Closed stevetorr closed 1 year ago
Similar issue is observed again on Travis. https://travis-ci.org/github/mir-group/flare/builds/677271523 . It seems any line with np.zeros or np array without declaring dtype=np.float64 or dtype=np.int32 may hang the unit tests...
Below is the list of libraries and version numpy(1.15.4) scipy(1.4.1) memory_profiler(0.57.0) numba(0.49.0) ase(3.19.1) pymatgen(2020.4.2) [nptyping(1.0.1) psutil(5.7.0) llvmlite 0.33.0.dev0, setuptools(40.8.0) matplotlib (3.2.1) sympy(1.5.1) spglib (1.15.0) tabulate(0.8.7) ruamel.yaml (0.16.10) pydispatcher (2.0.5) pandas(1.0.3) networkx (2.4) requests(2.23.0) plotly (4.6.0) monty (3.0.2) palettable (3.3.0) typish (1.6.0) pyparsing (2.4.7) kiwisolver (1.2.0) cycler (0.10.0) python-dateutil (2.8.1) mpmath (1.1.0) ruamel.yaml.clib 0.2.0) pytz (2019.3) decorator (4.4.2) idna (2.9) urllib3 (1.25.9) certifi (2018.11.29) chardet (3.0.4) six(1.11.0) retrying (1.3.3)
Solution provided by the comments above
Depending on murky and uncertain Conda package configurations it appears that sometimes the multiprocessing module hangs effectively indefinitely.
Ultimately Lixin and I were able to isolate it down to my conda environment. Mine below was:
Hers was: