Merged
Conversation
Member
Author
|
This pull request introduces 1 alert when merging 54a4165 into 05da357 - view on LGTM.com new alerts:
|
Member
Author
|
I have one question regarding
What do you think is the best approach? |
|
@matusvalo I get a deadlocks using this fix since my test script freezes for awhile and then does a few things, freezes again and get the familiar |
Member
Author
|
@stadelmanma could you provide that script? |
Member
Author
|
@stadelmanma I missed that you were reporting on py-amqp related issue :-). Is that the failing script one provided in this comment: celery/py-amqp#330 (comment)? I have used this script for debugging and fixing the issue. Before this PR the testing script was failing with timeouts and The only difference maybe from your setup can be that I am reaching RabbitMQ broker not in localhost (and hence with higher latency). But still this does not explain why it was failing before the fix and after ran with no issue on my setup. |
maybe-sybr
suggested changes
Oct 29, 2020
Contributor
maybe-sybr
left a comment
There was a problem hiding this comment.
Overall this change makes sense. I feel like there's probably a bunch of other places where we're unlikely to be thread-safe but this is a good place to start.
I think we could definitely stand to add a more clever @cached_thread_local_property decorator, like you mention. In theory that decorator could enforce that the methods it decorates belong to a class definition with some metaclass which can provide the actual thread local storage object. @auvipy, should we make an issue to track that on kombu? In the meantime, just changing it to be a property and not having setter/deleter support seems fine, but I think we'll need to fix up some tests which assume they can set app.backend and will get AttributeErrored when they try now - e.g. t/unit/tasks/test_chord.py::test_add_to_chord::test_add_to_chord() breaks at the moment.
celery/app/base.py
Outdated
| {0}="proj.celeryconfig" | ||
| """ | ||
|
|
||
| tlocal = threading.local() |
Contributor
There was a problem hiding this comment.
This needs to be moved to live in the Celery class instance so that it gets cleaned up by our fixtures during testing and there's no inter-app cross-talk.
mskogorevrmc
previously approved these changes
Nov 2, 2020
mskogorevrmc
reviewed
Nov 2, 2020
mskogorevrmc
left a comment
There was a problem hiding this comment.
I'm sorry, I didn't think I have enough rights to make the approval of this pull request.
Consider my comment and approval useless.
Member
|
any constructive review is OK |
Member
Author
|
Hi @stadelmanma, I have double check it via your script and still seems that this PR fixes the issue (I reran the test using only your script with RabbitMQ borker in localhost using docker) results from master branch
When PR branch with fix is used no issues were present
Can someone double check and confirm my observation. I would like to be 100% sure that PR is not breaking something and also is fixing the issue. |
|
This pull request fixes 2 alerts when merging 2b0af5b into 28ebcce - view on LGTM.com fixed alerts:
|
15f9e06 to
f6ea14a
Compare
|
This pull request fixes 2 alerts when merging f6ea14a into 60ba379 - view on LGTM.com fixed alerts:
|
f6ea14a to
75c2517
Compare
|
This pull request fixes 1 alert when merging 75c2517 into 60ba379 - view on LGTM.com fixed alerts:
|
75c2517 to
519c832
Compare
Codecov Report
@@ Coverage Diff @@
## master #6416 +/- ##
=======================================
Coverage 75.49% 75.50%
=======================================
Files 138 138
Lines 16383 16396 +13
Branches 2059 2059
=======================================
+ Hits 12369 12379 +10
- Misses 3792 3795 +3
Partials 222 222
Continue to review full report at Codecov.
|
|
There’s still deadlocks, this shouldn’t have been merged yet: #6521 |
Member
you can help find out & a possible fix for the remaining part. as no software is 100% perfect |
|
I know how it is! 😄 doing my part by investigating and spreading awareness, as threaded code isn’t my strong suit. I think you’re correct that a change like this should have gotten more test coverage. It’s a great feature, but touching a lot of sensitive code. |
|
I am getting the same issue on one worker, the others work just fine. Using celery 5.0.4 |
sihrc
added a commit
to IndicoDataSolutions/celery
that referenced
this pull request
Dec 23, 2020
* Remove defaults for unsupported Python runtimes. * Remove obsolete test. * Doc pytest plugin (celery#6289) * update to new pytest name * doc pytest plugin * trim heading to the length of the new pytest name * add warning against use of sort key on dynamodb table, closes celery#6332 * Remove celery.five and bump vine dep (celery#6338) * improv: Replace `five.values` with `dict.values` * improv: Use `time.monotonic()` in kombu tests Also in the docs where it is used to demonstrate `memcache` timeouts. * rm: Delete `celery.five` `vine.five` is no longer present in `vine >= 5`. * triv: Remove refs to `celery.five` in docs, &c * build: Bump `vine` dependency to 5.0+ * Wheels are no longer universal. * Remove failing before_install step. * Update changelog. * Bump version: 5.0.0rc2 → 5.0.0rc3 * Fix release date. * Remove unused import. * Correctly skip these tests when the relevant dependency is missing. * Expose retry_policy for Redis result backend Rather than adding a new top-level config option, I have used a new key in the already existing setting `result_backend_transport_options`. Closes celery#6166 * Update changelog for 4.3.1. * fix typo (celery#6346) * Travis CI: Test Python 3.9 release candidate 1 (celery#6328) * Travis CI: Test Python 3.9 release candidate 1 * fixup! Travis CI: matrix --> jobs * fixup! Fix indentation error * fixup! tox.ini: 3.9 --> 3.9-dev * Fix test failure in Python 3.9RC1. Co-authored-by: Omer Katz <omer.drow@gmail.com> * Fix the broken celery upgrade settings command. * Fix celery migrate settings options. * Remove Riak result backend settings. * Rephrase to mention that 3.5 is also EOL. * Add a note about the removal of the Riak result backend. * Fix examples of starting a worker in comments (celery#6331) * Remove deprecated function from app.log.Logging. * Migration guide. * Document breaking changes for the CLI in the Whats New document. * Add a "port code to Python 3" migration step. * Update supported Python versions in the introduction document. * Update bash completion. * Add note about new shell completion. * Update daemonization docs. * Remove amqp backend. (celery#6360) Fixes celery#6356. * Warn when deprecated settings are used (celery#6353) * Warn when deprecated settings are used. * Mention deprecation in docs. * Refer to the right place in the documentation. * Complete What's New. * Add wall of contributors. * Update codename. * Fix alt text. * isort. * PyPy 3.7 is currently in alpha. No need for that sentence. * Mention the new pytest-celery plugin. * Mention retry policy for the redis result backend. * Fix phrasing. * Mention ordered group results are now the default. * pyupgrade. * Complete release notes. * Bump version: 5.0.0rc3 → 5.0.0 * Happify linters. * Specify utf-8 as the encoding for log files. Fixes celery#5144. * Fixed some typos in readme * Fix custom headers propagation for protocol 1 hybrid messages * Retry after race during schema creation in database backend (celery#6298) * Retry after race during schema creation in database backend Fixes celery#6296 This race condition does not commonly present, since the schema creation only needs to happen once per database. It's more likely to appear in e.g. a test suite that uses a new database each time. For context of the sleep times I chose, the schema creation takes ~50 ms on my laptop. I did a simulated test run of 50 concurrent calls to MetaData.create_all repeated 200 times and the number of retries was: - 0 retries: 8717x - 1 retry: 1279x - 2 retries 4x * Add test for prepare_models retry error condition * Add name to contributors * Update daemonizing.rst Fix daemonizing documentation for issue celery#6363 to put `multi` before `-A` * Revert "Update daemonizing.rst" (celery#6376) This reverts commit 96ec6db. * bugfix: when set config result_expires = 0, chord.get will hang. (celery#6373) * bugfix: when set config result_expires = 0, chord.get will hang. `EXPIRE key 0` will delete a key in redis, then chord will never get the result. fix: celery#5237 * test: add testcase for expire when set config with zero. * Display a custom error message whenever an attempt to use -A or --app as a sub-command option was made. Fixes celery#6363 * Remove test dependencies for Python 2.7. * Restore the celery worker --without-{gossip,mingle,heartbeat} flags (celery#6365) In the previously used argparse arguments framework, these three options were used as flags. Since 5.0.0, they are options which need to take an argument (whose only sensible value would be "true"). The error message coming up is also (very) hard to understand, when running the celery worker command with an odd number of flags: Error: Unable to parse extra configuration from command line. Reason: not enough values to unpack (expected 2, got 1) When the celery worker is run with an even number of flags, the last one is considered as an argument of the previous one, which is a subtle bug. * Provide clearer error messages when app fails to load. * fix pytest plugin registration documentation (celery#6387) * fix pytest plugin registration documentation * Update docs/userguide/testing.rst Co-authored-by: Thomas Grainger <tagrain@gmail.com> Co-authored-by: Omer Katz <omer.drow@gmail.com> * Contains a workaround for the capitalized configuration issue (celery#6385) * Contains a workaround for the capitalized configuration issue * Update celery/apps/worker.py Co-authored-by: Omer Katz <omer.drow@gmail.com> * Update celery/apps/worker.py Co-authored-by: Omer Katz <omer.drow@gmail.com> Co-authored-by: Omer Katz <omer.drow@gmail.com> * Remove old explanation regarding `absolute_import` (celery#6390) Resolves celery#6389. * Update canvas.rst (celery#6392) * Update canvas.rst Tiny fixes. * Update docs/userguide/canvas.rst Co-authored-by: Omer Katz <omer.drow@gmail.com> Co-authored-by: Omer Katz <omer.drow@gmail.com> * Remove duplicate words from docs (celery#6398) Remove the duplicate usage of “required” in documentation (specifically, `introduction.rst`). * Allow lowercase log levels. (celery#6396) Fixes celery#6395. * Detach now correctly passes options with more than one word. (celery#6394) When specifying options such as `-E` the detached worker should receive the `--task-events` option. Instead it got the `--task_events` option which doesn't exist and therefore silently failed. This fixes celery#6362. * The celery multi command now works as expected. (celery#6388) * Contains the missed change requested by @thedrow * Added a some celery configuration examples. * fixed loglevel info->INFO in docs * return list instead set in CommaSeparatedList _broadcast method of kombu Mailbox. does not support set https://github.com/celery/kombu/blob/7b2578b19ba4b1989b722f6f6e7efee2a1a4d86a/kombu/pidbox.py#L319 * Rewrite detaching logic (celery#6401) * Rewrite detaching logic. * Ignore empty arguments. * Ensure the SystemD services are up to date. * fix: Pass back real result for single task chains When chains are delayed, they are first frozen as part of preparation which causes the sub-tasks to also be frozen. Afterward, the final (0th since we reverse the tasks/result order when freezing) result object from the freezing process would be passed back to the caller. This caused problems in signaling completion of groups contained in chains because the group relies on a promise which is fulfilled by a barrier linked to each of its applied subtasks. By constructing two `GroupResult` objects (one during freezing, one when the chain sub-tasks are applied), this resulted in there being two promises; only one of which would actually be fulfilled by the group subtasks. This change ensures that in the special case where a chain has a single task, we pass back the result object constructed when the task was actually applied. When that single child is a group which does not get unrolled (ie. contains more than one child itself), this ensures that we pass back a `GroupResult` object which will actually be fulfilled. The caller can then await the result confidently! * fix: Retain `group_id` when tasks get re-frozen When a group task which is part of a chain was to be delayed by `trace_task()`, it would be reconstructed from the serialized request. Normally, this sets the `group_id` of encapsulated tasks to the ID of the group being instantiated. However, in the specific situation of a group that is the last task in a chain which contributes to the completion of a chord, it is essential that the group ID of the top-most group is used instead. This top-most group ID is used by the redis backend to track the completions of "final elements" of a chord in the `on_chord_part_return()` implementation. By overwriting the group ID which was already set in the `options` dictionaries of the child tasks being deserialized, the chord accounting done by the redis backend would be made inaccurate and chords would never complete. This change alters how options are overridden for signatures to ensure that if a `group_id` has already been set, it cannot be overridden. Since group ID should be generally opaque to users, this should not be disruptive. * fix: Count chord "final elements" correctly This change amends the implementation of `chord.__length_hint__()` to ensure that all child task types are correctly counted. Specifically: * all sub-tasks of a group are counted recursively * the final task of a chain is counted recursively * the body of a chord is counted recursively * all other simple signatures count as a single "final element" There is also a deserialisation step if a `dict` is seen while counting the final elements in a chord, however this should become less important with the merge of celery#6342 which ensures that tasks are recursively deserialized by `.from_dict()`. * test: Add more integration tests for groups These tests are intended to show that group unrolling should be respected in various ways by all backends. They should make it more clear what behaviour we should be expecting from nested canvas components and ensure that all the implementations (mostly relevant to chords and `on_chord_part_return()` code) behave sensibly. * test: Fix old markings for chord tests * fix: Make KV-store backends respect chord size This avoids an issue where the `on_chord_part_return()` implementation would check the the length of the result of a chain ending in a nested group. This would manifest in behaviour where a worker would be blocked waiting for for the result object it holds to complete since it would attempt to `.join()` the result object. In situations with plenty of workers, this wouldn't really cause any noticable issue apart from some latency or unpredictable failures - but in concurrency constrained situations like the integrations tests, it causes deadlocks. We know from previous commits in this series that chord completion is more complex than just waiting for a direct child, so we correct the `size` value in `BaseKeyValueStoreBackend.on_chord_part_return()` to respect the `chord_size` value from the request, falling back to the length of the `deps` if that value is missing for some reason (this is necessary to keep a number of the tests happy but it's not clear to me if that will ever be the case in real life situations). * fix: Retain chord header result structure in Redis This change fixes the chord result flattening issue which manifested when using the Redis backend due to its deliberate throwing away of information about the header result structure. Rather than assuming that all results which contribute to the finalisation of a chord should be siblings, this change checks if any are complex (ie. `GroupResult`s) and falls back to behaviour similar to that implemented in the `KeyValueStoreBackend` which restores the original `GroupResult` object and `join()`s it. We retain the original behaviour which is billed as an optimisation in f09b041. We could behave better in the complex header result case by not bothering to stash the results of contributing tasks under the `.j` zset since we won't be using them, but without checking for the presence of the complex group result on every `on_chord_part_return()` call, we can't be sure that we won't need those stashed results later on. This would be an opportunity for optimisation in future if we were to use an `EVAL` to only do the `zadd()` if the group result key doesn't exist. However, avoiding the result encoding work in `on_chord_part_return()` would be more complicated. For now, it's not worth the brainpower. This change also slightly refactors the redis backend unit tests to make it easier to build fixtures and hit both the complex and simple result structure cases. * Update obsolete --loglevel argument values in docs * Set logfile, not loglevel. * Mention removed deprecated modules in the release notes. Fixes celery#6406. * Copy __annotations__ when creating tasks This will allow getting type hints. Fixes celery#6186. * test: Improve chord body group index freezing test Add more elements to the body so we can verify that the `group_index` counts up from 0 as expected. This change adds the `pytest-subtests` package as a test dependency so we can define partially independent subtests within test functions. * test: Use all() for subtask checks in canvas tests When we expect all of the tasks in some iterable to meet a conditional, we should make that clear by using `all(condition for ...)`. * test: Add more tests for `from_dict()` variants Notably, this exposed the bug tracked in celery#6341 where groups are not deeply deserialized by `group.from_dict()`. * fix: Ensure group tasks are deeply deserialised Fixes celery#6341 * Fix `celery shell` command * predefined_queues_urls -> predefined_queues * Update changelog. * Bump version: 5.0.0 → 5.0.1 * [Fix celery#6361] Fixing documentation for RabbitMQ task_queue_ha_policy * Fix _autodiscover_tasks_from_fixups function * fixup! Fix _autodiscover_tasks_from_fixups function * Correct configuration item: CELERY_RESULT_EXPIRES Related issue: celery#4050 celery#4050 (comment) * Flush worker prints, notably the banner In some cases (kubernetes, root) the banner is only printed at the end of the process execution, instead of at the beginning. * [Fix celery#6361] Remove RabbitMQ ha_policy from queue * ci: Fix TOXENV for pypy3 unit tests Fixes celery#6409 * ci: Move Python 3.9 test base from dev to release * docs: fix celery beat settings * move to travis-ci.com * fix: Ensure default fairness maps to `SCHED_FAIR` (celery#6447) Fixes celery#6386 * Preserve callbacks when replacing a task with a chain (celery#6189) * Preserve callbacks when replacing a task with a chain. * Preserve callbacks when replacing a task with a chain. * Added tests. * Update celery/app/task.py Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * Mark test as flaky. * Fix race condition in CI. * fix: Run linked tasks in original slot for replace This change alters the handling of linked tasks for chains which are used as the argument to a `.replace()` call for a task which itself has a chain of signatures to call once it completes. We ensure that the linked callback is not only retained but also called at the appropiate point in the newly reconstructed chain comprised of tasks from both the replacement chain and the tail of the encapsulating chain of the task being replaced. We amend some tests to validate this behaviour better and ensure that call/errbacks behave as expected if the encapsulating chain has either set. One test is marked with an `xfail` since errbacks of encapsulating chains are not currently called as expected due to some ambiguity in when an errback of a replaced task should be dropped or not (celery#6441). Co-authored-by: Asif Saif Uddin <auvipy@gmail.com> Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * Fix minor documentation omission (celery#6453) Co-authored-by: Lewis Kabui <lewisemm@users.noreply.github.com> * Fix max_retries override on self.retry (celery#6436) * Fix max_retries override * Fix max_retries override * Fix max_retries override * Update exceptions.py typo * Update autoretry.py typo * Update task.py Prevent exception unpacking for tasks without autoretry_for * Update test_tasks.py Unit test * Update test_tasks.py Added a new test * Update autoretry.py Fox for explicit raise in tasks * Update test_tasks.py * Update autoretry.py * Update task.py * Update exceptions.py * Update task.py * Happify linter. * Raise proper error when replacing with an empty chain. (celery#6452) Fixes celery#6451. * Update changelog. * Bump version: 5.0.1 → 5.0.2 * Update daemonizing.rst Improved systemd documentation for auto-start of the service, and mention the possibility to depend on RabbitMQ service. Also add Restart=always for Celery Beat example * Update celerybeat.service * Fix old celery beat variables Change made 5 days ago in 7c3da03 is faulty, the correct celery beat variables do start with `CELERYBEAT` and not `CELERY_BEAT` * Fix formatting. * Fix formatting. * fix: Make `--workdir` eager for early handling (celery#6457) This change makes the `--workdir` options an eager one which `click` will process early for us, before any of the others. At the same time, we add a callback which ensures that the `chdir()` is run during handling of the argument so that all subsequent actions (e.g. app loading) occur in the specified working directory. Fixes celery#6445 * Fix example. Fixes celery#6459. * When using the MongoDB backend, don't cleanup if result_expires is 0 or None. (celery#6462) Fixes celery#6450. * Add missing space (celery#6468) * Fix passing queues into purge command (celery#6469) In current wersion calling `celery --app my.celery_app purge -Q queue_name` is failing with following trace: ``` names = (queues or set(app.amqp.queues.keys())) - exclude_queues TypeError: unsupported operand type(s) for -: 'list' and 'list' ``` Becouse code is expecting set and `queues` is actually a list. Here is a fix. * Change donations sidebar to direct users to OpenCollective. * Added pytest to extras. Missed in 9a6c292. * Restore app.start() and app.worker_main() (celery#6481) * Restore `app.start()` and `app.worker_main()`. * Update celery/app/base.py Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * Fix spelling error. Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * fix: `node_format()` logfile before detaching Fixes celery#6426 * Multithreaded backend (celery#6416) * Cache backend to thread local storage instead of global variable * Cache oid to thread local storage instead of global variable * Improve code returning thread_local data * Move thread local storage to Celery class, introduced thread_oid and added unittests * Remove python2 compatibility code * Restore ability to extend the CLI with new sub-commands. * Adjust documentation to demonstrate how to introduce sub-command plugins in 5.x. Fixes celery#6439. * autopep8 & isort. * Linters now run using Python 3.9. * Fix apply_async() in Calling Tasks userguide * Fix dead links in contributing guide (celery#6506) * Fix inconsistency in documentation for `link_error` (celery#6505) * Make documentation of link_error consistent Fixes celery#4099 * Fix undefined variable in example * Add to contributors list * Update testing.rst (celery#6507) Use double back ticks for some code examples, so that quotes don't get converted into smart-quotes. celery#6497 * Don't upgrade click to 8.x since click-repl doesn't support it yet. Fixes celery#6511. Upstream issue: click-contrib/click-repl#72 * Update documentation on changes to custom CLI options in 5.0. Fixes celery#6380. * update step to install homebrew * redis: Support Sentinel with SSL Use the SentinelManagedSSLConnection when SSL is enabled for the transport. The redis-py project doesn't have a connection class for SSL+Sentinel yet. So, create a class in redis.py to add that functionality. * Revert "redis: Support Sentinel with SSL" (celery#6518) This reverts commit 18a0963. * Reintroduce support for custom preload options (celery#6516) * Restore preload options. Fixes celery#6307. * Document breaking changes for preload options in 5.0. Fixes celery#6379. * Changelog for 5.0.3. * Bump version: 5.0.2 → 5.0.3 * Added integration tests for calling a task (celery#6523) * DummyClient of cache+memory:// backend now shares state between threads (celery#6524) * isort. * Update changelog. * Bump version: 5.0.3 → 5.0.4 * Change deprecated from collections import Mapping/MutableMapping to from collections.abc ... (celery#6532) * fix celery#6047 * Fix type error in S3 backend (celery#6537) * Convert key from bytes to str * Add unit test for S3 delete of key with type bytes * events.py: Remove duplicate decorator in wrong place (celery#6543) `@handle_preload_options` was specified twice as a decorator of `events`, once at the top (wrong) and once at the bottom (right). This fixes the `celery events` commands and also `celery --help` * Update changelog. * Bump version: 5.0.4 → 5.0.5 * ADD: indico additions - trails * FIX: remove dev.txt dependencies * ADD: handle revoke failures * ADD: trailer_request support and better drain resolution * ADD: merge options was overriding link_error values * PATCH: DLX and reject behaviour * FIX: amqp dependencies Co-authored-by: Omer Katz <omer.drow@gmail.com> Co-authored-by: Thomas Grainger <tagrain@gmail.com> Co-authored-by: Martin Paulus <mpaulus@lequest.com> Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> Co-authored-by: Ash Berlin-Taylor <ash_github@firemirror.com> Co-authored-by: qiaocc <jasonqiao36@gmail.com> Co-authored-by: Christian Clauss <cclauss@me.com> Co-authored-by: Weiliang Li <to.be.impressive@gmail.com> Co-authored-by: Akash Agrawal <akashrocksha@gmail.com> Co-authored-by: Michal Kuffa <michal.kuffa@sentry.io> Co-authored-by: Frazer McLean <frazer@frazermclean.co.uk> Co-authored-by: Maarten Fonville <mfonville@users.noreply.github.com> Co-authored-by: laixintao <laixintaoo@gmail.com> Co-authored-by: Nicolas Dandrimont <nicolas@dandrimont.eu> Co-authored-by: Bas ten Berge <bas+github@tenberge-ict.nl> Co-authored-by: Zvi Baratz <z.baratz@gmail.com> Co-authored-by: Justinas Petuchovas <justinas.petuchovas@gmail.com> Co-authored-by: bastb <bas@tenberge-ict.nl> Co-authored-by: Artem Bernatskyi <artem.bernatskyy@gmail.com> Co-authored-by: ZubAnt <ya.zubarevanton@yandex.ru> Co-authored-by: Lewis Kabui <lewisemm@users.noreply.github.com> Co-authored-by: David Pärsson <david@parsson.se> Co-authored-by: Anthony Lukach <anthonylukach@gmail.com> Co-authored-by: Safwan Rahman <safwan.rahman15@gmail.com> Co-authored-by: Stepan Henek <stepan@henek.name> Co-authored-by: KexZh <hanbaobao2005@gmail.com> Co-authored-by: Thomas Riccardi <thomas@deepomatic.com> Co-authored-by: Egor Sergeevich Poderiagin <egor@crazyrussian.pro> Co-authored-by: Asif Saif Uddin (Auvi) <auvipy@gmail.com> Co-authored-by: Lewis M. Kabui <13940255+lewisemm@users.noreply.github.com> Co-authored-by: Ixiodor <Ixiodor@users.noreply.github.com> Co-authored-by: Mathieu Rollet <matletix@gmail.com> Co-authored-by: Mike DePalatis <depalatis@gmail.com> Co-authored-by: partizan <serg.partizan@gmail.com> Co-authored-by: Nick Pope <nick.pope@flightdataservices.com> Co-authored-by: Matus Valo <matusvalo@users.noreply.github.com> Co-authored-by: Matus Valo <matusvalo@gmail.com> Co-authored-by: henribru <6639509+henribru@users.noreply.github.com> Co-authored-by: Stuart Axon <stuaxo2@yahoo.com> Co-authored-by: Sonya Chhabra <sonyakc.2007@gmail.com> Co-authored-by: AbdealiJK <abdealikothari@gmail.com> Co-authored-by: František Zatloukal <Zatloukal.Frantisek@gmail.com> Co-authored-by: elonzh <elonzh@qq.com> Co-authored-by: Sven Koitka <razorx89@users.noreply.github.com> Co-authored-by: Arnon Yaari <wiggin15@yahoo.com>
This was referenced Feb 16, 2021
sihrc
added a commit
to IndicoDataSolutions/celery
that referenced
this pull request
Feb 17, 2021
* Remove defaults for unsupported Python runtimes. * Remove obsolete test. * Doc pytest plugin (celery#6289) * update to new pytest name * doc pytest plugin * trim heading to the length of the new pytest name * add warning against use of sort key on dynamodb table, closes celery#6332 * Remove celery.five and bump vine dep (celery#6338) * improv: Replace `five.values` with `dict.values` * improv: Use `time.monotonic()` in kombu tests Also in the docs where it is used to demonstrate `memcache` timeouts. * rm: Delete `celery.five` `vine.five` is no longer present in `vine >= 5`. * triv: Remove refs to `celery.five` in docs, &c * build: Bump `vine` dependency to 5.0+ * Wheels are no longer universal. * Remove failing before_install step. * Update changelog. * Bump version: 5.0.0rc2 → 5.0.0rc3 * Fix release date. * Remove unused import. * Correctly skip these tests when the relevant dependency is missing. * Expose retry_policy for Redis result backend Rather than adding a new top-level config option, I have used a new key in the already existing setting `result_backend_transport_options`. Closes celery#6166 * Update changelog for 4.3.1. * fix typo (celery#6346) * Travis CI: Test Python 3.9 release candidate 1 (celery#6328) * Travis CI: Test Python 3.9 release candidate 1 * fixup! Travis CI: matrix --> jobs * fixup! Fix indentation error * fixup! tox.ini: 3.9 --> 3.9-dev * Fix test failure in Python 3.9RC1. Co-authored-by: Omer Katz <omer.drow@gmail.com> * Fix the broken celery upgrade settings command. * Fix celery migrate settings options. * Remove Riak result backend settings. * Rephrase to mention that 3.5 is also EOL. * Add a note about the removal of the Riak result backend. * Fix examples of starting a worker in comments (celery#6331) * Remove deprecated function from app.log.Logging. * Migration guide. * Document breaking changes for the CLI in the Whats New document. * Add a "port code to Python 3" migration step. * Update supported Python versions in the introduction document. * Update bash completion. * Add note about new shell completion. * Update daemonization docs. * Remove amqp backend. (celery#6360) Fixes celery#6356. * Warn when deprecated settings are used (celery#6353) * Warn when deprecated settings are used. * Mention deprecation in docs. * Refer to the right place in the documentation. * Complete What's New. * Add wall of contributors. * Update codename. * Fix alt text. * isort. * PyPy 3.7 is currently in alpha. No need for that sentence. * Mention the new pytest-celery plugin. * Mention retry policy for the redis result backend. * Fix phrasing. * Mention ordered group results are now the default. * pyupgrade. * Complete release notes. * Bump version: 5.0.0rc3 → 5.0.0 * Happify linters. * Specify utf-8 as the encoding for log files. Fixes celery#5144. * Fixed some typos in readme * Fix custom headers propagation for protocol 1 hybrid messages * Retry after race during schema creation in database backend (celery#6298) * Retry after race during schema creation in database backend Fixes celery#6296 This race condition does not commonly present, since the schema creation only needs to happen once per database. It's more likely to appear in e.g. a test suite that uses a new database each time. For context of the sleep times I chose, the schema creation takes ~50 ms on my laptop. I did a simulated test run of 50 concurrent calls to MetaData.create_all repeated 200 times and the number of retries was: - 0 retries: 8717x - 1 retry: 1279x - 2 retries 4x * Add test for prepare_models retry error condition * Add name to contributors * Update daemonizing.rst Fix daemonizing documentation for issue celery#6363 to put `multi` before `-A` * Revert "Update daemonizing.rst" (celery#6376) This reverts commit 96ec6db. * bugfix: when set config result_expires = 0, chord.get will hang. (celery#6373) * bugfix: when set config result_expires = 0, chord.get will hang. `EXPIRE key 0` will delete a key in redis, then chord will never get the result. fix: celery#5237 * test: add testcase for expire when set config with zero. * Display a custom error message whenever an attempt to use -A or --app as a sub-command option was made. Fixes celery#6363 * Remove test dependencies for Python 2.7. * Restore the celery worker --without-{gossip,mingle,heartbeat} flags (celery#6365) In the previously used argparse arguments framework, these three options were used as flags. Since 5.0.0, they are options which need to take an argument (whose only sensible value would be "true"). The error message coming up is also (very) hard to understand, when running the celery worker command with an odd number of flags: Error: Unable to parse extra configuration from command line. Reason: not enough values to unpack (expected 2, got 1) When the celery worker is run with an even number of flags, the last one is considered as an argument of the previous one, which is a subtle bug. * Provide clearer error messages when app fails to load. * fix pytest plugin registration documentation (celery#6387) * fix pytest plugin registration documentation * Update docs/userguide/testing.rst Co-authored-by: Thomas Grainger <tagrain@gmail.com> Co-authored-by: Omer Katz <omer.drow@gmail.com> * Contains a workaround for the capitalized configuration issue (celery#6385) * Contains a workaround for the capitalized configuration issue * Update celery/apps/worker.py Co-authored-by: Omer Katz <omer.drow@gmail.com> * Update celery/apps/worker.py Co-authored-by: Omer Katz <omer.drow@gmail.com> Co-authored-by: Omer Katz <omer.drow@gmail.com> * Remove old explanation regarding `absolute_import` (celery#6390) Resolves celery#6389. * Update canvas.rst (celery#6392) * Update canvas.rst Tiny fixes. * Update docs/userguide/canvas.rst Co-authored-by: Omer Katz <omer.drow@gmail.com> Co-authored-by: Omer Katz <omer.drow@gmail.com> * Remove duplicate words from docs (celery#6398) Remove the duplicate usage of “required” in documentation (specifically, `introduction.rst`). * Allow lowercase log levels. (celery#6396) Fixes celery#6395. * Detach now correctly passes options with more than one word. (celery#6394) When specifying options such as `-E` the detached worker should receive the `--task-events` option. Instead it got the `--task_events` option which doesn't exist and therefore silently failed. This fixes celery#6362. * The celery multi command now works as expected. (celery#6388) * Contains the missed change requested by @thedrow * Added a some celery configuration examples. * fixed loglevel info->INFO in docs * return list instead set in CommaSeparatedList _broadcast method of kombu Mailbox. does not support set https://github.com/celery/kombu/blob/7b2578b19ba4b1989b722f6f6e7efee2a1a4d86a/kombu/pidbox.py#L319 * Rewrite detaching logic (celery#6401) * Rewrite detaching logic. * Ignore empty arguments. * Ensure the SystemD services are up to date. * fix: Pass back real result for single task chains When chains are delayed, they are first frozen as part of preparation which causes the sub-tasks to also be frozen. Afterward, the final (0th since we reverse the tasks/result order when freezing) result object from the freezing process would be passed back to the caller. This caused problems in signaling completion of groups contained in chains because the group relies on a promise which is fulfilled by a barrier linked to each of its applied subtasks. By constructing two `GroupResult` objects (one during freezing, one when the chain sub-tasks are applied), this resulted in there being two promises; only one of which would actually be fulfilled by the group subtasks. This change ensures that in the special case where a chain has a single task, we pass back the result object constructed when the task was actually applied. When that single child is a group which does not get unrolled (ie. contains more than one child itself), this ensures that we pass back a `GroupResult` object which will actually be fulfilled. The caller can then await the result confidently! * fix: Retain `group_id` when tasks get re-frozen When a group task which is part of a chain was to be delayed by `trace_task()`, it would be reconstructed from the serialized request. Normally, this sets the `group_id` of encapsulated tasks to the ID of the group being instantiated. However, in the specific situation of a group that is the last task in a chain which contributes to the completion of a chord, it is essential that the group ID of the top-most group is used instead. This top-most group ID is used by the redis backend to track the completions of "final elements" of a chord in the `on_chord_part_return()` implementation. By overwriting the group ID which was already set in the `options` dictionaries of the child tasks being deserialized, the chord accounting done by the redis backend would be made inaccurate and chords would never complete. This change alters how options are overridden for signatures to ensure that if a `group_id` has already been set, it cannot be overridden. Since group ID should be generally opaque to users, this should not be disruptive. * fix: Count chord "final elements" correctly This change amends the implementation of `chord.__length_hint__()` to ensure that all child task types are correctly counted. Specifically: * all sub-tasks of a group are counted recursively * the final task of a chain is counted recursively * the body of a chord is counted recursively * all other simple signatures count as a single "final element" There is also a deserialisation step if a `dict` is seen while counting the final elements in a chord, however this should become less important with the merge of celery#6342 which ensures that tasks are recursively deserialized by `.from_dict()`. * test: Add more integration tests for groups These tests are intended to show that group unrolling should be respected in various ways by all backends. They should make it more clear what behaviour we should be expecting from nested canvas components and ensure that all the implementations (mostly relevant to chords and `on_chord_part_return()` code) behave sensibly. * test: Fix old markings for chord tests * fix: Make KV-store backends respect chord size This avoids an issue where the `on_chord_part_return()` implementation would check the the length of the result of a chain ending in a nested group. This would manifest in behaviour where a worker would be blocked waiting for for the result object it holds to complete since it would attempt to `.join()` the result object. In situations with plenty of workers, this wouldn't really cause any noticable issue apart from some latency or unpredictable failures - but in concurrency constrained situations like the integrations tests, it causes deadlocks. We know from previous commits in this series that chord completion is more complex than just waiting for a direct child, so we correct the `size` value in `BaseKeyValueStoreBackend.on_chord_part_return()` to respect the `chord_size` value from the request, falling back to the length of the `deps` if that value is missing for some reason (this is necessary to keep a number of the tests happy but it's not clear to me if that will ever be the case in real life situations). * fix: Retain chord header result structure in Redis This change fixes the chord result flattening issue which manifested when using the Redis backend due to its deliberate throwing away of information about the header result structure. Rather than assuming that all results which contribute to the finalisation of a chord should be siblings, this change checks if any are complex (ie. `GroupResult`s) and falls back to behaviour similar to that implemented in the `KeyValueStoreBackend` which restores the original `GroupResult` object and `join()`s it. We retain the original behaviour which is billed as an optimisation in f09b041. We could behave better in the complex header result case by not bothering to stash the results of contributing tasks under the `.j` zset since we won't be using them, but without checking for the presence of the complex group result on every `on_chord_part_return()` call, we can't be sure that we won't need those stashed results later on. This would be an opportunity for optimisation in future if we were to use an `EVAL` to only do the `zadd()` if the group result key doesn't exist. However, avoiding the result encoding work in `on_chord_part_return()` would be more complicated. For now, it's not worth the brainpower. This change also slightly refactors the redis backend unit tests to make it easier to build fixtures and hit both the complex and simple result structure cases. * Update obsolete --loglevel argument values in docs * Set logfile, not loglevel. * Mention removed deprecated modules in the release notes. Fixes celery#6406. * Copy __annotations__ when creating tasks This will allow getting type hints. Fixes celery#6186. * test: Improve chord body group index freezing test Add more elements to the body so we can verify that the `group_index` counts up from 0 as expected. This change adds the `pytest-subtests` package as a test dependency so we can define partially independent subtests within test functions. * test: Use all() for subtask checks in canvas tests When we expect all of the tasks in some iterable to meet a conditional, we should make that clear by using `all(condition for ...)`. * test: Add more tests for `from_dict()` variants Notably, this exposed the bug tracked in celery#6341 where groups are not deeply deserialized by `group.from_dict()`. * fix: Ensure group tasks are deeply deserialised Fixes celery#6341 * Fix `celery shell` command * predefined_queues_urls -> predefined_queues * Update changelog. * Bump version: 5.0.0 → 5.0.1 * [Fix celery#6361] Fixing documentation for RabbitMQ task_queue_ha_policy * Fix _autodiscover_tasks_from_fixups function * fixup! Fix _autodiscover_tasks_from_fixups function * Correct configuration item: CELERY_RESULT_EXPIRES Related issue: celery#4050 celery#4050 (comment) * Flush worker prints, notably the banner In some cases (kubernetes, root) the banner is only printed at the end of the process execution, instead of at the beginning. * [Fix celery#6361] Remove RabbitMQ ha_policy from queue * ci: Fix TOXENV for pypy3 unit tests Fixes celery#6409 * ci: Move Python 3.9 test base from dev to release * docs: fix celery beat settings * move to travis-ci.com * fix: Ensure default fairness maps to `SCHED_FAIR` (celery#6447) Fixes celery#6386 * Preserve callbacks when replacing a task with a chain (celery#6189) * Preserve callbacks when replacing a task with a chain. * Preserve callbacks when replacing a task with a chain. * Added tests. * Update celery/app/task.py Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * Mark test as flaky. * Fix race condition in CI. * fix: Run linked tasks in original slot for replace This change alters the handling of linked tasks for chains which are used as the argument to a `.replace()` call for a task which itself has a chain of signatures to call once it completes. We ensure that the linked callback is not only retained but also called at the appropiate point in the newly reconstructed chain comprised of tasks from both the replacement chain and the tail of the encapsulating chain of the task being replaced. We amend some tests to validate this behaviour better and ensure that call/errbacks behave as expected if the encapsulating chain has either set. One test is marked with an `xfail` since errbacks of encapsulating chains are not currently called as expected due to some ambiguity in when an errback of a replaced task should be dropped or not (celery#6441). Co-authored-by: Asif Saif Uddin <auvipy@gmail.com> Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * Fix minor documentation omission (celery#6453) Co-authored-by: Lewis Kabui <lewisemm@users.noreply.github.com> * Fix max_retries override on self.retry (celery#6436) * Fix max_retries override * Fix max_retries override * Fix max_retries override * Update exceptions.py typo * Update autoretry.py typo * Update task.py Prevent exception unpacking for tasks without autoretry_for * Update test_tasks.py Unit test * Update test_tasks.py Added a new test * Update autoretry.py Fox for explicit raise in tasks * Update test_tasks.py * Update autoretry.py * Update task.py * Update exceptions.py * Update task.py * Happify linter. * Raise proper error when replacing with an empty chain. (celery#6452) Fixes celery#6451. * Update changelog. * Bump version: 5.0.1 → 5.0.2 * Update daemonizing.rst Improved systemd documentation for auto-start of the service, and mention the possibility to depend on RabbitMQ service. Also add Restart=always for Celery Beat example * Update celerybeat.service * Fix old celery beat variables Change made 5 days ago in 7c3da03 is faulty, the correct celery beat variables do start with `CELERYBEAT` and not `CELERY_BEAT` * Fix formatting. * Fix formatting. * fix: Make `--workdir` eager for early handling (celery#6457) This change makes the `--workdir` options an eager one which `click` will process early for us, before any of the others. At the same time, we add a callback which ensures that the `chdir()` is run during handling of the argument so that all subsequent actions (e.g. app loading) occur in the specified working directory. Fixes celery#6445 * Fix example. Fixes celery#6459. * When using the MongoDB backend, don't cleanup if result_expires is 0 or None. (celery#6462) Fixes celery#6450. * Add missing space (celery#6468) * Fix passing queues into purge command (celery#6469) In current wersion calling `celery --app my.celery_app purge -Q queue_name` is failing with following trace: ``` names = (queues or set(app.amqp.queues.keys())) - exclude_queues TypeError: unsupported operand type(s) for -: 'list' and 'list' ``` Becouse code is expecting set and `queues` is actually a list. Here is a fix. * Change donations sidebar to direct users to OpenCollective. * Added pytest to extras. Missed in 9a6c292. * Restore app.start() and app.worker_main() (celery#6481) * Restore `app.start()` and `app.worker_main()`. * Update celery/app/base.py Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * Fix spelling error. Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * fix: `node_format()` logfile before detaching Fixes celery#6426 * Multithreaded backend (celery#6416) * Cache backend to thread local storage instead of global variable * Cache oid to thread local storage instead of global variable * Improve code returning thread_local data * Move thread local storage to Celery class, introduced thread_oid and added unittests * Remove python2 compatibility code * Restore ability to extend the CLI with new sub-commands. * Adjust documentation to demonstrate how to introduce sub-command plugins in 5.x. Fixes celery#6439. * autopep8 & isort. * Linters now run using Python 3.9. * Fix apply_async() in Calling Tasks userguide * Fix dead links in contributing guide (celery#6506) * Fix inconsistency in documentation for `link_error` (celery#6505) * Make documentation of link_error consistent Fixes celery#4099 * Fix undefined variable in example * Add to contributors list * Update testing.rst (celery#6507) Use double back ticks for some code examples, so that quotes don't get converted into smart-quotes. celery#6497 * Don't upgrade click to 8.x since click-repl doesn't support it yet. Fixes celery#6511. Upstream issue: click-contrib/click-repl#72 * Update documentation on changes to custom CLI options in 5.0. Fixes celery#6380. * update step to install homebrew * redis: Support Sentinel with SSL Use the SentinelManagedSSLConnection when SSL is enabled for the transport. The redis-py project doesn't have a connection class for SSL+Sentinel yet. So, create a class in redis.py to add that functionality. * Revert "redis: Support Sentinel with SSL" (celery#6518) This reverts commit 18a0963. * Reintroduce support for custom preload options (celery#6516) * Restore preload options. Fixes celery#6307. * Document breaking changes for preload options in 5.0. Fixes celery#6379. * Changelog for 5.0.3. * Bump version: 5.0.2 → 5.0.3 * Added integration tests for calling a task (celery#6523) * DummyClient of cache+memory:// backend now shares state between threads (celery#6524) * isort. * Update changelog. * Bump version: 5.0.3 → 5.0.4 * Change deprecated from collections import Mapping/MutableMapping to from collections.abc ... (celery#6532) * fix celery#6047 * Fix type error in S3 backend (celery#6537) * Convert key from bytes to str * Add unit test for S3 delete of key with type bytes * events.py: Remove duplicate decorator in wrong place (celery#6543) `@handle_preload_options` was specified twice as a decorator of `events`, once at the top (wrong) and once at the bottom (right). This fixes the `celery events` commands and also `celery --help` * Update changelog. * Bump version: 5.0.4 → 5.0.5 * ADD: indico additions - trails * FIX: remove dev.txt dependencies * ADD: handle revoke failures * ADD: trailer_request support and better drain resolution * ADD: merge options was overriding link_error values * PATCH: DLX and reject behaviour * FIX: amqp dependencies Co-authored-by: Omer Katz <omer.drow@gmail.com> Co-authored-by: Thomas Grainger <tagrain@gmail.com> Co-authored-by: Martin Paulus <mpaulus@lequest.com> Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> Co-authored-by: Ash Berlin-Taylor <ash_github@firemirror.com> Co-authored-by: qiaocc <jasonqiao36@gmail.com> Co-authored-by: Christian Clauss <cclauss@me.com> Co-authored-by: Weiliang Li <to.be.impressive@gmail.com> Co-authored-by: Akash Agrawal <akashrocksha@gmail.com> Co-authored-by: Michal Kuffa <michal.kuffa@sentry.io> Co-authored-by: Frazer McLean <frazer@frazermclean.co.uk> Co-authored-by: Maarten Fonville <mfonville@users.noreply.github.com> Co-authored-by: laixintao <laixintaoo@gmail.com> Co-authored-by: Nicolas Dandrimont <nicolas@dandrimont.eu> Co-authored-by: Bas ten Berge <bas+github@tenberge-ict.nl> Co-authored-by: Zvi Baratz <z.baratz@gmail.com> Co-authored-by: Justinas Petuchovas <justinas.petuchovas@gmail.com> Co-authored-by: bastb <bas@tenberge-ict.nl> Co-authored-by: Artem Bernatskyi <artem.bernatskyy@gmail.com> Co-authored-by: ZubAnt <ya.zubarevanton@yandex.ru> Co-authored-by: Lewis Kabui <lewisemm@users.noreply.github.com> Co-authored-by: David Pärsson <david@parsson.se> Co-authored-by: Anthony Lukach <anthonylukach@gmail.com> Co-authored-by: Safwan Rahman <safwan.rahman15@gmail.com> Co-authored-by: Stepan Henek <stepan@henek.name> Co-authored-by: KexZh <hanbaobao2005@gmail.com> Co-authored-by: Thomas Riccardi <thomas@deepomatic.com> Co-authored-by: Egor Sergeevich Poderiagin <egor@crazyrussian.pro> Co-authored-by: Asif Saif Uddin (Auvi) <auvipy@gmail.com> Co-authored-by: Lewis M. Kabui <13940255+lewisemm@users.noreply.github.com> Co-authored-by: Ixiodor <Ixiodor@users.noreply.github.com> Co-authored-by: Mathieu Rollet <matletix@gmail.com> Co-authored-by: Mike DePalatis <depalatis@gmail.com> Co-authored-by: partizan <serg.partizan@gmail.com> Co-authored-by: Nick Pope <nick.pope@flightdataservices.com> Co-authored-by: Matus Valo <matusvalo@users.noreply.github.com> Co-authored-by: Matus Valo <matusvalo@gmail.com> Co-authored-by: henribru <6639509+henribru@users.noreply.github.com> Co-authored-by: Stuart Axon <stuaxo2@yahoo.com> Co-authored-by: Sonya Chhabra <sonyakc.2007@gmail.com> Co-authored-by: AbdealiJK <abdealikothari@gmail.com> Co-authored-by: František Zatloukal <Zatloukal.Frantisek@gmail.com> Co-authored-by: elonzh <elonzh@qq.com> Co-authored-by: Sven Koitka <razorx89@users.noreply.github.com> Co-authored-by: Arnon Yaari <wiggin15@yahoo.com>
This was referenced Feb 18, 2021
Closed
sihrc
added a commit
to IndicoDataSolutions/celery
that referenced
this pull request
Feb 23, 2021
* Remove defaults for unsupported Python runtimes. * Remove obsolete test. * Doc pytest plugin (celery#6289) * update to new pytest name * doc pytest plugin * trim heading to the length of the new pytest name * add warning against use of sort key on dynamodb table, closes celery#6332 * Remove celery.five and bump vine dep (celery#6338) * improv: Replace `five.values` with `dict.values` * improv: Use `time.monotonic()` in kombu tests Also in the docs where it is used to demonstrate `memcache` timeouts. * rm: Delete `celery.five` `vine.five` is no longer present in `vine >= 5`. * triv: Remove refs to `celery.five` in docs, &c * build: Bump `vine` dependency to 5.0+ * Wheels are no longer universal. * Remove failing before_install step. * Update changelog. * Bump version: 5.0.0rc2 → 5.0.0rc3 * Fix release date. * Remove unused import. * Correctly skip these tests when the relevant dependency is missing. * Expose retry_policy for Redis result backend Rather than adding a new top-level config option, I have used a new key in the already existing setting `result_backend_transport_options`. Closes celery#6166 * Update changelog for 4.3.1. * fix typo (celery#6346) * Travis CI: Test Python 3.9 release candidate 1 (celery#6328) * Travis CI: Test Python 3.9 release candidate 1 * fixup! Travis CI: matrix --> jobs * fixup! Fix indentation error * fixup! tox.ini: 3.9 --> 3.9-dev * Fix test failure in Python 3.9RC1. Co-authored-by: Omer Katz <omer.drow@gmail.com> * Fix the broken celery upgrade settings command. * Fix celery migrate settings options. * Remove Riak result backend settings. * Rephrase to mention that 3.5 is also EOL. * Add a note about the removal of the Riak result backend. * Fix examples of starting a worker in comments (celery#6331) * Remove deprecated function from app.log.Logging. * Migration guide. * Document breaking changes for the CLI in the Whats New document. * Add a "port code to Python 3" migration step. * Update supported Python versions in the introduction document. * Update bash completion. * Add note about new shell completion. * Update daemonization docs. * Remove amqp backend. (celery#6360) Fixes celery#6356. * Warn when deprecated settings are used (celery#6353) * Warn when deprecated settings are used. * Mention deprecation in docs. * Refer to the right place in the documentation. * Complete What's New. * Add wall of contributors. * Update codename. * Fix alt text. * isort. * PyPy 3.7 is currently in alpha. No need for that sentence. * Mention the new pytest-celery plugin. * Mention retry policy for the redis result backend. * Fix phrasing. * Mention ordered group results are now the default. * pyupgrade. * Complete release notes. * Bump version: 5.0.0rc3 → 5.0.0 * Happify linters. * Specify utf-8 as the encoding for log files. Fixes celery#5144. * Fixed some typos in readme * Fix custom headers propagation for protocol 1 hybrid messages * Retry after race during schema creation in database backend (celery#6298) * Retry after race during schema creation in database backend Fixes celery#6296 This race condition does not commonly present, since the schema creation only needs to happen once per database. It's more likely to appear in e.g. a test suite that uses a new database each time. For context of the sleep times I chose, the schema creation takes ~50 ms on my laptop. I did a simulated test run of 50 concurrent calls to MetaData.create_all repeated 200 times and the number of retries was: - 0 retries: 8717x - 1 retry: 1279x - 2 retries 4x * Add test for prepare_models retry error condition * Add name to contributors * Update daemonizing.rst Fix daemonizing documentation for issue celery#6363 to put `multi` before `-A` * Revert "Update daemonizing.rst" (celery#6376) This reverts commit 96ec6db. * bugfix: when set config result_expires = 0, chord.get will hang. (celery#6373) * bugfix: when set config result_expires = 0, chord.get will hang. `EXPIRE key 0` will delete a key in redis, then chord will never get the result. fix: celery#5237 * test: add testcase for expire when set config with zero. * Display a custom error message whenever an attempt to use -A or --app as a sub-command option was made. Fixes celery#6363 * Remove test dependencies for Python 2.7. * Restore the celery worker --without-{gossip,mingle,heartbeat} flags (celery#6365) In the previously used argparse arguments framework, these three options were used as flags. Since 5.0.0, they are options which need to take an argument (whose only sensible value would be "true"). The error message coming up is also (very) hard to understand, when running the celery worker command with an odd number of flags: Error: Unable to parse extra configuration from command line. Reason: not enough values to unpack (expected 2, got 1) When the celery worker is run with an even number of flags, the last one is considered as an argument of the previous one, which is a subtle bug. * Provide clearer error messages when app fails to load. * fix pytest plugin registration documentation (celery#6387) * fix pytest plugin registration documentation * Update docs/userguide/testing.rst Co-authored-by: Thomas Grainger <tagrain@gmail.com> Co-authored-by: Omer Katz <omer.drow@gmail.com> * Contains a workaround for the capitalized configuration issue (celery#6385) * Contains a workaround for the capitalized configuration issue * Update celery/apps/worker.py Co-authored-by: Omer Katz <omer.drow@gmail.com> * Update celery/apps/worker.py Co-authored-by: Omer Katz <omer.drow@gmail.com> Co-authored-by: Omer Katz <omer.drow@gmail.com> * Remove old explanation regarding `absolute_import` (celery#6390) Resolves celery#6389. * Update canvas.rst (celery#6392) * Update canvas.rst Tiny fixes. * Update docs/userguide/canvas.rst Co-authored-by: Omer Katz <omer.drow@gmail.com> Co-authored-by: Omer Katz <omer.drow@gmail.com> * Remove duplicate words from docs (celery#6398) Remove the duplicate usage of “required” in documentation (specifically, `introduction.rst`). * Allow lowercase log levels. (celery#6396) Fixes celery#6395. * Detach now correctly passes options with more than one word. (celery#6394) When specifying options such as `-E` the detached worker should receive the `--task-events` option. Instead it got the `--task_events` option which doesn't exist and therefore silently failed. This fixes celery#6362. * The celery multi command now works as expected. (celery#6388) * Contains the missed change requested by @thedrow * Added a some celery configuration examples. * fixed loglevel info->INFO in docs * return list instead set in CommaSeparatedList _broadcast method of kombu Mailbox. does not support set https://github.com/celery/kombu/blob/7b2578b19ba4b1989b722f6f6e7efee2a1a4d86a/kombu/pidbox.py#L319 * Rewrite detaching logic (celery#6401) * Rewrite detaching logic. * Ignore empty arguments. * Ensure the SystemD services are up to date. * fix: Pass back real result for single task chains When chains are delayed, they are first frozen as part of preparation which causes the sub-tasks to also be frozen. Afterward, the final (0th since we reverse the tasks/result order when freezing) result object from the freezing process would be passed back to the caller. This caused problems in signaling completion of groups contained in chains because the group relies on a promise which is fulfilled by a barrier linked to each of its applied subtasks. By constructing two `GroupResult` objects (one during freezing, one when the chain sub-tasks are applied), this resulted in there being two promises; only one of which would actually be fulfilled by the group subtasks. This change ensures that in the special case where a chain has a single task, we pass back the result object constructed when the task was actually applied. When that single child is a group which does not get unrolled (ie. contains more than one child itself), this ensures that we pass back a `GroupResult` object which will actually be fulfilled. The caller can then await the result confidently! * fix: Retain `group_id` when tasks get re-frozen When a group task which is part of a chain was to be delayed by `trace_task()`, it would be reconstructed from the serialized request. Normally, this sets the `group_id` of encapsulated tasks to the ID of the group being instantiated. However, in the specific situation of a group that is the last task in a chain which contributes to the completion of a chord, it is essential that the group ID of the top-most group is used instead. This top-most group ID is used by the redis backend to track the completions of "final elements" of a chord in the `on_chord_part_return()` implementation. By overwriting the group ID which was already set in the `options` dictionaries of the child tasks being deserialized, the chord accounting done by the redis backend would be made inaccurate and chords would never complete. This change alters how options are overridden for signatures to ensure that if a `group_id` has already been set, it cannot be overridden. Since group ID should be generally opaque to users, this should not be disruptive. * fix: Count chord "final elements" correctly This change amends the implementation of `chord.__length_hint__()` to ensure that all child task types are correctly counted. Specifically: * all sub-tasks of a group are counted recursively * the final task of a chain is counted recursively * the body of a chord is counted recursively * all other simple signatures count as a single "final element" There is also a deserialisation step if a `dict` is seen while counting the final elements in a chord, however this should become less important with the merge of celery#6342 which ensures that tasks are recursively deserialized by `.from_dict()`. * test: Add more integration tests for groups These tests are intended to show that group unrolling should be respected in various ways by all backends. They should make it more clear what behaviour we should be expecting from nested canvas components and ensure that all the implementations (mostly relevant to chords and `on_chord_part_return()` code) behave sensibly. * test: Fix old markings for chord tests * fix: Make KV-store backends respect chord size This avoids an issue where the `on_chord_part_return()` implementation would check the the length of the result of a chain ending in a nested group. This would manifest in behaviour where a worker would be blocked waiting for for the result object it holds to complete since it would attempt to `.join()` the result object. In situations with plenty of workers, this wouldn't really cause any noticable issue apart from some latency or unpredictable failures - but in concurrency constrained situations like the integrations tests, it causes deadlocks. We know from previous commits in this series that chord completion is more complex than just waiting for a direct child, so we correct the `size` value in `BaseKeyValueStoreBackend.on_chord_part_return()` to respect the `chord_size` value from the request, falling back to the length of the `deps` if that value is missing for some reason (this is necessary to keep a number of the tests happy but it's not clear to me if that will ever be the case in real life situations). * fix: Retain chord header result structure in Redis This change fixes the chord result flattening issue which manifested when using the Redis backend due to its deliberate throwing away of information about the header result structure. Rather than assuming that all results which contribute to the finalisation of a chord should be siblings, this change checks if any are complex (ie. `GroupResult`s) and falls back to behaviour similar to that implemented in the `KeyValueStoreBackend` which restores the original `GroupResult` object and `join()`s it. We retain the original behaviour which is billed as an optimisation in f09b041. We could behave better in the complex header result case by not bothering to stash the results of contributing tasks under the `.j` zset since we won't be using them, but without checking for the presence of the complex group result on every `on_chord_part_return()` call, we can't be sure that we won't need those stashed results later on. This would be an opportunity for optimisation in future if we were to use an `EVAL` to only do the `zadd()` if the group result key doesn't exist. However, avoiding the result encoding work in `on_chord_part_return()` would be more complicated. For now, it's not worth the brainpower. This change also slightly refactors the redis backend unit tests to make it easier to build fixtures and hit both the complex and simple result structure cases. * Update obsolete --loglevel argument values in docs * Set logfile, not loglevel. * Mention removed deprecated modules in the release notes. Fixes celery#6406. * Copy __annotations__ when creating tasks This will allow getting type hints. Fixes celery#6186. * test: Improve chord body group index freezing test Add more elements to the body so we can verify that the `group_index` counts up from 0 as expected. This change adds the `pytest-subtests` package as a test dependency so we can define partially independent subtests within test functions. * test: Use all() for subtask checks in canvas tests When we expect all of the tasks in some iterable to meet a conditional, we should make that clear by using `all(condition for ...)`. * test: Add more tests for `from_dict()` variants Notably, this exposed the bug tracked in celery#6341 where groups are not deeply deserialized by `group.from_dict()`. * fix: Ensure group tasks are deeply deserialised Fixes celery#6341 * Fix `celery shell` command * predefined_queues_urls -> predefined_queues * Update changelog. * Bump version: 5.0.0 → 5.0.1 * [Fix celery#6361] Fixing documentation for RabbitMQ task_queue_ha_policy * Fix _autodiscover_tasks_from_fixups function * fixup! Fix _autodiscover_tasks_from_fixups function * Correct configuration item: CELERY_RESULT_EXPIRES Related issue: celery#4050 celery#4050 (comment) * Flush worker prints, notably the banner In some cases (kubernetes, root) the banner is only printed at the end of the process execution, instead of at the beginning. * [Fix celery#6361] Remove RabbitMQ ha_policy from queue * ci: Fix TOXENV for pypy3 unit tests Fixes celery#6409 * ci: Move Python 3.9 test base from dev to release * docs: fix celery beat settings * move to travis-ci.com * fix: Ensure default fairness maps to `SCHED_FAIR` (celery#6447) Fixes celery#6386 * Preserve callbacks when replacing a task with a chain (celery#6189) * Preserve callbacks when replacing a task with a chain. * Preserve callbacks when replacing a task with a chain. * Added tests. * Update celery/app/task.py Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * Mark test as flaky. * Fix race condition in CI. * fix: Run linked tasks in original slot for replace This change alters the handling of linked tasks for chains which are used as the argument to a `.replace()` call for a task which itself has a chain of signatures to call once it completes. We ensure that the linked callback is not only retained but also called at the appropiate point in the newly reconstructed chain comprised of tasks from both the replacement chain and the tail of the encapsulating chain of the task being replaced. We amend some tests to validate this behaviour better and ensure that call/errbacks behave as expected if the encapsulating chain has either set. One test is marked with an `xfail` since errbacks of encapsulating chains are not currently called as expected due to some ambiguity in when an errback of a replaced task should be dropped or not (celery#6441). Co-authored-by: Asif Saif Uddin <auvipy@gmail.com> Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * Fix minor documentation omission (celery#6453) Co-authored-by: Lewis Kabui <lewisemm@users.noreply.github.com> * Fix max_retries override on self.retry (celery#6436) * Fix max_retries override * Fix max_retries override * Fix max_retries override * Update exceptions.py typo * Update autoretry.py typo * Update task.py Prevent exception unpacking for tasks without autoretry_for * Update test_tasks.py Unit test * Update test_tasks.py Added a new test * Update autoretry.py Fox for explicit raise in tasks * Update test_tasks.py * Update autoretry.py * Update task.py * Update exceptions.py * Update task.py * Happify linter. * Raise proper error when replacing with an empty chain. (celery#6452) Fixes celery#6451. * Update changelog. * Bump version: 5.0.1 → 5.0.2 * Update daemonizing.rst Improved systemd documentation for auto-start of the service, and mention the possibility to depend on RabbitMQ service. Also add Restart=always for Celery Beat example * Update celerybeat.service * Fix old celery beat variables Change made 5 days ago in 7c3da03 is faulty, the correct celery beat variables do start with `CELERYBEAT` and not `CELERY_BEAT` * Fix formatting. * Fix formatting. * fix: Make `--workdir` eager for early handling (celery#6457) This change makes the `--workdir` options an eager one which `click` will process early for us, before any of the others. At the same time, we add a callback which ensures that the `chdir()` is run during handling of the argument so that all subsequent actions (e.g. app loading) occur in the specified working directory. Fixes celery#6445 * Fix example. Fixes celery#6459. * When using the MongoDB backend, don't cleanup if result_expires is 0 or None. (celery#6462) Fixes celery#6450. * Add missing space (celery#6468) * Fix passing queues into purge command (celery#6469) In current wersion calling `celery --app my.celery_app purge -Q queue_name` is failing with following trace: ``` names = (queues or set(app.amqp.queues.keys())) - exclude_queues TypeError: unsupported operand type(s) for -: 'list' and 'list' ``` Becouse code is expecting set and `queues` is actually a list. Here is a fix. * Change donations sidebar to direct users to OpenCollective. * Added pytest to extras. Missed in 9a6c292. * Restore app.start() and app.worker_main() (celery#6481) * Restore `app.start()` and `app.worker_main()`. * Update celery/app/base.py Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * Fix spelling error. Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> * fix: `node_format()` logfile before detaching Fixes celery#6426 * Multithreaded backend (celery#6416) * Cache backend to thread local storage instead of global variable * Cache oid to thread local storage instead of global variable * Improve code returning thread_local data * Move thread local storage to Celery class, introduced thread_oid and added unittests * Remove python2 compatibility code * Restore ability to extend the CLI with new sub-commands. * Adjust documentation to demonstrate how to introduce sub-command plugins in 5.x. Fixes celery#6439. * autopep8 & isort. * Linters now run using Python 3.9. * Fix apply_async() in Calling Tasks userguide * Fix dead links in contributing guide (celery#6506) * Fix inconsistency in documentation for `link_error` (celery#6505) * Make documentation of link_error consistent Fixes celery#4099 * Fix undefined variable in example * Add to contributors list * Update testing.rst (celery#6507) Use double back ticks for some code examples, so that quotes don't get converted into smart-quotes. celery#6497 * Don't upgrade click to 8.x since click-repl doesn't support it yet. Fixes celery#6511. Upstream issue: click-contrib/click-repl#72 * Update documentation on changes to custom CLI options in 5.0. Fixes celery#6380. * update step to install homebrew * redis: Support Sentinel with SSL Use the SentinelManagedSSLConnection when SSL is enabled for the transport. The redis-py project doesn't have a connection class for SSL+Sentinel yet. So, create a class in redis.py to add that functionality. * Revert "redis: Support Sentinel with SSL" (celery#6518) This reverts commit 18a0963. * Reintroduce support for custom preload options (celery#6516) * Restore preload options. Fixes celery#6307. * Document breaking changes for preload options in 5.0. Fixes celery#6379. * Changelog for 5.0.3. * Bump version: 5.0.2 → 5.0.3 * Added integration tests for calling a task (celery#6523) * DummyClient of cache+memory:// backend now shares state between threads (celery#6524) * isort. * Update changelog. * Bump version: 5.0.3 → 5.0.4 * Change deprecated from collections import Mapping/MutableMapping to from collections.abc ... (celery#6532) * fix celery#6047 * Fix type error in S3 backend (celery#6537) * Convert key from bytes to str * Add unit test for S3 delete of key with type bytes * events.py: Remove duplicate decorator in wrong place (celery#6543) `@handle_preload_options` was specified twice as a decorator of `events`, once at the top (wrong) and once at the bottom (right). This fixes the `celery events` commands and also `celery --help` * Update changelog. * Bump version: 5.0.4 → 5.0.5 * ADD: indico additions - trails * FIX: remove dev.txt dependencies * ADD: handle revoke failures * ADD: trailer_request support and better drain resolution * ADD: merge options was overriding link_error values * PATCH: DLX and reject behaviour * FIX: amqp dependencies Co-authored-by: Omer Katz <omer.drow@gmail.com> Co-authored-by: Thomas Grainger <tagrain@gmail.com> Co-authored-by: Martin Paulus <mpaulus@lequest.com> Co-authored-by: maybe-sybr <58414429+maybe-sybr@users.noreply.github.com> Co-authored-by: Ash Berlin-Taylor <ash_github@firemirror.com> Co-authored-by: qiaocc <jasonqiao36@gmail.com> Co-authored-by: Christian Clauss <cclauss@me.com> Co-authored-by: Weiliang Li <to.be.impressive@gmail.com> Co-authored-by: Akash Agrawal <akashrocksha@gmail.com> Co-authored-by: Michal Kuffa <michal.kuffa@sentry.io> Co-authored-by: Frazer McLean <frazer@frazermclean.co.uk> Co-authored-by: Maarten Fonville <mfonville@users.noreply.github.com> Co-authored-by: laixintao <laixintaoo@gmail.com> Co-authored-by: Nicolas Dandrimont <nicolas@dandrimont.eu> Co-authored-by: Bas ten Berge <bas+github@tenberge-ict.nl> Co-authored-by: Zvi Baratz <z.baratz@gmail.com> Co-authored-by: Justinas Petuchovas <justinas.petuchovas@gmail.com> Co-authored-by: bastb <bas@tenberge-ict.nl> Co-authored-by: Artem Bernatskyi <artem.bernatskyy@gmail.com> Co-authored-by: ZubAnt <ya.zubarevanton@yandex.ru> Co-authored-by: Lewis Kabui <lewisemm@users.noreply.github.com> Co-authored-by: David Pärsson <david@parsson.se> Co-authored-by: Anthony Lukach <anthonylukach@gmail.com> Co-authored-by: Safwan Rahman <safwan.rahman15@gmail.com> Co-authored-by: Stepan Henek <stepan@henek.name> Co-authored-by: KexZh <hanbaobao2005@gmail.com> Co-authored-by: Thomas Riccardi <thomas@deepomatic.com> Co-authored-by: Egor Sergeevich Poderiagin <egor@crazyrussian.pro> Co-authored-by: Asif Saif Uddin (Auvi) <auvipy@gmail.com> Co-authored-by: Lewis M. Kabui <13940255+lewisemm@users.noreply.github.com> Co-authored-by: Ixiodor <Ixiodor@users.noreply.github.com> Co-authored-by: Mathieu Rollet <matletix@gmail.com> Co-authored-by: Mike DePalatis <depalatis@gmail.com> Co-authored-by: partizan <serg.partizan@gmail.com> Co-authored-by: Nick Pope <nick.pope@flightdataservices.com> Co-authored-by: Matus Valo <matusvalo@users.noreply.github.com> Co-authored-by: Matus Valo <matusvalo@gmail.com> Co-authored-by: henribru <6639509+henribru@users.noreply.github.com> Co-authored-by: Stuart Axon <stuaxo2@yahoo.com> Co-authored-by: Sonya Chhabra <sonyakc.2007@gmail.com> Co-authored-by: AbdealiJK <abdealikothari@gmail.com> Co-authored-by: František Zatloukal <Zatloukal.Frantisek@gmail.com> Co-authored-by: elonzh <elonzh@qq.com> Co-authored-by: Sven Koitka <razorx89@users.noreply.github.com> Co-authored-by: Arnon Yaari <wiggin15@yahoo.com>
Contributor
|
Note: although this merge request makes Lines 292 to 296 in d9d8250 Lines 322 to 331 in d9d8250 Lines 356 to 358 in d9d8250 are invoked at worker startup and capture backend and some bound methods ( Lines 448 to 452 in d9d8250 So, probably such functions should be re-written to use |
Member
are you up for that? |
Contributor
|
Yes, I'll check this code more thoroughly and try to create merge request in some days. |
vznncv
added a commit
to vznncv/celery
that referenced
this pull request
Jul 11, 2021
Add simple test with embedded worker that checks backend instance usage by threads. According merge request celery#6416 backends should be thread local.
vznncv
added a commit
to vznncv/celery
that referenced
this pull request
Jul 11, 2021
Add simple test with embedded worker that checks backend instance usage by threads. According merge request celery#6416 backends should be thread local.
vznncv
added a commit
to vznncv/celery
that referenced
this pull request
Jul 11, 2021
Fix backend capturing by closure during task creation in the function `celery.app.trace.build_tracer`, as different threads may create and use celery task. It complement changes in the pull request celery#6416.
vznncv
added a commit
to vznncv/celery
that referenced
this pull request
Jul 11, 2021
Fix backend capturing by closure during task creation in the function `celery.app.trace.build_tracer`, as different threads may create and use celery task. It complement changes in the pull request celery#6416.
vznncv
added a commit
to vznncv/celery
that referenced
this pull request
Jul 11, 2021
Fix backend capturing by closure during task creation in the function `celery.app.trace.build_tracer`, as different threads may create and use celery task. It complement changes in the pull request celery#6416.
auvipy
pushed a commit
that referenced
this pull request
Jul 11, 2021
* Add test of backend usage by threads Add simple test with embedded worker that checks backend instance usage by threads. According merge request #6416 backends should be thread local. * Fix backend captures in the `celery.app.trace.build_tracer` Fix backend capturing by closure during task creation in the function `celery.app.trace.build_tracer`, as different threads may create and use celery task. It complement changes in the pull request #6416. * Fix flake8 errors Fix flake8 errors from Celery/lint github workflow step
jeyrce
pushed a commit
to jeyrce/celery
that referenced
this pull request
Aug 25, 2021
* Cache backend to thread local storage instead of global variable * Cache oid to thread local storage instead of global variable * Improve code returning thread_local data * Move thread local storage to Celery class, introduced thread_oid and added unittests
jeyrce
pushed a commit
to jeyrce/celery
that referenced
this pull request
Aug 25, 2021
* Add test of backend usage by threads Add simple test with embedded worker that checks backend instance usage by threads. According merge request celery#6416 backends should be thread local. * Fix backend captures in the `celery.app.trace.build_tracer` Fix backend capturing by closure during task creation in the function `celery.app.trace.build_tracer`, as different threads may create and use celery task. It complement changes in the pull request celery#6416. * Fix flake8 errors Fix flake8 errors from Celery/lint github workflow step
18 tasks
4 tasks
fonhorst
pushed a commit
to aimclub/AutoTM
that referenced
this pull request
Jan 27, 2023
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
8 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
This PR contains two changes:
oid(Universally unique identifier) to be thread-based. Moreover, theoidis now different for each thread. This is causing that each thread has it's own reply-to queue (Before all threads shares one reply-to queue which was causing missing responses)This PR fixes #6414 and probably also issues linked to it.