| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-01-10 | |||
| 17:30:49 | melwitt | the infra folks just restarted gerrit | |
| 17:30:53 | melwitt | should be back soon I think | |
| 17:31:23 | kashyap | melwitt: Ah, thanks! | |
| 17:31:29 | kashyap | Yes, it's back now | |
| 17:33:49 | melwitt | dansmith: do you agree that this bp should have a spec at least because it's proposing to add a new flavor extra spec? or do you think it would be ok specless? https://blueprints.launchpad.net/nova/+spec/libvirt-disable-pmu-for-realtime-guests | |
| 17:34:21 | dansmith | will look in a sec | |
| 17:34:41 | melwitt | thanks. just want a sanity check | |
| 17:34:51 | kashyap | stephenfin: Worth respinning for the nit, and losing the +2s? Instead of a follow-up? | |
| 17:34:57 | kashyap | (For this, I mean: https://review.openstack.org/#/c/625216/) | |
| 17:35:18 | stephenfin | kashyap: narp. Follow up | |
| 17:35:26 | stephenfin | +W'd | |
| 17:35:36 | stephenfin | assuming melwitt didn't want to review it? ^ | |
| 17:35:56 | kashyap | Ah, okay, then. Thank you. I do have a follow-up patch in prep for the deprecation notes, etc. | |
| 17:37:42 | kashyap | FWIW, Mel did review the "mini spec" (blueprint), though. So I know she's onboard with the idea :-) | |
| 17:37:51 | kashyap | Alright, I get dinner, then. | |
| 17:39:02 | dansmith | melwitt: I dunno, I have a couple questions about the functionality, which might mean spec. Not sure if we always require a spec for new extra_specs, but generally they mean features which need some discussion | |
| 17:39:17 | dansmith | melwitt: sucks to say that on the spec deadline day, but this was also filed three days ago, so I dunno | |
| 17:40:45 | sean-k-mooney | dansmith: it was a small feature i was just pulling in form work i planned to do in train. i can write a mini spec if people want but i also wont be upset if we defer | |
| 17:40:58 | dansmith | sean-k-mooney: if it's train, then spec and defer | |
| 17:41:01 | dansmith | imho | |
| 17:41:24 | dansmith | oh, this isn't targeted at stein? I figured that's why melwitt was asking | |
| 17:41:39 | sean-k-mooney | it is targeted to stien | |
| 17:42:09 | melwitt | well, sean-k-mooney added it to the etherpad https://etherpad.openstack.org/p/nova-stein-blueprint-spec-freeze near the bottom so I was trying to make sure it got some consideration | |
| 17:42:12 | sean-k-mooney | but down stream it was not planned to be worked on until next cycle | |
| 17:42:24 | sean-k-mooney | i just taugh ti would have time to do it since it is small | |
| 17:42:31 | dansmith | sean-k-mooney: okay I don't see that in lo | |
| 17:42:32 | dansmith | *lp | |
| 17:43:01 | melwitt | yeah, I think only the nova-drivers group can target it | |
| 17:43:25 | dansmith | sean-k-mooney: are you saying that merely having the device exposed incurs the perf penalty? | |
| 17:43:28 | sean-k-mooney | so ya ill write a spec a and propose in train when the window reopens | |
| 17:43:34 | sean-k-mooney | yes | |
| 17:43:37 | dansmith | if so, that sucks, but also I don't really like having the tristate where =none could mean anything | |
| 17:44:07 | dansmith | IMHO, =none should mean =false, although I'm sure you're going to say that it may or may not be enabled today by default, | |
| 17:44:17 | sean-k-mooney | the over all latency cost of virtualiation was a minium latency hit of 60 micro second this redused it to 50 micoro second when turning it off | |
| 17:44:19 | dansmith | but regardless, seems better to be more deterministic to me | |
| 17:44:56 | sean-k-mooney | libvirt automaticaly tunes it on of you use cpu_mode=host-passthorugh and turns it off otherwise | |
| 17:45:09 | sean-k-mooney | so i was just tryign to keep that behavior | |
| 17:45:27 | dansmith | sounds confusing to me | |
| 17:45:45 | dansmith | libvirt or our libvirt driver? | |
| 17:45:56 | sean-k-mooney | well if the pmu is not present things like the linux perf tool dont work fully so i dont want disable by defualt | |
| 17:46:05 | sean-k-mooney | libvirt well qemu | |
| 17:46:43 | sean-k-mooney | if no element is defiend qemu make the decion based on if cpu=host or some other value | |
| 17:46:43 | dansmith | okay | |
| 17:48:40 | sean-k-mooney | basically my plan was if you dont set the extra spec we would not set the libivrt xml value which is what we do today and if you do set it we set the element to what you set explictly | |
| 17:49:41 | dansmith | yeah, understand | |
| 17:50:00 | dansmith | there's some benefit to that for sure, it just sucks to have that be ill-defined I think | |
| 17:50:07 | dansmith | but definitely needs to be documented | |
| 17:51:16 | sean-k-mooney | yep so do we want to leave this to train or woudl it be suffient to document it in the singel patch that impements the feature | |
| 18:18:37 | nicolasbock | Hi. Is it possible to PXE boot a Nova instance? Looking at https://review.openstack.org/#/c/133254/ this possibility was at least considered at some point, but abandoned in the Kilo cycle. The current Nova master doesn't seem to support PXE booting (or specifying a network device as boot device) if I understand the code correctly. But I wanted to make sure that I am not missing something. | |
| 18:19:17 | sean-k-mooney | nicolasbock: i belive you have to enable the boot menu via the image metadata and then hit f1 | |
| 18:19:26 | sean-k-mooney | in the vnc console to select pxe booting | |
| 18:19:37 | nicolasbock | Ah ok | |
| 18:19:49 | nicolasbock | So there is still a manual step involved | |
| 18:20:12 | sean-k-mooney | ya i dont think you can set it to auto pxe boot but i may be wronge | |
| 18:20:14 | nicolasbock | In other words, Nova can not configure the libvirt domain to try booting from net. | |
| 18:20:22 | nicolasbock | Ok | |
| 18:20:25 | nicolasbock | Thanks for the info sean-k-mooney ! | |
| 18:21:01 | sean-k-mooney | perhaps someone else can answer the netboot question but not that i am aware of | |
| 18:26:21 | openstackgerrit | sean mooney proposed openstack/nova-specs master: Add spec for sriov live migration https://review.openstack.org/605116 | |
| 18:26:54 | sean-k-mooney | jaypipes: artom stephenfin bauzas ^ nits addressed | |
| 18:36:35 | openstackgerrit | sean mooney proposed openstack/nova-specs master: Add spec for sriov live migration https://review.openstack.org/605116 | |
| 18:37:58 | openstackgerrit | sean mooney proposed openstack/nova-specs master: Add spec for sriov live migration https://review.openstack.org/605116 | |
| 18:39:00 | sean-k-mooney | ok i had a tab and a repated "as" which i also just fixed that really should be done now | |
| 18:40:05 | sean-k-mooney | https://review.openstack.org/#/c/605116/10..13/specs/stein/approved/libvirt-neutron-sriov-livemigration.rst will show you the delta from the previous version ye reviewed | |
| 18:59:47 | jaypipes | sean-k-mooney: +2 from me. stephenfin can send it on its merry way in a bit. | |
| 19:00:29 | sean-k-mooney | jaypipes: well he could if he was a specs core but he is not | |
| 19:00:39 | sean-k-mooney | jaypipes: thank you however for your +2 | |
| 19:00:39 | jaypipes | ah, then melwitt can :) | |
| 19:00:46 | sean-k-mooney | nope neither is she | |
| 19:00:52 | jaypipes | wha?! | |
| 19:01:00 | sean-k-mooney | jaypipes: dansmith: and bauzas can | |
| 19:01:14 | jaypipes | ok, bauzas it is then, since he's reviewed that already :) | |
| 19:01:40 | melwitt | yeah, I think you could +W it yourself though with a note that bauzas was previously +2 before nit fixing | |
| 19:01:52 | sean-k-mooney | jaypipes: nova-specs-core is a very select group https://review.openstack.org/#/admin/groups/302,members :) | |
| 19:02:07 | melwitt | in case bauzas is gone for the day | |
| 19:04:09 | jaypipes | melwitt: k, I will give bauzas a little while and will +W it along after waiting a bit. | |
| 19:04:45 | melwitt | k, sounds good | |
| 21:41:39 | artom | Booting an instance with cpu_policy=dedicated (and no other NUMA-y stuff) and 1GB of RAM in the flavor results in a rather bizarre libvirt message | |
| 21:41:55 | artom | total memory for NUMA nodes (0x100000) should equal RAM size (0x200000) | |
| 21:42:37 | artom | Even if <memory unit='KiB'>1024</memory> and <numa> <cell id='0' cpus='0' memory='1024' unit='KiB'/> </numa> | |
| 21:42:46 | artom | Same thing but with 2GB is fine. | |
| 21:43:06 | artom | Either I'm missing something way obvious, or there's a bug :/ | |
| 21:47:27 | artom | sean-k-mooney ^^ if your weird sleep schedule makes you still be here. You always know about these things ^_^ | |
| 22:05:53 | sean-k-mooney | artom i wasnt but i am now | |
| 22:06:52 | sean-k-mooney | artom: adding hw:cpu_policy=dedicated create an implict numa toplogy as if you also set hw:numa_nodes=1 | |
| 22:08:19 | sean-k-mooney | that message however is strange | |
| 22:10:35 | openstackgerrit | Inbar Stolberg proposed openstack/nova master: Fixes server rebuild if NumaTopologyFilter is used with CPU pinning https://review.openstack.org/629646 | |
| 22:13:09 | artom | sean-k-mooney, right, I knew about the implicit NUMA topology | |
| 22:13:25 | artom | But... Nova's XML is correct, as far as I can tell, units and everything | |
| 22:13:28 | sean-k-mooney | so i cant find that error upstream | |
| 22:13:33 | sean-k-mooney | what raised it | |
| 22:13:41 | artom | sean-k-mooney, it's from libvirt itsel | |
| 22:13:50 | artom | For all I know it's a bug in qemu/kvm | |
| 23:12:32 | sean-k-mooney | jaypipes: im just about to log off for the night. looks like bauzas didnt come back online. are you still ok to +w https://review.openstack.org/#/c/605116/? | |
| 23:13:13 | jaypipes | sean-k-mooney: yup | |
| 23:13:21 | sean-k-mooney | thanks :) | |
| 23:14:08 | jaypipes | sean-k-mooney: done | |
| 23:15:03 | sean-k-mooney | night all o/ | |
| 23:19:53 | openstackgerrit | Merged openstack/nova-specs master: Add spec for sriov live migration https://review.openstack.org/605116 | |
| #openstack-nova - 2019-01-11 | |||
| 00:42:05 | openstackgerrit | sean mooney proposed openstack/nova master: [WIP] libvirt: auto detach/attach sriov ports on migration https://review.openstack.org/629589 | |