Monday, 2018-07-09

*** edmondsw has joined #openstack-powervm01:04
*** edmondsw has quit IRC01:08
*** edmondsw has joined #openstack-powervm02:52
*** edmondsw has quit IRC02:57
*** edmondsw has joined #openstack-powervm04:40
*** edmondsw has quit IRC04:44
*** edmondsw has joined #openstack-powervm06:28
*** edmondsw has quit IRC06:33
openstackgerritChhavi Agarwal proposed openstack/nova-powervm master: iSCSI volume detach with no UDID
*** edmondsw has joined #openstack-powervm08:16
*** edmondsw has quit IRC08:21
*** edmondsw has joined #openstack-powervm12:24
*** edmondsw has quit IRC12:29
*** edmondsw has joined #openstack-powervm12:31
*** edmondsw has quit IRC12:35
*** edmondsw has joined #openstack-powervm12:37
*** edmondsw has quit IRC12:42
*** edmondsw has joined #openstack-powervm12:45
*** efried_off is now known as efried12:46
*** efried has quit IRC12:46
*** edmondsw has quit IRC12:49
*** edmondsw has joined #openstack-powervm12:51
*** edmondsw has quit IRC12:53
*** edmondsw has joined #openstack-powervm12:53
*** efried has joined #openstack-powervm13:02
openstackgerritMerged openstack/nova-powervm master: Fix requirements and lower constraints
openstackgerritMerged openstack/nova-powervm master: Bump pypowervm min 1.1.16
openstackgerritMerged openstack/nova-powervm master: Add support for rbd user
edmondswefried when you get a chance:
*** chhagarw has joined #openstack-powervm16:58
chhagarwedmondsw: can we close on this review:
chhagarwI updated the comments16:59
edmondswchhagarw looking17:03
edmondswchhagarw have you asked Gerald to review?17:04
chhagarwyeah, i reminded him as well17:04
edmondswchhagarw posted comments17:12
chhagarwso u mean we need to remove this with block completely
chhagarwno need to check mock_discover17:18
efriededmondsw: I've got that one queued up.  I want to pull it down and try it, but I wanted to clear out my sandbox first.17:32
edmondswefried tx17:33
efriededmondsw: Talk to me about support for attaching GPUs to existing instances.  That's a use case we want to support, nicht wahr?17:34
edmondswefried correct17:37
efriededmondsw: To my knowledge, there's no ComputeDriver interface for "attach PCI device".  Do you know of one?17:38
edmondswno, don't know of one17:38
edmondswand I would not get too hung up on that for the first go around17:38
edmondswI'm not sure if that can be done hot or would require shutdown17:39
edmondswand I don't think it will be super common... more an "eventually I expect this will be wanted" thing17:39
efriededmondsw: In the general case it can certainly be done hot as far as PowerVM is concerned, hot-attaching a slot to a VM.  Don't know about GPUs specifically or what it would take to discover and make usable in the guest.17:40
efriedBut hot or cold, it would be the same operation from a ComputeDriver interface point of view.17:40
edmondswright, powervm can definitely hot-attach PCI... but some individual cards support that and some don't, and I don't know which is the case for current GPUs17:41
efriedUnless now we'd be saying the only support would be via resize.17:41
efriedBut okay, scope of current work is spawn/destroy only.  Rolling with that...17:41
edmondswI had thought it would be via resize because that's just how nova works today, though it is odd that pci is handled differently from storage and networks in that way17:42
efriedI guess they care more about being able to hot-plug those things.17:42
edmondswand yeah, I guess we have to implement something either to make it work or block it, since someone could request more or less PCI via resize with current interface17:43
efriedwe should get it for free via resize, nah?17:44
edmondswthat'd sure be nice17:46
edmondswI don't know, honestly. Will phyp block a hot attach for cards that don't support that, in which case we have to handle some kind of failure?17:47
edmondswor... I don't know17:47
efriedNo, not hot attach.  Resize only applies to cold migrate.  Instance shut down.17:48
efriedYou can't specify a new flavor for live migrate.17:48
efriedThough it raises the question about whether we ought to block or support live migrate with attached GPUs.17:49
efriedGotta say no, because we can't migrate the state of the GPU.17:53
edmondswoh, right, there's a spec for live resize but not currently implemented17:57
edmondswand yes, I think we would have to block live migrate for VMs with PCI devices attachd17:58
openstackgerritChhavi Agarwal proposed openstack/nova-powervm master: iSCSI volume detach with no UDID
efriededmondsw: Should we provide a CONF option denoting the fs path to the yaml config for passthrough?18:11
edmondswefried I'm not sure that's necessary... a well-known name in /etc/nova is probably fine18:20
efriededmondsw: ack18:20
efriededmondsw: Shall we call it device-passthrough.yaml ?18:20
efriedI wouldn't think a powervm- prefix is necessary, since it'll be on a PowerVM compute node.18:21
efriedAnd with any luck, we may become the gold standard that all drivers wind up using :)18:21
edmondswyeah, that'd be nice. I don't see a reason to prefix with powervm-18:24
edmondswmaybe just passthrough.yaml?18:24
efriedshrug, okay.18:32
efriededmondsw: Where (in what doc) do we describe our flavor extra_specs stuff?18:33
efriedI'm not seeing it in nova-powervm/doc/source anywhere.  Do we maintain that somewhere separate?18:33
openstackgerritEric Fried proposed openstack/nova-powervm master: Spec: Device Passthrough
efriededmondsw: ^18:38
edmondswefried I don't see any docs for our extra specs :(18:38
edmondswwe need to fix that18:39
edmondswin the meantime, they're defined here in the code:
edmondswI'll add a TODO to the etherpad to include in docs18:40
edmondswefried hmm... we have a different list of supported extra_specs in-tree:
efriedyup, that's by design.  I remember having to scrape through them when we started porting.18:42
edmondswthese 318:43
edmondswdo you remember why?18:43
efriedenable metric because we... don't.  That would be a ceilometer thing?18:44
efriedI don't know what affinity check is.18:45
efriedAnd ppt ratio because... we don't care?18:45
edmondswI'm not satisfied with any of those answers... I'll try to dig into it18:45
edmondswI did find an existing todo for documenting extra specs18:45
edmondswmriedem had apparently commented on that before:
efriedyup, just found that myself :)18:49
efriedI do remember we were always looking for things to subset when we moved in tree.18:49
efriedTo reduce review footprint and avoid having to answer hard questions about "wtf is this and why do we need it?"18:49
edmondswsure, I just don't expect these would have been left out for those reasons18:50
edmondswppt and affinity are very recent additions... I bet they were only added OOT just because that's what PowerVC uses and the folks that contributed them weren't thinking about IT18:51
edmondswmay be the same for the other as well18:51
efriedah, that could be.  Looking...18:51
edmondswsomething we should be mindful of in future reviews18:51
efriedyup, those three came in after - at least after the in-tree port was proposed.18:53
efriedYeah, well after.18:53
openstackgerritMerged openstack/nova-powervm master: Fix coverage job
*** edmondsw has quit IRC23:26

Generated by 2.15.3 by Marius Gedminas - find it at!