Getting Started Guides Architecture API Changelog

Choose SDK

Python/Guides/

Reliability

Harden your task queue for production workloads.

← PreviousWorkflows Next →Retries

Retries & Dead LettersAutomatic retries with exponential backoff, dead letter queue Error HandlingTask failure lifecycle, error inspection, debugging patterns Delivery GuaranteesAt-least-once delivery, idempotency, and exactly-once patterns IdempotencyAuto-derived dedup keys — identical enqueues coalesce to one execution DebouncingCollapse a burst of enqueues into one run on a sliding deadline Flow controlDebounce vs throttle vs coalesce, and what to do with excess load Rate LimitingThrottle task execution with token bucket rate limits ConcurrencyPool size, cluster-wide task caps, per-worker fairness, admission control TimeoutsHard per-attempt deadlines and cooperative soft timeouts Dead-letter queueInspect, retry, and purge jobs that exhausted their retries Circuit BreakersProtect downstream services from cascading failures Distributed LockingMutual exclusion across workers with database-backed locks