| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2024-09-24 | |||
| 16:31:12 | bauzas | (gibi): Seeking approval for: Support hw_vif_model igb | |
| 16:31:18 | bauzas | gibi: your turn | |
| 16:31:19 | gibi | o/ | |
| 16:31:43 | gibi | so new enough libvirt and qemu supports to emulate a new nic type igb (old intel nic) | |
| 16:31:57 | gibi | the big benefit of it is for testing | |
| 16:32:08 | bauzas | yuuup | |
| 16:32:10 | gibi | as igb support SRIOV with 7 VFs per PF | |
| 16:32:35 | gibi | nova support hw_vif_model image property to ask for a nic type | |
| 16:32:44 | gibi | I'd like to extend the supported values there with ig | |
| 16:32:45 | gibi | ig | |
| 16:32:47 | gibi | igb | |
| 16:32:59 | gibi | https://blueprints.launchpad.net/nova/+spec/igb-vif-model | |
| 16:33:10 | bauzas | two questions | |
| 16:33:11 | gibi | https://review.opendev.org/q/topic:%22bp/igb-vif-model%22 | |
| 16:33:19 | bauzas | 1/ does it require a new os-vif release ? | |
| 16:33:22 | gibi | no | |
| 16:33:44 | bauzas | 2/ I guess you have a plan for rolling-upgrades ? | |
| 16:34:28 | gibi | yes. there is a new trait for the new igb vif model, libvirt driver will expose it only if both libvirt and qemu is new enough to support it | |
| 16:34:46 | gibi | and the image properies pre-filter adds the trait to the placement query | |
| 16:34:54 | bauzas | and then we'll schedule the instance with pre-filtering with that trait | |
| 16:35:02 | gibi | yepp | |
| 16:35:06 | bauzas | okay, sounds the right pattern to me | |
| 16:35:15 | tkajinam | +1 | |
| 16:35:22 | gibi | the code is up with documentation included | |
| 16:35:35 | bauzas | I don't see any other design concern | |
| 16:35:37 | gibi | it impacts os-traits and nova, + glance doc | |
| 16:35:46 | bauzas | yup, saw the series | |
| 16:36:11 | gibi | it is unfortunately non backportable due to ovo change | |
| 16:36:17 | bauzas | I honestly like the approach of a specless blueprint if we can avoid a spec just for the trait thing | |
| 16:36:39 | bauzas | there is no upgrade impact, since all the support will be made additive | |
| 16:37:02 | gibi | yepp | |
| 16:37:05 | bauzas | and there are no API changes, despite an image property additon | |
| 16:37:14 | gibi | yepp | |
| 16:37:42 | gibi | frickler was super nice and proposed cirros update to have the igb kernel module in cirros available | |
| 16:37:56 | bauzas | pardon my ignorance but do we have flavor extraspecs with hw:vif_model ? | |
| 16:38:13 | gibi | there is no flavor extra spec for that | |
| 16:38:20 | bauzas | perfect | |
| 16:38:43 | gibi | so I will be able to add a tempest test that actully proves that VFs can be created in the guest | |
| 16:38:57 | bauzas | okay, I don't see any reason to not accept that blueprint as specless, but is anyone having concerns with the design ? | |
| 16:39:50 | tkajinam | no | |
| 16:39:57 | bauzas | let's be it. | |
| 16:40:05 | gibi | \o/ | |
| 16:40:13 | bauzas | #agreed https://blueprints.launchpad.net/nova/+spec/igb-vif-model approved as specless | |
| 16:40:22 | bauzas | I'll do the paperwork right after the meeting | |
| 16:40:26 | gibi | I will ask for review on the os-traits patch first as I need a release there to be able to bump it in nova | |
| 16:40:31 | bauzas | ++ | |
| 16:40:39 | tkajinam | the feature sounds quite cool. | |
| 16:40:41 | gibi | https://review.opendev.org/c/openstack/os-traits/+/928582 | |
| 16:40:46 | tkajinam | I'm wondering if any of the node provider in CI infra and provide nodes with the flavor so that we can test sriov features in gate :-) | |
| 16:40:52 | bauzas | I honestly prefer we keep such features as specless as possible | |
| 16:40:59 | gibi | tkajinam: that is the long term goal | |
| 16:41:20 | gibi | tkajinam: if the node providers have this patch we can freely ask for SRIOV without using hardware | |
| 16:41:35 | bauzas | gibi: I assume noble is recent enough to ship the right libvirt versions? | |
| 16:41:52 | gibi | throughput is bad as it is emulated, but we can cover our SRIOV logic | |
| 16:41:55 | clarkb | johnsom has been discussing this with us and the main issue is the lack of support from nova aiui | |
| 16:42:10 | bauzas | clarkb: this will be a solved problem soon | |
| 16:42:17 | gibi | bauzas: I need to double check as I use debian sid | |
| 16:42:22 | clarkb | bauzas: not really because then we have to wait a few years for the clouds to upgrade | |
| 16:42:22 | tkajinam | I assume that is for sriov amphora. | |
| 16:42:22 | bauzas | cool | |
| 16:42:32 | gibi | tkajinam: yes | |
| 16:42:43 | clarkb | but yes the first step is fixing nova then seeing what we can do about upgrading nova (but that is unlikely to happen quickly) | |
| 16:42:53 | gibi | clarkb: understood. | |
| 16:43:15 | bauzas | clarkb: noted | |
| 16:43:24 | bauzas | and thanks, didn't thought about it | |
| 16:43:27 | gibi | btw without the nova fix one could use igb in local devstack to test SRIOV | |
| 16:43:35 | tkajinam | libvirt 10.0.0 and qemu 8.2.2 | |
| 16:43:44 | tkajinam | these are what package search for noble tells me | |
| 16:43:45 | bauzas | sounds noble | |
| 16:43:55 | bauzas | iirc | |
| 16:43:58 | gibi | tkajinam: good we need 9.3 and 8.0 | |
| 16:44:04 | clarkb | when discussing it with johnsom I think we decided that centos 9 stream is also new enough | |
| 16:44:11 | clarkb | for libvirt and qemu | |
| 16:44:16 | bauzas | yeah | |
| 16:44:31 | bauzas | c9s ships libvirt-10 IIRC | |
| 16:44:32 | tkajinam | yeah centos has been updating libvirt/qemu quite regularly | |
| 16:44:52 | bauzas | okay, I think we're done | |
| 16:45:10 | bauzas | thanks all, let me give you 15 mins back of your time | |
| 16:45:14 | tkajinam | can I ask one quick question about a different topic ? | |
| 16:45:19 | bauzas | shoot then, | |
| 16:45:20 | johnsom | Glad to hear there is progress on this. I opened a bug for the nova/igb issue in launchpad a while ago | |
| 16:45:38 | bauzas | I can drink my beer later | |
| 16:45:49 | tkajinam | I've re-proposed spec for SEV-ES support to E. The patches are up but haven't get enough review really https://review.opendev.org/c/openstack/nova-specs/+/928817 | |
| 16:46:02 | gibi | johnsom: I picket it up couple of weeks ago as I also need this for testing | |
| 16:46:10 | johnsom | +1 | |
| 16:46:11 | bauzas | tkajinam: I'll add it to the status etherpad | |
| 16:46:19 | johnsom | https://bugs.launchpad.net/nova/+bug/2077195 | |
| 16:46:32 | tkajinam | I wonder if it can be approved quickly or do I need any steps before that ? I can bring it to vPTG but I'm not aware of actual topics we can/should discuss | |
| 16:46:36 | bauzas | tkajinam: we still need to agree on some spec review days at the PTG | |
| 16:46:53 | bauzas | it was approved in D, right ? | |
| 16:47:03 | tkajinam | yes it was. | |
| 16:47:30 | bauzas | okay, there is a specific fast-approval process for pre-approved specs that are just straighly forwarded | |
| 16:47:39 | bauzas | I can surely help then that situation | |
| 16:48:07 | bauzas | particularly given you have patches up | |
| 16:48:20 | tkajinam | thanks. because this may still need some rounds of reviews if it can be approved early then that'd be helpful | |
| 16:48:41 | tkajinam | (so that it can be in the review list | |
| 16:48:45 | bauzas | if the spec is just reproposed without changes, this will be quick | |
| 16:49:27 | tkajinam | yeah. the reporposed spec is exactly same ... but I noticed I have to update the history section maybe. | |
| 16:49:38 | bauzas | oh you'll nee | |
| 16:49:40 | bauzas | need | |