Deployment Strategies

Table of Contents

🟡 P2 — mentioned in designs but rarely the crux; know the trade-offs

Strategies #

Strategy	Mechanism	Rollback	Risk
Rolling	Replace instances one-by-one	Stop and reverse	Low
Blue-Green	Two identical environments, swap traffic	Swap back	Medium
Canary	Route small % to new version, observe	Route back	Lowest
Shadow/Dark	Mirror traffic to new version, discard results	N/A (no impact)	Lowest

Instinct #

Canary for most production services. Route 1-5% of traffic to the new version, monitor RED metrics for 15-30 minutes, then gradually increase. Blue-Green for database migrations where you need instant rollback capability. Rolling for stateless services where canary infrastructure isn’t available.

Dark Launches (Shadow Traffic) #

Mirror production traffic to the new version but discard the results.

Production traffic
  ├──→ Live system (serves response to user)
  └──→ Shadow system (processes same request, response discarded)
       ↓
   Compare: latency, errors, output differences

Use cases: Validate migration handles real queries; load-test with production patterns; verify rewrite produces identical results.

Instinct: Dark launches are the safest validation because there’s zero user impact. Pair with Feature Flags for controlled rollout after validation.

References #

Dark Launching — Martin Fowler