[WIP][Don't merge] Comment out dmlc::SetEnv in pthread_atfork #13438

[larroy]

Description

Fixes #13438

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• [] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
• [] Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
• Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
• Code is well-documented:
• For user-facing API changes, API doc string has been updated.
• For new C++ functions in header files, their functionalities and arguments are documented.
• For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
• Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
• To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Changes

• Feature1, tests, (and when applicable, API doc)
• Feature2, tests, (and when applicable, API doc)

Comments

• If this change is a backward incompatible change, why must this change be made.
• Interesting edge cases to note here

[vandanavk]

@mxnet-label-bot add [pr-work-in-progress]

@larroy Please add "Fixes #13438" in the PR description to automatically close the issue when the PR is merged

[larroy]

@samskalicky what do you think of this? I think we should run a few examples to see if the number of threads remains constant when using CPU threaded engine.

[larroy]

@vandanavk done. Thanks for the suggestion.

[samskalicky]

Pedro, it would be great if the fix is as simple as commenting out the SetEnv. But what about the corresponding GetEnv in:
https://github.com/apache/incubator-mxnet/blob/266de6bef4da5769431557288d41fab2a02e52ca/src/engine/threaded_engine_perdevice.cc#L79
Shouldnt we need a mechanism to set the number of threads here to 1?

[larroy]

Yes most likely the fix is too simplistic. Wanted to have a fist test run to see how it looks.

[roywei]

@mxnet-label-bot add[par-awaiting-testing]

[cjolivier01]

setting to 1 effectively disabled OMP (assuming OMP looks at the environment variable after startup). What happens if the forked process tries to use OMP? There should be a test for this because I know libgomp doesn't play well with forking sometimes.

[larroy]

Thanks for your review Chris. Can you elaborate and provide more detail on what do you suggest to do?

[Roshrini]

@larroy Can you address comments and rebase this PR? Thanks

[stu1130]

@mxnet-label-bot update [pr-awaiting-response]

[pinaraws]

@larroy Can this be merged now? Or should we close this PR?

[piyushghai]

@larroy Gentle ping.

[larroy]

I think we need to do more testing to be confident that the solution is solid. I can do it during my next oncall. The change might be trivial but checking for correctness in the multi-threaded scenario and with interactions from other libraries is costly in terms of developer time.

[pinaraws]

@larroy Gentle ping.