Tuesday, 2026-01-20

mnasiadkatc-members: meeting in ~26 minutes here07:34
tonybmnasiadka: noted07:34
* frickler is around, but only for 30mins07:55
mnasiadka#startmeeting tc08:00
opendevmeetMeeting started Tue Jan 20 08:00:01 2026 UTC and is due to finish in 60 minutes.  The chair is mnasiadka. Information about MeetBot at http://wiki.debian.org/MeetBot.08:00
opendevmeetUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.08:00
opendevmeetThe meeting name has been set to 'tc'08:00
mnasiadkaWelcome to the weekly meeting of the OpenStack Technical Committee. A reminder that this meeting is held under the OpenInfra Code of Conduct available at https://openinfra.dev/legal/code-of-conduct.08:00
mnasiadkaToday's meeting agenda can be found at https://wiki.openstack.org/wiki/Meetings/TechnicalCommittee08:00
mnasiadka#topic Roll Call08:00
mnasiadkaO/08:00
gtemaO/08:00
frickler\o08:01
bauzaso/08:01
tonybo/08:01
mnasiadkacourtesy-ping: noonedeadpunk08:02
mnasiadka#topic Last Week's AIs08:05
mnasiadkaNo action items that need to be checked on, ongoing work is in the tracker08:05
mnasiadka#topic What after uWSGI?08:05
mnasiadka#link https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.org/thread/XPNDG4BL5M6EYG62MXMZKOQ24ZTNBLZZ/08:05
noonedeadpunko/08:06
noonedeadpunkooops08:06
mnasiadkaNoonedeadpunk:just in time, the thread says openstack-ansible has some concerns around uwsgi :)08:06
noonedeadpunkI frankly don't like idea of gunicorn as a replacement08:07
noonedeadpunkit's missing couple of crucial features08:07
mnasiadkaDo you have a written down analysis?08:07
noonedeadpunkwell, IIRC it was TLS implementation and routing08:07
noonedeadpunkso you pretty much have to use some web server together with it08:08
mnasiadkaCompared to uWSGI HTTP mode that usually just works?08:08
noonedeadpunkyup, and kinda feature-complete from my prespective08:08
noonedeadpunkthe only thing, is skyline08:08
noonedeadpunkand fastapi, which requires asgi08:09
noonedeadpunkbut if we'd be talking about replacement, I'd really would be thinking about granian08:09
noonedeadpunkdespite not being fan of rust, it supports both asgi and wsgi, and seems to be able to just replace uwsgi 08:10
noonedeadpunkthis is where I was going to do some research for over a year now, but never did08:10
noonedeadpunk#link https://github.com/emmett-framework/granian08:10
mnasiadkaCan you reply to the linked thread with your thoughts?08:10
fricklersounds to me like a topic for the PTG, or maybe a popup WG? I don't think we can properly handle this within the TC only08:11
noonedeadpunkah, and the missing feature of uwsgi is actually http/2 support08:11
noonedeadpunkwhich is a bottleneck for TLS connections08:11
noonedeadpunkyeah, I will compose an answer for the thread08:12
mnasiadkathanks08:12
mnasiadkaI agree with frickler, that probably it would need a popup WG08:13
noonedeadpunkfrankly, I don't think it's a TC/OpenStack issue08:13
noonedeadpunkwe provide proper wsgi apps, and it's up to users to select the web server08:13
noonedeadpunk(or up to deployment tools)08:13
mnasiadkaYes, but we sort of recommend uWSGI nowadays, due to devstack using it08:14
noonedeadpunkum. devstack != recommendation, imo08:14
noonedeadpunkwe should be clear about it and purpose of devstack08:14
noonedeadpunkwe want to have a reference and most simplistic implementation for CI tests08:14
tonybyeah technically it isn't needed but it's appropriate for the TC to set the direction and expectations for users 08:15
noonedeadpunkI'd even say it should be a list of things which users may expect to work08:15
tonyband yeah CI/devstack shouldn't be considered a recommendation but in the absence of anything else they will be 08:15
noonedeadpunkas any of these tools would be highly opinionated08:15
noonedeadpunklike "openstack APIs are wsgi-compatible, which can be leveraged with services like gunicorn, uwsgi, granian, etc"08:17
mnasiadkaAnyway, feels like it’s up to the deployers/users - but it should be stated what is tested today with devstack, but that’s not a recommendation08:17
noonedeadpunkand that is smth we can come up with as a tc08:17
fricklerthis might be getting a bit out of scope, but we need a framework where developers can tell deployers "this is how you would run new feature X". currently this is devstack mostly08:17
noonedeadpunkwhich is kinda wrong approach if you ask me....08:18
noonedeadpunkimo, devstack is pure mix of CI convenience and reference, which can not be really distinguisherd08:19
mnasiadkaWell, all the tests run using devstack, so basically it’s not a misconception to use what is most tested08:19
fricklerbut what would be a better one? like when you implemented skyline for osa, how did you do that08:19
noonedeadpunkwith that I'm pretty sure that devstack configures path-based URIs incorrectly for production deployments08:19
mnasiadkaSkyline uses unicorn in it’s documented deployment I think08:19
mnasiadkaAnd that’s what we use in Kolla-Ansible - but personally I have no clue if that’s right or wrong :)08:20
noonedeadpunkunless smth changed last 6m, which I doubt08:20
frickler"documented deployment" = devstack? or just their installation docs?08:20
mnasiadkaTheir installation docs08:21
mnasiadkaBut their devstack plugin uses gunicorn08:22
noonedeadpunkgunicorn should not work at all with fastapi?08:22
noonedeadpunkit's just not capable of it?08:23
frickleryeah, that's another option (install docs). sadly for many projects these are wildly incomplete, outdated or both08:23
mnasiadkaThat is starting to broaden the scope of the issue :)08:24
bauzasthis is a problem for deployment projects maybe, but that's not service project problem, right?08:25
bauzaslike, nova asks to use a WSGI server, that's it08:25
bauzasbut I understand the problem for openstack-ansible, for sure08:25
mnasiadka(And any other deployment projects)08:26
bauzasthat's why I wonder whether this is a TC concern, but sure we can discuss it08:26
noonedeadpunkI would count deployment projects out here at all, tbh08:26
bauzasthat's actually a cross-deployment project indeed08:26
noonedeadpunkI am most thinking of users who want to do their own thing/manual install/etc08:26
bauzasproject issue*08:26
noonedeadpunkand how to express what is the reference, and what is not08:26
noonedeadpunkpretty much just requirements for running the service08:27
mnasiadkaI think majority projects docs are focused on uWSGI config and mod_wsgi config08:27
noonedeadpunkand from this prespective, I think that should be install guides...08:27
mnasiadka#link https://docs.openstack.org/glance/latest/admin/apache-httpd.html08:27
mnasiadkaGlance says uWSGI or mod_wsgi are recommended :)08:27
mnasiadkaWell, even worse - uWSGI only08:27
mnasiadkaThis document describes the recommended deployment patterns for running Glance with Apache HTTPD with uWSGI.08:28
noonedeadpunkI don't see a problem with this I think....08:28
noonedeadpunkand also Takashi is right, that uWSGI still kinda maintained08:28
mnasiadkaI think the topic has been raised, because people are asking how a project that is ‘’in maintenance mode’’ can be recommended08:28
noonedeadpunklatest tag is 3m ago08:28
mnasiadka(At least I’ve seen such voices from the Scientific SIG community)08:29
mnasiadkaAnyway, we’re not going to solve it here08:30
* frickler sneaks out quietly, will catch up later08:30
mnasiadkaOutside of TC scope - noonedeadpunk will try to reply to the thread with some of his findings that might be beneficial for other people deploying OpenStack08:31
mnasiadkaBut since uWSGI is still sort of maintained (but not accepting new features like http/2) - it’s hard to justify any work in that subject08:31
mnasiadkaLet’s move on08:32
mnasiadka#topic TC Tracker, PTG Follow up08:32
mnasiadka#link https://etherpad.opendev.org/p/tc-2026.1-tracker08:32
mnasiadkaDon’t see any updates in the doc - I don’t think there’s anything to discuss regarding the tracker items08:33
mnasiadka#link https://etherpad.opendev.org/p/oct2025-ptg-os-tc-summary08:33
mnasiadka#link https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.org/message/RX3MZE33GEDV5JDHORZKRUKVDP47UMLJ/08:34
mnasiadka(OS/PTG summary)08:34
mnasiadkaoonedeadpunk will propose a community goal for project teams to support Character Set changes newer LTS releases of Mariadb/MySQL.08:34
mnasiadkanoonedeadpunk: did you get to that?08:34
mnasiadkaOk then, let’s move on to open discussion08:37
noonedeadpunkwell :)08:37
noonedeadpunkI have still that in my TODO list quite high08:38
mnasiadkaThat’s good!08:38
mnasiadka#topic Open Discussion and Reviews08:38
mnasiadkaSo - I’m off whole February, so I’m looking for a volunteer for running the meeting on 17th Feb08:39
opendevreviewMerged openstack/election master: Add configuration for 2026.2/"H" elections  https://review.opendev.org/c/openstack/election/+/97056508:39
mnasiadkaAnd these are the open reviews from gouthamr’s mail:08:40
mnasiadkaRevive os_freezer role for OSA | https://review.opendev.org/c/openstack/governance/+/97336408:41
mnasiadkaRevive os_watcher role for OSA | https://review.opendev.org/c/openstack/governance/+/97338508:41
mnasiadkaRemove rbd-iscsi-client from release management |https://review.opendev.org/c/openstack/governance/+/97231608:41
mnasiadkaRemove barbican-ui from release management | https://review.opendev.org/c/openstack/governance/+/97231508:41
mnasiadkaMove FIPS Compliance goal back to proposed | https://review.opendev.org/c/openstack/governance/+/96914508:41
noonedeadpunkfwiw, I'm potentially also will be unavailable in February, but that is not yet fully decided08:42
mnasiadkaThere’s no need to decide right now - I just wanted to call it out08:43
mnasiadkaOk then, I don’t see people having any topics for discussion08:43
mnasiadkaI’ll wait another 5 minutes and close the meeting08:43
bauzas-08:45
tonybnothing from me08:46
mnasiadkaThank you all for coming :)08:48
mnasiadka#endmeeting08:48
opendevmeetMeeting ended Tue Jan 20 08:48:33 2026 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)08:48
opendevmeetMinutes:        https://meetings.opendev.org/meetings/tc/2026/tc.2026-01-20-08.00.html08:48
opendevmeetMinutes (text): https://meetings.opendev.org/meetings/tc/2026/tc.2026-01-20-08.00.txt08:48
opendevmeetLog:            https://meetings.opendev.org/meetings/tc/2026/tc.2026-01-20-08.00.log.html08:48
fungii do find it frightening that the hype market has convinced users that maintained feature-complete software is more of a risk than software that's constantly changing and still being developed14:16
fungido they use grep? how do they feel about its stability? is grep a risk because it's not getting new features constantly? (hyperbole, there are many grep implementations and some of them do occasionally add new features, but the ones shipped in popular distros are generally stable/unchanging over spans of years)14:18
sean-k-mooneyfungi: is that in the context of uwsgi14:31
sean-k-mooneyor just in general14:31
fungiin general14:32
fungibut it was brought up during the meeting in the context of uwsgi14:32
fungi"...people are asking how a project that is 'in maintenance mode' can be recommended"14:33
fungii use an entire operating system made up of software components and utilities most of which could be described as in "maintenance mode" and i think that's a good thing14:34
cardoefungi: that's definitely something I've noticed as well.14:47
cardoefungi: but I will say uwsgi being in maint mode is problematic. Only because the Linux kernel APIs are evolving and they've made some funky choices about scheduling and memory choices that aligned with older cgroups v1 choices. Things that have been dead and gone from the kernel for a while. Which can make uWSGI do the wrong thing.14:48
cardoeAnd when someone proposed a patch that was incorrect it was merged without much of a review and shipped in a release which made things worse.14:49
cardoeSo that's where I agree that "maint mode" is bad.14:49
fungiyeah, i wouldn't frame a failure to keep pace with changes in underlying subsystems/protocols/standards as "maintenance mode" personally, that's part of maintaining software14:50
fungiit sounds closer to unmaintained, or at best "on life support" by that description14:51
fungii guess "maintenance mode" has become a kind of euphemism then14:51
fungi"i'm calling it abandoned without calling it abandoned, because there's one person occasionally answering questions on the mailing list who might get offended by that characterization"14:52
clarkbcardoe: fungi: uwsgi is also slow to add new python version support and it still doesn't compile reliably on arm6415:49
clarkbthe problem is that it really isn't in maintenance mode. It is effectively unmaintained and not keeping up with change around it15:50
cardoeExactly.15:51
fungiyeah, seems like "maintenance mode" has become a negative term due to being misapplied to nearly-abandoned projects in recent years15:52
opendevreviewMerged openstack/governance master: Revive os_watcher role for OSA  https://review.opendev.org/c/openstack/governance/+/97338517:09
opendevreviewMerged openstack/governance master: Revive os_freezer role for OSA  https://review.opendev.org/c/openstack/governance/+/97336417:11
jjung_"reached maturity" ?17:23
clarkbthe problem with that in this case is that python internal apis have been changing a lot with upstream efforts to improve python interpreter performance. So tools that build against python like uwsgi need regular maintenance to keep up17:24
clarkbif python also reached maturity and stopped making any updates it would probably be fine17:25
fungiyes, it's hard to stabilize software when the language ecosystem under it keeps shifting. of course that only gets worse with all the newer languages people are tempted to reimplement everything in17:30
jjung_Well, I've reached maturity. It does not mean I don't change, just that it's not crazy change like baby to adult :-)17:31
fungii've reached old age, but have never stopped being immature ;)17:32
clarkbright but uwsgi isn't even keeping up wit htaht17:32
clarkbit barely compiles on arm64 for example17:32
clarkbI don't think anyone is asking for fancy new features. More that it compile and work with the most recent python release17:33
clarkbI think building a more generic wsgi interface so that deployers can choose different servers is a good idea. The underlying issue here is that we're so closely tied to uwsgi aiui17:38
*** tosky_ is now known as tosky21:51

Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!