Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2026-06-01
16:55:33 phildree[m] Ok, thank you very much! :)
16:56:01 Uggla also try to ping people around Thursday, as this is a review day.
16:56:23 phildree[m] Good to know, will do!
16:57:25 Uggla They have a lot to review. But maybe they will find a slot before the spec freeze.
16:57:55 phildree[m] Understood :)
17:09:38 opendevreview Anton Iacobaeus proposed openstack/nova-specs master: Intel TDX support in libvirt driver https://review.opendev.org/c/openstack/nova-specs/+/979608
17:29:13 Uggla gibi, sambork --> https://blueprints.launchpad.net/nova/+spec/complete-eventlet-removal SLBP
17:30:17 cardoe Uggla: thanks. I’m on PTO so I missed the meeting. I’ll be out next week.
17:32:32 Uggla cardoe, no worries I was aware of it, so I have just skip your topic. But keep mentioning the spec to have it approved.
17:39:17 cardoe Thank you.
18:55:00 gmaan melwitt: graceful shutdown spec is ready for 2nd core review https://review.opendev.org/c/openstack/nova-specs/+/986447
18:55:44 melwitt gmaan: I am reading through it now :) thanks for reminding
18:56:00 gmaan melwitt: thanks
19:29:06 opendevreview Merged openstack/nova-specs master: Spec for the graceful shutdown part2: Task tracking https://review.opendev.org/c/openstack/nova-specs/+/986447
19:45:32 melwitt gmaan: we could use some help reviewing sean-k-mooney fix for unified limits quota enforcement https://review.opendev.org/c/openstack/nova/+/975872 the fix is basically to incorporate RequestSpec into the quota checking to handle cases where the flavor extra spec(s) is not a 1-to-1 mapping to the placement resource classes. if you might be able to take a look
19:46:49 gmaan melwitt: ack, will check.
19:47:44 melwitt thx
19:53:30 melwitt dansmith: I wondered if you might be able to help review this bug fix about stale file existence (false positive) on NFS https://review.opendev.org/c/openstack/nova/+/988547 this has come up in downstream CI intermittent failures on NFS and then more recently in a customer case. the fix itself is very small
20:04:47 dansmith okay will look in a sec
20:05:33 melwitt thx
20:08:23 dansmith melwitt: hmm, I get the hack but.. could we not do something similar by just trying to cache the disk and fail if it exists (using the same exclusive create)?
20:09:15 dansmith meaning,
20:09:29 dansmith instead of trying to create it, and removing it if we succeed,
20:09:54 melwitt dansmith: oh, maybe. the exclusive create is the key (obviously) I was kinda torn between putting it for everything vs knowing the swap disk is the uniquely problematic case
20:09:55 dansmith just create it (empty) and if we succeeded, continue to use that file for create_image()
20:10:11 dansmith and if we failed to create because EEXIST, then we don't call create_image
20:10:29 dansmith I mean, I think you can do this in _create_swap() right?
20:10:47 dansmith just always call that, and if we fail to create it because it exists, then return, else continue with create_image() ?
20:11:08 melwitt I think maybe not bc _create_swap() is only for creating the backing file
20:11:48 melwitt (and I know, I know, we shouldn't have backing files for swap and eph but we do .... that's a separate change we have talked about making)
20:11:50 dansmith ..._create_swap() i the thing you're calling right after, right?
20:12:01 dansmith ohhh
20:12:15 melwitt that's the fetch func
20:12:16 dansmith swap.cache() "fetches" the backing image by calling _create_swap() I guess
20:12:17 melwitt yeah. hah
20:12:53 dansmith okay gross but anyway
20:12:59 melwitt yes, _create_swap() is only for creating the backing file if it doesn't exist, then the file existence of the actual child disk is a separate thing
20:13:08 melwitt yes, it is gross
20:13:15 dansmith I guess my point is.. do we have to create the fake thing, then remove it, then re-create with swap.cache()?
20:13:19 dansmith can we not just leave it?
20:15:04 dansmith the os.path.exists that gets fooled is the one in cache() right?
20:16:45 melwitt yes
20:17:27 dansmith and you didn't want to do this in cache (replace os.exists with a test create) just because it will affect more than swap?
20:17:29 melwitt or in create_image possibly
20:18:26 dansmith this just seems like a really really specific thing where we tickle a bit of behavior that could change (or be different on some other NFS implementation) and making it more direct/in-the-critical-path would be better
20:18:29 melwitt kinda. was torn about it. bc this is kind of an NFS-only problem (i.e. shared storage that is not ceph, I guess) and a swap-only problem. to do the exclusive probe
20:18:57 dansmith right, but like, CIFS could behave similarly but not exactly the same, for example
20:20:48 melwitt yeah ... I'm not opposed, just wasn't sure about adding a more expensive thing to all paths. I think your point is fair
20:21:15 dansmith because we could use os.posix_fallocate() on the fd we get from open instead of spawning the fallocate command, and then it would be way less expensive than it is today
20:21:52 melwitt oh, that's interesting
20:21:54 dansmith anyway, I'm just not jazzed about the obscurity of it.. but I'll just approve with a not-so-passive-aggressive comment about maybe following up and exploring ^
20:23:36 melwitt I knew you would be good for this review :) I was blown away by the false positive, period. and my knowledge limited compared to what you're talking about. I am ok with changing the approach, I'm sure sean-k-mooney will understand it too
20:26:37 dansmith NFS has this sort of behavior all the time, and it has changed over the years,
20:26:47 dansmith and I suspect also differs based on your mount arguments
20:27:28 dansmith commented and +W'd
20:27:47 melwitt ok yeah confirmed it's in cache() where it does the wrong thing, I just re-read the comments in https://redhat.atlassian.net/browse/OSPRH-12784 where I first saw this issue
20:28:15 melwitt thanks dansmith
#openstack-nova - 2026-06-02
04:30:17 opendevreview Merged openstack/nova master: Disable interactive prompt on LVM image creation https://review.opendev.org/c/openstack/nova/+/990576
04:30:31 opendevreview Merged openstack/nova master: Use tempest_concurrency=1 for nova-vtpm job https://review.opendev.org/c/openstack/nova/+/984864
05:14:41 opendevreview Merged openstack/nova stable/2025.2: Fix resize to unpermitted flavors https://review.opendev.org/c/openstack/nova/+/988153
11:35:19 sean-k-mooney hi folks are we ok with running my cirros experiemtn for the next 2 weeks https://review.opendev.org/c/openstack/nova/+/990728 i have also created the revert https://review.opendev.org/c/openstack/nova/+/990729/1 im planning to update my pr https://github.com/cirros-dev/cirros/pull/132 to creat a seperate qcow image that is pre populated isntead of updating the existing one as
11:35:21 sean-k-mooney sugested by frickley but before we commit to that and look toward a release i want to actully confirm this fixes the issues
11:37:23 sean-k-mooney in terms of otehr reviews i woudl really love to close out https://review.opendev.org/c/openstack/nova/+/975872 and https://review.opendev.org/c/openstack/nova/+/984540 today if folks have time to reivew
11:37:48 sean-k-mooney gibi: stephenfin if time permits can ye add those to your queue
12:02:43 gibi sean-k-mooney: I trade https://review.opendev.org/c/openstack/nova/+/984540 for https://review.opendev.org/c/openstack/nova/+/986282
12:02:48 gibi :)
12:03:21 sean-k-mooney sure, oh the eventlet poison ya i can look at that now
12:04:26 sean-k-mooney this still works the way we dicussed right your faking an import expection in the poison so that oslo continues to work but new imporat that dont explcitly guard against that fail
12:08:15 gibi yepp it still raises ImportError
12:12:03 sean-k-mooney i would generally prefer if we didnt have get_eventlet() as that is allowing direct usage outside of the utils file
12:12:20 sean-k-mooney but that prefence is not enought to hold up the patch
12:12:41 sean-k-mooney can we consier if we can prevent leakign that as part of adressign the todo's you have in line
12:19:11 sean-k-mooney gibi: commented and +2w the overall change looks good to me jsut aksed that we consider the feedback inlien as an addtion todo for the followups
13:34:29 opendevreview Lajos Katona proposed openstack/nova master: Use SDK for Neutron security-groups https://review.opendev.org/c/openstack/nova/+/981141
14:02:05 opendevreview Anton Iacobaeus proposed openstack/nova-specs master: Intel TDX support in libvirt driver https://review.opendev.org/c/openstack/nova-specs/+/979608
14:31:41 stephenfin sean-k-mooney: done
14:32:37 stephenfin (the cirros change: gibi has +W'd the other two)
14:49:02 sean-k-mooney stephenfin: thanks
15:06:41 melwitt gibi: thanks for hitting the unified limits fix
15:11:10 sean-k-mooney from a purely upsteam point of view is that somethign we would want to backport?
15:11:17 sean-k-mooney or master only
15:12:22 sean-k-mooney makeing unified limits work for cybrog/pci-passhtough/neutiron qos is both a bug and a feature
15:12:33 sean-k-mooney so while im happy that tis a bug that that was overlooked on master
15:12:45 sean-k-mooney im not sure it qualifies for the stabel policy as a backportable bug
15:13:15 sean-k-mooney so i was going to not backport premetivly and wait for someoen to ask for it
15:14:11 sean-k-mooney i do know there was at least some operator intest in this so im fine with either approch
15:21:26 melwitt yeah I guess that's true it was a known issue but also I think important for cloud operators, so I would lean toward being reasonable to backport upstream. that's just me though
15:25:40 sean-k-mooney ya im not sure which operaors woudl find more suprising the knwo broken behvior ro that it now works if we backprot it. if elodilles has any opions on the mater i would be happy to hear them but i guess we can wait for it to actully land and then dicuss next steps
15:26:20 sean-k-mooney one of those is i am planing to extend the cyborg tempest plugin to have some quota tests eventually.
15:27:05 sean-k-mooney but before i do that i need to intoduce the idea of serial tests to that plugin and build out some geneic testing infra in it
15:30:23 opendevreview dalekseev proposed openstack/nova master: Skip ironic instances in machine type check https://review.opendev.org/c/openstack/nova/+/991137
15:37:40 elodilles sean-k-mooney: sorry, i don't have the context o:) though as you say, if something is more like a feature, then it should not be backported upstream. but if it can be considered as a bug fix and the patch is not that invasive and the stable cores agree, then the possibility is there to backport. o:)
15:38:25 sean-k-mooney elodilles: the tl;dr is our unified limits check didnt consider resouces form pci_passhtough, cybrog or neutron when it was computing the resouce requests
15:38:39 sean-k-mooney elodilles: so if you condiufred the limits they were not being enfoced
15:38:58 sean-k-mooney elodilles: i fixed that but it not clear if that is backportable or not
15:39:37 sean-k-mooney these resouce shoudl have been condiserd in the past but its been that way for quite a while
15:40:20 sean-k-mooney what this means in partice is the pci_in_placement spec assuemed/asserted that unified limite woudl provide quota for pci passhtough device when using pci in placment
15:40:25 sean-k-mooney in practice it never has
15:41:46 sean-k-mooney https://specs.openstack.org/openstack/nova-specs/specs/zed/approved/pci-device-tracking-in-placement.html#dependencies """The unified limits feature exists in an opt-in, experimental state and will allow defining limits for the new PCI resources if enabled.""" but we eiarl defiend makign tha twork out of scope of this spec
15:42:11 sean-k-mooney """Device quotas would require unified limits to be implemented. Implementing quotas is out of the scope of this spec beyond enabling the use case by modeling PCI devices in Placement."""

Earlier   Later