HoloClean / holoclean

A Machine Learning System for Data Enrichment.
http://www.holoclean.io
Apache License 2.0
515 stars 131 forks source link

PyTorch multiprocessing pool causes OSError Too Many Files Open on Linux #8

Closed minafarid closed 5 years ago

minafarid commented 5 years ago

Currently, using torch.multiprocessing in InitAttFeaturizer fails on linux and not on macOS. The reported error is OSError Too Many Files Open. Similar issues have been reported on PyTorch github repo:

Documentation for sharing strategies:

From the documentation:

file_descriptor is the default strategy (except for macOS and OS X where it’s not supported).