multiprocessing.Condition.notify_all() blocks indefinitely if a process waiting on it has died

[mickp]

BPO	30975
Nosy	@pfmoore, @pitrou, @tjguk, @zware, @zooba, @applio, @mickp
Files	WaitersSleepers.py: A script to demonstrate to the issue. WaitersSleepers.py

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2017-07-20.11:29:48.884>
labels = ['type-bug', '3.7', 'OS-windows']
title = 'multiprocessing.Condition.notify_all() blocks indefinitely if a process waiting on it has died'
updated_at = <Date 2017-07-20.14:20:00.088>
user = 'https://github.com/mickp'

bugs.python.org fields:

activity = <Date 2017-07-20.14:20:00.088>
actor = 'pitrou'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Windows']
creation = <Date 2017-07-20.11:29:48.884>
creator = 'mickp'
dependencies = []
files = ['47025', '47027']
hgrepos = []
issue_num = 30975
keywords = []
message_count = 4.0
messages = ['298713', '298719', '298720', '298722']
nosy_count = 8.0
nosy_names = ['paul.moore', 'pitrou', 'tim.golden', 'sbt', 'zach.ware', 'steve.dower', 'davin', 'mickp']
pr_nums = []
priority = 'normal'
resolution = None
stage = 'needs patch'
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue30975'
versions = ['Python 2.7', 'Python 3.5', 'Python 3.6', 'Python 3.7']

[mickp]

If a process was waiting on an Event but dies for some reason, nothing else can set the Event. I think the Event._cond lock can not be acquired. See attached for a demonstration.

I see the same behaviour in Python 2.7.5 and Python 3.5.2. Is this expected behaviour?

[pitrou]

Hi Mick, thanks for reporting this issue. You're right that the behaviour isn't expected.

[mickp]

Thanks for confirming. If anyone finds this and is looking for a workaround, here's what I'm using:

while not event.is_set()
    try:
        time.sleep(timeout)
    except (KeyboardInterrupt, IOError):
        pass

[pitrou]

As a matter of fact, the bug is in Condition (which is used by Event.wait() but not by Event.is_set(), hence your workaround). Here is a reproducer script.

This looks like a hard issue to solve. There is no way to know whether a process that used to be waiting for a Condition was killed.