[NCCL] Timeout Loop Thread for Async Error Handling #41050
Closed
Conversation
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
Initial code for adding WorkNCCL objects to a vector when any NCCL Collective operation is called (similar to ProcessGroupGloo), and a timeout thread that busy-polls the vector of WorkNCCL objects and removes them upon completion. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
This was referenced Jul 7, 2020
💊 CI failures summary and remediationsAs of commit 8f07a92 (more details on the Dr. CI page):
🕵️ 1 new failure recognized by patternsThe following CI failures do not appear to be due to upstream breakages:
|
Initial code for adding WorkNCCL objects to a vector when any NCCL Collective operation is called (similar to ProcessGroupGloo), and a timeout thread that busy-polls the vector of WorkNCCL objects and removes them upon completion. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
Initial code for adding WorkNCCL objects to a vector when any NCCL Collective operation is called (similar to ProcessGroupGloo), and a timeout thread that busy-polls the vector of WorkNCCL objects and removes them upon completion. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
mrshenli
reviewed
Aug 19, 2020
Initial code for adding WorkNCCL objects to a vector when any NCCL Collective operation is called (similar to ProcessGroupGloo), and a timeout thread that busy-polls the vector of WorkNCCL objects and removes them upon completion. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
**This Commit:** We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion. **This Stack:** The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
**This Commit:** We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion. **This Stack:** The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
jiayisuse
reviewed
Aug 21, 2020
**This Commit:** We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion. **This Stack:** The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
**This Commit:** We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion. **This Stack:** The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
pritamdamania87
approved these changes
Sep 2, 2020
Contributor
pritamdamania87
left a comment
pritamdamania87
left a comment
There was a problem hiding this comment.
There seem to be some DDP tests failing, please check those before landing.
**This Commit:** We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion. **This Stack:** The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
**This Commit:** We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion. **This Stack:** The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
**This Commit:** We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion. **This Stack:** The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
**This Commit:** We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion. **This Stack:** The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! [ghstack-poisoned]
Contributor
|
This pull request has been merged in 1df24fd. |
loadbxh
pushed a commit
to loadbxh/Torch
that referenced
this pull request
Sep 23, 2020
Pull Request resolved: pytorch/pytorch#41050 **This Commit:** We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion. **This Stack:** The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic. Differential Revision: [D21916637](https://our.internmc.facebook.com/intern/diff/D21916637/) **NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D21916637/)! ghstack-source-id: 111301603
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Stack from ghstack:
This Commit:
We introduce a workVector to track live workNCCL objects corresponding to collective operations. Further, we introduce a workCleanupLoop, which busy-polls the vector of workNCCL objects and removes them upon completion.
This Stack:
The purpose of this stack is to fix the hanging behavior observed in when using PyTorch DDP training with NCCL. In various situations (desynchronization, high GPU utilization, etc.), NCCL collectives may hang due to waiting on an unresponsive worker. This stack detects such hanging behavior and aborts timed-out collectives by throwing a user-visible exception, all with minimal perf regression. Training can then be restarted from a previous checkpoint with something like torchelastic.
Differential Revision: D21916637
NOTE FOR REVIEWERS: This PR has internal Facebook specific changes or comments, please review them on Phabricator!