14:02:24 #startmeeting PowerVM Driver Meeting 14:02:25 Meeting started Tue Apr 17 14:02:24 2018 UTC and is due to finish in 60 minutes. The chair is edmondsw. Information about MeetBot at http://wiki.debian.org/MeetBot. 14:02:26 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 14:02:29 The meeting name has been set to 'powervm_driver_meeting' 14:02:42 #link https://etherpad.openstack.org/p/powervm_driver_meeting_agenda 14:03:28 efried esberglu ut? 14:03:40 yep 14:03:49 ō/ 14:03:51 #topic In-Tree Driver 14:03:54 #link https://etherpad.openstack.org/p/powervm-in-tree-todos 14:04:05 I understand we're in a runway now 14:04:18 excited to see how that goes 14:04:20 https://etherpad.openstack.org/p/nova-runways-rocky 14:04:20 Everything through localdisk is +1 from edmondsw, +2 from efried, and in the runway 14:04:40 any comments yet? 14:05:06 nope 14:05:09 (just checked) 14:05:10 Nope. Target date for that is 4/30 so I'm expecting them to start coming in soon 14:05:34 anything else to note here then? 14:05:34 We've been doing runways for all of two weeks, but tbh I haven't really noticed it prompting a significant increase in reviewer focus. 14:05:58 I'm thinking of suggesting a "champion" idea 14:06:14 efried what would that be? 14:06:14 I've been prioritizing multinode CI over finishing the cold mig/resize patch, nothing to report there 14:06:31 esberglu that makes sense... CI is a prereq for that 14:06:41 where we assign one reviewer (core or not, I guess, but not the owner of the series) to go pester reviewers to look at the changes, and pester the owner to follow up as needed. 14:07:06 kinda like we champion invention disclosures on IDTs. 14:07:13 who pesters folks to be a champion? ;) 14:07:29 Touché 14:07:42 Part of the process; add it to runway, assign champion at that time. 14:07:50 just an idea I'm toying with. 14:08:00 yep, I think that can work 14:08:13 can become a nova meeting agenda item 14:08:52 #topic Out-of-Tree Driver 14:09:03 #link https://etherpad.openstack.org/p/powervm-oot-todos 14:09:16 we've made some progress on the iscsi changes 14:09:41 I just merged one, which put the others into merge conflict since they're not in a series 14:10:00 I told prashkre to rebase the other that I +2/+A'd and when he does that I'll fast approve it 14:10:30 I did give him a comment to go address on the 3rd iscsi commit 14:10:49 he's off and running on that 14:11:10 once those all merge, I intend to tag 6.0.1 14:11:11 prashkre proposed openstack/nova-powervm master: Return iSCSI Initiator for VIOSes https://review.openstack.org/557800 14:11:52 I also just setup a meeting with the PowerVC storage guys to talk about community contribution for multiple initiator support, among other things 14:12:08 anything else to discuss around iscsi? 14:12:45 the volume refactor will need to be rebased and updated once these changes merge 14:13:02 and I still need to dig into the concerns that the PowerVC team raised on that 14:13:16 so probably more updates there 14:13:25 tjakobs FYI ^ 14:13:36 anything else for OOT? 14:14:17 some other things need to happen, like adding MSP support, but I haven't gotten to anything there 14:14:23 We could have done the iscsi changes in a series and not had to rebase. 14:14:38 efried I suggested that a couple times, but nobody did it, so... 14:14:50 Not sure how much more prashkre is going to be doing for us, but might be worth a little tutorial on that. 14:15:27 efried I'd love to see something written up on that for some wiki somewhere 14:15:34 and then just link it with things like that 14:15:37 I'm sure there are documents galore. 14:15:46 I'm sure... question is where, and how good are they 14:15:50 There's a whole man page for git restack 14:15:55 which is what I use *daily* 14:16:04 if we find a good one, great, just need to keep in our back pocket and start linking 14:16:47 maybe just link that then 14:16:58 I'm familiar with rebase, but not restack 14:17:19 which is probably why I'm confused about managing series myself 14:18:05 esberglu said he tried restack and couldn't get the hang of it. 14:18:19 but I love it. 14:18:32 you play with it enough to figure out the quirks, probably 14:18:38 efried: I think it's because I didn't check out the latest patch. I would check out the latest in the series, git restack 14:18:48 Then change anything I want to edit to edit 14:18:54 And go from there? 14:19:20 yes, using git rebase --continue whenever you're ready to move to the next one up. 14:19:38 Okay I see where I went wrong now, thanks 14:19:49 But yeah, it's key to check out the top patch in the series. 14:20:06 top = the first one that should merge? 14:20:13 or is that bottom? 14:20:22 Top = last one that should merge 14:20:45 ^ 14:21:00 ok, so if I want to change something 3 patches down... 14:21:11 I check out the top, and then what? 14:22:02 git restack 14:22:08 it puts you into an editor just like git rebase -i 14:22:20 You change 'pick' to 'edit' for any patches in the series you want to change. 14:22:36 Save/quit. You'll be shoved down the chain into the first 'edit' patch. 14:22:45 cool 14:23:04 Make your changes, and run git rebase --continue. It'll commit your changes and put you into the commit message editor. 14:23:17 Save/quit that guy and you'll be shunted to the next 'edit' patch... 14:23:23 ...unless there's a merge conflict with one in between 14:23:29 in which case regular merge conflict process 14:23:32 so you don't need to 'git commit' at all, it does that for you with rebase --continue? 14:23:33 when done, git rebase --continue. 14:23:45 yes. You can do the commit yourself if you like 14:23:48 just need to git add after you change something, and then rebase --continue? 14:23:57 either way. 14:24:05 cool, got it 14:24:13 moving on 14:24:15 note that it doesn't come in the box. 14:24:19 you have to pip install git-restack 14:24:25 it's something the os infra folks wrote, I think. 14:24:26 oh, good to now 14:24:35 #topic Device Passthrough 14:24:44 efried what's the latest? 14:24:57 Jay and I had a knock-down drag-out; now we're asking for an impartial referee to decide who wins. 14:25:13 you couldn't take him? 14:25:17 ;) 14:25:33 So many possible responses to that. 14:25:42 I was trying to set you up easy... 14:25:54 At this point I would be content either way tbh. We've wrangled both of our proposals into compromise territory. 14:26:37 I still have a preference for mine, but I won't be seriously mad if it goes the other way. 14:26:45 k 14:26:48 I made progress on the granular patch. 14:27:04 I've got the bugs worked out to where it has parity with the existing functionality 14:27:19 That is, if you're not using granular syntax, all the old scenarios still work when they're funneled through the new algorithm. 14:27:29 I'll be in microversion rebase hell until it merges. 14:27:53 Next step will be to write tests; and once the above argument is settled, to fold in that functionality. 14:28:01 Because it's looking likely that we're going to want to do it in the same microversion. 14:28:11 Jay admonished me for not waiting until his NRP stuff was done 14:28:16 and then started working on it again 14:28:21 Which was kind of the goal. 14:28:24 :) 14:28:26 so goodness there. 14:28:58 So progress is being made overall. Otherwise, nothing new to report specifically on device passthrough as it pertains to PowerVM. 14:29:13 oh 14:29:21 except I don't remember if I mentioned this 14:29:44 I wrote a couple of patches that use upt in libvirt to solve the age-old problem of sharing providers being double-reported. 14:30:03 I wouldn't have done it, except they took like 15 minutes and basically validated everything I wrote last cycle. 14:30:34 bhagyshris is going to take over the patches to fix up the tests and get everything ready to merge. 14:30:42 cool 14:30:50 https://review.openstack.org/560444 14:30:50 https://review.openstack.org/560459 14:30:52 I'm sure the KVM folks thank you 14:30:57 we'll see :) 14:31:27 I think that's all I have for now. 14:31:45 I'm more than happy to see things fixed for libvirt, since Power cares about that as well as PowerVM 14:32:02 one thing I was going to mention 14:32:53 I think we've said before that we need to consider infiniband and other types of passthrough as well as GPU 14:33:07 RoCE is another 14:33:49 and more than just passthrough, virtualization like (or using, I hope) SR-IOV 14:34:00 We're getting set up for allll of that. 14:34:04 except that I want to do it the right way, not the way we currently do SR-IOV 14:34:15 so just throwing that out there... stay tuned 14:34:19 yep yep 14:34:41 #topic PowerVM CI 14:34:51 #link https://etherpad.openstack.org/p/powervm_ci_todos 14:34:53 esberglu ? 14:35:30 I've got a change out for the powervm-ci side of multinode. Just adds local.conf files and updates prep_devstack.sh to work with AIO, control, and compute cases 14:35:30 6470 14:35:58 It doesn't affect any existing jobs so I'm ready to merge if you guys are okay with it, will make neo-os-ci dev easier to have that in 14:36:19 esberglu ack 14:36:36 We gonna get Mujahid in here moving forward? 14:36:57 On neo-os-ci I've got to the point where I can have the control node stack, then kick off another job to have the compute stack 14:37:10 efried yeah, I'll ask him to join, though I expect we'll have more CI discussions in slack going forward 14:37:25 ight 14:37:33 Still not sure what I have to do to force them to be on the same host, but staging CI only has 1 compute node right now 14:38:08 I want to bring neo4 back into staging, but I still have that set up for IT work, and might need it again there 14:38:25 I may just steal a node from production for a little while 14:38:36 esberglu have you asked the Power KVM CI guys if they've faced this problem? 14:38:49 No 14:38:50 do they do multinode in their CI? 14:38:59 I don't think so, I can check 14:39:03 k 14:39:21 The big blocker for multinode is the jenkins dropping connections on the staging environment 14:40:00 Which I'm completely in the dark on, I spent a TON of time trying to figure this out when I was working the queens upgrade 14:40:05 have you asked for help from infra? 14:40:22 it sounds like a possible jenkins issue 14:40:37 Yeah I guess that's the next step 14:40:38 I know they're not hitting it, but they have experience tracking down these problems, right? 14:41:06 edmondsw: Problem with that error message is that it could be a billion different things 14:41:12 and it may very well be a jenkins bug they just haven't hit because of environment differences or something 14:41:19 esbergly yep exactly 14:41:21 But I'll see if they can shed some light 14:41:59 But that is 100% blocking multinode CI right now and is my top priority 14:42:45 Nothing else from me 14:42:58 #topic Open Discussion 14:43:06 anything? 14:43:16 efried: You interested in being included in the PowerVM CI education sessions at all? 14:43:28 um 14:43:33 lol 14:43:42 Does it increase the probability that I will be called upon to work on the CI? 14:43:49 cause then no 14:44:11 I'll leave you be :) 14:44:24 I don't see that happening... but you might still want to be able to plead ignorance anyway :) 14:44:49 better use of your time and all that 14:45:00 alright, if nothing else... 14:45:05 #endmeeting