| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-06-05 | |||
| 09:14:11 | kashyap | I'll link you shortly, unless someone beats me to it | |
| 09:25:52 | frickler | xiaolin: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-discuss , just send a mail with the subject above. you may want to subscribe before to avoid your mail being held in moderation | |
| 09:26:21 | gibi | aarents: I don't think nova can add pci-bridges today. If the limit can be avoided by the q35 machine type then I would take that direction instead | |
| 09:28:32 | bauzas | I do agree with gibi | |
| 09:28:39 | bauzas | the 'pc' machine type is old | |
| 09:28:47 | bauzas | very old, even | |
| 09:28:56 | bauzas | and we shouldn't modify nova because of it | |
| 09:33:23 | xiaolin | frickler: Thanks, I got it :-) | |
| 09:38:37 | aarents | bauzas: gibi oh by simply switch from pc to q35 increase slot availiable? I was thinking we need to 'declare' add some pci-bridge device into root-brigde before have benefit of mode slots, maybe libvirt do it magicly will do some test | |
| 09:39:00 | bauzas | aarents: try with this machine type | |
| 09:39:10 | aarents | yep | |
| 09:55:02 | slaweq | gibi: hi | |
| 09:56:18 | slaweq | ok | |
| 09:56:18 | slaweq | gibi: works for me | |
| 09:56:18 | gibi | slaweq: let's use the nova room. some of the folks had problem with jitsi so we are using zoom | |
| 09:56:18 | slaweq | gibi: do You want to meet today in Nova or Neutron's room? | |
| 09:56:18 | gibi | slaweq: hi! | |
| 09:56:19 | gibi | cool, thanks | |
| 10:23:44 | kashyap | frickler: Thanks! :) | |
| 10:25:10 | stephenfin | sean-k-mooney: Look what I just found https://github.com/openstack/nova/blob/master/nova/virt/libvirt/host.py#L1104-L1105 | |
| 10:25:31 | stephenfin | https://stackoverflow.com/a/10429736 | |
| 10:25:37 | stephenfin | \o/ | |
| 11:42:33 | sean-k-mooney | wait what does that even do | |
| 11:43:14 | sean-k-mooney | do we return 0 on kernel 4.x and 5.x? | |
| 11:44:11 | sean-k-mooney | stephenfin: i assume we are just going to nuke that code from orbit and pretend it was never there | |
| 11:46:39 | sean-k-mooney | oh its based on the linux kernel version used to build the python package | |
| 11:46:58 | sean-k-mooney | that even worse | |
| 11:48:34 | gibi | bauzas: left feedback on https://review.opendev.org/#/c/733703 | |
| 11:52:25 | sean-k-mooney | stephenfin: i guess we technically could do sys.platform.startswith('linux') but since we only support linux for the libvirt dirver we should just remove that code. good find | |
| 12:26:40 | bauzas | gibi: ack, thanks | |
| 12:26:51 | gibi | thanks for starting the spec | |
| 12:33:19 | gibi | nova PTG will continue in less than 30 minutes in the Juno Zoom room with neutron-nova cross project session first and the with cyborg topics | |
| 12:58:17 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Correct reported system memory https://review.opendev.org/733827 | |
| 14:14:14 | bauzas | man, a cross-project session on a Friday late afternoon and discussing about placement workflows, what did I made wrong to have such punishment ? :) | |
| 14:24:07 | dansmith | heh | |
| 14:27:16 | gibi | bauzas: I'm sorry. I owe you a free beer next time we have a face to face PTG | |
| 14:27:40 | bauzas | nah, I have a beer machine at home, I can fix this issue | |
| 14:27:59 | gibi | then we should have the next PTG at your place :D | |
| 14:28:43 | gibi | and then free beer for everybody | |
| 14:29:30 | dansmith | bauzas: I was going to ask if your wife realizes that you're broadcasting her jamming out on her earphones to the world.. | |
| 14:29:54 | bauzas | you did hear her ? | |
| 14:29:59 | bauzas | well, it's encrypted | |
| 14:30:08 | dansmith | no, but we see her | |
| 14:30:11 | bauzas | we call this specific encryption 'French' | |
| 14:30:16 | dansmith | hah | |
| 14:30:24 | bauzas | this way, we're sure nobody can understand us | |
| 14:30:26 | gibi | :) | |
| 14:31:09 | dansmith | hmm, that was some encryption we just heard | |
| 14:31:39 | dansmith | my ciphertext analysis leads to a general message of "No, get out of here, I'm on a call!" | |
| 14:33:15 | bauzas | LOL, yeah, our door doesn't lock unfortunately | |
| 15:21:17 | sean-k-mooney | also sorry to monoplise the cross project so much | |
| 15:21:39 | johnthetubaguy | sorry I missed you all for today, I was in keystone land, and got stuck there | |
| 15:22:01 | gibi | sean-k-mooney: thank you for providing input for all these discussions. | |
| 15:22:28 | sean-k-mooney | my brain is slightly hurting after it all | |
| 15:23:35 | gibi | johnthetubaguy: no worries I think we had proper nova representation on the session | |
| 15:23:51 | TheJulia | random question, are there any thoughts regarding letting a compute driver throttling such that we don't have timeout failures ? Thinking once in a blue moon someone trying to create a 1000 physical machine deployments in ironic and they just want to batch them all up at once. | |
| 15:23:56 | johnthetubaguy | gibi: all good :) | |
| 15:24:58 | sean-k-mooney | TheJulia: there is a config option you can set that kind of does that | |
| 15:25:08 | sean-k-mooney | TheJulia: not sure if it works with ironic | |
| 15:25:11 | TheJulia | oh? | |
| 15:25:27 | sean-k-mooney | https://docs.openstack.org/nova/latest/configuration/config.html#DEFAULT.max_concurrent_builds | |
| 15:25:35 | sean-k-mooney | max concurent builds | |
| 15:25:43 | sean-k-mooney | its a compute agent config | |
| 15:26:08 | johnthetubaguy | until recent releases, the scheduler used to fall over first though, in theory better now, but hope to be testing that more soon | |
| 15:26:10 | sean-k-mooney | so you could set that per ironic conpute service | |
| 15:26:28 | sean-k-mooney | johnthetubaguy: multi create still sucks | |
| 15:26:42 | sean-k-mooney | johnthetubaguy: it has less races with plamcnet | |
| 15:27:00 | sean-k-mooney | but unless we change how retires work anti affingity server groups break things | |
| 15:27:14 | TheJulia | sean-k-mooney: I feel like it works, but I'm wondering if someone has that set to like 100... | |
| 15:27:19 | TheJulia | I guess I need more information | |
| 15:28:15 | sean-k-mooney | TheJulia: in the ironic case there is going to be one hell of a power spike too | |
| 15:28:29 | sean-k-mooney | 1000 servers booting at once | |
| 15:29:28 | TheJulia | sean-k-mooney: some people have the power for this | |
| 15:33:36 | sean-k-mooney | TheJulia: yep but i once rebooted half the server on a rack and triped its local breaker | |
| 15:33:43 | sean-k-mooney | that was fun... | |
| 15:33:55 | TheJulia | yeaaaah | |
| 15:34:05 | TheJulia | Been there, I really overengineered power for every rack I built | |
| 15:34:52 | sean-k-mooney | that was when we discover the electian that install it connected it to the wrong ups and it was signifcaly more over subsribed then the lab team tought | |
| 15:35:09 | sean-k-mooney | the 3 new rack were all on the old ups not the new one in the room next door | |
| 15:35:56 | sean-k-mooney | as a result i was not allowed to power them back on until we got someone in to swap them | |
| 16:08:03 | TheJulia | I'm suddenly reminded of the one time I literally burned out a breaker | |
| 16:08:40 | johnthetubaguy | sean-k-mooney: yeah, anti affinity totally breaks things still, +1 that | |
| 16:09:18 | sean-k-mooney | its because we calusate the alternitive host up front so once one thing files and retire it can cause another thing to fail | |
| 16:09:39 | johnthetubaguy | it gets worse with PCI passthrough | |
| 16:09:48 | johnthetubaguy | or seems to | |
| 16:09:56 | sean-k-mooney | so unless we call the sculder again instead when we do the retry your screwed | |
| 16:10:04 | sean-k-mooney | johnthetubaguy: yes it will | |
| 16:10:58 | sean-k-mooney | johnthetubaguy: that just because we dont track the pci device in placment so when we claim the new candiate we dont calim the pci device and that can cause a reschdule | |
| 16:11:08 | johnthetubaguy | ack | |
| 16:11:20 | sean-k-mooney | johnthetubaguy: we dont need placement to fix that we could allocate the deivce in the conductor be we dont today | |
| 16:12:35 | sean-k-mooney | the pci devices table has 3 states available allocated and claimed i think and the difference between the last too is one is a reservation for a vm that has not yet been created on the host and the other is for when the vm is actully using it | |
| 16:12:52 | johnthetubaguy | oh my... custom resources using the provider yaml totally helps with those races, I think | |
| 16:13:07 | sean-k-mooney | it could | |
| 16:13:21 | sean-k-mooney | but we have no way to associate the RPs with the pci devcices | |
| 16:13:27 | sean-k-mooney | so its close but wont work | |
| 16:13:32 | sean-k-mooney | at least not today | |
| 16:13:42 | johnthetubaguy | because we pick the specific one too early? | |
| 16:14:08 | sean-k-mooney | no we pick the specific on on the compute node but that code has no knolage of placment | |
| 16:14:19 | sean-k-mooney | so it would not pick form the device claimed in palcment | |
| 16:14:25 | sean-k-mooney | it would pick form all devices on the host | |