Deployment Orchestration - Shared Skill

Overview

This skill provides standardized deployment orchestration patterns for managing deployments across multiple services, companies, or cloud instances with zero-downtime and safe rollback capabilities.

When to Use

• When coordinating multi-service deployments
• When managing deployment pipelines
• When implementing canary/phased rollouts
• When setting up rollback procedures
• When automating deployment workflows

Core Procedures

Deployment Pipeline Structure

deployment_pipeline:
  stages:
    - name: build
      actions:
        - lint
        - test
        - build
    
    - name: staging
      actions:
        - deploy to staging
        - integration tests
        - smoke tests
    
    - name: production
      actions:
        - backup current version
        - deploy canary (5%)
        - health check
        - gradual rollout (25%, 50%, 100%)
        - monitoring

Zero-Downtime Deployment

deployment_strategy:
  type: rolling_update
  
  rolling_update:
    max_surge: 1
    max_unavailable: 0
   健康检查: /health endpoint
    
  blue_green:
    - maintain parallel environments
    - switch traffic after validation
    - instant rollback capability
    
  canary:
    - initial: 5% traffic
    - check: 30 minutes
    - increment: 25%, 50%, 100%
    - automatic rollback on failure

Pre-Deployment Checklist

pre_deployment:
  - [ ] All tests passing
  - [ ] Code review approved
  - [ ] Change log updated
  - [ ] Database migrations tested
  - [ ] Rollback plan documented
  - [ ] Stakeholders notified
  - [ ] Maintenance window scheduled
  - [ ] Monitoring alerts configured

Post-Deployment Verification

post_deployment:
  health_checks:
    - GET /health → 200
    - GET /api/health → 200
    - Database connectivity
    - Cache connectivity
  
  smoke_tests:
    - User login flow
    - Core business logic
    - Critical integrations
  
  monitoring:
    - Error rate baseline vs post-deploy
    - Latency comparison
    - Resource usage

Rollback Procedures

Instant Rollback

rollback:
  strategy: blue_green
  
  process:
    1. detect_failure:
       - error_rate > threshold
       - health check failed
       - latency spike
  
    2. trigger_rollback:
       - switch traffic to previous version
       - zero-downtime if blue-green
  
    3. verify_rollback:
       - health checks pass
       - error rate normalized
  
    4. notify:
       - stakeholders
       - post-mortem scheduled

Database Rollback

database_rollback:
  backup_strategy:
    - pre-deployment backup
    - point-in-time recovery enabled
  
  migration_rollback:
    - down migrations ready
    - data integrity verified
    - rollback tested in staging

Multi-Instance Deployment

Staged Rollout

staged_rollout:
  stages:
    - name: internal
      targets:
        - dev.internal.example.com
      percentage: 0%
      duration: 1 hour
  
    - name: canary
      targets:
        - prod.example.com
      percentage: 5%
      duration: 30 minutes
  
    - name: gradual
      percentage: 25%
      duration: 1 hour
  
    - name: full
      percentage: 100%
      duration: immediate

Cross-Region Deployment

cross_region:
  strategy: regional_rollback
  
  regions:
    - us-east-1 (primary)
    - eu-west-1 (GDPR)
    - ap-southeast-1
  
  deployment_order:
    - regions:
        - eu-west-1 (compliance first)
        - us-east-1
        - ap-southeast-1
  
  rollback_order: reverse

Monitoring & Alerts

Deployment Metrics

deployment_metrics:
  - deployment_duration
  - time_to_health_check_pass
  - rollback_frequency
  - success_rate
  
comparison_metrics:
  - error_rate (before/after)
  - latency_p95 (before/after)
  - availability (before/after)

Alerting Rules

alerts:
  critical:
    - deployment_failed
    - health_check_failed
  
  warning:
    - error_rate_increased
    - latency_degraded
    - partial_rollout_stalled

Best Practices

1. Always have rollback plan before deployment
2. Deploy during low traffic windows
3. Monitor closely for first 30 minutes
4. Communicate with stakeholders
5. Document all deployments and outcomes
6. Automate where possible
7. Test rollback in staging first