Thursday, 2026-02-26

opendevreviewMaxim Sava proposed openstack/glance master: Add tests with selective store deletion and invalid store  https://review.opendev.org/c/openstack/glance/+/97373612:47
croelandt#startmeeting glance14:00
opendevmeetMeeting started Thu Feb 26 14:00:37 2026 UTC and is due to finish in 60 minutes.  The chair is croelandt. Information about MeetBot at http://wiki.debian.org/MeetBot.14:00
opendevmeetUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.14:00
opendevmeetThe meeting name has been set to 'glance'14:00
croelandt#topic roll call14:00
croelandt#link https://etherpad.openstack.org/p/glance-team-meeting-agenda14:00
croelandto/14:00
mheno/14:00
sakumbhao/14:00
* fungi waves14:01
abhishekko/14:01
croelandt#topic Release/periodic job updates14:02
croelandtAll good \o/14:02
abhishekkcool14:02
croelandtRegarding jobs, tehre is apparently an issue with the client jobs right now14:02
croelandtso patches will fail the checks on gerrit14:02
abhishekkI hope issue is not at our end14:02
croelandtit is due to a tox upgrade, this should be fixed soon, and applies to all projects14:03
abhishekkgr814:03
croelandt#topic Image encryption14:03
croelandtOne big development for image encryption is that dansmith has a PoC for inspecting encrypted images14:03
croelandtmhen: I commented on the spec to make sure this becomes part of it14:03
croelandtwe want to keep Glance as the first line of defense, even with LUKS encrypted images14:04
croelandtdoes that make sense?14:04
mhengiven the CVE history so far I guess we don't have any other choice, do we?14:05
abhishekknope14:05
fungido you have a link for the poc, or is it not up for review yet?14:06
fungii'm not immediately finding it14:06
abhishekkno its not up yet14:06
croelandtfungi: no, dansmith has been working on it offline btu it's not ready to beshared14:06
fungiis it just about checking plaintext headers, or decrypting the images prior to validation?14:07
croelandtmhen: I think dansmith may provide the implementation, so on your end it would be a matter of updating the spec and thtat's it14:07
croelandtfungi: I think the PoC decrypts the image to validate it like it would an unencrypted one14:07
fungigot it, so glance will need access to the decryption keys for images14:08
croelandtI think it will access them through Barbican, right?14:08
mhenyea it already can14:08
fungipresumably, i just mean it's one more place the keys end up14:08
croelandtOr was there a proposal to make the keys unavailable to Glance?14:08
mhenbecause of secret consumers and key deletion policy stuff14:08
croelandtmhen: yes, it is my understanding that it had access to the keys already14:08
fungiah, okay, i missed that detail. thanks14:09
croelandtmhen: Ok so I'll re-review the spec once you remove the bit I commented14:09
fungiif glance already needed access to the keys anyway then it doesn't seem like any additional security risk14:09
croelandtyes14:09
mhenI am not thrilled about Glance being an in-the-middle between the actual image producers and connsumers that can actually fully decrypt the image and has access to the keys but I understand it is necessary because of several reasons (key_deletion_policy stuff and the defender)14:10
croelandtwe're also thinking of allowing users to tell the OSC: "hey, this is my unencrypted image, please upload it to Glance with this key" and that would maybe be an additional risk14:10
croelandtmhen: it's a bit of a tradeoff14:10
mhenso if we deem this method (dansmith's PoC) feasible and functional, you would want it to be part of both the existing spec and the implementation patchsets for image encryption, correct?14:11
croelandtso the spec mentions it *cannot* be done, so this part should at least be removed14:12
croelandtas for the implementation, I think dansmith would contribute his PoC rather than have you write it yourself14:12
mhensure14:12
croelandtespecially as this is not trivial14:12
mhenthat would be much appreciated14:12
croelandtso maybe it's two patches, maybe you work on the same patch14:12
croelandtwe should ask him in about 2 hours when he's on :D14:13
mhenfeel free to ping me14:13
croelandtI *think* it would probably be one more function call in the current flow14:13
croelandtso nothing super disruptive14:13
croelandtok I think we can move on14:13
croelandt#topic Bridging the Gap: Flamingo Cycle Retrospective 14:14
croelandtfungi: ^ you got the microphone14:14
fungiildiko posted openstack-wide 2025.2 (flamingo) cycle retrospective contributor/maintainer survey results and metrics to openstack-discuss at the end of last year:14:14
fungi#link https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.org/message/XZZYFHMUCB3IZU5AXM366AY7WJXQMTBX/ Bridging the Gap Flamingo Cycle Retrospective Survey Results14:14
fungi#link https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack.org/message/FD4JALJN7OB2YNBUVWCPZZ6YNJMIL2JT/ Bridging the Gap Flamingo Cycle Retrospective Metrics Analysis14:14
fungii and the other community managers on the openinfra foundation staff have also been digging into team-specific details and i'm doing a round of outreach similar to last cycle, to go over how things may have changed14:15
fungithe glance team had 3 contributor survey responses (and no maintainers), so while it could be better this is still more engagement than we saw for the epoxy retrospective surveys14:15
fungi2 of the contributors self-identified as established while the other considered themself new to glance, with all contributing to 2 or more other open source projects14:15
fungiaverage ratings on the contributor survey questions were positive overall, with useful automated testing results coming in highest at 4.33 out of 5 and timely reviews the lowest with 3.00 (typical of overall openstack response averages)14:15
fungiit's worth noting that the new contributor ranked everything a 5 out of 5, which pushed the averages up a bit, but implies they had a very positive experience so that definitely counts for something14:15
fungithe biggest challenges reported by contributors were getting reviewer attention to changes, scope creep requested by reviewers, and apparent lack of consensus between reviewers, as well as struggling to implement cross-project features14:15
fungithe new survey questions about priorities indicated all of them (3/3) followed prioritization from irc and mailing list discussions, and most (2/3) looked at specs, but then there was a long tail (1/3) for every other option too14:16
fungias for metrics, active reviewer count fell by 12% and active maintainer count was down 33% in flamingo as compared to epoxy, but both were about at the same level as they were a few cycles earlier in caracal14:16
fungiaverage (mean) time to first review and merge improved slightly in flamingo, but median times for both increased, indicating review activity may have become more consistent overall while struggling to deal with the team shrinkage14:16
fungiglance maintainers closed a whopping 30% more changes than were opened during the flamingo cycle, however, far better than any of the prior 5 cycles (next best was 21% over during antelope)14:16
fungithe community managers have been distilling feedback from last year's discussions, and are working on compiling a concise set of techniques/recommendations for improving communication and efficiency, which i can get into more another week14:16
fungiwe're also hoping to get some case studies done in concert with successful contributors and maintainers to highlight how specific practices and behaviors help them achieve better throughput, in hopes of being able to replicate them14:17
fungianyway, that was a quick dump, i know it's a lot to take in but i didn't want to eat up too much of your meeting, so i'll put this on the agenda again in two weeks to give everyone time to digest and come up with questions or ideas14:17
fungithough i'm happy to answer any immediate feedback now if there's time14:17
croelandtIs there a link to Glance-specific results/metrics? 14:17
abhishekkwe need time to process it :D14:17
croelandtyeah I'll have to read that in details14:18
fungii don't have sanitized team-specific breakdown documents like the openstack-wide ones, but i can put some more specific details together for glance if that's desirable14:18
fungior if there are specific numbers you want to know i can paste those in now14:18
croelandtyou said "glance maintainers closed a whopping 30% more changes than were opened during the flamingo cycle" so I thought you had them14:18
fungiyeah, just a sec14:19
croelandtif we keep closing more than we open, we might be done with the backlog around 203514:19
croelandt#hope14:19
fungiin the flamingo cycle, there were 153 changes opened for glance repositories and 199 closed14:19
fungias opposed to epoxy where 159 were opened and 161 closed14:20
fungiso the team is more than keeping up with incoming review volume, even with fewer maintainers active during the cycle14:20
croelandtinteresting, since I keep feeling like I'm drowning in reviews14:20
fungithat's part of why we analyze this stuff, perception is often different from reality (though both have a valid basis)14:21
croelandtyes14:21
fungiit's helpful to try to dig into why people feel buried in work even when they're more than keeping up14:21
croelandtthanks a lot, I'll dive into the emails linked above later 14:21
abhishekkalso @croelandt can you dump this in etherpad?14:22
croelandtOne must imagine Sisyphus happy14:22
fungiand yeah, as i said, feel free to ask me more detailed questions whenever you think of them14:22
abhishekkack14:22
fungiand i'll bring it up on the agenda again in two weeks (i have something conflicting next week at this time or i'd cover it then)14:22
croelandtabhishekk: just dumped comments by fungi 14:22
croelandtthanks a lot fungi14:23
croelandtmoving on14:23
fungiyw14:23
croelandt#topic Open Discussion14:23
abhishekkthanks14:23
croelandtamnik: you got the mic14:23
amnikhello, I want to share an idea with you before any specs14:23
abhishekkgo ahead14:24
amnikIn  interoperable image import we apply configured plugins for all request14:24
amnikI wonder If add parameter to the API that user can choose which plugin apply on request14:24
amnikchoose which plugin set apply on request.14:25
abhishekkI am not recovering it but we thought of this while interoperable image import was introduced and it was rejected14:25
croelandtwhat is the use case here?14:26
rajivhi14:26
amnikI think it's good to enable user choose which plugins apply on image 14:27
amnikinstead of static list of plugins apply on all images14:27
croelandtyeah but does this come up in real life?14:27
croelandtis anyone complaining about it?14:28
abhishekkwe can always use override way14:28
abhishekkif user sepcifies then apply those plugins only14:28
abhishekkand ignore plugins configured by system14:28
abhishekkBUT as cyril says we need solid use case for that14:29
amnikok for now I not say any special use cases.14:30
croelandtyeah if no one using a real deployment complains about it, I'd rather not touch it14:30
amnikI think it can be good feature14:30
croelandtoh it could probably be14:30
croelandtbut it's not something I'd be willing to spend time on if no one is bothered by the current implementation14:30
abhishekk+114:30
abhishekkamnik: if you found any good use case then we can discuss this topic during PTG14:31
croelandt^ this14:31
amnikyes ok, Thank you14:31
abhishekkthank you!!14:31
croelandtok, I think we can close this meeting14:33
croelandtunless someone else has anything to add?14:33
rajivi wanted to follow up on last weeks review request, i am unsure if we can fix the zuul error14:33
rajivhttps://review.opendev.org/c/openstack/glance_store/+/97732314:33
croelandtah let me check14:33
croelandthm interesting14:35
rajivsecondly, would this be fixed https://review.opendev.org/c/openstack/glance_store/+/578549 ?14:35
croelandttesttools should probably require typing_extensions14:35
rajivokay, last, would this be of any interest if i start working on this again : https://review.opendev.org/c/openstack/glance_store/+/78823314:36
croelandtso, is the swiftclient really retrying like I wrote back in 2022?14:37
croelandtparallel uploads to Swift seems interesting14:37
croelandtabhishekk: ^ what do you think?14:37
abhishekkI need to go through the spec14:38
abhishekkhttps://review.opendev.org/c/openstack/glance-specs/+/78717914:38
abhishekkThis was my last comment on it, I need to recollect why I said that :D14:38
abhishekkAs I told and discuss during the PTG, this is not how we want to modify the swift driver. We don't want condition based implementations anymore. Feel free to discuss this in weekly glance meeting.14:38
abhishekkI think we decided on creating a sub-driver based14:39
rajivyep, hence wanted to bring this up :) 14:39
rajivah ok ok14:39
croelandtok so we agree with the feature, it's just an implementation detail?14:39
abhishekkyes14:40
croelandtok14:40
croelandtso I think rajiv you could write a PoC and then we'd discuss this at PTG, probably14:40
abhishekkwe need spec update as well based on previous comments14:40
rajivokay, i think i wrote a glance specs for this but unable to find it14:41
rajivcroelandt: regarding the 1st review request, the testtools thing is something for me ?14:42
croelandtI'll look into it14:42
rajivthank you!14:42
croelandtit's weird that it fails in 310 but not in 31314:42
rajivyes14:42
croelandtso yeah, gotta figure it out14:42
croelandtbut I think it will be an issue for other patches as well14:42
croelandtso might as well debug it sooner than later14:43
rajiv:) 14:43
croelandtok, thanks everyone for joining14:43
croelandtsee you next week!14:43
croelandt#endmeeting14:43
opendevmeetMeeting ended Thu Feb 26 14:43:39 2026 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)14:43
opendevmeetMinutes:        https://meetings.opendev.org/meetings/glance/2026/glance.2026-02-26-14.00.html14:43
opendevmeetMinutes (text): https://meetings.opendev.org/meetings/glance/2026/glance.2026-02-26-14.00.txt14:43
opendevmeetLog:            https://meetings.opendev.org/meetings/glance/2026/glance.2026-02-26-14.00.log.html14:43
dansmithwait14:43
abhishekkthank you14:43
dansmithaww14:43
croelandtTOO LATE14:43
abhishekkTOO FAST14:43
croelandtwe shall never know what you had to say14:43
rajiv:D 14:44
dansmithmhen: croelandt yep, I'll be pushing up a PoC and I will handle getting that part working14:44
mhendansmith: thank you!14:44
dansmithmhen: we currently allow operators to disable deep inspection today, which I suppose we could extend to "just don't look inside the encrypted ones" if you want - the implementation allows for that14:45
croelandt\o/ perfect14:45
dansmithhowever, nova cinder (and as you noted) glance already have to access the key so I see very little reason to just not inspect things14:45
dansmiththey're never stored, and only the first block or two will be decrypted in memory14:45
dansmithqemu-img allows me to hide a qcow file inside a luks in a pretty straightforward way, so...14:46
fungii recall there was some debate as to whether qemu-img's backend auto-selection actually got used in that situation14:47
fungisince if you have to tell it to treat the plaintext payload as a specific type of image then there's no risk of accidentally passing it to one of the dangerous drivers14:48
dansmithfungi: but nova has a mode where we unwrap the luks layer with dm-crypt and pass the inner device to qemu14:49
fungiah, and then yes it becomes the original problem14:49
dansmithand any of our image manipulation things during snapshots and things get more complicated14:49
dansmitheither way, I'm trying to avoid providing a way to expose qemu to these things even if we don't have a specific exploit in mind14:50
dansmiththe recent nova one was majorly less severe because glance refused to let nova exfiltrate host data14:50
fungiyep, makes sense14:53
opendevreviewDan Smith proposed openstack/glance master: DNM: Test glance against LUKS images  https://review.opendev.org/c/openstack/glance/+/97808815:42
dansmiththis ^ is just to get a baseline of what happens with luks images today and then I'll depends-on the oslo code15:46
opendevreviewDan Smith proposed openstack/glance master: DNM: Test glance against LUKS images  https://review.opendev.org/c/openstack/glance/+/97808818:05
dansmiththis ^ should now run with modified oslo.utils and be able to upload the luks images, but it won't detect the bogus qcow2 ones because no inspection18:12
dansmithwe'll need tempest to register secrets in barbican for those and set the metadata on the images to find them18:13
dansmithah, nope, because those tempest tests don't know to set the container_format instead of disk format19:48

Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!