Resilience Engineering

Checklist et patterns pour le code backend qui appelle des services externes (APIs, bases de donnees, services tiers).

Declencheurs

• Code qui appelle des APIs externes, bases de donnees, ou services tiers
• "resilience", "circuit breaker", "retry", "timeout", "fallback"
• Backend work avec appels reseau

Checklist (avant de marquer du code backend comme termine)

• Timeouts — Every external call has an explicit timeout (no infinite waits)
• Retry + backoff — Transient failures retry with exponential backoff + jitter
• Fallback chain — Critical paths degrade gracefully (primary → secondary → cached → default)
• Stale-on-error — Cache last successful response; serve stale data on upstream failure
• Circuit breaker — Failing endpoints are short-circuited (cooldown before retry)
• In-flight dedup — Concurrent identical requests share a single upstream call
• Negative caching — Failed lookups are cached briefly to prevent retry storms
• Rate-aware requests — Rate-limited APIs use staggered sequential calls, not parallel
• Backpressure — Producers slow down when consumers can't keep up

Pattern Reference

Request → Timeout guard → Circuit breaker check
  → Cache lookup (hit → return)
  → In-flight dedup check (pending → await existing)
  → Upstream call (with retry + backoff)
    → Success → Update cache → Return
    → Failure → Serve stale cache → Trip circuit breaker

When NOT to apply

• Internal function calls (no network boundary)
• Local file reads (OS handles retries)
• Development/test environments (keep it simple)
• One-shot scripts (no long-running process)