Codestin Search App

rtimpe · 2026-01-27T22:16:29Z

Stack from ghstack (oldest at bottom):

This is an alternative approach to #118387 using context vars

cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @kadeng @muchulee8 @amjames @chauhang @aakhundov @coconutruben @jataylo @Lucaskabela @azahed98

[ghstack-poisoned]

This is an alternative approach to #118387 using config vars ghstack-source-id: 43801c9 Pull-Request: #173568

pytorch-bot · 2026-01-27T22:16:36Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/173568

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 6e3c08b with merge base 4026771 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[ghstack-poisoned]

This is an alternative approach to #118387 using config vars ghstack-source-id: 5fe67b8 Pull-Request: #173568

[ghstack-poisoned]

This is an alternative approach to #118387 using context vars ghstack-source-id: 244c0d0 Pull-Request: #173568

[ghstack-poisoned]

This is an alternative approach to #118387 using context vars ghstack-source-id: 1e7fc7c Pull-Request: #173568

rtimpe · 2026-01-29T01:06:16Z




+speech_transformer,pass,16


I'm not sure why this is passing now, but I was getting CI failures before since the expected result was missing

rtimpe · 2026-01-29T01:10:22Z

        return cls._instance

    def submit(self, fn, *args, **kwargs):
+        ctx = contextvars.copy_context()


This is needed since the spawned threads need access to the config. If we want to avoid doing this manually, another option would be to only use contextvars in python 3.14 (or even 3.14t), where PYTHON_THREAD_INHERIT_CONTEXT is available

This seems like a good idea to me. Even with PYTHON_THREAD_INHERIT_CONTEXT, ThreadPoolExecutor tasks don't inherit the context from their submitter -- the threads inherit their context from the the thread that created the pool. (This is something we are considering changing for 3.15)

colesbury · 2026-01-29T19:32:23Z

This looks like a good approach to me, but I'm not very familiar with the code

ezyang · 2026-02-10T13:31:43Z

Whoever accepts and shepherds this, please note the last time we switched config to thread local it caused a SEV, so you're going to have to watch deployment very carefully.

rtimpe · 2026-02-10T19:05:37Z

@pytorchbot rebase

pytorchmergebot · 2026-02-10T19:07:51Z

@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here

[ghstack-poisoned]

pytorchmergebot · 2026-02-10T19:08:06Z

Successfully rebased gh/rtimpe/44/orig onto refs/remotes/origin/viable/strict, please pull locally before adding more changes (for example, via ghstack checkout https://github.com/pytorch/pytorch/pull/173568)

This is an alternative approach to #118387 using context vars ghstack-source-id: a190dd3 Pull-Request: #173568

This reverts commit c904bdc. Reverted pytorch#173568 on behalf of https://github.com/georgehong due to tests failing with torch._dynamo hit config.recompile_limit (8) ([comment](pytorch#173568 (comment)))

[ghstack-poisoned]

This is an alternative approach to #118387 using context vars ghstack-source-id: 4215140 Pull-Request: #173568

rtimpe · 2026-04-01T22:17:17Z

@pytorchbot merge

pytorchmergebot · 2026-04-01T22:19:30Z

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

@williamwen42

This was triggered by an internal test failure on #173568 (thanks to @williamwen42 for help debugging). See https://gist.github.com/williamwen42/a31bee4f3175c0f540d4170c2dd7ceae for more context/detail Authored with Claude. Differential Revision: [D99142314](https://our.internmc.facebook.com/intern/diff/D99142314) Pull Request resolved: #178950 Approved by: https://github.com/williamwen42

This is an alternative approach to pytorch#118387 using context vars Pull Request resolved: pytorch#173568 Approved by: https://github.com/williamwen42

This reverts commit c904bdc. Reverted #173568 on behalf of https://github.com/georgehong due to tests failing with torch._dynamo hit config.recompile_limit (8) ([comment](#173568 (comment)))

@williamwen42

This was triggered by an internal test failure on pytorch#173568 (thanks to @williamwen42 for help debugging). See https://gist.github.com/williamwen42/a31bee4f3175c0f540d4170c2dd7ceae for more context/detail Authored with Claude. Differential Revision: [D99142314](https://our.internmc.facebook.com/intern/diff/D99142314) Pull Request resolved: pytorch#178950 Approved by: https://github.com/williamwen42

This is an alternative approach to pytorch#118387 using context vars Pull Request resolved: pytorch#173568 Approved by: https://github.com/williamwen42 ghstack dependencies: pytorch#178950

This is an alternative approach to pytorch#118387 using context vars Pull Request resolved: pytorch#173568 Approved by: https://github.com/williamwen42

This reverts commit c904bdc. Reverted pytorch#173568 on behalf of https://github.com/georgehong due to tests failing with torch._dynamo hit config.recompile_limit (8) ([comment](pytorch#173568 (comment)))

@williamwen42

This was triggered by an internal test failure on pytorch#173568 (thanks to @williamwen42 for help debugging). See https://gist.github.com/williamwen42/a31bee4f3175c0f540d4170c2dd7ceae for more context/detail Authored with Claude. Differential Revision: [D99142314](https://our.internmc.facebook.com/intern/diff/D99142314) Pull Request resolved: pytorch#178950 Approved by: https://github.com/williamwen42

This is an alternative approach to pytorch#118387 using context vars Pull Request resolved: pytorch#173568 Approved by: https://github.com/williamwen42 ghstack dependencies: pytorch#178950

@williamwen42

This was triggered by an internal test failure on pytorch#173568 (thanks to @williamwen42 for help debugging). See https://gist.github.com/williamwen42/a31bee4f3175c0f540d4170c2dd7ceae for more context/detail Authored with Claude. Differential Revision: [D99142314](https://our.internmc.facebook.com/intern/diff/D99142314) Pull Request resolved: pytorch#178950 Approved by: https://github.com/williamwen42

This is an alternative approach to pytorch#118387 using context vars Pull Request resolved: pytorch#173568 Approved by: https://github.com/williamwen42 ghstack dependencies: pytorch#178950

@williamwen42

This was triggered by an internal test failure on pytorch#173568 (thanks to @williamwen42 for help debugging). See https://gist.github.com/williamwen42/a31bee4f3175c0f540d4170c2dd7ceae for more context/detail Authored with Claude. Differential Revision: [D99142314](https://our.internmc.facebook.com/intern/diff/D99142314) Pull Request resolved: pytorch#178950 Approved by: https://github.com/williamwen42

This is an alternative approach to pytorch#118387 using context vars Pull Request resolved: pytorch#173568 Approved by: https://github.com/williamwen42 ghstack dependencies: pytorch#178950

There's potential deadlock related to _stash_obj_in_tls. As of #173568 the context object is stored in the calling thread TLS during _engine_run_backward. Normally this object deallocated after the backward pass completes, but there's a race where the autograd device thread can end up holding the last reference. If this happens the context object gets deallocated by the device thread, which needs to acquire the GIL. If another thread happens to be holding the GIL while blocked on the device thread the result is a deadlock. See #173568 for more detail The fix here avoids aquiring the GIL in the device thread by clearing the GraphTask's shared_ptr to the context object during cleanup. Since the TLS in the calling thread also has a shared_ptr, this avoids calling the SafePyObject destructor (and therefore acquiring the GIL). I'm not a huge fan of this approach since it introduces a manual cleanup step, but I wasn't able to come up with something better. Another option would be to use SafePyHandle instead of SafePyObject and introduce a `_remove_obj_from_tls` python API. But that makes the python side responsible for refcounting, which feels wrong. I'm also not sure exactly how we end up with a thread holding the GIL and also blocked on an autograd device thread. That feels like a bug, but it seems to be triggered by Meta-internal code, so I'm not able to debug it. All of which is to say that this is the best approach I can come up with, but I'm very open to other suggestions. ghstack-source-id: 81c9703 Pull-Request: #180700

Update

812ffce

[ghstack-poisoned]

rtimpe added a commit that referenced this pull request Jan 27, 2026

Implement thread-local config with contextvars

bf6ba79

This is an alternative approach to #118387 using config vars ghstack-source-id: 43801c9 Pull-Request: #173568

pytorch-bot Bot added ciflow/inductor module: dynamo module: inductor labels Jan 27, 2026

pytorchbot added the open source label Jan 27, 2026

rtimpe added the topic: not user facing topic category label Jan 27, 2026

Update

63c20fa

[ghstack-poisoned]

github-actions Bot deleted a comment from pytorch-bot Bot Jan 27, 2026

rtimpe added a commit that referenced this pull request Jan 27, 2026

Implement thread-local config with contextvars

cb737e1

This is an alternative approach to #118387 using config vars ghstack-source-id: 5fe67b8 Pull-Request: #173568

rtimpe mentioned this pull request Jan 27, 2026

Make thread-local config layer #170468

Closed

Update

a73a16f

[ghstack-poisoned]

rtimpe added a commit that referenced this pull request Jan 28, 2026

Implement thread-local config with contextvars

a2af602

This is an alternative approach to #118387 using context vars ghstack-source-id: 244c0d0 Pull-Request: #173568

Update

290cb35

[ghstack-poisoned]

rtimpe added a commit that referenced this pull request Jan 28, 2026

Implement thread-local config with contextvars

a03fa3f

This is an alternative approach to #118387 using context vars ghstack-source-id: 1e7fc7c Pull-Request: #173568

rtimpe commented Jan 29, 2026

View reviewed changes

rtimpe marked this pull request as ready for review January 29, 2026 01:12

rtimpe requested review from colesbury, ezyang, guilhermeleobas and williamwen42 January 29, 2026 01:12

williamwen42 approved these changes Feb 10, 2026

View reviewed changes

Update

8286b71

[ghstack-poisoned]

pytorchmergebot pushed a commit that referenced this pull request Feb 10, 2026

Implement thread-local config with contextvars

874822f

This is an alternative approach to #118387 using context vars ghstack-source-id: a190dd3 Pull-Request: #173568

Update

6e3c08b

[ghstack-poisoned]

rtimpe added a commit that referenced this pull request Apr 1, 2026

Implement thread-local config with contextvars

cb78044

This is an alternative approach to #118387 using context vars ghstack-source-id: 4215140 Pull-Request: #173568

rtimpe mentioned this pull request Apr 1, 2026

Fix use after free in ensureCUDADeviceGuardSet #178950

Closed

rtimpe added a commit that referenced this pull request Apr 1, 2026

Implement thread-local config with contextvars

7644a62

This is an alternative approach to #118387 using context vars ghstack-source-id: 4215140 Pull-Request: #173568

rtimpe added a commit that referenced this pull request Apr 1, 2026

Implement thread-local config with contextvars

7f440a3

This is an alternative approach to #118387 using context vars ghstack-source-id: 4215140 Pull-Request: #173568

pytorchmergebot added the merging label Apr 1, 2026

pytorchmergebot closed this in f6ef243 Apr 1, 2026

pytorchmergebot removed the merging label Apr 1, 2026

rtimpe mentioned this pull request Apr 17, 2026

Fix deadlock related to _stash_obj_in_tls #180700

Closed

rtimpe mentioned this pull request Apr 17, 2026

Fix deadlock related to _stash_obj_in_tls #180701

Closed

williamwen42 mentioned this pull request May 1, 2026

_remove_obj_from_tls to fix deadlock involving _stash_obj_in_tls (#182083) #182083

Closed

dxqb mentioned this pull request Jun 7, 2026

torch 2.12 regression: torch.compile thread-local config - even backward thread has separate config #186537

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Implement thread-local config with contextvars#173568

Implement thread-local config with contextvars#173568
rtimpe wants to merge 12 commits into
gh/rtimpe/44/basefrom
gh/rtimpe/44/head

rtimpe commented Jan 27, 2026 •

edited by pytorch-bot Bot

Loading

Uh oh!

pytorch-bot Bot commented Jan 27, 2026 •

edited

Loading

Uh oh!

rtimpe Jan 29, 2026

Uh oh!

rtimpe Jan 29, 2026

Uh oh!

colesbury Jan 29, 2026

Uh oh!

colesbury commented Jan 29, 2026

Uh oh!

ezyang commented Feb 10, 2026

Uh oh!

rtimpe commented Feb 10, 2026

Uh oh!

pytorchmergebot commented Feb 10, 2026

Uh oh!

pytorchmergebot commented Feb 10, 2026

Uh oh!

rtimpe commented Apr 1, 2026

Uh oh!

pytorchmergebot commented Apr 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants




		speech_transformer,pass,16

Uh oh!

Conversation

rtimpe commented Jan 27, 2026 • edited by pytorch-bot Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot Bot commented Jan 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/173568

✅ No Failures

Uh oh!

rtimpe Jan 29, 2026

Choose a reason for hiding this comment

Uh oh!

rtimpe Jan 29, 2026

Choose a reason for hiding this comment

Uh oh!

colesbury Jan 29, 2026

Choose a reason for hiding this comment

Uh oh!

colesbury commented Jan 29, 2026

Uh oh!

ezyang commented Feb 10, 2026

Uh oh!

rtimpe commented Feb 10, 2026

Uh oh!

pytorchmergebot commented Feb 10, 2026

Uh oh!

pytorchmergebot commented Feb 10, 2026

Uh oh!

rtimpe commented Apr 1, 2026

Uh oh!

pytorchmergebot commented Apr 1, 2026

Merge started

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants

rtimpe commented Jan 27, 2026 •

edited by pytorch-bot Bot

Loading

pytorch-bot Bot commented Jan 27, 2026 •

edited

Loading