Description

I noticed a big memory hog when running a certain number of iterations of pyhf.infer.hypotest. This strange behaviour has been confirmed with a small code that repeatedly executes hypotest. A similar behavior has been noticed when switching backends.

Expected Behavior

Memory profile should be constant when using pyhf.infer.hypotest or switching backends.

Actual Behavior

The following behavior was observed by using the python memory_profile library :

• When running pyhf.infer.hypotest 10000 times in a row, memory consumption is constantly growing.
• Similarly, when switching backends with pyhf.set_backend() repeatedly, an anomalous memory consumption can be observed.

Steps to Reproduce

pyhf version 0.4.1 was used. In attachment, you can find the following files :

• memPyhf.py : a code that executes pyhf.infer.hypotest 10000 times in a row using a simple json definition
• mprofile_hypotest.dat : the memory profile of memPyhf.py (made with mprof run memPyhf.py)
• switchBackends.py : a code that switches backends a 1000 times in a row
• mprofile_switchBackends.dat : memory profile of switchBackends.py

The memory profiles can be reproduced by running mprof run memPyhf.py or mprof run switchBackends.py with the two python scripts.

Checklist

• Run git fetch to get the most up to date version of master
• Searched through existing Issues to confirm this is not a duplicate issue
• Filled out the Description, Expected Behavior, Actual Behavior, and Steps to Reproduce sections above or have edited/removed them in a way that fully describes the issue