Policies & decorators

Policies

Policy and AsyncPolicy are the unified containers. Configure retries with Retry or use RetryPolicy / AsyncRetryPolicy as sugar:

from redress import Policy, Retry, default_classifier
from redress.strategies import decorrelated_jitter

policy = Policy(
    retry=Retry(
        classifier=default_classifier,
        strategy=decorrelated_jitter(max_s=5.0),
        deadline_s=60,
        max_attempts=6,
    )
)

result = policy.call(lambda: do_work(), operation="sync_op")

Async mirrors the same API and uses asyncio.sleep under the hood.

You can add a circuit breaker at the policy level:

from redress import CircuitBreaker, ErrorClass, Policy, Retry

policy = Policy(
    retry=Retry(...),
    circuit_breaker=CircuitBreaker(
        failure_threshold=5,
        window_s=60.0,
        recovery_timeout_s=30.0,
        trip_on={ErrorClass.TRANSIENT, ErrorClass.SERVER_ERROR},
    ),
)

Context managers let you reuse hooks/operation:

with policy.context(operation="batch") as retry:
    retry(task1)
    retry(task2)

Decorator

@retry auto-chooses sync/async policy and defaults to default_classifier + decorrelated_jitter(max_s=5.0):

from redress import retry

@retry
def fetch_user():
    ...

@retry
async def fetch_user_async():
    ...

You can override classifiers/strategies and hooks via decorator arguments.