Open Stack

On Thu, Jan 6 2022 at 09:53:06 AM -0800, melanie witt 
<melwittt at gmail.com> wrote:
> On Thu Jan 06 2022 08:41:13 GMT-0800 (Pacific Standard Time), Matthew 
> Thode <mthode at mthode.org> wrote:
>> Subject:
>> Re: [nova][requirements] fasteners===0.16.3 held back by nova
>> From:
>> Matthew Thode <mthode at mthode.org>
>> Date:
>> 1/6/22, 08:41
>> 
>> To:
>> Ben Nemec <openstack at nemebean.com>
>> CC:
>> openstack-discuss at lists.openstack.org
>> 
>> 
>> That seems right, I seem to remember eventlet in there.
>> 
>> -- Matthew Thode On 22-01-06 10:17:56, Ben Nemec wrote:
>>> Is it still this:https://github.com/harlowja/fasteners/issues/36  ?
>>> 
>>> I did some investigation about a year ago with one of the fasteners
>>> maintainers, but we never really came up with a definite answer as 
>>> to what
>>> is going on. :-/
>>> 
>>> On 1/5/22 20:33, Matthew Thode wrote:
>>>> This one is simple, and iirc is blocked on upstream fixing 
>>>> something
>>>> (but cannot find the reference).
>>>> 
>>>> fasteners===0.16.3
>>>> 
>>>> https://review.opendev.org/823470
>>>> and
>>>> https://review.opendev.org/804246
>>>> both test this change.
> 
> I worked on the fasteners thing for Too Long of a Time last October 
> and found what is happening. It is indeed the same issue from 2019 
> [1] and I explain the problem (it's long) in a new eventlet issue I 
> opened [2].
> 
> I proposed a patch to "fix" the problem in nova [3], it was initially 
> nacked because it has to do with eventlet, but it's the simplest, 
> smallest change IMHO that will address the issue.

It was me who originally nacked PS1. I still think that PS1 is a hack, 
but after the many weeks of investigation Melanie did and the 
discussions in the github issues I have to accept that we have no 
better option in our hands at the moment. So I'm +2 on [3].

Cheers,
gibi

> 
> I also went on a wild goose chase trying to change all our spawn_n() 
> calls with spawn() in PS2 and PS3 but it ended in a dead end. There 
> are comments detailing that attempt in the review if anyone is 
> curious.
> 
> So, based on that dead end and seeing this come up on the ML, I have 
> reverted [3] to PS1 if anyone can review and give feedback on what 
> approach they would prefer if they think the current approach is not 
> suitable.
> 
> Note: the reason we pull in fasteners is through oslo.concurrency, 
> the lockutils use it.
> 
> Cheers,
> -melanie
> 
> [1] https://github.com/harlowja/fasteners/issues/36
> [2] https://github.com/eventlet/eventlet/issues/731
> [3] https://review.opendev.org/c/openstack/nova/+/813114
> 
>