Skip to content

Across processes (Redis)

A circuit_breaker lives in one process. Behind a load balancer with 40 workers, each worker discovers a dead dependency on its own, and a rate_limit(100) lets 4,000 calls through. nopanic-redis keeps that state in Redis, so the fleet trips together and shares one quota.

pip install nopanic-redis

import redis
from nopanic_redis import redis_circuit_breaker, redis_rate_limit

r = redis.Redis()   # redis.asyncio.Redis() for async functions

payments = redis_circuit_breaker(r, name="payments", min_calls=20, reset_timeout=15.0)
openai_quota = redis_rate_limit(r, name="openai", rate=500, per=60.0, max_wait=5.0)

@openai_quota
@payments
def charge(order): ...

Semantics

  • Same model as the local policies. redis_circuit_breaker takes the parameters of circuit_breaker (failure threshold over a bucketed sliding window, min_calls, reset_timeout, half_open_max_calls, on); redis_rate_limit behaves like rate_limit (waiting, max_wait, RateLimited.retry_after, deadlines, refunds for cancelled waiters).
  • Identity is the name. Every instance with the same name and prefix is the same breaker or bucket, whichever process created it. Keys carry a Redis Cluster hash tag, so one policy's keys share a slot.
  • One atomic script per decision. A breaker costs two round trips per call (admit, record), a rate limit one.
  • Wall-clock time. Decisions use the callers' time.time() (injectable via clock=). Keep hosts NTP-synced; skew only shifts window boundaries slightly.
  • Redis down is not your outage. By default (fail_open=True) calls go through unguarded and a breaker.store_error / ratelimit.store_error event is emitted. fail_open=False rejects with CircuitOpen / RateLimited instead.
  • Client flavour decides call flavour. A redis.Redis client serves sync functions, a redis.asyncio.Redis client async ones. To guard both, create two instances with the same name: they share state.
  • get_state() and reset() act fleet-wide (they return awaitables with an async client). Streams are not supported.

Combine freely with local policies: a local bulkhead per process and a shared redis_rate_limit per fleet is a common pairing.