Joblib silently ignores pickle exceptions and does not use cache

[p-himik]

I'm trying to use memory.cache on a function that returns a bunch of Pandas named tuples made with list(dataframe.itertuples()).
Apparently, such tuples cannot be pickled. I'm fine with that, however I'm not fine with the fact that Joblib doesn't even log the issue. Because of the empty except at https://github.com/joblib/joblib/blob/master/joblib/_store_backends.py#L193 the pickle exception is just ignored, and the hint about race condition doesn't really help to debug the issue.

[tomMoral]

Indeed, the except to catch the directory creation race condition should probably be moved in the create_location method. I also run into some race conditions when calling in parallel a cached function that has just been modified. The workers are all trying to delete the directory at once, creating errors. I will try to fix this this week.

[tomMoral]

Btw, if you have a reproducing case, I would be very interested, to make sure we solve the problem at hand.

[p-himik]

Should be something as simple as

import joblib as jb
import pandas as pd

memory = jb.memory.Memory()

@memory.cache
def f():
  return list(pd.DataFrame({'a': [1]}).itertuples())

f()

[tomMoral]

Note that if you do not provide an argument location= in Memory, there will be no caching in joblib.

But indeed, this example reproduce the problem.

[p-himik]

Right, sorry - I was just reproducing it from my head. In the real code, I have location specified.

[jeremiedbb]

Fixed by #1359