Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2022-08-16
20:46:57 sean-k-mooney its not recheck the clears it
20:47:01 sean-k-mooney its zuul
20:47:09 sean-k-mooney it only does it when the build set starts
20:47:28 sean-k-mooney or ends i cant rememebr
20:47:47 sean-k-mooney but it will clear the result when the job end i think
20:48:14 sean-k-mooney its a little od because i know the gate pipeline clear it at the start of the jobs
20:51:11 JayF interesting, I'll file that knowledge away
20:51:31 sean-k-mooney i think this is defiend by the pipeline config
20:52:17 sean-k-mooney https://github.com/openstack/project-config/blob/master/zuul.d/pipelines.yaml#L90-L92
20:52:54 sean-k-mooney JayF: if you look at the check pipline above it it does not have the start action for the gerrit trigger defiend
20:53:05 sean-k-mooney so verifed is only update on success or failure
20:53:19 JayF that makes sense
20:56:12 opendevreview sean mooney proposed openstack/nova master: add sorce dev parsing for vdpa interfaces https://review.opendev.org/c/openstack/nova/+/841016
#openstack-nova - 2022-08-17
00:34:49 opendevreview melanie witt proposed openstack/nova master: Adapt websocketproxy tests for SimpleHTTPServer fix https://review.opendev.org/c/openstack/nova/+/853379
07:06:53 gibi melwitt: thank you for taking the triage baton
07:14:38 whoami-rajat hey bauzas , jfyi I've rebased my rebuild feature to use mv 2.94
10:06:53 gibi melwitt, sean-k-mooney, stephenfin: with frickler's help the https://review.opendev.org/c/openstack/nova/+/826523 is unblocked now (the procedural -2 is removed). I let some of you +A the bottom when you feel it can start landing
10:08:17 sean-k-mooney ah sorry had to load context frickler removed sylvains procedual -2
10:08:20 sean-k-mooney got it
10:09:40 sean-k-mooney looks like the first third to half of the feature is ready to merge so ill kick off the start of the serries thanks for that
10:11:31 sean-k-mooney gibi: done
10:11:45 gibi sean-k-mooney: cool
10:12:37 sean-k-mooney that should merge everything up until https://review.opendev.org/c/openstack/nova/+/760456/12
10:14:07 sean-k-mooney ah i was +1 on that while we were debating the terminology
10:19:02 sean-k-mooney ok upgraded my votes on 2 other so that should bring us up to https://review.opendev.org/c/openstack/nova/+/826529/9 and later which i have not really reviewd. thats just over half way i think so that will help manage the patch seires review
10:19:57 sean-k-mooney gibi: im not sure ill have time today but am i right in assuming there are some patches ready to review in the pci series
10:20:26 gibi sean-k-mooney: yes, until https://review.opendev.org/c/openstack/nova/+/850468 it is ready
10:21:08 gibi patches top of that is under development (have TODOs in the commit message)
10:21:26 sean-k-mooney ok cool ill see if i can make some time for those
10:30:10 gibi thanks
10:31:41 sean-k-mooney do you have the link to the etherpad for the feature reviews by the way
10:32:59 sean-k-mooney i have the microversion plan https://etherpad.opendev.org/p/nova-zed-microversions-plan but cant recall what the other one was called
10:34:15 sean-k-mooney oh its proably on the meeting wiki
10:34:34 sean-k-mooney ah yes https://etherpad.opendev.org/p/nova-zed-blueprint-status
10:43:54 gibi sorry I was distracted
10:44:02 gibi but yes you found the right one
11:14:07 opendevreview Merged openstack/nova master: Add limitation to docs about bug 1983570 https://review.opendev.org/c/openstack/nova/+/852168
11:18:48 sean-k-mooney artom: o/
11:19:15 artom ~o~
11:19:55 sean-k-mooney artom: will you have time to work on updating bauzas's mdev parseing seriese this week
11:20:59 artom sean-k-mooney, yeah, I didn't forget
11:21:21 sean-k-mooney thanks
13:06:39 opendevreview Merged openstack/nova stable/train: [CI] Fix gate by using zuulv3 live migration and grenade jobs https://review.opendev.org/c/openstack/nova/+/795435
14:17:17 opendevreview Merged openstack/nova stable/xena: Reproducer unit test for bug 1934094 https://review.opendev.org/c/openstack/nova/+/842922
15:14:49 opendevreview Merged openstack/nova stable/xena: Fix instance's image_ref lost on failed unshelving https://review.opendev.org/c/openstack/nova/+/842923
15:27:53 melwitt gibi: \o/ thanks!
15:29:28 gibi melwitt: :)
15:34:57 melwitt elodilles: thank you for getting stable/train unblocked. it's a miracle! 😂
15:54:19 opendevreview Balazs Gibizer proposed openstack/nova master: Reproduce bug 1986838 https://review.opendev.org/c/openstack/nova/+/853516
15:54:39 gibi sean-k-mooney: ^^ another interesting bug fall out from the pci work
15:56:44 gibi the scheduler expects the compute will fail and the compute says the scheduler should have done a better job
15:57:06 gibi but eventually nobody does its job and we have broken instances
15:57:08 gibi story of nova :D
16:08:49 elodilles melwitt: np, it was your patch anyway o:) so thanks too! :)
16:09:55 elodilles i still don't know what is causing the original problem, but at least with zuul v3 jobs the gate works \o/
16:16:36 melwitt same. glad you thought to try that patch again, I don't think I would have thought of it :)
16:19:34 elodilles well, it was not even my idea as neutron team fixed the same issue with the same solution: changing to zuul v3, so that is why i also tried it o:)
16:24:21 melwitt :D
16:53:25 sean-k-mooney gibi: i tought we had a check to prevent two alisis beign the same
16:54:08 gibi sean-k-mooney: we dont have. And with the traits in alias we can create different aliases that matches the same device from nova perspective
16:54:49 gibi but at least with placement we in the above case there would be zero allocation candidates
16:54:58 gibi so the scheduling would fail accordingly
16:55:06 gibi anyhow I will propose a simple fix that is backportable
16:55:22 gibi and on master this could never happen with placement
16:55:29 gibi after the pci in placement lands
16:55:52 sean-k-mooney https://github.com/openstack/nova/blob/master/nova/tests/unit/pci/test_request.py#L190-L231
16:56:01 sean-k-mooney this is what i was remembering
16:56:29 sean-k-mooney we prevent the same alis beign defiend twice with confliging numa_policies or device types
16:57:04 gibi yeah we prevent defining the same alias twice
16:57:10 gibi but we allow two aliases matching the same device
16:57:18 sean-k-mooney yep the second is valid
16:57:27 sean-k-mooney what is not valid is having 2 claims agaisnt one device
16:57:50 sean-k-mooney the two alisas shoudl have created two pci request objects
16:58:12 sean-k-mooney the bug is that we dont ensure the all pci_request objects are fullfiled by unique host pci adresses
16:59:18 gibi there will be two InstancePCIRequest objects that is OK. the bug is either a) the scheduler does not fail when it fails to consume both requests b) pci_claim does not fail when fails to consume both requests
16:59:26 sean-k-mooney for example if the alias name is differnt its fine for the addres/(product_id/vendor_id) to be the same with different numa polcies
16:59:48 sean-k-mooney ya so its A
17:00:06 sean-k-mooney we should not get to the claim on the compute host if there is only one device
17:00:28 sean-k-mooney by the way the code for the schdluler and the comptue node is basically the same
17:00:38 sean-k-mooney excpt the schduler path uses a copy of the data
17:00:56 sean-k-mooney when its checking if you can consume the pci device
17:01:10 sean-k-mooney so if you fix it it will fix it for both
17:01:12 gibi but even if #a) passes as there is two free devices at that time. #b) shouldt detect and fail if a parallel claim request consumed one of the devices and the new request cannot be fulfilled
17:01:25 sean-k-mooney yes
17:01:38 sean-k-mooney but a is impletnend by calling the code for b on a copy of the data
17:01:47 sean-k-mooney so we fix the later it fixes both
17:02:27 gibi while in general the logic of the scheduler and the pci claim is the same for pci, in reality this part differs. I will push the fix and show that we need to fix this in two separate places
17:04:28 gibi I think we are the same as using _filter_pools but then that is called from two differnt paths by the two service
17:04:33 sean-k-mooney https://github.com/openstack/nova/blob/master/nova/pci/stats.py#L622-L647
17:04:44 sean-k-mooney support_resuts just calls filter_pools
17:04:58 sean-k-mooney correct
17:05:06 sean-k-mooney on the compute node it calls consume_requests
17:05:20 sean-k-mooney https://github.com/openstack/nova/blob/master/nova/pci/stats.py#L226-L269
17:05:21 gibi btw, https://github.com/openstack/nova/blob/master/nova/pci/stats.py#L645 is broken as it individually matches requests to pools and does not consume pools by requests
17:05:47 sean-k-mooney but fixing it in _filter_pools is what i ment by fixing it in one place to fix both
17:05:48 gibi the scheduler calls apply_requests that detects the double consumption
17:06:14 gibi sean-k-mooney: fixing it in filter_pools is tricky as filter_pools does not meant to consume things
17:06:26 gibi can be done though if needed
17:06:33 sean-k-mooney right but apply is not ment to be called in the sculer
17:06:43 sean-k-mooney it will work but we need to make sure not to commit to it on the db

Earlier   Later