| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-08-19 | |||
| 11:20:21 | sean-k-mooney | althoguh i think we more or less had covered the pci stuff with numa | |
| 11:20:25 | sean-k-mooney | yes | |
| 11:20:30 | sean-k-mooney | that is what i ment | |
| 11:20:32 | stephenfin | with the neutron fixture mostly, yeah | |
| 11:20:41 | stephenfin | we had no SR-IOV with neutron coverage | |
| 11:20:46 | sean-k-mooney | correct | |
| 11:20:59 | sean-k-mooney | that is why i did not add the functional tests in the first plance | |
| 11:21:27 | sean-k-mooney | it was trying to get it laned befor m3 and it then got pushed to early train | |
| 11:21:56 | sean-k-mooney | anyway got a link i can review | |
| 11:24:57 | stephenfin | just touching up the last few bits, making sure stuff passes | |
| 11:25:45 | sean-k-mooney | we dont have a good way to simultate different config right i know i did an example that everyone hated | |
| 11:26:01 | sean-k-mooney | but can we easily simulate different hardware toplogies on idfferent nodes | |
| 11:26:13 | sean-k-mooney | we would just need different pci info objects right | |
| 11:26:22 | stephenfin | yup, exactly | |
| 11:27:46 | sean-k-mooney | i was wondering how easy it would be to create a repoducer for https://review.opendev.org/#/c/738432/3 | |
| 11:29:00 | sean-k-mooney | i feel like it will be a little tricky as i dont think we mock the eth device lookup but it also might be simipale if that all i have to mock | |
| 11:29:30 | sean-k-mooney | we just need the netdev names to be different on each host which i think we can do with mocks | |
| 11:30:11 | stephenfin | that's the main reason for doing these tests | |
| 11:30:22 | stephenfin | I haven't got a test case for that yet though | |
| 11:30:25 | stephenfin | baby steps and all that | |
| 11:30:52 | sean-k-mooney | i know how that bug happens i repoduced it on my local setup and the v1 patch fixed it | |
| 11:30:59 | sean-k-mooney | i should proably test v3 | |
| 12:09:49 | openstackgerrit | Merged openstack/os-vif master: update tox envs and support pdf docs https://review.opendev.org/728037 | |
| 12:21:52 | openstackgerrit | Lee Yarwood proposed openstack/nova master: rbd: Move rbd_utils out of libvirt driver under nova.storage https://review.opendev.org/746904 | |
| 12:21:52 | openstackgerrit | Lee Yarwood proposed openstack/nova master: Add ability to download Glance images into the libvirt image cache via RBD https://review.opendev.org/574301 | |
| 12:28:51 | stephenfin | lyarwood: I know you've started that already, but does this rbd direct download scheme make sense for other virt drivers? | |
| 12:29:05 | stephenfin | i.e. did you give thought to my suggestion on making this virt driver-configurable? | |
| 12:29:42 | stephenfin | sean-k-mooney: what's the expected behavior rn if you try to live migrate an instance with PCI devices attached? | |
| 12:30:01 | sean-k-mooney | via an alias | |
| 12:30:05 | stephenfin | yes | |
| 12:30:09 | sean-k-mooney | it should be rejected in the conductor | |
| 12:30:16 | openstackgerrit | Hao Sun proposed openstack/nova master: Add checks for volume status when rebuilding https://review.opendev.org/734734 | |
| 12:30:35 | stephenfin | Hmm, I'm seeing the request request fail with NoValidHost | |
| 12:30:44 | sean-k-mooney | yep | |
| 12:30:50 | sean-k-mooney | well let me check | |
| 12:30:54 | stephenfin | oh, so it's not validated at the API layer | |
| 12:30:55 | stephenfin | ? | |
| 12:30:55 | sean-k-mooney | i think that could happen | |
| 12:31:08 | sean-k-mooney | i dont think so | |
| 12:31:27 | sean-k-mooney | https://github.com/openstack/nova/blob/master/nova/conductor/tasks/live_migrate.py#L229-L264 | |
| 12:31:39 | lyarwood | stephenfin: yeah it is possible to do this in other virt drivers | |
| 12:31:47 | sean-k-mooney | we dont know at the api if the hosts are new enough | |
| 12:31:53 | sean-k-mooney | lyarwood: actully its not | |
| 12:31:58 | lyarwood | yeah it is | |
| 12:32:02 | sean-k-mooney | lyarwood: it would be possible to do it in other deivers | |
| 12:32:11 | sean-k-mooney | but you cant live migrate with pci device period | |
| 12:32:28 | lyarwood | sean-k-mooney: I'm replying to stephenfin from earlier about rbd imports | |
| 12:32:36 | sean-k-mooney | the pci-sig sriov spec does not provide a fasicalty for it | |
| 12:32:38 | sean-k-mooney | oh :) | |
| 12:32:38 | lyarwood | sean-k-mooney: not whatever you're talking about now | |
| 12:32:40 | lyarwood | :D | |
| 12:32:43 | sean-k-mooney | ok ignore me then | |
| 12:32:51 | stephenfin | we always do 0:) | |
| 12:33:20 | lyarwood | oh look it's princess eugenie | |
| 12:33:52 | sean-k-mooney | stephenfin: anyway _check_can_migrate_pci is the only check we do and thats in the conductor | |
| 12:34:05 | lyarwood | https://i.pinimg.com/originals/c6/42/73/c64273010126b4ca904a776dc7a079e7.jpg <- context sorry that was a very British joke | |
| 12:34:28 | stephenfin | heh | |
| 12:34:37 | sean-k-mooney | silly hats are silly | |
| 12:35:06 | sean-k-mooney | stephenfin: you could add a check for pci aliase safely in the api | |
| 12:36:28 | sean-k-mooney | stephenfin: we currently do the check in this loop after schduling https://github.com/openstack/nova/blob/cff7382fb3de4eef7aeddc34b6ae3409cbe50ea0/nova/conductor/tasks/live_migrate.py#L523-L567 | |
| 12:36:50 | sean-k-mooney | which eliminates all the hosts and results in a no valid hosts i guess | |
| 12:38:45 | sean-k-mooney | eventually _check_not_over_max_retries fires a MaxRetriesExceeded exception which gets translated to no valid host | |
| 13:21:15 | openstackgerrit | Hao Sun proposed openstack/nova master: Add checks for volume status when rebuilding https://review.opendev.org/734734 | |
| 13:24:20 | gmann | lyarwood: Bionic will not be tested by default any more so we can say it is not tested so might work or might not. devstack drop the support (remove from supported distro list ) in next cycle from when we upgrade the distro so W cycle will be to drop the Bionic support in devstack. | |
| 13:28:31 | sean-k-mooney | for vitoria we need it since we need it for grenade? | |
| 13:28:43 | lyarwood | gmann: kk, so for Nova we'd only bump our MIN_LIBVIRT_VERSION to the focal version in W | |
| 13:28:49 | sean-k-mooney | but form W on it will be 20.04 on both sides right | |
| 13:29:38 | sean-k-mooney | lyarwood: i think kasyap has a workflow for this. when ever i have tried to change it i have got shouted at but apparently he is blessed | |
| 13:30:01 | sean-k-mooney | lyarwood: are you thinkihng of bumping the min version to 6 | |
| 13:30:02 | gmann | sean-k-mooney: yes, in V it will be Bionic->bionic upgrade testing and in W Focal->Focal | |
| 13:30:51 | sean-k-mooney | gmann: yep because we dont do distro upgrades in grenade so we need to support bionic for one release longer | |
| 13:31:15 | gmann | lyarwood: if no compatible issue then it is fine to bump in W which give upgrade compatibility also | |
| 13:31:27 | sean-k-mooney | by the way that is not a suggestion to start doing disto upgrade in grenade :) | |
| 13:34:56 | lyarwood | sean-k-mooney: yeah, would be nice to move to >= 6.0.0 etc in W | |
| 13:35:24 | lyarwood | gmann: yeah there shouldn't be any issues moving to 6.0.0 in W assuming we also drop support for el7 distros | |
| 13:35:51 | gmann | ok | |
| 13:35:56 | sean-k-mooney | i think we need to bump https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L228 so to 6 | |
| 13:36:18 | aarents | Hi there, lyarwood FYI, I resummit with the missing tempest config : https://review.opendev.org/#/c/743708/ thks! | |
| 13:36:45 | sean-k-mooney | lyarwood: sorry https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L235 | |
| 13:38:01 | lyarwood | aarents: looking | |
| 13:38:35 | sean-k-mooney | https://github.com/openstack/nova/commit/1b76c0f72ee1c1175a089b83a41e4ada94570820 | |
| 13:38:51 | sean-k-mooney | so we should be moving to 5.0.0 in v ithink based on this | |
| 13:39:16 | sean-k-mooney | we bumped NEXT_MIN_LIBVIRT_VERSION in ussuri | |
| 13:39:18 | sean-k-mooney | to 5.0.0 | |
| 13:39:38 | sean-k-mooney | lyarwood: we need a similar patch in V for W | |
| 13:42:04 | sean-k-mooney | looking at https://wiki.openstack.org/wiki/LibvirtDistroSupportMatrix and https://repology.org/project/libvirt/versions | |
| 13:42:15 | sean-k-mooney | 6.0.0 should be ok for W | |
| 13:43:48 | sean-k-mooney | the only really concern i see would be debian buster(10) has 5.0.0 but bullseye(11) has 6.0.0 | |
| 13:52:06 | lyarwood | sean-k-mooney: yeah doesn't look like bullseye actually has a confirmed release date | |
| 13:52:23 | lyarwood | sean-k-mooney: still raising NEXT_MIN_LIBVIRT_VERSION in V is still correct | |
| 13:52:41 | sean-k-mooney | lyarwood: yes raising next min is correct | |
| 13:52:55 | sean-k-mooney | you can also technicaly bump the min version to 5.0.0 | |
| 13:53:34 | sean-k-mooney | we have 5.4 from ubuntu cloud archive on bionic | |
| 13:54:34 | sean-k-mooney | lyarwood: actully we have 6.0.0 | |
| 13:54:45 | sean-k-mooney | in the ussuri repos | |
| 13:54:48 | sean-k-mooney | http://ubuntu-cloud.archive.canonical.com/ubuntu/dists/bionic-proposed/ussuri/main/binary-amd64/Packages | |
| 13:54:50 | lyarwood | we do? | |
| 13:54:52 | lyarwood | ah | |
| 13:54:54 | sean-k-mooney | yep | |
| 13:55:48 | sean-k-mooney | its in both proposed and updates repos | |