Earlier  
Posted Nick Remark
#openstack-nova - 2021-03-11
11:05:26 sean-k-mooney so ya +9-14 lines
11:07:48 lyarwood gibi: https://review.opendev.org/c/openstack/nova/+/735947 - btw I -1'd this earlier, more for my own education and understanding as the code LGTM but the commit wasn't clear on the reasoning for this.
11:09:40 gibi artom: ack, looks good to me
11:10:39 gibi lyarwood: I have to get back to that
11:12:23 artom gibi, mcuh thanks!
11:12:24 sean-k-mooney off the top of my head i dont rememebr all the detail but the highlevle reasoning is that hyperthreading affect the entire host ans sicne cpu are part of the unname greoup these should also be. it would be invalid today to have 2 named groups one that requested HT on and the other HT off for the same host
11:12:44 openstackgerrit Merged openstack/nova master: apidb: Compact Rocky database migrations https://review.opendev.org/c/openstack/nova/+/759405
11:13:45 sean-k-mooney i rememebr suggesting that we shoudl perhaps only ally HW_CPU_* in general in the unamed group becasue at the time i did not think there was ever a valid usecase for varing them in depently in the named groups
11:13:51 sean-k-mooney lyarwood:^
11:15:11 sean-k-mooney https://review.opendev.org/c/openstack/nova/+/735947/5/nova/api/validation/extra_specs/traits.py#26
11:21:33 gibi I think I agree with sean-k-mooney here
11:21:36 sean-k-mooney lyarwood: actully i remember more of the context now filling it in to the patch
11:25:34 lyarwood k thanks I also need to drop in a min to pick up some new glasses, I'll get back to the review this afternoon
11:26:20 sean-k-mooney lyarwood: https://review.opendev.org/c/openstack/nova/+/735947/5//COMMIT_MSG#13
14:01:52 gibi yonglihe: finshed reading the main smartnic patch I still have issues. Also I think you missed my earlier request to add a set of functional tests
14:09:04 gibi alex_xu: ^^ fyi
14:20:17 gibi is there any patch that needs my immediate attention?
14:22:42 stephenfin sean-k-mooney: question: is this change necessary? https://review.opendev.org/c/openstack/nova/+/770533/10/nova/virt/libvirt/host.py#1262
14:23:10 stephenfin I can't figure out what uses it? I don't think we ever finished the NIC capability feature and that's the only thing that code was used for previously
14:23:29 sean-k-mooney the elif?
14:23:34 sean-k-mooney yes the elif is
14:23:49 stephenfin adding the 'vdpa' field to the 'capabilities'
14:23:52 stephenfin dic
14:23:54 stephenfin *dict
14:24:09 sean-k-mooney its used to get the path to add to the xml
14:24:11 stephenfin which will be saved in the 'extra_info' field of the 'PciDevice' table
14:25:02 stephenfin I don't think it is
14:25:04 stephenfin https://review.opendev.org/c/openstack/nova/+/770532/10/nova/virt/libvirt/vif.py#501
14:25:05 sean-k-mooney when generating the xml for the interface we add the vdpa path not the pci addres or the parent
14:25:19 stephenfin that's calling the 'get_vdpa_device_path'
14:25:53 stephenfin which is pretty self-contained and inspect the XML from libvirt directly https://review.opendev.org/c/openstack/nova/+/770533/10/nova/virt/libvirt/host.py#1333
14:26:02 sean-k-mooney ach right
14:26:16 stephenfin so is that something you forgot to cleanup?
14:26:26 sean-k-mooney both will work
14:26:41 stephenfin personally, I'd rather we didn't have to pass through an instance of 'Host'
14:26:46 stephenfin if that's an option
14:27:14 sean-k-mooney we do not have the pci request here
14:27:41 sean-k-mooney so we cant get if form the db like i planned orginally
14:27:54 sean-k-mooney thats why im looking it up i think
14:28:10 sean-k-mooney i will need to look at this in context again sorry
14:28:25 stephenfin okay
14:28:38 stephenfin I'll play around with it a little more, see if I can make sense of it :)
14:30:12 sean-k-mooney so we dont have the pci devices form the db in the virt driver
14:30:40 sean-k-mooney and at present we do not have the mapping form the neutorn port to the pci request ourside of the network_request object
14:31:09 sean-k-mooney so to remove passing in host we would have to pass in the network requiest to lookup the requite id then do a db lookup
14:31:15 sean-k-mooney that would be much more expensive
14:31:45 stephenfin can we pass information through as part of the binding profile, like we do for SR-IOV?
14:31:50 sean-k-mooney i personally have been using the vdpa path in the pci_devices table to debug this and ensure the xml is useing the correct path for the device tha was saved
14:32:01 sean-k-mooney stephenfin: i really dont want to do that
14:32:15 sean-k-mooney because i want to stop using the binding procile for the xml generation entirely soon
14:32:16 gibi somebody opened ~80 new bugs to nova based on fault injection. e.g. https://bugs.launchpad.net/nova/+bug/1918340
14:32:17 openstack Launchpad bug 1918340 in OpenStack Compute (nova) "Fault Injection #1 - improve unit test effectiveness" [Undecided,New]
14:32:40 stephenfin e.g. https://github.com/openstack/nova/blob/master/nova/virt/libvirt/vif.py#L242-L246
14:32:52 sean-k-mooney close them all with we do not consider bugs cause by falut injectoin to be valid?
14:33:13 sean-k-mooney stephenfin: ya i want to remove that use of the profile
14:33:34 stephenfin sean-k-mooney: could we do that wholesale though?
14:33:36 sean-k-mooney we never should have done that in the first place it was a hack
14:33:58 stephenfin I'm sure there's a good reason not to do it, but it's a bit weird having two different ways to do things now
14:34:13 sean-k-mooney right now im using that
14:34:24 sean-k-mooney to get the pci addres of the parent vf
14:34:47 sean-k-mooney but in the futre i want to store this info in our own data stucure in nova db somewhere
14:35:10 sean-k-mooney and pass it in to build/attach ectra
14:35:51 yonglihe gibi, then i got work to do. -:)
14:47:10 sean-k-mooney stephenfin: by the way im currently writing two patches. one to block booting vms with ports of type vdpa until its supproted and a second to block unsupported operations in general
14:47:53 sean-k-mooney similar to how we block api calls with cyborg if they are not supported yet with a 403.
14:49:34 stephenfin makes sense
14:59:21 kashyap stephenfin: For later, you might also find this useful to know about: https://github.com/rhuefi/qemu-ovmf-secureboot
14:59:55 kashyap (Fedora and other distros already ship a variant of this tool we wrote in the past.)
15:16:05 openstackgerrit Stephen Finucane proposed openstack/nova master: tests: Add functional tests for UEFI, secure boot https://review.opendev.org/c/openstack/nova/+/776682
15:16:06 openstackgerrit Stephen Finucane proposed openstack/nova master: docs: Document UEFI secure boot feature https://review.opendev.org/c/openstack/nova/+/776684
15:16:06 openstackgerrit Stephen Finucane proposed openstack/nova master: tests: Remove duplicated 'start_compute' helper https://review.opendev.org/c/openstack/nova/+/776683
15:16:39 stephenfin lyarwood: removed that unnecessary ADMIN_API from the functional test ^
15:17:28 openstackgerrit Lee Yarwood proposed openstack/nova stable/train: compute: Lock by instance.uuid lock during swap_volume https://review.opendev.org/c/openstack/nova/+/758733
15:17:29 openstackgerrit Lee Yarwood proposed openstack/nova stable/train: [stable-only] gate: Skip ceph tests after plugin moved to Octopus release https://review.opendev.org/c/openstack/nova/+/780014
15:17:38 lyarwood elod / melwitt ; ^ fun times with ceph on stable/train btw
15:17:41 lyarwood stephenfin: ack thanks
15:21:20 sean-k-mooney lyarwood: you could pin the plugin to an older commit no?
15:22:19 sean-k-mooney lyarwood: you just do https://github.com/SeanMooney/ci-sean-mooney/blob/main/zuul.d/jobs.yaml#L36
15:22:26 sean-k-mooney required-projects:
15:22:28 sean-k-mooney - name: openstack/devstack
15:22:30 sean-k-mooney override-checkout: master
15:22:40 sean-k-mooney but in this case override the ceph plugin
15:23:05 sean-k-mooney so you can continue to deploy what ever it had before
15:23:37 elod lyarwood: yikes. thanks for looking it and fixing! is this blocking train and older branches?
15:23:42 lyarwood yeah the issue is these jobs aren't zuul native on stable/train and manually check out the plugin
15:24:27 lyarwood elod: I assume <=stable/train with the nova-live-migration failures
15:24:40 lyarwood elod: given grenade isn't running prior to stable/train right?
15:25:03 lyarwood sean-k-mooney: that's a neat hack btw, I wansn't aware of it.
15:25:20 lyarwood sean-k-mooney: we could limit the checkout in the hook script to a known good commit to keep ceph coverage I guess
15:25:22 sean-k-mooney we use it in a number of places for stable branches
15:25:49 sean-k-mooney lyarwood: tempest is one case sicne it branchless
15:26:36 sean-k-mooney lyarwood: but yes pinning to a know good commit or actully taging it and pinnging to that tag would proably be the way to go
15:27:05 sean-k-mooney if it was zuul native i think that can be anything that is vaild for a git checkout command
15:27:19 sean-k-mooney so commit tag or branch
15:27:48 lyarwood right, let me do it that way actually so we don't drop coverage
15:28:15 sean-k-mooney where is the job defiend
15:28:27 sean-k-mooney the enable plugin line optionally takes a git ref
15:28:39 sean-k-mooney which is the branch/tag/commit to use
15:29:09 sean-k-mooney you can do it trivally in the lcoal.conf if you can change the enable_plugin line
15:30:21 sean-k-mooney oh ... https://github.com/openstack/nova/blob/stable/train/gate/live_migration/hooks/ceph.sh#L13

Earlier   Later