| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2022-03-29 | |||
| 08:59:25 | opendevreview | Stephen Finucane proposed openstack/nova master: objects: Remove 'NovaObjectDictCompat' from 'Service' https://review.opendev.org/c/openstack/nova/+/835595 | |
| 13:37:08 | bauzas | folks, I have a problem with the nova meeting today | |
| 13:37:33 | bauzas | since Europe is now on DST, our meeting will be at 4pm UTC but 6pm for me | |
| 13:37:54 | bauzas | but I need to go to someway by 7pm | |
| 13:38:41 | sean-k-mooney | ack we coudl just cancel it since the ptg is next week | |
| 13:38:54 | bauzas | so either we have a small meeting until 4.15pm UTC or someone else could continue to discuss it | |
| 13:38:55 | dansmith | ++ | |
| 13:38:57 | sean-k-mooney | unless there is imporant topics that cant wait | |
| 13:38:59 | dansmith | TC also canceled this week | |
| 13:39:13 | bauzas | I have a small agenda at the time | |
| 13:39:31 | bauzas | at least for asking folks to provide their PTG topics | |
| 13:39:38 | bauzas | https://wiki.openstack.org/wiki/Meetings/Nova#Agenda_for_next_meeting | |
| 13:39:53 | sean-k-mooney | ok do you want to just skip the normal standing itmes then and have a short meetign to cover that topic and wrap early | |
| 13:40:04 | bauzas | sean-k-mooney: we could do this | |
| 13:40:35 | bauzas | sean-k-mooney: just opening the meeting for the PTG | |
| 13:40:51 | bauzas | and then I'll provide a email thread | |
| 13:45:36 | noonedeadpunk | hey there! I was wondering - any reason we do this https://opendev.org/openstack/nova/src/branch/master/nova/virt/libvirt/driver.py#L986 ? | |
| 13:47:14 | noonedeadpunk | Why I'm asking is bassically after talk in #virt regarding MPX flag, that makes IceLake CPUs being identified as Broadwell..... | |
| 13:47:47 | noonedeadpunk | I mean - there's masking of flags being already implemented | |
| 13:48:08 | noonedeadpunk | so technically comparing provided model to virsh capabilities is not needed | |
| 13:50:06 | sean-k-mooney | if you have configured a cpu model that is not compatibale with the host the agent shoudl refuse to start | |
| 13:50:19 | sean-k-mooney | so we have to check cpu compatiabity on agent start up | |
| 13:50:57 | sean-k-mooney | noonedeadpunk: we have some patches in flight to use the newer libvirt api to do this which takes into account which flags are masked or emulated | |
| 13:51:03 | noonedeadpunk | So I have intel xeon gold 6338 and trying to provide Icelake, but only Broadwell is available for it | |
| 13:51:13 | noonedeadpunk | oh | |
| 13:51:54 | sean-k-mooney | presumable icelake reference cpu feature flags that are not present on the host correct | |
| 13:52:03 | sean-k-mooney | perhaps tsk | |
| 13:52:06 | sean-k-mooney | *tsz | |
| 13:52:10 | noonedeadpunk | well, the only "streight" way is to add dozen of extra flags to cover difference... | |
| 13:52:10 | sean-k-mooney | ... tsx | |
| 13:53:04 | noonedeadpunk | sean-k-mooney: so the falg that is missing is MPX which is sunset since 2019... But #virt said they can't jsut drop it for $reasons | |
| 13:53:36 | noonedeadpunk | and they just suggested to mask it as they don't have other option | |
| 13:55:01 | sean-k-mooney | ya https://paste.opendev.org/show/bmfNxhnzVYp54W9LQK0J/ so mpx is listed in the cpu model | |
| 13:55:56 | noonedeadpunk | and I mean - https://lore.kernel.org/lkml/tip-eb012ef3b4e331ae479dd7cd9378041d9b7f851c@git.kernel.org/ and Intel also said that since 2019 no CPU will have it | |
| 13:56:53 | noonedeadpunk | so I'm kind of... either use cpu model from 2014 or have dozens of flags specified in extras... | |
| 13:57:05 | noonedeadpunk | or manually edit cpu_map which is too bad | |
| 13:57:22 | sean-k-mooney | well that is a abi break on intels part which libvirt could accoutn for by adding a no_MPX verison of the cpu model | |
| 13:57:32 | sean-k-mooney | but looking at https://opendev.org/openstack/nova/src/branch/master/nova/virt/libvirt/driver.py#L982-L1008 | |
| 13:57:56 | sean-k-mooney | we coudl rework this os that we apply the extra cpu flags to the model before we do the compat test | |
| 13:58:21 | sean-k-mooney | that woudl allow you to add cpu_model_extra_flags=-mpx | |
| 13:59:01 | sean-k-mooney | and provide that combination was valid it would proceed | |
| 13:59:28 | noonedeadpunk | well, self._get_cpu_info() is jsut `virsh capabilities`? | |
| 13:59:29 | sean-k-mooney | right now it checks that the each model is valid on its own and then the default mode + the extra flags is valid | |
| 13:59:47 | noonedeadpunk | and `virsh capabilities` falls back to Broadwell | |
| 14:00:02 | noonedeadpunk | because all else have that damn MPX | |
| 14:00:24 | sean-k-mooney | yes again its an abi break by intel effectivly | |
| 14:00:59 | sean-k-mooney | this might be address by the new cpu compare api that we are moving too | |
| 14:01:07 | sean-k-mooney | but if not we can rework that validation | |
| 14:01:17 | sean-k-mooney | the curertn two pahse apprch is not really correct | |
| 14:01:25 | sean-k-mooney | it predates the ablity to remove flags | |
| 14:01:42 | sean-k-mooney | when we added the ability to remove flags this shoudl have been factored into one loop | |
| 14:02:09 | sean-k-mooney | for now yes your only option is to ude broadwall and add the missing flags | |
| 14:04:00 | noonedeadpunk | sean-k-mooney: ok, thanks a lot for help) | |
| 14:04:16 | noonedeadpunk | hopefully it will be better soon ) | |
| 14:04:52 | sean-k-mooney | well the currnt patch has been draging on for 3 cycle it might be betere to just rework this check as a simple backportable bugfix | |
| 14:05:47 | noonedeadpunk | tbh that would be perfect but I can't say I see easy way how to re-work it | |
| 14:07:46 | sean-k-mooney | something like this https://paste.opendev.org/show/b6PI9EzL6ucJUN7FKIBL/ | |
| 14:08:28 | sean-k-mooney | that would allow you to set cpu_model=Icelake cpu_model_extra_flags=-mpx | |
| 14:08:47 | sean-k-mooney | at least i think that would work | |
| 14:09:13 | sean-k-mooney | the important thing is we apply the flag modification to the model before we ask libvirt to see if its compatible | |
| 14:11:24 | sean-k-mooney | that quick refactor woudl loose some granualrity in the error message but i think it would allow better flexiblity | |
| 14:12:36 | sean-k-mooney | we could just print the requested feature and the host features and let it to the reader to do the intersection to get back most of the the previously useful info form the error | |
| 14:13:57 | noonedeadpunk | But I think it won't affect _get_cpu_info result? | |
| 14:14:24 | noonedeadpunk | as libvirt still would provide Broadwell as an aswer to that | |
| 14:14:55 | noonedeadpunk | and we're comparing flags requested vs cpu model provided but virsh capabilities iirc | |
| 14:15:05 | sean-k-mooney | _get_cpu_info i think shoudl provide the host cpu info | |
| 14:18:54 | noonedeadpunk | not really I guess | |
| 14:18:59 | noonedeadpunk | ant least not on Xena | |
| 14:21:03 | noonedeadpunk | sean-k-mooney: I added some logging https://paste.opendev.org/show/b0AFUQdxdydGrXHQHn6o/ | |
| 14:21:52 | noonedeadpunk | so _get_cpu_info for me looks exactly as `virsh capabilities` output | |
| 14:22:28 | sean-k-mooney | yes we use libvirt as our interface to collect host infomation | |
| 14:22:48 | noonedeadpunk | or welll | |
| 14:22:55 | noonedeadpunk | flags are not from broadwell | |
| 14:23:11 | sean-k-mooney | the flags should be the same or simlar to lscpu | |
| 14:23:25 | noonedeadpunk | ok, yes, then your suggestion can make sense | |
| 14:24:11 | sean-k-mooney | the model that is listed is the closet model that matches the cpu if i undersand correctly | |
| 14:25:51 | noonedeadpunk | yeah, I think you're right here | |
| 14:53:27 | ade_lee | dansmith, so what changed to make https://review.opendev.org/c/openstack/nova/+/831844 pass? | |
| 14:53:51 | dansmith | ade_lee: the wait-for-sshable I think | |
| 14:54:14 | ade_lee | ah ok | |
| 14:54:26 | dansmith | there were a number of changes to various tests to make that work, but that seemed to be the problem with the quick boot-attach-detach workflow which was in a number of places | |
| 14:56:01 | ade_lee | dansmith, excellent - we're going to run the cinder and glance fips tests again now to see if they pass | |
| 15:01:59 | dansmith | I was seeing those qemu crashes in cs8, so I'm interested to see if those still pop up in cs9 now that this always-fail stuff is resolved | |
| 15:13:25 | gmann | ade_lee: dansmith device tag test also is now ssh-able so may be that making it pass. | |
| 15:13:43 | gmann | recently ssh-able | |
| 15:36:05 | bauzas | reminder (special for EU folks) : nova meeting in 25 mins here | |
| 15:36:18 | bauzas | we'll have a very short meeting | |
| 16:00:06 | opendevmeet | The meeting name has been set to 'nova' | |
| 16:00:06 | opendevmeet | Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. | |
| 16:00:06 | opendevmeet | Meeting started Tue Mar 29 16:00:06 2022 UTC and is due to finish in 60 minutes. The chair is bauzas. Information about MeetBot at http://wiki.debian.org/MeetBot. | |
| 16:00:06 | bauzas | #startmeeting nova | |
| 16:00:12 | bauzas | hola everybody | |
| 16:00:23 | gmann | o/ | |
| 16:00:25 | bauzas | let's do a very quick meeting (up to 15 mins) as I need to move on | |
| 16:00:27 | elodilles | o/ | |
| 16:00:30 | gmann | sure | |
| 16:00:37 | bauzas | DST change here, lovely | |
| 16:00:40 | gibi | o/ | |
| 16:00:56 | bauzas | if someone wants to continue discussing, I can pass the chair baton | |
| 16:01:19 | bauzas | but given next week will be PTG, I don't think we have a lot of things to discuss | |