What are blue-green, canary, and rolling deployment strategies? When would you use each?

Deployment Strategies

1. Rolling Update (Default in Kubernetes)

Old pods gradually replaced by new pods
[v1][v1][v1] → [v1][v1][v2] → [v1][v2][v2] → [v2][v2][v2]

✅ Zero downtime, simple
❌ Old and new versions run simultaneously during rollout
❌ Can't easily rollback if v2 has a data migration issue

2. Blue-Green Deployment

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

Switch traffic: Blue → Green (instant)
If issues: Switch back: Green → Blue (instant rollback)

✅ Instant switch and rollback
✅ Full testing before switch
❌ Requires 2x infrastructure
❌ Database migrations need backward compatibility

3. Canary Deployment

[v1][v1][v1][v2]
 80% traffic ↑↑↑  20% ↑

Monitor metrics for 5-30 minutes:
  - Error rate normal? → Increase v2 to 50% → 100%
  - Error rate spike? → Route all traffic back to v1

✅ Real user testing with blast radius limited
✅ Gradual confidence building
❌ Requires traffic splitting (Istio, load balancer)
❌ More complex setup

4. Feature Flags (safest)

// Deploy v2 code with flag OFF
if (featureFlags.isEnabled("new-checkout", user.id)) {
  return newCheckout(cart);
}
return legacyCheckout(cart);

// Enable for 1% of users → 5% → 25% → 100%
// Kill switch: disable flag = instant "rollback" without redeployment

Decision Matrix

Strategy	Downtime	Cost	Rollback	Complexity
Rolling	None	1x	Slow	Low
Blue-Green	None	2x	Instant	Medium
Canary	None	1.2x	Fast	High
Feature Flags	None	1x	Instant	Medium

Start with: Feature flags + rolling updates. Add canary when you have monitoring mature enough to auto-detect issues.