type_inference_provider does not properly clean up child processes on timeout

[lisroach]

If the subprocess call to in type_inference_provider times out the child processes do not get immediately cleaned up:

ERROR:1:1: X002 raise TimeoutExpired(
ERROR:1:1: X002 subprocess.TimeoutExpired: Command 'pyre --noninteractive query "types(path='python/fixit/flake8/test.py')"' timed out after 0 seconds
ERROR:1:1: X002 .../lib/python3.8/subprocess.py:942: ResourceWarning: subprocess 881287 is still running
ERROR:1:1: X002 _warn("subprocess %s is still running" % self.pid,
ERROR:1:1: X002 ResourceWarning: Enable tracemalloc to get the object allocation traceback
ERROR:1:1: X002 sys:1: ResourceWarning: unclosed file <_io.BufferedReader name=3>
ERROR:1:1: X002 sys:1: ResourceWarning: unclosed file <_io.BufferedReader name=5>

When running over a large directory a new process is spawned per path so this can become problematic for scaling.

I believe the processes don't get killed by the TimeoutExpired being raised because shell=True is in use (

LibCST/libcst/metadata/type_inference_provider.py

Line 106 in 27ef2bf

command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=True

).
StackOverflow suggests it's because the blocking process is a child of the shell process (https://stackoverflow.com/questions/48763362/python-subprocess-kill-with-timeout).

I'm trying to get a repro of this externally but haven't done so yet.

I plan to work on this, but am opening the issue for discussion and tracking.

[zsol]

Easiest way to repro is probably to add a fake pyre wrapper on your path that sleeps for a minute before delegating to real pyre :)

[lisroach]

@zsol that is a great idea, I got something working before I noticed it though:

<test_libcst.py>

from pathlib import Path
from libcst.metadata.type_inference_provider import TypeInferenceProvider

cache = TypeInferenceProvider.gen_cache(Path("."), ["test.py"], 0)

And running with:
python3 -Wdefault test_libcst.py

Triggers the warning like:

(libcst-env) lisroach@lisroach-mbp libCST % python3 -Wdefault test_libcst.py
Traceback (most recent call last):
  File "test_libcst.py", line 4, in <module>
    cache = TypeInferenceProvider.gen_cache(Path("."), ["test.py"], 0)
  File "/Users/lisroach/LibCST/libcst/metadata/type_inference_provider.py", line 64, in gen_cache
    raise exc
  File "/Users/lisroach/LibCST/libcst/metadata/type_inference_provider.py", line 62, in gen_cache
    stdout, stderr, return_code = run_command(cmd, timeout=timeout)
  File "/Users/lisroach/LibCST/libcst/metadata/type_inference_provider.py", line 108, in run_command
    stdout, stderr = process.communicate(timeout=timeout)
  File "/opt/homebrew/Cellar/python37/3.7.5_3/Frameworks/Python.framework/Versions/3.7/lib/python3.7/subprocess.py", line 964, in communicate
    stdout, stderr = self._communicate(input, endtime, timeout)
  File "/opt/homebrew/Cellar/python37/3.7.5_3/Frameworks/Python.framework/Versions/3.7/lib/python3.7/subprocess.py", line 1710, in _communicate
    skip_check_and_raise=True)
  File "/opt/homebrew/Cellar/python37/3.7.5_3/Frameworks/Python.framework/Versions/3.7/lib/python3.7/subprocess.py", line 1011, in _check_timeout
    stderr=b''.join(stderr_seq) if stderr_seq else None)
subprocess.TimeoutExpired: Command 'pyre --noninteractive query "types(path='test.py')"' timed out after 0 seconds
/opt/homebrew/Cellar/python37/3.7.5_3/Frameworks/Python.framework/Versions/3.7/lib/python3.7/subprocess.py:883: ResourceWarning: subprocess 35820 is still running
  ResourceWarning, source=self)
sys:1: ResourceWarning: unclosed file <_io.BufferedReader name=3>
sys:1: ResourceWarning: unclosed file <_io.BufferedReader name=5>

Which leads to a zombie process like:
lisroach 35820 0.0 0.0 0 0 s001 Z+ 4:03PM 0:00.00 (Python)