[C10D] Report detected failures when emitting collective end events.#109739
Closed
kumpera wants to merge 16 commits intogh/kumpera/64/basefrom
Closed
[C10D] Report detected failures when emitting collective end events.#109739kumpera wants to merge 16 commits intogh/kumpera/64/basefrom
kumpera wants to merge 16 commits intogh/kumpera/64/basefrom
Conversation
We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/109739
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 1 Unrelated FailureAs of commit 152d8c0 with merge base c151163 ( NEW FAILURES - The following jobs have failed:
BROKEN TRUNK - The following job failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
added 2 commits
October 2, 2023 08:31
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
wconstab
reviewed
Oct 2, 2023
wconstab
reviewed
Oct 2, 2023
wconstab
reviewed
Oct 2, 2023
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
4a43294 to
c4601d7
Compare
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
Contributor
Author
|
There's one major issue left on this PR: it doesn't work with gloo. |
added 2 commits
October 5, 2023 10:42
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
fduwjj
reviewed
Oct 5, 2023
Contributor
fduwjj
left a comment
There was a problem hiding this comment.
How we ensure WatchDog does not abort the program and let us return error message?
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
This was referenced Oct 13, 2023
wconstab
reviewed
Oct 13, 2023
| evt.operation = c10d::opTypeToString(work.retrieveOpType()); | ||
| evt.drop_count = 0; | ||
| // isCompleted is mutable :facepalm: | ||
| if (const_cast<Work&>(work).isCompleted() && !work.isSuccess()) |
Contributor
There was a problem hiding this comment.
@fduwjj I think we should discuss this part
we're calling work.exception() which will query some cuda APIs.
we're doing this from the main watchdog thread.
i think that in today's watchdog design this is already something we're doing in the main loop, so i don't have a big reservation about it.
but if you're redesigning the loops to put cuda accesses on another thread, you'd also want to move this part. would be good to think through how we'll handle the callback reporting error message safely, or else maybe not add the error to the callback in the first place. (I think it should be possible to do though)
…nd events." We leverage the fact that errors get tracked in Work to report them. The error message by itself is not crazy useful and we might need to include some form of error code in the mix (nccl failure, timeout, etc). [ghstack-poisoned]
Contributor
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
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
4 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.
Stack from ghstack (oldest at bottom):
We leverage the fact that errors get tracked in Work to report them.
The error message by itself is not crazy useful and we might need to
include some form of error code in the mix (nccl failure, timeout, etc).