| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-08-30 | |||
| 08:52:13 | bauzas | :p | |
| 08:54:00 | bauzas | stephenfin: FWIW, I'll start looking over the rest of the series later in the afternoon | |
| 08:55:23 | sean-k-mooney | by the way thanks to donnyd we now have a multi numa lable in the gate with nested vert adn teh ablity to use hugepages,dpdk,and pinning testing in the gate... os i have a ci job to go right | |
| 08:55:40 | sean-k-mooney | *write | |
| 08:56:34 | bauzas | cool | |
| 08:57:41 | sean-k-mooney | oh by the way i mentioned this to artom alreay but i also figured out how to do nested pci passthough over the weekend | |
| 08:57:59 | sean-k-mooney | e.g. pci passthough from a host all the way to an l2 guest | |
| 08:58:35 | sean-k-mooney | and i might be able to use virtual device in the l1 guest as pci deivce to the l2 guest also. | |
| 08:59:21 | sean-k-mooney | nova cant do it yet due to how we currently generate teh xml but i might be able to use that to test sriov or pci passthough at some point. | |
| 08:59:58 | kashyap | sean-k-mooney: On nested PCI passthrough, have you seen this past blog from DanPB: https://www.berrange.com/posts/2017/02/16/setting-up-a-nested-kvm-guest-for-developing-testing-pci-device-assignment-with-numa/ | |
| 09:00:05 | stephenfin | bauzas: Not always terrible. Just mostly terrible ;) | |
| 09:00:32 | bauzas | like proposing to ski over a weekend ? | |
| 09:00:54 | bauzas | I can assume that | |
| 09:01:04 | stephenfin | that was one of the non-terrible ones | |
| 09:01:05 | sean-k-mooney | kashyap i think a few years ago | |
| 09:01:46 | sean-k-mooney | kashyap: once we got a virtula iommu and q35 we could technically do it | |
| 09:02:07 | sean-k-mooney | but the issue is getin ghte right pci tree such that the passthough devices are in there own iommu group | |
| 09:03:03 | sean-k-mooney | and ya i have been advocating creating a pci root per numa node for years at this point | |
| 09:03:21 | kashyap | sean-k-mooney: On virtual IOMMU, there was a talk on its implementation earlier this year at FOSDEM: | |
| 09:03:24 | kashyap | https://archive.fosdem.org/2019/schedule/event/vai_iommu_implementation_using_hw_nested_paging/ | |
| 09:03:54 | kashyap | (It's a different approach to existing impl.) | |
| 09:04:07 | sean-k-mooney | ok | |
| 09:08:04 | sean-k-mooney | at this point if qemu supported a virtual device that had sriov support we would have everything we need to be able to test sriov in the gate | |
| 09:08:40 | sean-k-mooney | well assuming we lanned all the reuqired changes in openstack and a cloud provier or two ran that version and hadn the correct flavors | |
| 09:10:00 | kashyap | sean-k-mooney: QEMU supporting an emulated SRIOV device is the key bit -- needs someone familiar with a card to implement it. | |
| 09:10:11 | kashyap | I won't count on it, though | |
| 09:10:30 | sean-k-mooney | there were patches for it before | |
| 09:10:34 | sean-k-mooney | they just never landed | |
| 09:11:32 | sean-k-mooney | kashyap: https://github.com/knuto/qemu | |
| 09:11:50 | kashyap | sean-k-mooney: Yes, I know the author. He works for Oracle | |
| 09:12:00 | kashyap | Last I checked he had a student working on it...but the student left | |
| 09:12:23 | kashyap | Knut ('knuto') introduced the supporting infrastructure for adding an emulated SRIOV card | |
| 09:22:53 | openstackgerrit | Luyao Zhong proposed openstack/nova master: object: Introduce Resource and ResouceList objs https://review.opendev.org/678448 | |
| 09:22:54 | openstackgerrit | Luyao Zhong proposed openstack/nova master: Add resources dict into _Provider https://review.opendev.org/678449 | |
| 09:22:54 | openstackgerrit | Luyao Zhong proposed openstack/nova master: Retrive the allocations early https://review.opendev.org/678450 | |
| 09:22:55 | openstackgerrit | Luyao Zhong proposed openstack/nova master: Claim resources in resource tracker https://review.opendev.org/678452 | |
| 09:22:55 | openstackgerrit | Luyao Zhong proposed openstack/nova master: libvirt: Enable driver configuring PMEM namespaces https://review.opendev.org/678453 | |
| 09:22:56 | openstackgerrit | Luyao Zhong proposed openstack/nova master: libvirt: report VPMEM resources by provider tree https://review.opendev.org/678454 | |
| 09:22:56 | openstackgerrit | Luyao Zhong proposed openstack/nova master: libvirt: Support VM creation with vpmems and vpmems cleanup https://review.opendev.org/678455 | |
| 09:22:57 | openstackgerrit | Luyao Zhong proposed openstack/nova master: Parse vpmem related flavor extra spec https://review.opendev.org/678456 | |
| 09:22:57 | openstackgerrit | Luyao Zhong proposed openstack/nova master: Add functional tests for virtual persistent memory https://review.opendev.org/678470 | |
| 09:26:10 | openstackgerrit | Luyao Zhong proposed openstack/nova master: Retrive the allocations early https://review.opendev.org/678450 | |
| 09:26:10 | openstackgerrit | Luyao Zhong proposed openstack/nova master: Claim resources in resource tracker https://review.opendev.org/678452 | |
| 09:26:11 | openstackgerrit | Luyao Zhong proposed openstack/nova master: libvirt: Enable driver configuring PMEM namespaces https://review.opendev.org/678453 | |
| 09:26:11 | openstackgerrit | Luyao Zhong proposed openstack/nova master: libvirt: report VPMEM resources by provider tree https://review.opendev.org/678454 | |
| 09:26:12 | openstackgerrit | Luyao Zhong proposed openstack/nova master: libvirt: Support VM creation with vpmems and vpmems cleanup https://review.opendev.org/678455 | |
| 09:26:12 | openstackgerrit | Luyao Zhong proposed openstack/nova master: Parse vpmem related flavor extra spec https://review.opendev.org/678456 | |
| 09:26:13 | openstackgerrit | Luyao Zhong proposed openstack/nova master: Add functional tests for virtual persistent memory https://review.opendev.org/678470 | |
| 09:30:56 | openstackgerrit | Balazs Gibizer proposed openstack/nova master: Add cold migrate and resize to nova-grenade-multinode https://review.opendev.org/679210 | |
| 09:37:18 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Provide HW_CPU_X86_AMD_SEV trait when SEV is supported https://review.opendev.org/638680 | |
| 09:39:59 | sean-k-mooney | just tested the v22 iteration of artoms patches and its still working correctly | |
| 09:49:23 | openstackgerrit | Brin Zhang proposed openstack/nova master: Add operator user_id/project_id to the migrations https://review.opendev.org/679413 | |
| 10:08:10 | sean-k-mooney | for those not on the neutron channel i just confrimed that cold migration and live migration with vnic_type=direct-physical is broken | |
| 10:09:40 | sean-k-mooney | there are at least two related issues, 1 we do not update the mac on migration and second we do not correctly find the device to detach it on live migration | |
| 10:09:58 | sean-k-mooney | live migration also like has the first issue but its failing before it gets that far | |
| 10:12:55 | sean-k-mooney | it looks like detach works correctly for suspend and the live migration code should be useing the same fucntion i think but maybe its not | |
| 10:14:35 | sean-k-mooney | im going to try and fix this next week and ill file the upstream bug shortly. i think the cold migration behavior is broken all the way back to queens. the live migration behavior is obviouly new since we added that this cycle | |
| 10:25:12 | stephenfin | aspiers: Hit that whole SEV series. Couple of comments left, but ping me when you've tackled them and I'll re-review | |
| 10:36:51 | artom | sean-k-mooney, hey, so how feasible would be to hack together a tempest job that runs https://review.rdoproject.org/r/#/c/18832/ on that new FN flavor? | |
| 10:36:57 | artom | ('morning!) | |
| 10:37:34 | sean-k-mooney | artom: well my nfv test job did actully work a few months ago so i could make it work on FN pretty quick | |
| 10:37:42 | sean-k-mooney | as in today | |
| 10:37:54 | sean-k-mooney | but im currently reworking functional test | |
| 10:38:07 | sean-k-mooney | want me to give it ago when im done | |
| 10:38:52 | artom | sean-k-mooney, that'd be awesome :) whitebox needs a bunch of configs to be set, I can help with those | |
| 10:39:26 | artom | https://review.opendev.org/#/c/678887/ failed with KVM errors, so I'm trying think of other possibilities | |
| 10:39:54 | sean-k-mooney | ya well first thing would be to just get it stacking multinode with a few smoketests. then we can add the plugin | |
| 10:40:47 | artom | Right, baby steps :) | |
| 10:40:58 | sean-k-mooney | artom: i dont see errors in the n-cpu log | |
| 10:41:15 | sean-k-mooney | oh multinode | |
| 10:41:40 | sean-k-mooney | still no erros on the other node | |
| 10:42:03 | artom | sean-k-mooney, no, they're in the qemu logs | |
| 10:42:06 | artom | ex: https://zuul.opendev.org/t/openstack/build/905b52e0b1e84532944f0606b5455381/log/logs/subnode-2/libvirt/qemu/instance-00000005.txt.gz | |
| 10:42:33 | sean-k-mooney | oh thats the kernel bug i mentioned to donnyd | |
| 10:42:54 | sean-k-mooney | actully may be not | |
| 10:43:12 | sean-k-mooney | it could be but that looks like an issue with the cpu model | |
| 10:43:29 | donnyd | i am using host-passthrough on nova | |
| 10:43:41 | sean-k-mooney | its checking for svm with is amds version of vt-x | |
| 10:43:54 | sean-k-mooney | donnyd: ya this is on vexhost | |
| 10:44:01 | sean-k-mooney | but i think this is a different issue | |
| 10:44:07 | donnyd | oh | |
| 10:44:25 | sean-k-mooney | ya so | |
| 10:44:27 | sean-k-mooney | [libvirt] | |
| 10:44:28 | sean-k-mooney | live_migration_uri = qemu+ssh://stack@%s/system | |
| 10:44:30 | sean-k-mooney | cpu_mode = none | |
| 10:44:32 | sean-k-mooney | virt_type = kvm | |
| 10:45:04 | sean-k-mooney | we should be setting cpu_mode=custom and cpu_model=kvm64 in devstack | |
| 10:45:22 | sean-k-mooney | with cpu_mode=none we default to host-model | |
| 10:45:40 | sean-k-mooney | but host-model does not always detect the capablity proably espcailly with nested virt | |
| 10:45:58 | openstackgerrit | Brin Zhang proposed openstack/nova master: Filter migrations by user_id/project_id https://review.opendev.org/674243 | |
| 10:46:02 | artom | sean-k-mooney, I'll just add that to Matt's devstack nuke | |
| 10:46:10 | sean-k-mooney | so this host whihc is likel intel x86 was detected as support svm which from the kvm error it is not | |
| 10:46:22 | sean-k-mooney | cool | |
| 10:46:32 | artom | But coffee first | |
| 10:46:43 | sean-k-mooney | kvm64 is the most portable cpu model that work with kvm | |
| 10:47:02 | artom | Daughter started school. I'm guess I'm just excited/stressed for her. | |
| 10:47:20 | sean-k-mooney | first time | |
| 10:47:24 | artom | Gym attendance took a hit as well | |
| 10:47:26 | artom | Yeah | |
| 10:47:58 | kashyap | sean-k-mooney: IIRC, there was a reason we went with existing CPU model 'qemu64' | |