Thursday, 2025-09-04

*** mhen_ is now known as mhen02:04
mhensorry, I will not be able to attend today's meeting13:59
croelandt#startmeeting glance14:01
opendevmeetMeeting started Thu Sep  4 14:01:45 2025 UTC and is due to finish in 60 minutes.  The chair is croelandt. Information about MeetBot at http://wiki.debian.org/MeetBot.14:01
opendevmeetUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.14:01
opendevmeetThe meeting name has been set to 'glance'14:01
croelandt#topic roll call14:01
croelandto/14:01
dansmitho/14:02
croelandtabhishek_: around?14:02
croelandt#link https://etherpad.openstack.org/p/glance-team-meeting-agenda14:03
abhishek_o/14:03
croelandtNot much on today's agenda14:03
croelandt#topic Release/periodic job updates14:03
croelandtglance-multistore-cinder-import-fips fails... differently :)14:03
croelandtwe finally moved to CentOS 10 to have a recent version of Python14:03
abhishek_ack14:04
croelandtNow the jobs fails while setting up fips14:04
croelandt[Errno 2] No such file or directory: b'fips-mode-setup'"14:04
dansmithawesome14:04
croelandtI kind of need to figure out how ansible handles this but I cannot find the role for fips enablement14:04
croelandtI need to dig a bit deeper14:04
croelandt#topic Open Discussion14:05
croelandtAbhishek is on PTO starting next week, so I encourage everybody to throw stuff to review at him, because we probably will merge way fewer patches in the next three weeks as we usually do14:06
abhishek_I have one, related to s3 checksum validation14:06
croelandtdo you have a link?14:06
abhishek_I think right now we should add a release note stating set the environment variables14:06
croelandtok14:07
abhishek_and next cycle fix it properly14:07
croelandtand later release a proper fix?14:07
abhishek_let me share the idea14:07
croelandtHm, why not fix in this cycle?14:07
abhishek_https://paste.openstack.org/show/bIn61aqEtUVII8nnij9S/14:08
croelandtoh you're turning this into a spec :)14:09
abhishek_yes14:09
croelandtwith very fine-grained control over the behaviour14:09
croelandtyes in that case it's a discussion for G14:09
dansmithso this is new boto change, which has an envar workaround for the time being, and we can address it later correct?14:09
abhishek_right14:10
croelandtyes14:10
croelandtthe issue is that for some users things are going to start failing14:10
croelandtwhich is really a boto issue, to be fair :)14:10
abhishek_and only ceph rgw is having issue with this, with swift it is working14:10
croelandtI think Cinder *might* have issues with it14:10
croelandtso I am going to try and talk to ratailor Rajat or rosmaita about this, see if we can check14:11
abhishek_I will submit the spec when I will be back and we can discuss this during PTG14:11
croelandtI also wonder how easily this fix can be backported14:11
croelandtok14:11
dansmithyeah, so envar workaround seems fine to make sure we handle it properly14:11
croelandtyeah at first14:11
abhishek_ack14:11
croelandtideally things would be a bit smoother than "wow, it's failing now? OK gotta set some vars"14:12
croelandtthis is so annoying14:12
dansmithwe could also just pin boto for the time being right?14:12
croelandthm, interesting14:13
abhishek_hmm, but that might not allow us to use other features if there are any?14:13
croelandtno but we pin it to <1.36.0 for F14:13
croelandtand unpin it for G14:13
abhishek_either env workaround or pin boto I am ok for anything14:13
croelandtbut I think that is a change we'd have to implement in openstack/requirements14:14
croelandtso by pinning, we make sure the users never run into the issue14:14
abhishek_yes14:14
croelandtand once they upgrade to G, they get glance_store with the fix and an up-to-date boto14:14
croelandtso it's worth considering14:14
abhishek_ack14:14
croelandtok I'll consult first with Cinder and see if they have the same issue14:14
croelandtbecause it would be nice for the both of us to implement the same fix for F14:14
abhishek_Ok14:15
dansmithI mean.. this is not unusual for us to do14:15
dansmithunless there's some burning reason for us to have the newer thing14:15
croelandtdansmith: yeah, makes sense14:15
croelandtespecially if we pin it temporarily14:15
croelandtI don't want this to turn into "the year is 2037, boto3 is still pinned to 1.35"14:15
croelandt:D14:15
abhishek_:P14:16
croelandtOK, moving on14:16
croelandt#topic https://etherpad.opendev.org/p/2026.1-ptg-glance-planning14:16
dansmithwe should be so lucky to still be relevant in 2037 :)14:16
croelandtYou may start adding topics for the pTG at the end of october14:16
croelandtdansmith: the industry moves *slowly* :D14:16
croelandtnothing's urgent, but it would be nice not to fill that Etherpad in the last week :)14:17
croelandt#topic https://review.opendev.org/c/openstack/glance/+/95871514:17
croelandtLast but not least, we should keep an eye on this patch14:17
croelandtthere is a Cinder counterpart14:17
abhishek_you can merge it once cinder is good to go14:18
croelandtyeah waiting on Cinder :)14:18
croelandtAnything else?14:18
abhishek_nope, I am good14:19
abhishek_I will be back in October 1st week, will be checking mail frequently14:19
dansmithabhishek_: have a good one!14:19
croelandtplease don't check email too frequently14:20
abhishek_thank you :D14:20
croelandtPlease go to places with no 4G14:20
abhishek_croelandt: ack :P14:20
croelandt#endmeeting14:20
opendevmeetMeeting ended Thu Sep  4 14:20:52 2025 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)14:20
opendevmeetMinutes:        https://meetings.opendev.org/meetings/glance/2025/glance.2025-09-04-14.01.html14:20
opendevmeetMinutes (text): https://meetings.opendev.org/meetings/glance/2025/glance.2025-09-04-14.01.txt14:20
opendevmeetLog:            https://meetings.opendev.org/meetings/glance/2025/glance.2025-09-04-14.01.log.html14:20
abhishek_croelandt: so are you going to look at pinning 1.36 boto ?15:15
abhishek_< 1.36 i mean15:15
abhishek_In that case we need to do it for master and then for epoxy as well15:16
croelandtabhishek_: trying to get Cinder's attention right now so both projects do the same thing15:34
croelandtI don't want users to have to do different things for different projects15:35
dansmithyou can't..15:35
dansmithyou need to pin it in u-c AFAIK, which will affect everyone15:35
clarkbyes upper constraints should win during installations in the gate15:36
croelandtdansmith: yes but if we do that, once we unpin it, Cinder starts failing15:46
croelandtI'm trying to figure out whether they have the same issue (quite likely) and how we end up fixing it (leaving the user to set env variables? Providing config options?)15:47
dansmithcroelandt: that's how this always works.. the unpin helps us find out if anything is broken and we don't unpin until it's fixed15:47
croelandtok wait a minute15:49
croelandtpatch incoming15:49
croelandthttps://review.opendev.org/c/openstack/requirements/+/95956016:00
dansmithcroelandt: yeah so unless that causes cinder heartache it seems like that's a good plan to me16:09
croelandtyeah, let's get the requirements patch merged16:09
clarkbdo we use any problematic backends in the gate? (like maybe swift's s3 backend or cephs?) It would probably be helpful to specify those details too so we know why we need to modify thinsg for the gate16:11
clarkbmakes it easier to unpin later as it will be more clear what the original conflict was16:12
dansmithclarkb: u-c is not just for the gate right?16:12
croelandtclarkb: I don't think we broke the gate, but as dansmith says, u-c is not just for the gate? Or do I need to amend that patch?16:13
clarkbdansmith: that is what it is for aiui. To ensure we test all the services with a consistent set of dependencies and to reduce dep solving time16:13
clarkbfungi is constantly trying to get the deployment projects to stop relying on it or manage their own version of it16:14
dansmithclarkb: I thought the only reason we really still have it is because of the distros not knowing what their bounds are16:14
clarkbu-c exists because once or twice a week a python dep would update and break all of openstack's gate and untangling that was a mess so now we update things in lockstep as they are made to work to keep the gate working16:14
croelandtso should I set < 1.36.0 directly in requirements.txt for glance?16:14
clarkbdansmith: requirements.txt is supposed to set the bounds16:14
clarkbconstraints is for the gate. That is my interpretation anyway16:15
croelandtclarkb: should I ask for clarification on #openstack-infra?16:15
clarkbcroelandt: no we don't manage the constraints. The requirements team can weigh in on it and can probably do so via the change16:15
croelandtI mean, is there a channel where I can ask whether your interpretation is the right one?16:16
clarkbI think they have an irc channel too yes16:16
dansmithokay, well, if that's the case, then we should pin in requirements at least16:16
dansmithand I guess it seems weird to have u-c be higher than the upper of requirement16:16
clarkbdansmith: no u-c should be within the requirements bounds16:17
croelandtyeah so we need to change in both places16:17
clarkbbut all of this is sort of tertiary to what I was trying to convey16:17
clarkbthe change as proposed does not provide sufficient detail to know when to unpin things16:17
clarkbthat is a bug16:17
croelandtwell16:17
clarkbespecially if the gate doesn't give us the signal for "things are ok now"16:17
croelandtwe unpin when glance_store is fixed? :)16:18
croelandtI have no better answer for now16:18
clarkbcroelandt: yes but 12 months from now the people looking at this may not be us and not have this context anymore16:18
croelandtyeah I tried to convey context in the commit log16:18
croelandtthere is no test that is going to go from red to green16:18
clarkbright and I suggested that a critical piece of info was missing16:18
croelandtbecause I don't think we have that kind of backend in the gate16:18
clarkbspecifically which backends are affected16:19
croelandtoh the S3 backend16:19
clarkbso that they can be tested one way or another or support can be dropped for them etc16:19
croelandtisn't that clear from the commit log?16:19
croelandtI thought it was 16:19
croelandtand *possibly* the Cinder S3 storage backend, but we're still trying to figure this out16:19
clarkbcroelandt: its not just the s3 backend16:19
clarkbits the s3 backend with implementations that are not s3. Which implementations?16:20
clarkbif you use the s3 backend with boto and s3 everything should work fine16:20
croelandtyes16:20
croelandtso Ceph RGW would be one of them16:20
dansmithclarkb: the commit explains the reason, to me16:20
croelandtbut I don't really have a list of S3 implementations so it's hard to say16:20
croelandtit's basically "whatever S3-looking backend people might want to use"16:21
clarkbcroelandt: ya so i would list ceph rgw s3 api as that is something specific that can be tested16:21
croelandtI see16:21
clarkbthen 12 months from now when we want to unpin someone can test that and if it works make a better educated decision on whether or not it is safe to unpin16:21
croelandtYes, I wish it were simpler and just a matter of running a test :/16:22
fungiincluding an adjacent todo comment linking to a bug url or fixing commit that needs to appear in a new release is preferable, just to help future-you not need to fall back on git blame16:22
clarkbit is easy for these nebulous "some backends don't work" pins to never get unpinned out of fear for breaking something. Then when a situation makes it so that we really want to update (api changes or security fixes whatever) its tough to make a good decision on which break is more acceptable16:23
clarkbso I was merely trying to convey that providing as much detail as possible including the currently known broken backends may be helpful in the future16:23
croelandtok, updated with a Launchpad link16:25
croelandtand example of failing backend16:25
clarkbthanks!16:25
clarkbpython version updates are another common reason we end up needing to bump things16:25
croelandtanother layer of complexity16:30

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