Avoid 'Connection refused' when using gthread with keep-alive

[ddzialak]

Fixes #1698

[benoitc]

Thanks for the patch! Looking :)

[ddzialak]

any update?

[ddzialak]

I've added two commits to make added attribute corresponding to it's name, these commits don't affect solution for connection refused problem. Now is_active should be True if it's processed and should not be present on self._keep and False if processing is done and should be present on self._keep. Please squash these commits if everything is ok.

[tilgovi]

Does any of this get simpler to reason about or handle if the murder_keepalived method held the lock for the whole body of the loop? With the lock being taken in both branches of the if statement on line 168, I wonder whether it's worth the overhead of locking twice and whether holding the lock throughout might remove the race condition without introducing this new variable.

[ddzialak]

@tilgovi That was my first try and after that received connection refused again but most probably I've done something wrong with testing..

[ddzialak]

Now it's clean why moving whole murder_keepalived under lock doesn't solved that issue. Other important condition is that bug may happen only with Python2 because __cmp__ in python2 is used to override == operator while function __lt__ in Python3 doesn't.

[ddzialak]

any update?

[benoitc]

Sorry for the delay. looks good, merging it now :) Thanks !

[ddzialak]

Thanks ☺️

…

On Mar 1, 2018 1:56 PM, "Benoit Chesneau" ***@***.***> wrote: Sorry for the delay. looks good, merging it now :) Thanks ! — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1699 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHXbDCYRKTduea0_Ti0DdIhKMqWADwwlks5tZ-_0gaJpZM4R7JdS> .

[ddzialak]

@benoitc What is the estimated time when this change is pushed to PyPI?