DAOS-19058 pydaos: torch surface worker errors in parallel_list (#18414)#18488
Open
enakta wants to merge 1 commit into
Open
DAOS-19058 pydaos: torch surface worker errors in parallel_list (#18414)#18488enakta wants to merge 1 commit into
enakta wants to merge 1 commit into
Conversation
Worker processes spawned by _Dfs.parallel_list may raise exceptions that never reached the calling process. This results in indefinite hang during Dataset and IterableDataset construction with no surfaced error to the user. Replacing manual Process + Queue scheme and its queued/processed counter with a multiprocessing.Pool driven by imap_unordered. Pool re-raises worker exceptions in the parent when their results are consumed, so a worker error now propagates as a raised OSError instead of a deadlock, and the Pool context manager reaps all workers on any exit path. `concurrent.futures.ProcessPoolExecutor` would be even better but its initializer/initargs arguments are unavailable before Python 3.7, and the target runtime includes EL8.8 / Python 3.6. Signed-off-by: Denis Barakhtanov <dbarahtanov@enakta.com>
enakta
added
the
clean-cherry-pick
|
Ticket title is 'pytorch parallel_list does not surface worker process errors, causing silent hangs' |
Collaborator
|
Test stage Functional on EL 9 completed with status FAILURE. https://jenkins-3.daos.hpc.amslabs.hpecorp.net//job/daos-stack/job/daos/view/change-requests/job/PR-18488/1/execution/node/1085/log |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Worker processes spawned by _Dfs.parallel_list may raise exceptions that never reached the calling process. This results in indefinite hang during Dataset and IterableDataset construction with no surfaced error to the user.
Replacing manual Process + Queue scheme and its queued/processed counter with a multiprocessing.Pool driven by imap_unordered. Pool re-raises worker exceptions in the parent when their results are consumed, so a worker error now propagates as a raised OSError instead of a deadlock, and the Pool context manager reaps all workers on any exit path.
concurrent.futures.ProcessPoolExecutorwould be even better but its initializer/initargs arguments are unavailable before Python 3.7, and the target runtime includes EL8.8 / Python 3.6.Steps for the author:
After all prior steps are complete: