15:02:51 #startmeeting scheduler 15:02:52 Meeting started Tue Sep 3 15:02:51 2013 UTC and is due to finish in 60 minutes. The chair is n0ano. Information about MeetBot at http://wiki.debian.org/MeetBot. 15:02:53 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 15:02:56 The meeting name has been set to 'scheduler' 15:03:09 we can start the meeting, just on the off chance some others arrive 15:03:16 Sure 15:04:27 on the mail list you mentioned you wanted to speak about ideas for sessions at summit 15:05:00 yeah, I thought maybe we should coordinate a little, see what ideas people have 15:05:14 #topic session topics for the Icehouse summit 15:05:26 hi 15:05:50 'future of scheduler'? :-) 15:05:55 cool. myself and a few colleagues have been discussing extending the image properties filter and would like to expand on this to be able to deploy templates 15:05:56 one thought was should we have a big `futures` session to talk about long term directions 15:06:20 glikson, looks like we're on the same page here 15:06:35 that is, have the image properties, couple with host support drive the scheduling decisions 15:07:01 another candidate would be around scheduling policies 15:07:04 Would be good if we can avoid last year's crush of trying to do 11 subjects in one session 15:07:08 garyk, seem like something close to what my group has been calling enhanced platform awareness 15:07:37 n0can you please elaborate a little more on ethat 15:07:46 I think part of the reason it was hard to land soem changes in H was that they didn't really get enough time at the summit 15:07:49 PhilDay: agree.. PTL sets the agenda, right? 15:07:58 PhilDay, partly that was due to a lot of late submissions, if we can be more prompt this year it might be better 15:08:28 the instance group feature has been in the queue since the beginning of H. 15:08:39 glikson, I would imagine there's also only so many sessions available, hard to fit everything in 15:09:08 n0ano: sure, so, it is also a matter of priorities 15:09:11 the last summit there were a ton of scheduling sessions. i think that we did well to collaborate on presenting the data. 15:09:22 phil was instrumental in getting everyone together before and syncing 15:09:44 i hope that this time we will also have a platform to present ideas and discuss them 15:10:04 I am not saying anything went wrong in Portland.. Just saying that maybe this time there will be more awareness. 15:10:05 Feels like we need a session to bottom out the whole "to what extend should the scheduler depend on or just feed into celiometer" issue 15:10:29 agreed. 15:10:35 PhilDay, that's an important area that will warrant a separate session I belive 15:10:38 PhilDay: yep, metrics, etc 15:10:58 but this is certainly something that we would need to try and sync with the ceilometer guys and have a combined session 15:11:22 it could be also related to the question of using DB versus RPC versus whatever to keep/deliver metrics.. 15:11:34 sorry for joining the chat late - are we trying to use ceil for communicating the "state" that a scheduler can use later on or just for metrics 15:11:54 So I;m seeing at least - future directions, metrics(Ceilometer), image properties&host capabilities - at least 3 sessions so far 15:12:18 n0ano: policies 15:12:24 Does Future directiosn cover teh proposal from Boris et al ? 15:12:38 glikson, oops, that's 4 - DB vs. fanout 15:12:43 glikson: yeah, things we want to add to instances 15:12:53 +1 15:13:17 also would be good to revisit the topic of scheduling across nova/cinder/etc 15:13:19 I want to do a session on the pcloud / whole host stuff - only partially related to scheduler though 15:13:38 PhilDay: i think that the performance is very important. Not sure if implementation details is where the discussion is at the moment (that is my two cents) 15:14:03 @garyk - sorry, not sure i follow ? 15:15:08 PhilDay: in the few times we have spoken about boris's proposal the discussion seems to be drawn into the implementaion details. i think that there are design issues that we need to address first. 15:15:53 well, the big design issues is fanout vs. DB, until we resolve that it hard to go any further 15:16:04 Ok, yes agreed. a shared view of the design is the key thing we need to come away with 15:16:07 n0ano: agreed 15:16:54 boris-42: you around? 15:17:08 yes I am here 15:17:11 garyk ^ 15:17:20 whats up? 15:17:58 boris-42, we're talking about potential summit sessions, are you planning on doing one on your scheduler proposal 15:18:02 garyk: +1 about hte design comments .... 15:18:02 same same. we are talking possible scheduler sessions ad the performance issue arose. i guess we should have a session on that too. 15:18:10 n0ano yeah it will be nhice 15:18:20 n0ano we are going to present real nubmers 15:18:34 n0ano between current and our approaches 15:19:11 regarding DB versus fanout.. would it be feasible to keep both? e.g., with a separate HostManager implementation? 15:19:21 there is no fanout 15:19:28 w don't want to use fanaout at all 15:19:41 guys, isn't the hong-kong sessions voting ended already, or is this for a different scope? 15:19:43 boris-42, one issue is design vs. implementation, we're still lacking concensus on the basic design, how do we achieve that 15:19:54 just another way to store in key value storage all information 15:19:58 boris and I have started merging the 2 documents 15:20:01 n0ano we already achive it 15:20:21 n0ano but we should make some other cleanups 15:20:31 https://docs.google.com/document/d/1cR3Fw9QPDVnqp4pMSusMwqNuB_6t-t_neFqgXA98-Ls/edit# and boris' implementation details into one 15:20:33 n0ano to remove compute_nodes table and compute_nodes update 15:20:33 doron, no, that's for presentations, working sessions are still open 15:20:54 n0ano: thanks. I'd love to join 15:21:21 boris-42, I'm not sure we have concensus, I think jog0 is still unconvinced at minimum 15:21:40 n0ano we will show numbers from real deployment 15:22:18 n0ano btw here is patch 15:22:19 https://review.openstack.org/#/c/43151/ 15:22:29 that improves performance in safe way 15:22:34 numbers are good but they don't trump all other considerations 15:23:08 note that I'm really Devil's Advocate here, I happen to mostly agree with you but I don't think everybody does 15:24:33 i agree partially but there is a still a lot of cases to be discussed. 15:24:59 which is why I think real resolution will have to happen at the summit 15:25:06 n0ano numbers that could be repeated are not just numbers 15:25:16 i think if we have a session at the summit it will be great. number also help, but use cases and understanding the exact tests is also very important 15:25:33 n0ano we are going to add new project Rally (that will be benchmark system for OS) 15:25:50 garyk, +1 15:25:52 can we have a list of requirements for the scheduker improvements from everyone and collate a bunch of approaches 15:26:04 debo_os: +1 15:26:20 +1 15:26:27 yes this should be done before session 15:26:30 have a etherpad for now 15:26:37 yeah it needs to be done now 15:26:42 that is a good approach 15:26:56 boris-42, curious about Rally, would like more info on that (but later) 15:27:03 coming with crystalized ideas will enable us to hash out the details 15:28:07 for example one clear requirement is to make placement decisions based on cinder nova neutron 15:28:23 also scheduler = placement + state machine to shedule 15:29:01 if we can collect simple requirements and then hash out pre summit we might have conclusions by nov 8 15:29:44 debo_os, I think you're a little optmistic, if we can just get the requirements out before then I'd be happy (conclusions come later) 15:30:00 +1 to setting up either pads now for the set of sessions that non captures from this session (or maybe one pad to plan the sessions?) If we can offer Russell a list of 4-5 scheduler session we want to have that might be a good way to avoid last year's problems 15:30:02 n0ano: optimism is cheap ;) 15:30:25 PhilDay: +1 15:30:50 PhilDay, I like the idea of 1 etherpad for planning, does someone want to signup for creating/maintaing that pad? 15:31:14 one a practical level it would also be nice if we can get core reviewers to look at the various scheduling changes (some have been around for longer than forever) 15:31:29 * n0ano would sing up but I'm on vacation for the next 2 months (eat your hearts out :-) 15:31:45 Sure - I can set up the pad 15:32:00 PhilDay: i am happy to help out to 15:32:09 #action PhilDay to setup up scheduler planning etherpad for the Icehouse summit 15:32:31 Thanks Gary 15:32:33 #action `everyone` to maintain the planning pad 15:33:49 let's move on a little bit 15:33:58 #topic multiple-scheduler-drivers 15:34:21 glikson, I believe there is some controvery on the mailing list about this, is there anything that can be done to resolve things? 15:34:30 i like the idea :) 15:34:55 So, almost nothing left from the original idea -- but we did submit several small incremental patches which should not be controversial.. 15:35:27 at least based on my discussion with jog0 the other day 15:35:44 so does that mean the ultimate design goal has changed? 15:36:17 basically improving the existing AggregateCoreFilter and AggregateRamFilter -- making the overriding mechanism generic, and then extending to additional filters/option types 15:36:55 well, the ultimate goal has not changed -- but the goal for Havana has 15:36:59 so kind of a step wise progression, which doesn't sound all that bad, just takes a little longer 15:37:06 the rest requires discussion at the summit 15:37:14 yep 15:37:34 much longer :-) 15:37:38 glikson, refer back to the last action item, update the planning etherpad :-) 15:37:50 yep, will certianly contribute to that 15:38:08 OK, onward 15:38:21 #topic temporary moderator 15:38:30 as a general thought.. I think it would be very helpful to engage people from core early. 15:38:38 +1 15:38:53 Note this is my last meeting for 2 months, is there anyone willing to moderate this meeting while I'm gone? 15:39:56 even not having someone from core in this forum is not a good sign.. 15:40:40 glikson, just incentive for all of us here to become core members :-) 15:41:15 anyway, I don't need an answer right now, think about moderating this meeting, it's not that hard. 15:41:22 #topic opens 15:41:37 Anyone have anything new they want to bring up today? 15:42:25 n0ano: i can moderate whilst you are away 15:42:39 garyk, excellent, tnx much 15:42:52 n0ano: ok. from next week? 15:43:09 #action garyk to moderate this meeting for the next 2 months, start on 9/10 15:43:51 OK, hearing nothing else, let's close for today 15:44:11 tnx everyone, hopefully I'll see you all at the summit 15:44:17 #endmeeting