests
• Environment variables set in target — missing secrets cause silent failures
• Database migrations run before code deploy — new code expecting new schema fails
• Rollback plan ready — know exactly how to revert before you need to

Deployment Strategies

• Rolling: update instances one by one — safe, slower, no extra resources
• Blue-green: full parallel environment, instant switch — fast rollback, 2x resources
• Canary: route percentage to new version — catch issues early, complex routing
• Choose based on risk tolerance and resources — no universal best

Zero-Downtime Deploys

• Health checks must pass before traffic routes — unhealthy instances stay out
• Graceful shutdown: finish in-flight requests before terminating
• Database changes must be backwards compatible — old code still running during deploy
• Session handling: sticky sessions or external session store — don't lose user state

CI/CD Pipeline

• Build once, deploy everywhere — same artifact to staging and prod
• Cache dependencies between builds — save minutes per deploy
• Parallel steps where possible — tests, linting, security scans
• Fail fast: quick checks first — don't wait for slow tests to catch typos
• Pin action versions with SHA — tags can change unexpectedly

Environment Management

• Staging mirrors prod — different configs cause "works in staging" bugs
• Secrets in secret manager, not environment files — rotation without redeploy
• Feature flags decouple deploy from release — ship dark, enable later
• Config as code in version control — except secrets

Database Migrations

• Migrations must be backwards compatible during deploy window
• Add columns nullable first, then backfill, then add constraint
• Never rename columns in one step — add new, migrate data, remove old
• Test migrations on prod-size data — 10 rows is fast, 10 million isn't
• Rollback script for every migration

Rollback

• Automated rollback on health check failure
• Keep previous version artifacts available — can't rollback what you deleted
• Database rollbacks are hard — design migrations to not need them
• Feature flags for instant rollback of functionality without deploy
• Document rollback procedure — panic time is not learning time

Monitoring Post-Deploy

• Watch error rates for 15 minutes after deploy — most issues surface quickly
• Compare key metrics to pre-deploy baseline
• Alerting on anomalies: latency spike, error rate increase
• Log correlation: trace requests through systems
• User-facing smoke tests after deploy

Platform-Specific

Containers

• Image tagged with git SHA — know exactly what's running
• Health check endpoint that verifies dependencies
• Resource limits set — prevent runaway containers

Serverless

• Cold start optimization — keep bundles small
• Provisioned concurrency for latency-sensitive paths
• Timeout set appropriately — default is often too short

Static Sites

• CDN cache invalidation after deploy
• Immutable assets with content hashes — cache forever
• Preview deploys for PRs

Common Mistakes

• Deploying Friday afternoon — issues surface when nobody's watching
• No rollback plan — hoping nothing goes wrong isn't a strategy
• Mixing code and migration deploys — one thing at a time
• Manual deploy steps — if it's not automated, it's wrong sometimes
• Deploying without monitoring — you won't know it's broken until users complain

这个Skill质量较好，专注于讲解如何可靠地部署应用。优点是内容全面系统，从部署前准备到上线后监控都有覆盖，还包含多种部署方式的对比指导，语言简洁易懂。主要不足是内容偏理论化，缺乏具体操作示例或模板参考资料，实际使用时需要自己摸索。适合想系统了解部署最佳实践的开发者快速学习，但落地执行层面还需要额外查阅具体工具文档。