16:00:12 #startmeeting cinder-nova-api-changes 16:00:13 Meeting started Thu Dec 14 16:00:12 2017 UTC and is due to finish in 60 minutes. The chair is ildikov. Information about MeetBot at http://wiki.debian.org/MeetBot. 16:00:14 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 16:00:17 The meeting name has been set to 'cinder_nova_api_changes' 16:00:22 johnthetubaguy jaypipes e0ne jgriffith hemna mriedem patrickeast smcginnis diablo_rojo xyang1 raj_singh lyarwood jungleboyj stvnoyes 16:01:04 @! 16:01:05 <_pewp_> jungleboyj (*´・д・)ノ 16:01:08 o/ 16:01:30 good morning gentlemen :) 16:01:45 Good morning. 16:02:06 we can quickly go through things I think 16:02:37 we have a bug fixing round that has bits and pieces in both Nova and Cinder for the new attach code 16:02:48 #link https://review.openstack.org/#/q/topic:bug/1737779+(status:open+OR+status:merged) 16:03:14 there is a cinder one now too 16:03:18 #link https://review.openstack.org/#/c/527852/ 16:03:22 https://review.openstack.org/#/c/527852/ 16:03:28 ahead of me 16:03:37 mriedem: yep, just didn't have the link handy :) 16:03:46 i'll update ^ once i'm out of meetings 16:03:48 and ML threads 16:03:51 and tc channels 16:03:53 oh my 16:03:57 mriedem: Great and then I will review. 16:03:59 mriedem: thanks for stepping up on that one, I wasn't sure whether or not we would want to drop attachment_specs 16:04:53 mriedem: sorry about the meeting :( :) 16:05:23 mriedem: as for the Nova side, do we have anything else beyond testing on those patches? 16:05:42 just needs reviews 16:05:49 i already asked for that in the nova meeting earlier today 16:05:56 i'll bug melwitt later 16:06:03 ah ok, missed that part 16:06:05 thanks! 16:06:46 as for reviews, we have the Cinder side spec for multi-attach still open: https://review.openstack.org/#/c/523608/ 16:07:09 with a recent comment to move it Rocky from tommylikehu 16:07:35 jungleboyj: what's your take on this? 16:08:59 ildikov: In multiple meetings. 16:09:27 jungleboyj: ok, no worries, we can get back to it, when you have a minute 16:09:39 mriedem: I started to update the libvirt patch 16:09:40 ildikov: Why was that put in there? 16:10:11 jungleboyj: I'm not sure I understand your question 16:10:17 jungleboyj: the spec is not approved yet 16:10:29 i haven't read the cinder spec yet 16:10:37 i wasn't sure why there was a cinder spec honestly 16:10:43 It's up to jungleboyj, but I think we will still want that spec even though it's past spec freeze. 16:10:46 jungleboyj: and it seems we're over some deadline now that would suggest to move it Rocky now? 16:10:54 i thought nova just needed the shared_targets field on the volume to disconnect safely 16:10:54 mriedem: Someone told jgriffith to write one. 16:10:57 * smcginnis looks at ildikov 16:11:15 I was just following orders 16:11:15 is there anything in the cinder spec that nova actually needs to add multiattach support? 16:11:23 mriedem: we talked about that spec two weeks ago or three, johnthetubaguy had comments that got fixed 16:11:55 mriedem: it's mainly about the policies and how to specify 'multiattach' at volume creation and after 16:12:00 I don't think the spec was that far off and I think that we were close to having the pieces in place to integrate for Rocky. 16:12:02 ok 16:12:14 if it's mostly policy knobs, sure 16:12:16 I belive the idea was to make sure to get written down some of our assumptions and decisions so we didn't have to rehash it 100 times. 16:12:18 mriedem: not a long read, if you have a minute to glance through if you see anything odd, I think it should be ok b now 16:12:19 Just 50. 16:12:22 Long story short, I think we need to get the spec in for Queens. 16:12:35 smcginnis: 75 16:12:35 you need the policy rules for sure 16:12:52 because some nova backends don't support multiattach so deployments will need to turn it off 16:12:57 jgriffith: 60, that's my final offer. 16:13:05 smcginnis: jgriffith: my bad, I like having some notes on what we agreed to do :) 16:13:06 smcginnis: sold 16:13:23 ildikov: you weren't the only one that asked for it so don't worry about it 16:13:30 ildikov: No, it does make sense. We just need someone to approve it. 16:13:34 60 is a nice number :) 16:13:49 and it is good to have it written down so somebody can work it out 16:14:09 jgriffith: yeah, I guess I wanted notes I can follow... :) 16:15:13 mriedem: the only argument we had was handling 'multiattach' as volume type at creation time and then blocking the ability to modify it once the volume is attached 16:15:27 I will go approve it if that means we can move on. 16:15:43 jgriffith: ildikov if I approve it do we think we can still get it in, in queens? 16:15:45 i think definitely on the latter 16:16:02 changing a multiattach volume that is in-use likely screws up stuff on the nova side 16:16:03 jungleboyj: that would be the idea 16:16:17 409 IMO 16:16:23 jungleboyj: mriedem has been missing some sleep for a few days/weeks now to make it happen for Queens 16:16:25 ildikov: Ok, I will review and we should target approving today. 16:16:39 mriedem is the best! 16:16:50 mriedem: yeah, we cannot play with the domain xml, or at least I don't want to 16:17:24 mriedem: so no modifying that value once the volume is attached, if it's back to available then I don't care who does what with it until the next attach 16:17:35 jungleboyj: cool, thanks 16:17:48 ildikov: same 16:17:54 and yes, mriedem is the best!!! 16:17:56 i'll have to read the spec about the volume type thing 16:18:09 i thought we said last time that the existing boolean flag on the volume was enough 16:18:16 even though the volume type might not support it 16:18:26 maybe fail the volume create in that case? 16:18:33 or fail the PUT /volumes change 16:18:40 mriedem: it is still a flag on the volume from the volume object perspective 16:18:40 Can we retrype a volume when it is attached? 16:18:46 jgriffith: ^^^ 16:18:47 jungleboyj: yes 16:18:51 that's the swap volume API in nova 16:18:55 mriedem: jgriffith clarified it in the spec as I got confused as well 16:19:04 Holy shit. Didn't know that. 16:19:12 retype in cinder posts to the swap volume api in nova 16:19:29 to attach the new block device in the guest and remove the old one 16:19:50 Ok, so we don't want to allow that on multi-attached volumes? 16:20:13 jungleboyj: we don't want to allow that for multi-attach type 16:20:17 jungleboyj: read the spec :) 16:20:29 jgriffith: Going. That is fine with me. 16:21:24 mriedem: That's just retype with migration, right? 16:22:27 smcginnis: That should be the case yes 16:22:36 smcginnis: which you may recall :) 16:22:38 just saying 16:22:43 * jungleboyj has the spec open and is going to approve. 16:22:59 after reading again. 16:23:06 mriedem: Any concerns with that? 16:24:42 sorry, in tc 16:24:57 smcginnis: it's also migration w/o retype 16:25:16 swap volume is called from cinder if you (1) retype a volume or (2) don't retype but migrate to another backend 16:25:21 as far as i understand 16:25:23 But just plain old retype is fine and would make it very simple to change the flag. 16:25:33 i think you' can't do 1 without 2 16:25:37 but you can do 2 without 1 yes? 16:25:43 mriedem: migrate is a form of retype, but yes, we're saying the same thing 16:25:50 There's 1) retype, 2) retype with --migration-policy on-demand, and there's 3) migration. 16:28:53 smcginnis: +1 to easily change the flag when the conditions are met 16:29:21 i'll try to read the spec this afternoon 16:29:22 Ok, the spec says flat out that you can't retype. I think that is the safest thing. 16:29:25 if jungleboyj can hold off on the approval 16:29:30 smcginnis: I left the migrate command out because it shouldn't include any changes to the multiattach type setting 16:29:31 mriedem: Yep, will do. 16:29:33 i'm way overcommitted atm 16:29:40 and need to start actually doing some stuff 16:29:59 smcginnis: there's no retype involved, so it should just keep the same setting etc, but maybe not 16:30:11 mriedem: fair enough 16:30:13 jgriffith: Yeah, I think the same settings in that case. 16:30:34 mriedem: I think the libvirt patch is mainly ready except moving up the support-matrix.ini changes 16:30:55 So #1 should be no big deal, #2 would have to raise SorryCantDoThat(), and migrate isn't involved. 16:31:12 mriedem: there was a concern on how to puch down the 'multiattach' info to the libvirt driver, which I sorted out by putting it into the connection_info dict 16:31:17 smcginnis: yeah, that's what I came up with at least 16:31:34 jgriffith: ++ 16:31:35 mriedem: that might be the only questionable thing in it 16:32:07 smcginnis: jgriffith: +1 16:32:21 puch down? 16:32:24 is that hungarian? 16:32:41 smcginnis: ++ 16:32:42 doesn't seem to be :) 16:32:42 pass down? 16:32:56 ildikov: will have to look 16:33:06 ildikov: maybe leave a comment on the patch where you have a question? 16:33:08 to highlightit 16:33:50 I am good on the Spec unless we want to clarify the retype considerations but I think being conservative there is fine. Don't need to respin. 16:34:05 jungleboyj: We can always do an update later. 16:34:23 ++ 16:34:27 mriedem: I will add a comment to be sure 16:35:00 mriedem: I just would like to be sure we don't have concerns with that part anymore 16:35:07 jungleboyj: line #100 - 110 16:35:53 actually just 107 16:36:16 tab opened to https://review.openstack.org/#/c/523608/ for later 16:36:31 jgriffith: I saw that but I thought that was only a problem if it is retype with migration ? 16:36:47 jgriffith: But I am fine with leaving it as is. 16:37:38 mriedem: thanks! 16:38:07 jungleboyj: ahh, I see; yeah we can add some clarification in there; but in-use should cover that case IMO 16:38:43 jgriffith: Ok. I think staying on the safe side is best. 16:38:56 retyping an in-use volume is the problem, not necessarily migrating, migrating an in-use volume however is no bueno, but that's a result of retype. Geesh... so simple 16:40:02 Yep! 16:45:18 ok, it seems that we're on the same page with this 16:45:26 or are there any further concerns? 16:45:49 I don't think so. 16:46:00 Just need to get the spec through and keep working things. 16:46:03 ok, cool 16:46:32 jgriffith: I guess you're out of bandwidth regarding anything written in the Cinder spec 16:46:46 jgriffith: is that a valid assumption? 16:46:57 ildikov: I'm just out of bandwidth to spend a year polishing the turd 16:47:37 jgriffith: fair enough I guess 16:47:38 so yes, that's valid; I won't be picking up that work any time in the next few weeks 16:48:03 I started it last week, but I won't get back to it any time soon I don't think 16:48:37 jgriffith: if there are any bits and pieces I can start from plz point me to it 16:48:45 jgriffith: otherwise I'll figure it out 16:49:04 ildikov: yeah, I'll shoot you some info in Cinder channel 16:49:24 jgriffith: sounds good 16:49:28 jgriffith: thank you 16:49:31 ildikov: Thank you. 16:50:08 ok, I don't have anything more for today 16:50:14 anything from anyone else? 16:50:52 ildikov: I don't think so. Thank you for all you do! 16:51:12 jgriffith: Thank you for continuing to help. I know this has been a big turd to polish. 16:51:19 jungleboyj: just trying to keep it together :) 16:51:31 :-) 16:51:39 ok, let's close this for today and do some work 16:51:45 thanks everyone! 16:51:46 ildikov: ++ 16:51:57 #endmeeting