[R-package] Temporary workaround for not modifying global OMP config

[david-cortes]

ref #4705

This is a temporary workaround to avoid modifying the global OpenMP user configuration for the whole R process when running lgb.train. This PR should be reverted once issue #4705 is fixed properly.

[jameslamb]

Thanks for your proposal, but I don't support introducing a new required dependency to the R package for this purpose.

Instead of this fix that benefits only the R package, and even then only one part of it (i.e. not other multi-threaded operations like prediction and Dataset construction), if you're interested in seeing LightGBM's behavior here change, please work with @guolinke and @shiyu1994 on #4705 to implement something at the C++ level that all users of LightGBM will benefit from.

[david-cortes]

In #4705, @guolinke suggested a workaround like this. It does BTW fix the issue for dataset construction when it happens inside lgb.train.

[jameslamb]

If you mean #4705 (comment)

A quick solution is to have a context manager. For example, when start running LightGBM, it records current omp nthreads, then when exiting LightGBM, it resets the omp nthreads by the previous recorded values.

I believe that was a recommendation for how LightGBM's C++ code would manage getting and setting the the number of threads, not a recommendation that every wrapper of that code do this itself.

[shiyu1994]

I agree that such changes should be made in C++ so that both R and Python packages can leverage the workaround. I can help to do that.

[jameslamb]

both R and Python packages

Thanks @shiyu1994 . For the benefit of everyone reading this PR, I just want to clarify...when I refer to wrappers of LightGBM's C++ code, I don't only mean the R and Python package in this repo. I also mean SynapseML (formerly MMLSpark), the Java interface generated with SWIG, LightGBM.NET, and the other projects listed at https://github.com/microsoft/LightGBM#external-unofficial-repositories.

Based on this discussion, I'm going to close this PR.

@david-cortes thanks very much for documenting #4705 and for helping this project try to address it!

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.