🛡️Circuit Breaker / Retry / Bulkhead

Use complementary resilience patterns instead of one-size-fits-all retries.

📌 Key IdeaRetries recover transient failures, circuit breakers fail fast on sustained outages, and bulkheads isolate resources so one bad dependency cannot exhaust the whole service.

🧩

Problem it solves

A failing downstream dependency can cascade and take down otherwise healthy services.
Blind retries during outages often increase load and make recovery slower.
Circuit breaker + retry + bulkhead split recovery, fail-fast, and isolation concerns cleanly.

✨

Why It Matters

Resilience patterns prevent one dependency outage from crashing the whole service.
Correct policies protect latency SLOs during partial failures and traffic spikes.
They convert random failure behavior into controlled, observable degradation.

🖼

Visual

🛡️ Resilience trio in call path

Request -> Retry policy -> Circuit Breaker -> Bulkhead-isolated dependency

Retry      : handles transient failures
Breaker    : stops hammering unhealthy dependency
Bulkhead   : limits blast radius by resource partitioning

✅

When to Use

Networked microservice calls with dependency failures.
Critical APIs where cascading failures must be contained.
Systems with variable downstream latency and intermittent outages.

🚫

When NOT to Use

Blind retries on non-idempotent operations.
Huge retry counts without jitter/backoff (retry storms).

💻

Code Variations

Resilience4j sketchjava

Supplier<Response> guarded = Decorators.ofSupplier(() -> client.call())
    .withRetry(retry)
    .withCircuitBreaker(circuitBreaker)
    .withBulkhead(bulkhead)
    .decorate();

Order and policy tuning matter; observe rejection/timeout metrics.

📊

Pattern purpose matrix

Aspect	Retry	Circuit Breaker	Bulkhead
Primary problem solved	Transient error recovery	Fail-fast during sustained failure	Resource isolation / blast-radius control
Main risk	Amplified load under outage	Too aggressive open thresholds	Over-partitioning resources
Must pair with	Backoff + jitter + idempotency	Health probes/half-open tuning	Capacity planning + queueing policy

🎤

Interview Q&A

🧾

Quick Cheat Sheet

RetryTransient failures only; exponential backoff + jitter.

BreakerOpen on failure threshold; half-open probes for recovery.

BulkheadSeparate pools/queues per dependency path.

🔗