14:31:48 <edmondsw> #startmeeting PowerVM Driver Meeting
14:31:49 <openstack> Meeting started Tue Apr  3 14:31:48 2018 UTC and is due to finish in 60 minutes.  The chair is edmondsw. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:31:50 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
14:31:52 <openstack> The meeting name has been set to 'powervm_driver_meeting'
14:32:00 <edmondsw> https://etherpad.openstack.org/p/powervm_driver_meeting_agenda
14:32:15 <edmondsw> #topic In-Tree Driver
14:33:09 <edmondsw> efried esberglu ready?
14:33:15 <esberglu> yep
14:33:16 <efried> hit it
14:33:25 <edmondsw> so https://review.openstack.org/#/c/547169/ merged
14:33:33 <edmondsw> but none of the actual function yet
14:33:49 <edmondsw> we have one +2 on the network hotplug
14:34:18 <edmondsw> and there is a runway queued up to look at that, vscsi, and snapshot
14:34:33 <esberglu> vSCSI, Snapshot, base DiskAdapter, and Localdisk are all ready for review
14:34:55 <esberglu> Only other one is resize/cold migrate which is still WIP
14:35:01 <edmondsw> esberglu I think you said yesterday that you rebased something?
14:35:08 <edmondsw> I didn't get a chance to look
14:35:27 <esberglu> edmondsw: I just made everything one commit chain, made the resize/cold migrate stuff easier
14:35:39 <edmondsw> ah cool
14:36:19 <edmondsw> vscsi -> snapshot -> diskadapter -> localdisk -> migr/resize
14:36:43 <esberglu> And responded to the couple comments you left on snapshot and disk adapter, those should be ready for your +1
14:36:57 <edmondsw> esberglu will try to look this afternoon
14:37:07 <esberglu> tx
14:37:07 <edmondsw> I did +1 the vscsi changes
14:37:16 <esberglu> yep
14:37:29 <esberglu> Resize is coming along
14:37:44 <esberglu> Need to figure out if boot from cinder volume works, some of the code paths depend on that
14:37:54 <esberglu> And then UT, Live test, and get CI rolling
14:38:32 <edmondsw> esberglu testing boot from volume would also be good for the vscsi commit, right?
14:39:02 <edmondsw> I don't recall whether there is a separate op for that in the support matrix that we need to mark "complete"
14:39:15 <edmondsw> if it works
14:39:27 <esberglu> edmondsw: Yep I'm stacking right now to test it out
14:39:33 <edmondsw> cool, let us know
14:39:46 <esberglu> It for me
14:40:10 <edmondsw> efried no word on core yet?
14:40:57 <edmondsw> any day now I assume
14:40:58 <efried> nope.  The one week "speak now or forever hold your peace" period technically ended last night.  If melwitt is thinking about it, hopefully she'll flip the switch when she gets on today.
14:41:12 <edmondsw> yep
14:41:16 <edmondsw> ok moving on
14:41:23 <edmondsw> #topic Out-of-Tree Driver
14:41:39 <edmondsw> https://etherpad.openstack.org/p/powervm-oot-todos
14:42:10 <edmondsw> we got pypowervm bump merged and the 2 commits waiting on that as well
14:42:42 <edmondsw> chhagarw is working on iscsi stuff
14:42:59 <chhagarw> yes
14:43:19 <chhagarw> it should be mostly done by tomorrow, adding UTs
14:43:20 <edmondsw> there is a commit about setting fuse_limit for fileio that I'm trying to dig into
14:43:27 <edmondsw> chhagarw cool
14:43:55 <edmondsw> got a response from thorst about fuse_limit that "fileio seems to max at 15…I think"
14:44:17 <edmondsw> so I don't think they're doing that for perf reasons
14:44:44 <edmondsw> so it might actually make sense in nova-powervm and not just in NovaLink... but I need to get more clarity there
14:45:19 <edmondsw> tjakobs rebased the volume refactor but then it has another merge conflict now with other things having merged
14:45:47 <edmondsw> I think that covers OOT unless anyone has anything to add?
14:45:49 <efried> edmondsw: "seems to max" leads me to think this is the result of the same kind of testing the other consumer was seeing.
14:46:01 <tjakobs> edmondsw: noted
14:46:21 <edmondsw> efried you and I should catch hsien today about that
14:46:34 <edmondsw> and get to the bottom of it
14:46:41 <efried> We need to get all the players together to make sure we're really talking about two different things before we send hsien off to fix it one way and drop code on our layer to avoid the bug.
14:47:00 <edmondsw> sure
14:47:11 <efried> edmondsw: Can you find out who in pvc observed the issue?  I think svenkat referenced an RTC bug
14:47:12 <edmondsw> I'm told the other player would be thorst
14:47:25 <efried> thorst who?
14:47:36 <edmondsw> lol
14:47:36 <efried> Oh THAT guy - haven't seen him in ages.
14:48:01 <edmondsw> alright, moving on
14:48:05 <esberglu> Oh one other thing, haven't figured out how to handle duplicate opts yet, but also haven't looked into it at all since last week
14:48:15 <edmondsw> #topic Device Passthrough
14:48:35 <edmondsw> this has been languishing from my perspective... been caught up in other things
14:48:39 <edmondsw> how's it going with you efried?
14:48:48 <edmondsw> update on nova?
14:49:09 <efried> Designs and patches churning along frantically.
14:49:23 <efried> It is becoming clear that we've got way, way, way too much content for Rocky.
14:49:40 <edmondsw> across Nova, or re: placement?
14:49:45 <efried> yes
14:49:49 <edmondsw> :)
14:49:51 <efried> upt is starting to merge.
14:49:59 <edmondsw> \o/
14:50:02 <efried> Still verrry slow progress on nrp-in-alloc-cands.
14:50:31 <efried> I'm thinking I'm going to start grinding on the granular work, because everybody is depending on it; even though I won't be able to tie it off until we get nrp-in-alloc-cands.
14:50:55 <edmondsw> can/should you help more on nrp-in-alloc-cands?
14:51:19 <efried> double-edged sword.  I want to be able to +2 that stuff.
14:51:30 <edmondsw> yeah
14:51:40 <efried> One of our main cores who looks at placement stuff is out for 2w
14:51:45 <efried> (gibi on honeymoon)
14:52:14 <efried> Progress is being made; it's just slower than I would like.
14:52:26 <edmondsw> k
14:52:30 <efried> BTW, there's also been some argument over a design point for granular.
14:52:52 <efried> Per design, if you specify two separate numbered request groups, there's no restriction on whether they will or will not land on the same provider.
14:53:09 <efried> Now people are demanding that separate numbered groups *must* land on separate providers.
14:53:23 <efried> which I think is a mistake for several reasons.  But I'm being outvoted.
14:53:24 <edmondsw> why?
14:53:39 <edmondsw> not why you think it's a mistake
14:53:48 <edmondsw> why others disagree?
14:54:10 <efried> I'm actually not terribly sure, to be honest.  I think it's mainly to satisfy NUMA use cases.
14:54:22 <edmondsw> hmmm
14:54:28 <efried> I mean, I concede that we have to be able to handle both things eventually.
14:54:45 <edmondsw> could you add a way to say THESE providers have to go together but others don't?
14:55:00 <efried> Yeah.  And we're going to have to do that eventually.
14:55:27 <edmondsw> maybe need to accelerate that
14:55:38 <efried> It's a question of whether that option is gonna be called allow_same_provider=<this group, that group> or separate_providers=<this group, that group>
14:55:40 <edmondsw> or at least articulate it
14:55:52 <efried> oh, believe me, I've been articulating my ass off.
14:55:58 <edmondsw> :)
14:56:19 <efried> I'm not sure it's going to affect us, to be honest.
14:56:33 <efried> We can make dev passthrough work whichever way the default shakes out.
14:56:48 <efried> I'm just arguing from a purely architectural standpoint.  And I'm right.  But I'm gonna lose.
14:56:59 <edmondsw> boo
14:57:05 <efried> them's the breaks.
14:57:10 <edmondsw> yeah
14:57:12 <edmondsw> anything else?
14:57:23 <efried> The only thing I still have on my side is that the spec, which is approved, already has it my way.
14:57:31 <efried> so someone is going to have to propose a delta to it.
14:57:42 <edmondsw> -2 it ;)
14:57:42 <efried> And I'm conspicuously dragging my feet on that leetle work item.
14:58:01 <edmondsw> #topic: PowerVM CI
14:58:12 <edmondsw> https://etherpad.openstack.org/p/powervm_ci_todos
14:58:14 <edmondsw> esberglu?
14:58:22 <esberglu> Upgraded all the systems last week
14:58:30 <esberglu> Everything is running really smoothly
14:58:38 <edmondsw> yay
14:58:53 <esberglu> Only failure that is hitting with any sort of consistency is the vopt naming conflict thing
14:58:59 <esberglu> But even that is really low %
14:59:42 <edmondsw> I'm gonna have to join another mtg in a min... esberglu take over this one?
14:59:44 <esberglu> I got some multinode CI runs going for OOT, all manual, nothing automated yet
14:59:50 <esberglu> edmondsw: Sure
14:59:51 <esberglu> http://184.172.12.213/manual/2node/
15:00:03 <esberglu> But it's mostly passing, hitting a few errors on resize I haven't looked into yet
15:00:10 <edmondsw> :)
15:00:19 <esberglu> "Specified UUID is already used by other partitions"
15:01:15 <esberglu> edmondsw: Wanted to talk about my move to PowerAI in open discussion, but we can do it later today so you can be there
15:01:26 <efried> nice
15:01:29 <edmondsw> I just got a 5 min reprieve so go ahead
15:01:39 <efried> Do we want to do that here or internal?
15:01:39 <esberglu> That's all for CI
15:02:00 <esberglu> Actually internal may be better
15:02:23 <edmondsw> esberglu any more progress on the move to queens?
15:02:38 <esberglu> edmondsw: Haven't touched it, working multinode has taken priority
15:02:48 <edmondsw> sure, agreed on that
15:03:07 <edmondsw> #topic Open Discussion
15:03:08 <edmondsw> anything?
15:03:18 <esberglu> nada
15:03:53 <edmondsw> alright, thanks guys
15:03:54 <edmondsw> #endmeeting