| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2021-01-08 | |||
| 10:12:07 | openstackgerrit | Stephen Finucane proposed openstack/nova master: tests: Clean up 'test_hypervisors' https://review.opendev.org/c/openstack/nova/+/769903 | |
| 10:20:54 | openstackgerrit | Elod Illes proposed openstack/nova stable/stein: [stable-only] Cap bandit and make lower-constraints job non-voting https://review.opendev.org/c/openstack/nova/+/766487 | |
| 10:21:47 | gibi | sean-k-mooney: hi! There is a SRIOV NIC, couple of VF is created on the PF and then both PF and VFs are whitelisted in nova compute. Then booting a VM with a direct-physical port nova will allocated the PF and makes the VFs unavailable. But which module (nova, neutron, libvirt) makes sure that the VFs are deleted? | |
| 10:22:10 | gibi | Does it work like the following? libvirt detaches the PF from the host and that makes the VFs deleted? | |
| 10:22:14 | openstackgerrit | Elod Illes proposed openstack/nova stable/stein: WIP [stable-only] Fix lower-constraints job https://review.opendev.org/c/openstack/nova/+/769809 | |
| 11:43:59 | openstackgerrit | Elod Illes proposed openstack/nova stable/rocky: [stable-only] Cap bandit and make lower-constraints job non-voting https://review.opendev.org/c/openstack/nova/+/766492 | |
| 11:44:00 | openstackgerrit | Elod Illes proposed openstack/nova stable/rocky: WIP [stable-only] Fix lower-constraints job https://review.opendev.org/c/openstack/nova/+/769910 | |
| 11:46:40 | yonglihe | gibi, look libvirt detach it from host if use managed mode. | |
| 11:47:10 | yonglihe | <hostdev mode='subsystem' type='pci' managed='yes'>, if this is what you looking for. | |
| 11:47:27 | gibi | yonglihe: thanks that is what I was looking for | |
| 11:51:59 | xinranwang | gibi: Hi gibi, as we discussed in nova spec, a new vnic type is needed for device managed by cyborg. I create a neutron RFE for this, and there is some question posed. Could you please check it ? https://bugs.launchpad.net/neutron/+bug/1909100 | |
| 11:52:03 | openstack | Launchpad bug 1909100 in neutron "[RFE]add new vnic type "cyborg"" [Wishlist,Confirmed] - Assigned to Rodolfo Alonso (rodolfo-alonso-hernandez) | |
| 11:54:03 | gibi | xinranwang: I deferr to sean-k-mooney and the neutron folks here. From my perspective what nova needs is a way to know that the port needs a special handling (talking to cyborg) during boot. | |
| 11:56:34 | xinranwang | gibi: yes, nova can check the vnic type and trigger the interaction with cyborg then. | |
| 11:56:37 | gibi | commented on the RFE | |
| 11:56:44 | gibi | xinranwang: yes | |
| 11:56:56 | gibi | a new vnic_type is OK for me | |
| 11:57:13 | gibi | but maybe the neturon folks has other idea | |
| 11:58:30 | xinranwang | Neutron will have irc meeting at 14:00 UTC today. We will discuss this topic. Do you and sean-k-mooney have time to join the discussion? | |
| 11:58:50 | stephenfin | lyarwood, gibi, sean-k-mooney: If you've a chance, could you take a look at https://review.opendev.org/c/openstack/python-openstackclient/+/762396/ again at some point today? | |
| 11:59:02 | stephenfin | (OSC change to "Add 'server shelve --offload', 'server shelve --wait' options") | |
| 11:59:28 | gibi | xinranwang: I will join, please ping me if I did not show up | |
| 11:59:44 | gibi | stephenfin: added to my review list | |
| 12:00:34 | xinranwang | gibi: cool, thanks | |
| 12:02:27 | lyarwood | stephenfin: queued | |
| 12:07:27 | stephenfin | gibi: I've closed this bug as WONTFIX. I don't think we have access to PPC nodes in zuul, which means we can't say we support anything other than x86/x86_64. Lemme know if you disagree https://bugs.launchpad.net/nova/+bug/1909972 | |
| 12:07:28 | openstack | Launchpad bug 1909972 in OpenStack Compute (nova) "a number of tests fail under ppc64el arch" [Low,Won't fix] | |
| 12:08:05 | gibi | stephenfin: I saw that and I have the same problem, I cannot start fixing those tests as I have no way to reproduce the problem | |
| 12:08:19 | gibi | (no I did not tried spinning up a ppc VM with qemu) | |
| 12:08:47 | stephenfin | Yeah, I've suggested that zigo simply mark the package as arch-specific. If it's not tested, it's broken | |
| 12:08:49 | gibi | stephenfin: did we declare support for ppc somewhere in nova doc? | |
| 12:08:55 | stephenfin | I don't think so | |
| 12:09:04 | stephenfin | We don't block it but https://github.com/openstack/nova/blob/46899968619e4ea0ff2ab380977619bb29578d43/nova/virt/libvirt/driver.py#L572-L581 | |
| 12:10:00 | lyarwood | odd why don't we list aarch64 there? | |
| 12:10:05 | lyarwood | I thought we had jobs now | |
| 12:10:14 | lyarwood | ah third party ignore me | |
| 12:10:23 | stephenfin | someone's working on it but it's not merged yet, no | |
| 12:10:26 | gibi | stephenfin: good point | |
| 12:10:43 | gibi | stephenfin: if we even log a warning then that should be enough to close the bug | |
| 12:12:23 | gibi | let's see what zigo's oppinion about this | |
| 12:19:37 | sean-k-mooney | xinranwang: i should be able to join too | |
| 12:20:05 | zigo | stephenfin: gibi: My opinion is: if you can't test with PPC, and that the unit test is reported as broken on non-x86_64, then skip it on non-x86_64 rather than just closing the bug ... | |
| 12:20:09 | sean-k-mooney | stephenfin: yes i can take a look at that shortly | |
| 12:20:53 | stephenfin | zigo: what about when someone adds the next test that doesn't work on x86_64 | |
| 12:21:13 | sean-k-mooney | gibi: the vf are deleted by the kernel when the PF is passhtough to the vm | |
| 12:21:16 | zigo | stephenfin: These issues are *very* rare. | |
| 12:21:39 | zigo | So we can act on a case-by-case ... | |
| 12:21:57 | gibi | sean-k-mooney: thanks | |
| 12:22:48 | stephenfin | I disagree. You skip tests because they're not applicable. These tests are applicable. They simply don't work on PPC | |
| 12:22:53 | sean-k-mooney | the managed=yes part of <hostdev mode='subsystem' type='pci' managed='yes'> i blelive tells libvirt to manage bining the pf to vfio-pci form its current nic driver and to restore it when the vm is done | |
| 12:22:54 | stephenfin | who knows what else doesn't work | |
| 12:23:13 | sean-k-mooney | by which i mean its stop and the pf returns to the host | |
| 12:23:48 | sean-k-mooney | technically its udev coupled with the kernel config that decide what driver it goes back too | |
| 12:23:49 | zigo | stephenfin: As much as I understand, they would work if s/ide/scsi/ ... | |
| 12:24:42 | stephenfin | Perhaps. I genuinely don't know | |
| 12:24:43 | sean-k-mooney | if you used udev rules to allocate the vfs then that will recreate them | |
| 12:24:46 | stephenfin | which is kind of the point | |
| 12:25:20 | sean-k-mooney | on the ppc front | |
| 12:25:35 | sean-k-mooney | we have support power kvm for years | |
| 12:25:46 | sean-k-mooney | which is libvirt/kvm on ppc | |
| 12:25:54 | sean-k-mooney | ibm has a ci for it | |
| 12:26:06 | sean-k-mooney | that runs against nova | |
| 12:26:10 | stephenfin | sean-k-mooney: clearly they don't, or these tests would be failing there | |
| 12:26:25 | stephenfin | or their CI job matrix isn't complete | |
| 12:26:39 | sean-k-mooney | it only runs integration tests ie tempest | |
| 12:27:28 | sean-k-mooney | https://oplab9.parqtec.unicamp.br/pub/ppc64el/openstack/nova/68/767368/1/check/tempest-dsvm-full-focal-py3/ef10362/ | |
| 12:27:39 | sean-k-mooney | that a run from the IBM PowerKVM CI | |
| 12:27:48 | sean-k-mooney | ppc64el | |
| 12:28:21 | sean-k-mooney | stephenfin: gibi zigo we have no unit or func test coverage | |
| 12:29:01 | sean-k-mooney | but we cant mark the package as not supported on ppc or it will break the ablity to install nova on ppc which has been supported for 4+ years | |
| 12:30:45 | sean-k-mooney | powervm support started in ocata and was finished in queens. powerkvm i think predates powervm which is there own hypervior | |
| 12:34:12 | sean-k-mooney | stephenfin: i think we should reopen https://bugs.launchpad.net/nova/+bug/1909972 | |
| 12:34:13 | openstack | Launchpad bug 1909972 in OpenStack Compute (nova) "a number of tests fail under ppc64el arch" [Low,Won't fix] | |
| 12:35:15 | sean-k-mooney | stephenfin: we cant just declare ppc64el as unsupported when it has been supported for years | |
| 12:35:40 | sean-k-mooney | stephenfin: we even have a mulit arch dfg downstream that is workign to get support for power into osp | |
| 12:38:17 | sean-k-mooney | stephenfin: we event ship ppc version of osp in osp 16 https://access.redhat.com/documentation/en-us/red_hat_openstack_platform/16.0/html/release_notes/chap-introduction | |
| 12:38:59 | sean-k-mooney | ppc64le to be specific | |
| 12:46:11 | zigo | At least, LibvirtBlockInfoTest.test_get_disk_mapping_stable_rescue_ide_cdrom can be fixed... | |
| 12:46:24 | zigo | LibvirtBlockInfoTest.test_get_disk_mapping_rescue_with_config too. | |
| 12:46:58 | zigo | Test result is {'bus': 'scsi', 'dev': 'sda', 'type': 'cdrom'} while it's expecting {'bus': 'ide', 'dev': 'hda', 'type': 'cdrom'} | |
| 12:47:25 | sean-k-mooney | yes | |
| 12:47:47 | lyarwood | yeah that's my bad, smells like I didn't mock out the arch check somewhere | |
| 12:47:49 | gibi | sean-k-mooney: could we ask IBM to include unit and functional test in their CI? then we could have coverage and properly fix the test | |
| 12:48:14 | sean-k-mooney | gibi: we can always ask but not sure if it will happen :) | |
| 12:48:17 | zigo | This is also a problem on arm, FYI. | |
| 12:48:27 | zigo | (the above 2 unit tests failing...) | |
| 12:48:34 | sean-k-mooney | i might reach out to our multi arch folks downstream and see if they can help | |
| 12:49:04 | sean-k-mooney | we do package nova for ppc64el downstream and we have limited downstream ci for it | |
| 12:49:54 | sean-k-mooney | presumably that shoudl included unit and functional test but they may still be priamlary working with 16 which is train or relying on x86 runs | |
| 12:50:00 | sean-k-mooney | for unit and functional | |
| 12:50:50 | sean-k-mooney | zigo: for arm we can test that in the gate | |
| 12:51:07 | sean-k-mooney | zigo: we have limited ablity to run thigns in the arm check pipeline | |
| 12:51:24 | sean-k-mooney | well we can run anything we want in that pipeline just limited capsity | |
| 12:51:34 | sean-k-mooney | running unit and funcitonal there i think would be ok | |
| 12:51:55 | sean-k-mooney | ill ask infra and can submit a patch to do that | |
| 12:52:25 | sean-k-mooney | actully openstack-qa woudl be better | |
| 12:55:51 | zigo | Great ! | |
| 12:56:24 | lyarwood | what was the bug for that again? | |
| 12:57:04 | lyarwood | nvm | |
| 12:57:52 | openstackgerrit | Lee Yarwood proposed openstack/nova master: libvirt: Mock get_arch during some stable rescue unit tests https://review.opendev.org/c/openstack/nova/+/769916 | |