13:00:26 #startmeeting senlin 13:00:27 Meeting started Tue Jul 19 13:00:26 2016 UTC and is due to finish in 60 minutes. The chair is Qiming. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:00:29 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:00:31 The meeting name has been set to 'senlin' 13:01:00 hello 13:01:09 Good evening. 13:01:17 good evening 13:01:32 good evening 13:01:51 evening everyone 13:02:04 #topic agenda items 13:02:29 if you have any topics for discussion, please add them here: https://wiki.openstack.org/wiki/Meetings/SenlinAgenda 13:03:23 let's start with the progress etherpad 13:03:29 #topic newton work items 13:03:33 #link https://etherpad.openstack.org/p/senlin-newton-workitems 13:04:00 seems yan yan is not online 13:04:23 the rally tests are pending review 13:04:49 hopefully, with some minor tweaks, they will get merged 13:05:05 lixinhui_, any update from scaling tests? 13:05:49 Intel will loan vmware a super large test bed 13:06:06 if time permits, we can try to run senlin on it 13:06:15 wow, what do you mean by "super large"? 13:06:37 around thousand machines 13:06:46 physical hosts 13:07:01 now it is used to running vio long run tests 13:07:02 it would be super great if we can get some perf data on a large scale deployment 13:07:17 after that if the time permits, we can try to run senlin 13:07:24 great 13:07:50 that belongs to OpenStack foundation 13:07:58 please take notes on all aspects of senlin behavior 13:08:11 all the machines come from different aspects 13:08:13 not just performance, but also concurrency, consistency etc 13:08:21 hope so 13:08:33 please share me details of china mobile deployment 13:08:46 hope we can remove basic problems 13:08:54 by these experience 13:09:05 I don't have lot to share other than their public presentation 13:09:15 ok 13:09:36 little bit pity if most of time spent on small problems 13:09:43 I believe you have got some doc from liu junwei, correct? 13:10:08 not really 13:10:08 I just got his chart 13:10:12 but no detailed configurations on nova and so on 13:10:12 about message queue, about nova api etc 13:10:32 do not have these deyails 13:10:34 if you do a search 13:10:39 if possible 13:10:45 I think he has some presentation on last summit 13:10:45 I want to get these info 13:10:56 about how they solved the bottleneck problem 13:11:13 I once learned that 13:11:17 it is something you can find online 13:11:21 but still need details 13:11:48 the talk only mention the problem side but need details to reproduce 13:11:53 you know 13:12:01 just like paper and deployment document 13:12:29 then you have to contact liu junwei directly 13:12:40 see if they have details that meet your request 13:14:11 moving on 13:14:15 health management 13:14:21 performance/salability really need more time 13:14:52 yes, I know that, I have been working on performance for at least 3 years in a row, you know 13:15:13 yes, I see 13:15:29 any progress would be good to know for the team 13:15:38 not just the final result 13:16:09 the LB side, what's the status of the bug? 13:16:12 will yanyan or eldon can get the deployment document from china mobile? 13:16:28 eldon is from china mobile 13:16:52 you can just ask for this doc for sharing, if there is such a doc 13:16:57 you guys have a talk on this 13:16:57 it is not secret 13:17:00 for summit 13:17:08 we didn't 13:17:28 we don't have any prep on that yet 13:18:02 then can not put this into schedule 13:18:07 will need more time to prepare 13:18:16 just like the cluster do, we don't have implementation there 13:18:16 if I got this test bed, only one week maybe 13:18:18 for us 13:18:20 correct? 13:18:42 deadline for the test bed is end of July 13:18:58 okay ... 13:19:24 maybe eldon is the proper name for the scalability tests 13:19:48 at least, he can summarize all the configs of china mobile 13:19:55 as the first step 13:20:00 can we do some perf test if we cannot get physical machines? 13:20:02 do you agree 13:20:06 ? 13:20:07 yes sure 13:20:16 could have changed the name to someone else 13:20:43 the item has been there for quite some time 13:20:50 okay 13:20:54 if we have the first doc 13:20:55 if priority is not high, we can move it back to TODO.rst 13:21:12 then we can run further tests to reproduce or optimize 13:21:26 I see now 13:21:28 seems 600 or 800 nodes 13:21:41 in china mobile's cloud 13:21:44 right? 13:21:51 I said I don't know 13:21:56 ok 13:21:59 let it go 13:22:11 just trying to know more information about it 13:22:14 you can either contact eldon 13:22:25 or you can ask yanyan to do the communication 13:22:27 whatever 13:22:40 let it go 13:22:49 moving on 13:23:00 the health management topic 13:23:15 no comment on the etherpad during the past weeks 13:23:19 fine 13:23:36 the bug is reviewed 13:23:36 LBaaS bug 13:23:45 marked as incomplete 13:24:10 https://review.openstack.org/#/c/325624/ 13:24:53 Stephen hope to add tests 13:25:12 but feel no tests need to add if we follow the original flavor 13:25:13 finally someone is reviewing it 13:25:20 enen 13:25:32 then argue with the reviewer 13:25:42 but if he reviewed the code really, he will know now test should be added 13:25:49 try convince him/her that a test is not necessary 13:27:02 still I agree to the reviewer that some tests would be good 13:27:27 e.g. make sure BasePoolManager is a subclass of driver_mixins.BaseStatusMixin 13:27:44 sounds reasonable 13:27:51 if, for some reasons, this subclassing is changed in future, your tests there will detect it 13:28:29 pls continue work with those guys to get the bug fixed 13:28:31 there is no test for original BasePoolManager even 13:28:43 * Qiming is still wondering why that bug is marked incomplete 13:28:52 then you can add one 13:29:01 it is just a few lines code 13:29:31 then someone review to suggest add more tests 13:29:32 ... 13:29:39 for the whole functions 13:29:44 anyway 13:29:48 some times, people get picky when reviewing code ... 13:29:59 we can try if anything helpful 13:30:57 if it is not relevant to your patch, you can tell the reviewers why a separate patch is needed to cover all those functions 13:31:31 moving on 13:31:33 :) 13:31:37 fencing code 13:31:50 we don't have anything to commit into our code base? 13:32:30 let us do thing one by one 13:32:52 no much progress on this 13:32:58 okay, moving on 13:33:05 no update from me on documentation 13:33:22 haiwei is online? 13:33:28 yes 13:33:38 hi 13:33:47 posted some comments to your etherpad 13:33:57 hi, saw your comment on the ether pad 13:34:02 https://etherpad.openstack.org/p/senlin-container-cluster-issues 13:34:16 any more questions? 13:34:37 okay 13:34:42 I commented it below 13:34:48 if you have many nova servers created in a single heat stack 13:35:02 that heat stack is not designed for this purpose 13:35:14 ok 13:35:26 we cannot force people to write a heat template in a particular way 13:35:47 but we can say that we only accept a template that has a single ip_address output 13:36:08 make sense 13:36:22 if your template looks that way, senlin can somehow treat it as a "virtual" nova server 13:36:35 "virtual" <=> "conceptual" 13:36:58 I don't know much about ip namespace 13:37:15 in my simple deployment, I am not using namespace 13:37:32 I can ssh to each and every nova servers I created 13:37:55 it really depends on the network configuration you are using 13:38:04 so ... I think it is not a senlin-specific problem 13:38:19 we don't have to worry about it too much 13:38:27 for example, when you want to ssh to a nova server by private ip address, you need to specify namespace to do it, right? 13:38:36 no 13:38:38 I don't have to 13:38:59 just 'ssh cirros@10.0.0.2', for example 13:39:14 that will work? 13:39:17 yes 13:39:41 if you have complicated network setting, things may become not so straightforward 13:39:51 but still, you will have a way to do that 13:40:05 let's assume the basics 13:40:14 I can find a email thread on this for you 13:40:41 it is not a problem specific to senlin 13:40:46 ok, I will investigate it to see in which case we need nameapace 13:40:56 ... 13:41:15 I'd assume we don't need a namespace and continue innovate our solution, ;) 13:41:31 but anyway, it is up to you 13:41:40 I will test in your way 13:41:53 okay 13:42:01 ping me if helps needed 13:42:04 moving on 13:42:16 engine rework for NODE_CREATE/NODE_DELETE 13:42:19 this item was wrong 13:42:38 when I started walking through the code of those actions 13:42:52 I noticed that policy checking are also on those paths 13:43:23 so it is a matter of checking which policy should be impoved to handle NODE_CREATE/NODE_DELETE 13:43:30 rather than rework the action logic 13:43:54 I'm revising the work item and I'll start check each policy and see how they impact these two actions 13:44:54 moving on 13:45:02 the zaqar receiver work 13:45:12 yanyan has been pushing hard at the sdk side 13:45:14 Hi 13:45:22 hi, qwebirc33749 13:45:40 Not able to login for https://review.openstack.org 13:46:00 their api doc is a little bit unstable 13:46:09 hopefully things can be improved soon 13:46:22 we have a zaqar basic driver merged now 13:46:43 event/notifications generalization 13:46:55 em ... no progress from me on this either 13:47:14 Qiming 13:47:21 I'm even wondering if we can finish it by newton-3 13:47:26 I have one question 13:47:27 yes? 13:47:35 about listener 13:47:43 of vm-lifecyle-event 13:48:05 today I configured nova but got nothing from health_manager 13:48:05 moving on to the last few lines 13:48:14 proposal are all submitted 13:48:20 deleting those lines 13:48:52 I should propose this question on open disucssion 13:49:01 what do you mean you 'configured nova'? 13:49:16 enable nova notification 13:49:42 have you cross-verified with ceilometer? 13:49:44 then attach health policy with the cluster 13:50:11 I saw nova print the notifications out in n-cpu.log 13:50:33 can you try a script that monitors the message queue and see if you can get any notification there? 13:51:50 i just tried health manager 13:52:12 even I put the filter to quiet common rule, still nothing to get 13:52:28 so? 13:52:56 so I wanna know if the filter is not right for new oslo.messging versio 13:52:58 what's your filter look like? 13:53:34 do you have a simple script watching the message queue? 13:53:49 to isolate problems between components 13:54:07 never try that before 13:54:17 it will be nice if you can share me one 13:54:23 it worth trying 13:54:28 or I will try to work out one tomorrow 13:54:35 okay 13:54:44 still one thing on agenda 13:54:54 but ... I don't think we have time for it today 13:55:03 it is about senlin delieverables for newton release 13:55:29 we still have one milestone (newton-3) to cut the release 13:56:01 team please push your patches if you want them to be part of this 2.0 version 13:56:59 for example, we planned to do policy-validate, profile-validate, but we haven't touched that yet 13:57:20 we may have to postpone that to next release I guess 13:58:34 the cluster-collect and cluster-do apis are not yet completed, due to lazy reviews in openstacksdk 13:58:42 sigh . 13:59:02 we are introducing micro-versioning only to block our own progress 13:59:39 anyway 13:59:46 thank you all for joining today 14:00:01 we are running out of time 14:00:06 #endmeeting