Set up production deployment pipeline (GitHub Actions)

[fusion94]

Description

Configure a GitHub Actions deployment workflow that deploys to the GCP VM after CI passes.

Deployment Flow

Push to main
  → GitHub Actions: lint + test (existing CI)
    → On success: Deploy job
      → SSH to GCP VM
      → git pull (or rsync built artifacts)
      → npm ci --production
      → Run database migrations (Drizzle migrate)
      → PM2 restart cellarsync
      → Health check (curl /api/health)
      → Notify on failure

GitHub Actions Workflow

• New job deploy in CI workflow (or separate deployment workflow)
• Triggered only on push to main (not on PRs)
• Depends on all test jobs passing
• Uses SSH action to connect to GCP VM

SSH Configuration

• GitHub Secrets: SSH private key, VM IP address, SSH user
• SSH key restricted to deployment user on VM
• Known hosts configured to prevent MITM

Deployment Script (on VM)

cd /opt/cellarsync
git pull origin main
npm ci --production
npm run db:migrate
pm2 restart cellarsync
sleep 2
curl -f http://localhost:3001/api/health || (pm2 restart cellarsync && exit 1)

Rollback Strategy

• git revert + re-deploy for code rollback
• Database migrations are forward-only
• Previous PM2 process snapshot as emergency fallback
• Document rollback procedures

Health Check

• POST-deploy health check: GET /api/health returns 200
• If health check fails, restart PM2 and alert
• GitHub Actions step fails if health check fails (blocks marking deployment as successful)

Notifications

• GitHub Actions failure notification (email or Slack webhook)
• Deployment status badge in README

Acceptance Criteria

• Push to main triggers deployment after CI passes
• SSH connection to VM works securely
• Migrations run automatically
• PM2 restart completes successfully
• Health check verifies deployment
• Failed deployment doesn't leave app in broken state
• Rollback procedure documented and tested

References

• PRD Section 7: Deployment Pipeline, Deployment Strategy