NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI, #452 most downloaded on PyPI
Distributed Task Queue.
Latest stable
5.7.0
Last release 17 days ago
05 Nov 2026
Release timing varies
gaps range from 2 weeks to 6 months
Some releases are documented
notes for 28 of the last 60 stable releases
2 versions withdrawn
withdrawn after publishing
18 years old
228 releases · first in 2009
Drag across the ruler to compare, or tap a release to open it.
Removed deprecated arguments from Redis connection calls.
We're pleased to announce Celery 5.7.0, a new stable release of Celery.
This release brings substantial improvements to task workflows, worker reliability, result handling, scheduling, and broker connectivity. It also includes new integrations, improved Python compatibility, performance optimizations, and a large number of bug fixes contributed by the community.
A significant part of this release focuses on making complex Canvas workflows more reliable and efficient, improving worker behavior during broker disconnections and shutdown, and making result-backend state handling more consistent.
ignore_result behavior, and concurrent AsyncResult tracking.pydantic=True can accept Pydantic model instances.pymemcache as a result backend.Celery 5.7 includes extensive improvements to Canvas execution and composition, particularly for nested workflows and complex chains and chords.
| operator.GroupResult.as_tuple() after groups in chains.GroupResult fan-out behavior in chains.Ignore or Reject.link and link_error options on chords so they are applied to the chord body.map and starmap option merging.These changes make complex Canvas compositions more predictable and reduce unnecessary broker traffic and message growth.
This release includes a broad set of improvements to worker lifecycle management, process pools, connection recovery, and shutdown behavior.
spawn start method with the asynchronous prefork pool.--concurrency=auto.task_retry_on_worker_lost to control task requeues after a worker process is lost.worker_process_init signal.These changes improve worker resilience in environments with frequent broker reconnections, process failures, and graceful shutdown requirements.
Task publishing and execution behavior received several important correctness improvements.
acks_on_failure and acks_on_timeout, allowing these acknowledgment behaviors to be configured independently.retry_kwargs mutation in automatic retry handling.send_task().ignore_result defaults in send_task() and Task.apply().ignore_result overrides across task retries.RETRY after worker loss.task_failure.REVOKED while another worker is still executing it.Result handling has received significant reliability and consistency improvements across several backends.
result_chord_expires to control the expiration of chord-related Redis keys.result_failure_expires to configure expiration for failure results.decode_responses is enabled.DriverInfo support.pymemcache as a result backend.result_serializer.date_done index to the database backend.engine_callback to the database backend SessionManager.SUCCESS result in the database, MongoDB, and Cassandra backends.forget() across supported backends.result_extended.AsyncResult.exists() to check whether a result exists.AsyncResult instances waiting for the same task ID.ResultSet.add() so results can be added after the set has already become ready.0 is respected rather than treated as an indefinite wait.wait_for, ResultSet.join(), get_many(), and result draining.Celery 5.7 includes extensive fixes to scheduling and timezone handling, especially around daylight-saving-time transitions.
crontab.remaining_delta() handling for timezone-aware datetimes.make_aware().beat_cron_starting_deadline.inspect ping support to Celery Beat.celery_beat_task header to task options.These changes improve scheduling correctness and make Beat more resilient when individual scheduled tasks or schedule entries fail.
Several improvements were made to queue selection, routing, and broker reconnection behavior.
add_consumer is used.Tasks configured with pydantic=True can now accept Pydantic model instances when called.
This improves the integration between Celery tasks and applications that use Pydantic models for data validation and structured task arguments.
ConfigurationView behavior so membership checks, key lookups, and iteration respect configuration mapping order.ConfigurationView.swap_with() handling.config_from_object(silent=True).add_defaults() behavior when configuration loading fails on an already configured application.backends.by_url() accepts backend classes.@shared_task is applied more than once.__class_getitem__ support to generic classes.task_repr_maxlevels to control the depth of task argument and keyword-argument representations.datefmt support to ColorFormatter and TaskFormatter.saferepr truncation to preserve balanced closing brackets.celery multi node-index handling, including index lists for named nodes.celery multi failures with node names containing dashes.--workdir handling by expanding ~ for spawned workers.codecs.open() usage in the settings upgrade command.regen iteration behavior, including generator slicing and source replacement.BufferMap handling of bufmaxsize and its total count.LimitedSet timestamp and sequence-number handling when items are re-added or copied.firstmethod() so it does not swallow AttributeError.apply_async() CPU overhead.State._tasks_to_resolve to prevent unbounded memory growth.Python 3.10 is now the minimum supported Python version.
Python 3.9 support has been removed from the package and its associated development and CI configurations.
This release also includes compatibility improvements for newer Python versions, including fixes for Python 3.14 annotation evaluation and function introspection behavior.
>=4.3.1.5.7.0.click-repl to 0.4.0.brotlipy with brotlicffi in the Brotli extra, with support for both CPython and PyPy.Optional integrations and development dependencies have also received compatibility updates.
Before upgrading, review the following points:
>=4.3.1 and uses Kombu 5.7.0.acks_on_failure and acks_on_timeout are now decoupled, so applications that rely on their previous combined behavior should review their task acknowledgment settings.result_chord_expires and result_failure_expires independently.brotlicffi instead of brotlipy.Consult the [Celery documentation](https://docs.celeryq.dev/en/stable/) for configuration details and upgrade guidance.
Thank you to everyone who contributed code, tests, documentation, reviews, and feedback to this release.
This release includes contributions from many first-time contributors as well as long-standing Celery maintainers and community members. We appreciate every contribution that helped make Celery 5.7 more reliable and maintainable.
See the automatically generated contributor list and individual pull requests in the full changelog.
For the complete list of changes between Celery 5.6.3 and 5.7.0, including individual pull requests and contributors, see:
[Compare v5.6.3...v5.7.0](v5.6.3...v5.7.0)
__class_getitem__ to generic classes by @brianhelba in #10038TypeError during exception handling by @ChickenBenny in #10178reconnect_on_error to BaseResultConsumer by @ChickenBenny in #10189engine_callback to database backend SessionManager by @avolant in #10135Note truncated.
Your coding agent can read these notes before it upgrades. Set up the MCP server →
Celery v5.7.0 Stable Release is now available for testing and deploy to production. Please help us test this version and report any issues.
Chore(deps): Bump cryptography from 50.0.1 to 50.0.2 by @dependabot [bot] in #10815
| by @LD-RW in #10840Full Changelog: v5.7.0b1...v5.7.0rc1
This section summarizes the changes made since the Celery 5.7.0rc1 pre-release.
Run a replaced task's callbacks once when it is replaced with a chain (`10927).
Fix chord link and link-error options so they are passed to the body only (`10778).
Fix stamped header revocation (`10884).
Count each chord header task only once in the Redis backend (`10919).
Fix Redis chord claim cleanup ownership (`10998).
Fail the chord when chord_unlock retry is refused by the broker (`10959).
Fix retry argument conflicts when switching calling conventions (`10895).
Pass values to backend hooks by position again (`10918).
Don't mark a task REVOKED while another worker is running it (`10922).
Keep the first SUCCESS in the database, MongoDB, and Cassandra backends (`10920, 10944, 10945).
Don't record FAILURE when on_success raises after SUCCESS was stored (`10759).
Don't fail a task whose SUCCESS was stored before the worker was lost (`10752).
Fix a worker hang in WorkerLostError when result_backend_always_retry=True (`10960).
Fix send_task() overriding ignore_result defaults (`10953).
Preserve ignore_result overrides across task retries (`10958).
Make Task.apply() respect the task.ignore_result default (`11001).
Don't store RETRY for an ignored result when retrying after a lost worker (`11009).
Track every AsyncResult waiting on a task ID (`10956).
Allow ResultSet.add() after the set's results are ready (`10930).
Use the documented retry defaults in the database backend (`10921).
Fix Manager warning output during retries (`10948).
Wait between retries when the broker refuses a callback publish (`10975).
Merge signature options in map and starmap apply_async() (`10787).
Stop gating task publishing signals on cached receiver lists (`10791).
Ensure worker shutdown state is set even if the shutdown handler's logging or signal dispatch raises (`10738).
Cover the second call of _ensure_after_fork in a test (`10924).
Cancel only per-connection timer entries on broker reconnect instead of clearing the worker timer (`10901).
Keep Celery Beat running when a schedule entry fails, and reject impossible crontab dates (`10896).
Fix solar schedule timing across daylight-saving time changes (`10915).
Enhance timezone localization tests (`10996).
Use Cuba's 2025 fall-back in the midnight DST tests (`11011).
Don't expire or evict re-added LimitedSet items using their old timestamp (`10758).
Give entries copied by LimitedSet.update() this set's sequence numbers (`10943).
Keep the closing bracket when saferepr truncates the last element (`10771).
Fix MultiParser to honor its configured Node class (`10938).
Raise a clear error when @shared_task is applied twice (`10889).
Fix an incorrect sentence in the user guide (`10902).
Document the quorum queue delivery limit (`10926).
Improve signature documentation (`10978).
Update the pull request template with AI-spamming guidelines (`10979).
Add guidance for AI agents in the Celery project (`10999).
Update click-plugins requirement to >=1.1.1.2 (`10899).
Update Django requirement to >=3.2.1 (`10816).
Replace brotlipy with brotlicffi in the Brotli extra (`10992).
Update Brotli dependencies for PyPy and CPython (`10994).
Update Billiard and click-repl versions in requirements (`11002).
Update grpcio to 1.85.1 (`11006).
Update Kombu to 5.7.0 (`11007).
Update GitHub CodeQL action to 4.39.0 and 4.39.1 (`10900, 10974).
Update Pydantic to 2.14.1 (`10903).
Update tox to 4.65.0 (`10935).
Update mypy to 2.4.0 (`10984).
Update grpcio to 1.84.1 (`10973).
Update azure-storage-blob to 12.31.1 (`11005).
Update google-cloud-storage to 3.17.1 (`11013).
Update azure-identity to 1.27.0 (`11020).
Assert exact task replacement nesting levels (`10925).
Run mock assertions inside pytest.raises (`11004).
This section summarizes the changes made since the Celery 5.7.0rc1 pre-release.
Run a replaced task’s callbacks once when it is replaced with a chain ( :ghpull:`10927 ).
Fix chord link and link-error options so they are passed to the body only ( :ghpull:`10778 ).
Fix stamped header revocation ( :ghpull:`10884 ).
Count each chord header task only once in the Redis backend ( :ghpull:`10919 ).
Fix Redis chord claim cleanup ownership ( :ghpull:`10998 ).
Fail the chord when chord_unlock retry is refused by the broker ( :ghpull:`10959 ).
Fix retry argument conflicts when switching calling conventions ( :ghpull:`10895 ).
Pass values to backend hooks by position again ( :ghpull:`10918 ).
Don’t mark a task REVOKED while another worker is running it ( :ghpull:`10922 ).
Keep the first SUCCESS in the database, MongoDB, and Cassandra backends ( :ghpull:10920 , :ghpull:10944 , :ghpull:10945` ).
Don’t record FAILURE when on_success raises after SUCCESS was stored ( :ghpull:`10759 ).
Don’t fail a task whose SUCCESS was stored before the worker was lost ( :ghpull:`10752 ).
Fix a worker hang in WorkerLostError when result_backend_always_retry=True ( :ghpull:`10960 ).
Fix send_task() overriding ignore_result defaults ( :ghpull:`10953 ).
Preserve ignore_result overrides across task retries ( :ghpull:`10958 ).
Make Task.apply() respect the task.ignore_result default ( :ghpull:`11001 ).
Don’t store RETRY for an ignored result when retrying after a lost worker ( :ghpull:`11009 ).
Track every AsyncResult waiting on a task ID ( :ghpull:`10956 ).
Allow ResultSet.add() after the set’s results are ready ( :ghpull:`10930 ).
Use the documented retry defaults in the database backend ( :ghpull:`10921 ).
Fix Manager warning output during retries ( :ghpull:`10948 ).
Wait between retries when the broker refuses a callback publish ( :ghpull:`10975 ).
Merge signature options in map and starmap apply_async() ( :ghpull:`10787 ).
Stop gating task publishing signals on cached receiver lists ( :ghpull:`10791 ).
Ensure worker shutdown state is set even if the shutdown handler’s logging or signal dispatch raises ( :ghpull:`10738 ).
Cover the second call of _ensure_after_fork in a test ( :ghpull:`10924 ).
Cancel only per-connection timer entries on broker reconnect instead of clearing the worker timer ( :ghpull:`10901 ).
Keep Celery Beat running when a schedule entry fails, and reject impossible crontab dates ( :ghpull:`10896 ).
Fix solar schedule timing across daylight-saving time changes ( :ghpull:`10915 ).
Enhance timezone localization tests ( :ghpull:`10996 ).
Use Cuba’s 2025 fall-back in the midnight DST tests ( :ghpull:`11011 ).
Don’t expire or evict re-added LimitedSet items using their old timestamp ( :ghpull:`10758 ).
Give entries copied by LimitedSet.update() this set’s sequence numbers ( :ghpull:`10943 ).
Keep the closing bracket when saferepr truncates the last element ( :ghpull:`10771 ).
Fix MultiParser to honor its configured Node class ( :ghpull:`10938 ).
Raise a clear error when @shared_task is applied twice ( :ghpull:`10889 ).
Fix an incorrect sentence in the user guide ( :ghpull:`10902 ).
Document the quorum queue delivery limit ( :ghpull:`10926 ).
Improve signature documentation ( :ghpull:`10978 ).
Update the pull request template with AI-spamming guidelines ( :ghpull:`10979 ).
Add guidance for AI agents in the Celery project ( :ghpull:`10999 ).
Update click-plugins requirement to >=1.1.1.2 ( :ghpull:`10899 ).
Update Django requirement to >=3.2.1 ( :ghpull:`10816 ).
Replace brotlipy with brotlicffi in the Brotli extra ( :ghpull:`10992 ).
Update Brotli dependencies for PyPy and CPython ( :ghpull:`10994 ).
Update Billiard and click-repl versions in requirements ( :ghpull:`11002 ).
Update grpcio to 1.85.1 ( :ghpull:`11006 ).
Update Kombu to 5.7.0 ( :ghpull:`11007 ).
Update GitHub CodeQL action to 4.39.0 and 4.39.1 ( :ghpull:10900 , :ghpull:10974` ).
Update Pydantic to 2.14.1 ( :ghpull:`10903 ).
Update tox to 4.65.0 ( :ghpull:`10935 ).
Update mypy to 2.4.0 ( :ghpull:`10984 ).
Update grpcio to 1.84.1 ( :ghpull:`10973 ).
Update azure-storage-blob to 12.31.1 ( :ghpull:`11005 ).
Update google-cloud-storage to 3.17.1 ( :ghpull:`11013 ).
Update azure-identity to 1.27.0 ( :ghpull:`11020 ).
Assert exact task replacement nesting levels ( :ghpull:`10925 ).
Run mock assertions inside pytest.raises ( :ghpull:`11004 ).
Replace deprecated codecs.open in celery upgrade settings by @Barry-Lenhart in #10706
Full Changelog: v5.7.0a1...v5.7.0b1
Celery v5.7.0b1 Beta 1 is now available for testing. Please help us test this version and report any issues.
Add opt-in remote control (inspect ping) support to celery beat (#10422)
Add pymemcache result backend support (#9977)
Make --concurrency=auto aware of cgroup CPU limits (#10328)
Fix Couchbase missing-document handling to raise instead of returning None (#10680)
Preserve propagate when joining nested native results (#10689)
Keep forget() idempotent on Elasticsearch and CouchDB backends (#10697)
Keep forget() idempotent on Azure Blob Storage and CosmosDB (#10714)
Handle missing ArangoDB documents when deleting results (#10701)
Upsert CosmosDB results so later task states are stored correctly (#10746)
Publish the Cassandra session only once the connection is ready (#10657)
Avoid holding the Redis pubsub lock while waiting for messages (#10715)
Defer re-entrant cancel_for to avoid Redis pubsub deadlocks (#10716)
Preserve EagerResult.name when copying or pickling (#10773)
Raise TaskRevokedError from EagerResult.get() after revoke() (#10772)
Finish a half-read child result on prefork shutdown and acknowledge it during the join (#10768)
Apply countdown and eta to chord headers (#10735)
Fail chords after a chord whose header failed (#10767)
Fix geometric message bloat in chains of implicit chords (#10743)
Resolve the app of a chain or chord through its tasks (#10749)
Avoid raising in chain.app when the first task's canvas contains a dict (#10792)
Flush buffered events with hub.call_soon() (#10736)
Synchronize EventDispatcher shutdown with publishing (#10745)
Preserve pending operation order in the consumer (#10734)
Fix consumer reconnect prefetch reduction for green pools (#10804)
Fix Beat scheduler connection cleanup on shutdown (#10776)
Fix countdown, expiration, and Beat intervals across DST transitions (#10769)
Limit Django fixup signal handlers to its own app (#10731)
Tell apart signal receivers bound to different instances (#10699)
Use the merged retry policy for before_task_publish and task-sent events (#10708)
Avoid calling isoformat() on protocol v1 ETA and expiration strings (#10720)
Apply index lists such as -c:1,2 to named nodes in celery multi (#10661)
Expand ~ in --workdir for spawned workers (#10739)
Replace deprecated codecs.open in celery upgrade settings (#10706)
Fix escaped percent signs in node formatting (#10663)
Fix BufferMap handling of bufmaxsize and total size accounting (#10705)
Fix regeneration iteration when consumed elsewhere during the loop (#10756)
Stop firstmethod() from swallowing AttributeError (#10774)
Handle AttributeError in iterate_file_descriptors_safely() after connection loss (#10318)
Fix cache directory ownership in Docker builds (#10707)
Update the billiard dependency to 4.3.1 or later (#10802)
Update the Kombu dependency to 5.7.0b1 (#10809)
Remove version constraints for Redis requirements (#10810)
One column per quarter.