When passing big arrays to joblib.Parallel in docker container, parallel processing does not start immediately.
Joblib uses the folder specified by "JOBLIB_TEMP_FOLDER" for memmap of the arrays. Without specifying, JOBLIB_TEMP_FOLDER is set to /dev/shm
, which usually has small size and is not enough for the big arrays.
https://pythonhosted.org/joblib/generated/joblib.Parallel.html
- Set JOBLIB_TEMP_FOLDER