CI fix CircleCI failures by making it run sequencially #25809
Conversation
There was a problem hiding this comment.
For the CI, I think it's better to set SPHINX_NUMJOBS in https://github.com/scikit-learn/scikit-learn/blob/main/.circleci/config.yml itself. This way, the sphinx build is still parallel on local development environments.
|
I'd be curious if having it serial locally would also make it faster. For |
There was a problem hiding this comment.
We might be oversubscribing in a few places.
There is some oversubscribing, but it does speed things up on CircleCI. I recall benchmarking on CircleCI and -j2 was ~50% faster than -j1.
In any case, I think it's okay to make it one anyways. Parallelizing Sphinx has lead to issues in the past.
…t-learn#25809)" This reverts commit ac24ebb.
Investigate the CircleCI failure (i.e.
EOFErrorandOSError) that could be linked to some hardware limitation (i.e. memory).The first thing to try is to limit the number of jobs to build the documentation. This could limit memory consumption.