Phase 1 - Internal (Day 1):
  - 100% to internal team
  - Test thoroughly
  - Fix critical bugs

Phase 2 - Beta (Day 2-3):
  - 5% to beta users
  - Monitor errors/performance
  - Collect feedback

Phase 3 - Gradual (Day 4-7):
  - 25% of users
  - Watch metrics closely
  - 50% of users if good
  - 100% if still good

Phase 4 - Full Release:
  - 100% of users
  - Remove feature flag
  - Announce publicly

// Feature flag implementation
const featureFlags = {
  newDashboard: {
    enabled: true,
    rollout: 0.25, // 25% of users
    userGroups: ['beta-testers'], // Always on for beta
  }
}

function isFeatureEnabled(feature, user) {
  const flag = featureFlags[feature]

  // Check user group
  if (user.groups.some(g => flag.userGroups.includes(g))) {
    return true
  }

  // Check rollout percentage
  const hash = hashUserId(user.id)
  return (hash % 100) < (flag.rollout * 100)
}

// Usage
{isFeatureEnabled('newDashboard', user) ? (
  <NewDashboard />
) : (
  <OldDashboard />
)}

Process: 1. Deploy to "green" environment
  2. Test green thoroughly
  3. Switch traffic to green
  4. Keep blue as rollback

Pros: Instant rollback
Cons: 2x infrastructure cost

Process: 1. Deploy to 5% of servers
  2. Monitor for 1 hour
  3. If good, deploy to 25%
  4. Monitor for 1 hour
  5. If good, deploy to 100%

Pros: Gradual, safe
Cons: Slower rollout

Criteria for Rollback:
  - Error rate > 1%
  - Performance degradation > 20%
  - Critical bug discovered
  - Negative user feedback

Rollback Process: 1. Disable feature flag immediately
  2. Notify team
  3. Investigate issue
  4. Fix and redeploy

Key Metrics During Release:
  - Error rate
  - Response time p95
  - CPU/memory usage
  - User-reported issues

Alerts:
  - Error rate > 1% → Pause rollout
  - Response time > 2s → Investigate
  - Memory spike > 90% → Rollback

Internal:
  - Slack announcement
  - Deploy log updated
  - Engineering team notified

External:
  - Changelog updated
  - Email to power users (if major)
  - Blog post (if significant)