Timeouts, Retries, Bulkheads, and Backpressure
A service is only as reliable as its dependencies allow. A slow database or API can consume all request resources if there are no timeouts or concurrency limits.
Resilience techniques include deadlines, bounded retries with jitter, circuit breakers, bulkheads, rate limits, queues, and backpressure.
Request
|
v
Timeout --> Retry policy --> Downstream
| |
| +--> bounded attempts
v
Fallback / error
Observability should include latency, error rate, saturation, dependency timings, and retry counts.
Key Takeaways
- Timeouts are mandatory for remote dependencies.
- Retries can amplify outages if uncontrolled.
- Backpressure and concurrency limits protect downstream systems.