arch-circuit-breaker

One-line summary

Stop calling a failing dependency for a while so it can recover and your service doesn't drown in timeouts. Combine with bulkheads (isolate resources per dependency), retries with backoff (handle transient blips), and timeouts (bound the failure window) for a complete resilience story.

When to use this skill

• Service-to-service calls (REST, gRPC, message broker) where the downstream can fail or slow down.
• External dependencies (databases, third-party APIs, payment providers) that can degrade.
• Any system where one slow dependency can cascade: threads block on it, the thread pool fills, the service stops accepting requests.
• Anywhere you'd otherwise write "I'll just retry until it succeeds" — that's the cascade you're about to cause.

When NOT to use this skill

• All dependencies are in-process — no network failure mode to protect against.
• A simple synchronous failure that should just propagate to the caller — circuit breaker hides the failure, which isn't always desirable.
• Total volume is low and a slow call won't cascade — the operational overhead exceeds the benefit.

Core content