14:11:16 #startmeeting Nova Live Migration 14:11:17 Meeting started Tue Apr 5 14:11:16 2016 UTC and is due to finish in 60 minutes. The chair is PaulMurray. Information about MeetBot at http://wiki.debian.org/MeetBot. 14:11:18 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 14:11:21 The meeting name has been set to 'nova_live_migration' 14:11:28 is anyone still here ? 14:11:40 o/ 14:11:50 o/ 14:11:58 o/ 14:12:34 Sorry I was late starting the meeting 14:12:49 First time - remarkably 14:12:54 agenda here: https://wiki.openstack.org/wiki/Meetings/NovaLiveMigration 14:13:05 #topic Bugs 14:13:30 nothing in the agenda for bugs, anyone got one to bring up? 14:14:04 #topic Specs 14:14:22 There is a subteam section 14:14:33 in the review pad 14:14:34 https://etherpad.openstack.org/p/newton-nova-priorities-tracking 14:14:36 * bauzas lurks at the door 14:15:08 We have a request to discuss one today 14:15:11 #link Adds post-copy live migration support to Nova: https://review.openstack.org/#/c/301509 14:15:19 luis5tb, yours I think 14:15:26 yep 14:15:42 I submitted the bp and the spec early today 14:16:43 Good, I don't know if anyone has had a chance to look yet ? 14:16:54 are there any questions you want to bring up ? 14:17:17 yep, I would like to know your opinion in a couple of items 14:17:50 based on the reply I got, seems that it is preferred to make nova automatically choose pre- or post copy as well as the switch between them 14:18:16 when it is best for the migrations, i.e, when the transfer rate is lower than the amount of memory being dirtied 14:18:59 but the challenge is that you can't always use postcopy 14:19:04 luis5tb, I think there are two options 14:19:11 as I'm a new in this, I just wanted to define first the basis, i.e, a way of telling nova to start the migration in post-copy ready mode 14:19:18 and then when to switch 14:19:20 luis5tb, in the last cycle we introduced "force-complete" 14:19:37 luis5tb, and it was implemented using pause 14:19:56 we expected that other ways to do the completion would come along 14:20:05 was that triggered by the admin? 14:20:07 so this can be an alternative to pause 14:20:19 luis5tb, yes 14:20:33 luis5tb, the second option would be automatic 14:21:01 I agree on having a "switch" option similar to force-complete and the automatic one 14:21:04 that would be an option for the provider to configure in their config 14:21:11 just wanted to set first the basis and then go for more advance features 14:21:35 I would guess the force-complete would be the place to start 14:21:52 and you can come up with ideas on how to decide automatically 14:21:59 does that make sense ? 14:22:16 I agree, to me it seems fairly similar (and I took a look at force-complete specs to write mine) 14:22:56 that is focusing on the second API (the switch), any input about the first one? extending the nova live-migration command with a post-copy flag? 14:23:59 In my opinion this should be something to be dynamically chosen by the admin, as you don't want to have all the migrations ready for being switched to post-copy due the the overhead it may suppose 14:24:49 or do we? in that case, (as D. Berrange suggested) we can avoid any new APIs for post-copy live migration 14:25:21 ...just looking at comments 14:26:00 I can see dan's point 14:26:15 we do avoid anything that is this specific 14:26:39 maybe it should just be automatic and an option the provider can configure 14:26:56 the admin can always use force-complete to make it happen on demand 14:27:02 for the switch? or for both APIs? 14:27:29 I was thinking that it should be possible as an option on force-complete 14:27:50 might need a little discussion on it with people like dan 14:27:51 yep, maybe we can extend force-complete 14:28:05 and depend on some configuration, you either do pause or switch 14:28:10 PaulMurray: Except it's something that needs enabling at the start of migration (the ability to later switch) and it's only available on some tranports and there are other restrictions; so have to be a little careful 14:28:13 force-complete has the same problem that we don't want to be specific about the way its doen 14:28:46 davidgiluk, right 14:29:02 I the only real issue is how to expose it in the API 14:29:11 yep, and for the post-copy enable, we need to start the migration with that flag already, as davidgiluk pointed out 14:29:54 it has certain properties as a mechanism 14:30:04 it can have some overhead on the vm 14:30:10 and it is a little more volitile 14:30:21 (i.e. a failure after switch is a problem) 14:30:36 but its guaranteed to finish 14:31:01 so something that uses that as a category rather than saying "post-copy" 14:32:41 luis5tb, you could try getting input on ML as not all people with an opinion are in this meeting 14:33:06 ML? 14:33:13 mail list 14:33:17 openstack-dev 14:33:59 ok 14:34:08 luis5tb, are you on the mail list ? 14:34:16 yep, I am 14:34:56 We can review you spec and look out for the ML post to get some discussion going 14:35:01 (that is how I discovered this meetings about live migrations) 14:35:29 I dont' think there is much contention except these details 14:35:41 So, should I write an email with the two side of the coin and see what people think? 14:36:03 yes, and give a link to the spec review 14:36:37 we can try to settle is quite quickly 14:37:16 Anyone one else have a spec to bring up ? 14:38:15 #topic Summit sessions 14:38:35 regular ref to newton summit ideas: https://etherpad.openstack.org/p/newton-nova-summit-ideas 14:39:09 #topic Open Discussion 14:39:14 we have something here 14:39:41 Use target VIF information on Migration (scheuran) 14:39:49 but I don't see scheuran 14:40:44 #link Use target VIF information on live migration https://review.openstack.org/#/c/301090/ 14:40:50 anyone know anything about this ? 14:41:57 Something to look at then.... 14:42:07 Anyone want to bring anything else up ? 14:42:42 I don't have time to look at it now but it seems there was no progress on CI 14:43:31 tdurakov, is there anything planned ? 14:44:00 hey, sorry missed meeting 14:44:09 no update for ci now 14:45:07 tdurakov, would be good to get a plan together for coming months 14:45:17 tdurakov, are you continuing with it ? 14:45:21 sure 14:45:25 tdurakov, do you need help 14:45:39 nope, will get back for ci next week 14:45:52 tdurakov, I'll put it on the agenda :) 14:45:58 ok 14:46:23 I think we are done ? 14:46:57 I am very sorry for being late... thanks jlanoux for reminding me 14:47:08 thanks for coming 14:47:12 #endmeeting