On Thu, 2023-02-23 at 11:21 +0100, Sylvain Bauza wrote:
> Le jeu. 23 févr. 2023 à 11:04, Dmitriy Rabotyagov <noonedeadpunk@gmail.com>
> a écrit :
>
> > At the very least I'd say that the wsgi server used in projects should
> > be standardized and be part of PTI. Previous TC meeting was already
> > hot and full of time-consuming topics, so I've added uwsgi topic for
> > the next meeting agenda.
> >
> >
> I'm not exactly asking for a standard wsgi server across the service
> projects, I'm first inclined to sit down with all the projects team members
> and come up with a discussion about the uswsgi support.
> anyway, happy to hear the TC is taking the ball.
well the point of using wsgi is that the server should not matter
we shoudl not have a depency on any of them and you should be able touse any wsgi
complaint implamantion.
so i dont think we shoudl be prescibien a server.
we coudl try and standarise on which framework we have as we have at least 4 wsgi frameworks
in use to implement the wsgi application.
for what its worth the nova-api console scipt is still a thing if you want to use it with the eventlet
webserver via the oslo server integration hwoever that has been deprecated for a long time.
we try to make sure however that you can bring your own wsgi server and keep our wsgi app portable.
form a PTI point of view it proably would make sense to agree on some that we test and support but we
shoudl not change the generic wsgi apps to only run on one server.
i will have limited upstream time for the next 6 months so i dont want to commit to anything but im
personlaly intereested in seeing if nova/palcement can run under gunicorn well as a uswigi alternitive.
i may even try to add support for that to devstack or other installers if i get time but i cant commit to that.
personally i ahve wanted ot have a lighter weight alternitie to mod_wsgi for years for contaierised openstack
installs and ci where in both cases reducing disk and memory footprint woudl be a benifit.
>
I can give a try for Nova/Placement. TBC, here my point is that I'd appreciate some cross-project discussion (surely during a vPTG if we may do it, or under the umbrella of the TC) about the possibilities we have. Like I said on #openstack-tc channel, I count three of those:
- consider uwsgi being no longer maintained isn't a concern.
- consider keeping uwsgi as our devstack main wsgi server and put the maintainance burden on our shoulders (we did that in the past with various outcomes)
- consider that keeping uswsgi on devstack is nearly unreasonable and identify the potential alternatives with a solig migration plan if so.
That's only once we agree on the direction we may follow that we should define a migration plan and a strong candidate.
In the meantime, like I said, I'm OK with pushing a few DNMs for testing gunicorn on nova-api, nova-metadata-api and placement-api services if thay may help the decision-making.
-Sylvain
>
> > чт, 23 февр. 2023 г. в 10:21, Sylvain Bauza <sylvain.bauza@gmail.com>:
> > >
> > >
> > >
> > > Le jeu. 23 févr. 2023 à 09:30, Thomas Goirand <zigo@debian.org> a écrit
> > :
> > > >
> > > > On 2/22/23 22:55, Pete Zaitcev wrote:
> > > > > On Wed, 22 Feb 2023 11:14:24 +0100
> > > > > Thomas Goirand <zigo@debian.org> wrote:
> > > > >
> > > > > > It's been a few Debian release that all OpenStack APIs are served
> > using
> > > > > > uwsgi. Unfortunately:
> > > > >
> > > > > Swift uses eventlet, doesn't it?
> > > > >
> > > > > -- Pete
> > > >
> > > > Yeah. We tried hard to make both Swift and Glance use uwsgi, but these
> > > > services simply fail in very subtle ways. It's not obvious that they
> > > > fail at first, but they really do. I very much believe this is a defect
> > > > in both projects, and that they should invest time working on this.
> > > >
> > > > Anyways, for the time being, I believe I'll continue using uwsgi for as
> > > > long as the project is still in maintenance mode. Uwsgi is by far the
> > > > best, as much as I can tell.
> > > >
> > >
> > > Shouldn't we somehow identify whether DevStack is at risk given our hard
> > dependency on uwsgi and should we as service project teams somehow kick-off
> > some coordinated brainstorming efforts to identify the risks and mitigate
> > them ?
> > >
> > > Sounds a goal to me.
> > >
> > > > Cheers,
> > > >
> > > > Thomas Goirand (zigo)
> > > >
> > > >
> >
> >