Thursday, 2026-08-20

*** jlejeune5 is now known as jlejeune03:57
masahitoabhishekk: thanks for the quick review. applied the comment.07:02
abhishekkmasahito: ack, currently CI is failing so waiting for it to resolve the issue so we can get clean run09:02
masahitoabhishekk: got it. thanks for the kindly reply.09:04
abhishekknp, welcome!!09:04
opendevreviewMaxim Sava proposed openstack/glance master: ci: Discover S3 credentials for tempest tests  https://review.opendev.org/c/openstack/glance/+/99719111:15
abhishekk#startmeeting glance14:00
opendevmeetMeeting started Thu Aug 20 14:00:58 2026 UTC and is due to finish in 60 minutes.  The chair is abhishekk. 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
abhishekk#topic roll call14:00
abhishekk#link https://etherpad.openstack.org/p/glance-team-meeting-agenda14:00
abhishekko/14:01
mklejno/14:01
abhishekkPTL is not around and we have no agenda so open for discussion14:01
mklejnperfect timing :D14:01
abhishekkyeah14:01
abhishekkyou have anything to highlight?14:02
mklejnok, so I'm back with topic of splitting api/data in glance. let's just say that this is cursed :) 307 redirect on PUT is dead idea, 2-part protocol is the only way on this one14:02
dansmitho/14:03
abhishekkack, I think anyway we need to discuss this in upcoming PTG 14:03
mklejnand for host->s3 part, this is possible, but non-trivial. I had to employ surprising amount of tricks to have 95% of draft idea on this one14:03
mklejnack, yes, topic is huge and needs high bw discussion IMO14:04
abhishekkmay be you can list it here, https://etherpad.opendev.org/p/2027.1-ptg-glance-planning14:04
mklejnin the meantime, for next week I'll create etherpad with all of my loose notes rewritten into something paletable14:04
abhishekk++14:04
mklejnso I can have at least some initial comments14:04
mklejnabhishekk, ++ for PTG entry14:05
abhishekkgood idea14:05
abhishekkwe will discuss more next week if needed 14:07
mklejnI _think_ that's about it for rough update, if you don't have specific questions or hard blockers -- don't let me detain you :) 14:07
abhishekkatm, I have nothing on this14:07
mklejnyes, it will be easier to have proper discussion with file that outlines all ideas/issues14:07
abhishekkexactly14:07
mheno/14:08
abhishekkhey14:08
abhishekkI was about to close14:08
mklejnthat's it from me then, I'm staying for mhen update because that's also something that we're interested in :) 14:08
abhishekkmhen I see there is no progress on cinder blocker yet14:09
mhenthere is kinda14:09
abhishekkthen I missed it, I saw your detail comment on one patch though14:09
mhenso I dug deeper into the Cinder topic and discovered that they seem to have more fundamental issues right now concerning backing formats in the RemoteFS-based drivers (such as NFS): https://bugs.launchpad.net/cinder/+bug/216452614:10
mhenI discussed this in yesterday's midcycle14:10
mhenit seems like it's gonna take a while to get that sorted14:10
abhishekkack, means we might be moving to next cycle with this 14:11
mhenso I asked if we maybe could exclude this category of drivers for now in the image encryption patchset14:11
mhenand they said yes, we could do that14:11
abhishekklike document as know issue or something?14:11
mhenso I will try to push the patchset again14:11
dansmithwhat does that mean? snapshot is supported but just not for files on NFS?14:12
mhendansmith: can you elaborate on your question? what do you mean by snapshot?14:13
dansmiththis is about uploading snapshots of encrypted volumes, no?14:14
mhenno14:14
mhennot exactly14:14
mhenthat is a part of it14:14
dansmiththat's not the issue on the cinder side, but what you're saying is we'd be moving forward with encrypted snapshots of everything other than files on NFS...I thought14:15
mhencore problem is, Cinder aims to create an encrypted qcow2 backing file for RemoteFS-based volumes using encryption but once attached to a Nova instance, it gets reformatted as raw LUKS on first use14:15
mhenin the image encryption spec we do not account for encrypted qcow2 in Cinder since we do not expect Cinder to use this format at all currently14:16
mhendepending on which way Cinder will go about fixing this, that assumption might change14:16
mhenand all processes involving transfer between image and volume might need adjustment14:16
rosmaitawell, cinder is going to have to continue using qcow2-with-luks-inside because otherwise you can't snapshot volumes on an nfs-based backend14:17
dansmithI guess my point is it seems risky to move forward with implementation stuff that may not apply to the other and require yet more changes.. 14:17
rosmaitai think what we could maybe do in cinder is only allow upload as a "regular" luks container14:18
dansmithrosmaita: you mean luks inside qcow2, just as you would LUKS in a raw image, and *not* using qcow2's native luks mode where it's sort of blended, is that right?14:18
rosmaitaholy shit, give me a minute to digest that sentence14:18
abhishekk:D14:19
mhendansmith: no, right now, Cinder uses the qcow2 native LUKS14:19
dansmithum... okay14:19
mhenbut that gets overwritten by accident (?) in os-brick14:19
mhen... making it work for Nova magically14:19
dansmithmhen: ack, okay14:19
rosmaitawell, volumes on an nfs based backend are a bit crazy14:19
rosmaitathey are sometimes just flat files, and sometimes qcow214:20
rosmaitadepending on what's happening to them14:20
dansmithmhen: so what is the part of the scheme that is not well-defined.. the glance side described LUKS and qcow2+LUKS, so what about the cinder part is ill-defined specifically?14:20
rosmaitaso if on first use, a volume is raw, that's probably what you'd expect14:21
rosmaitabut if you snapshot it, it turns into qcow214:21
rosmaitabut if you delete the snapshot, it may turn back to raw14:21
mhencinder formats the empty NFS volume with qemu-img as qcow2-native LUKS14:21
dansmithrosmaita: wow.14:21
mhenos-brick on attachment does not recognize the qcow2 header and reformats as raw LUKS14:22
mhenthe actual backing file entirely changes type after its first attachment14:22
mhenqcow2 is discarded14:22
mhenlook at the bug I linked above14:22
dansmithmhen: I'd sort of hope it wasn't trying to inspect the disk at all and was going based on some volume metadata to know what to do/use :/14:22
dansmithbecause...we've seen this movie before14:22
mhenos-brick does not define any encryptor for qcow2+luks14:23
mhenand cinder only says "luks" in metadata14:23
* dansmith has yet to get launchpad to load this morning, FWIW14:23
mhenthis maps to the raw LUKS encryptor in os-brick14:23
rosmaitawell, i'm not claiming that nfs encryption works, i'm just saying that it's going to have to have the same workflow in nova as un-encrypted nfs volumes14:24
mhenCinder establishes a format I don't think os-brick and nova are prepared to handle currently14:24
rosmaitaor else you can't do snapshots or anything14:24
mhenwhen Nova crafts the libvirt XML it attaches it as raw LUKS using QEMU definitions14:24
mhenso if Cinder wants to actually make use of the qcow2 it tries to establish, I think it needs extension for this in os-brick and Nova14:25
rosmaitawell, i really wish that no one had had the bright idea to use nfs to represent block storage devices14:26
mhenhttps://opendev.org/openstack/os-brick/src/commit/379b7a5fe27ee22148fe48b68c454059485b8ec9/os_brick/encryptors/luks.py#L190-L19514:26
dansmithyeah, that's what I'm saying - brick should be looking at some communicated detail about the format of the file and not trying to guess for sure.. if LUKS is already being communicated (to mean regular LUKS) then surely it's as "simple" as just adding another format14:26
* abhishekk need to go afk urgently, making chair to rosmaita so that he can end the meeting once discussion is over14:27
rosmaitawhy pick on me?14:27
abhishekk#chair rosmaita14:27
opendevmeetCurrent chairs: abhishekk rosmaita14:27
rosmaita:D14:27
abhishekkyou can end this quickly :D14:27
* abhishekk going afk, thank you and sorry!!14:27
mhenthe link shows the code in os-brick that makes the reformat happen and silently fixes the thing, making it usable in Nova14:27
mhenanyway, for the image encryption we'd need to limit the implementation to backends that use raw LUKS currently, excluding the RemoteFS-based ones14:29
rosmaitamhen: are you still hoping to land this in H ?14:30
mhenI have my doubts tbh14:30
dansmithhonestly I can't see how it would happen at this point14:31
mhenI will need to place some safeguards in the Cinder patchsets and then push it again14:31
rosmaitawe need to spend some time on this at the PTG14:31
rosmaitawe do have cinder-tempest-plugin tests around encryption14:32
rosmaitaand i believe they run as part of the nfs job14:32
mhenI have a hunch we will need a PTG session with Nova, Cinder and os-bricks folks for the backing format topic14:32
rosmaitabut, i don't know if we have a scenario that goes deep enough14:32
rosmaitamhen: i think we need a working group so people actually spend time looking into this before the PTG14:33
mhenwho is able to organize this?14:33
rosmaitawell, it can be informal14:34
rosmaitait would be good to have some checkpoints14:34
rosmaitabecause the tempest tests i mentioned have come up before, but i don't think anyone has had/made time to actually check14:34
rosmaitamhen: when you say "who is able to organize", do you mean "who has time to organize" or "who is the appropriate person to organize"?14:35
mhenboth14:36
mhenhaha14:36
rosmaitai was thinking you :D14:36
mhenwould it help if I were to sent a message to the ML, describe the topic briefly (with link to the bug) and some questions to be answered?14:36
rosmaitai think that would help ... we need to get melwitt in on this discussion, she has done a lot of work fixing some nfs-related issues in nova14:37
rosmaitahopefully, not so long ago that she's forgotten the details14:37
mhenwould be good if some TC folks or PLs could push my ML entry then, I dunno how much weight it carries coming out of nowhere from some random guy14:39
dansmithuh, more weight than random TC or PTL members :)14:39
rosmaitai think you are hardly a random guy, you have been working on this for like 5 years14:39
dansmiththere's no "stick" approach to success here14:40
mhenokay, I will try14:40
rosmaitayou can begin the message, "I have been working on this for over 5 years now, and we have hit one more major issue we need to figure out before landing this feature"14:41
rosmaita"Please help me get this figured out so we can finally land it in Indri"14:42
rosmaitaanything else?  (I need to get my dog outside, she got into something, and her stomach is making weird noises)14:44
mhenthat's all from me14:44
rosmaitaok, thanks ... sorry to dump more work on you, but hopefully we can get this moving14:44
rosmaitaby the powers vested in my by abhishekk, i pronounce this meeting closed14:45
rosmaita#endmeeting14:45
opendevmeetMeeting ended Thu Aug 20 14:45:20 2026 UTC.  Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)14:45
opendevmeetMinutes:        https://meetings.opendev.org/meetings/glance/2026/glance.2026-08-20-14.00.html14:45
opendevmeetMinutes (text): https://meetings.opendev.org/meetings/glance/2026/glance.2026-08-20-14.00.txt14:45
opendevmeetLog:            https://meetings.opendev.org/meetings/glance/2026/glance.2026-08-20-14.00.log.html14:45
melwittmhen: yes I think it would be good to make a ML post on the issue and I will reply if I know anything heh15:45
mhenI'm preparing one right now; I think I will post it tomorrow15:46

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