13:00:43 #startmeeting senlin 13:00:44 Meeting started Tue Jul 5 13:00:43 2016 UTC and is due to finish in 60 minutes. The chair is yanyanhu. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:00:45 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:00:47 The meeting name has been set to 'senlin' 13:00:52 hi 13:00:56 Hi! 13:00:57 hi 13:01:07 hi 13:01:08 hi :) 13:01:11 Qiming, you are here 13:01:16 seems so 13:01:21 watching 13:01:21 :P 13:01:34 free to hold the meeting? 13:01:45 pls go ahead 13:01:49 ok 13:01:53 https://wiki.openstack.org/wiki/Meetings/SenlinAgenda#Weekly_Senlin_.28Clustering.29_meeting 13:02:03 plz feel free to add items to agenda 13:02:13 #topic newton workitem 13:02:29 ok, lets go through newton workitem first 13:02:30 https://etherpad.openstack.org/p/senlin-newton-workitems 13:02:41 no new progress in testing I think 13:02:52 but the functional test migration has been done 13:03:15 hi 13:03:53 so will remove this item from etherpad 13:03:54 I guess also no progress in performance test 13:03:54 so lets skip it 13:04:01 hi, lixinhui_ 13:04:15 next item is about HA 13:04:35 hi, lixinhui_, so you mentioned the fencing part has been done last week? 13:04:43 yes 13:04:56 manually set up 13:05:21 need to consider how to automatically enable the scenario 13:05:29 * Qiming raises his thumb, both ... 13:05:44 and the bug of nuetron-lbaas... 13:05:45 Great. Since we have proposed a topic about end-to-end HA solution based on Senlin HA to summit, we may need to finish some basic support to build a PoC for it 13:05:53 still need some time to get review 13:06:00 lixinhui_, yes, I saw you have assigned the bug to yourself :) 13:06:06 yes 13:06:30 hi, Qiming, any idea about this work item 13:06:43 or something we need to pay attention 13:06:43 it is fine 13:06:58 we may need a full story 13:07:16 even though we still have something missing 13:07:19 yes, full story is very important, not only for a demo 13:07:39 https://blueprints.launchpad.net/senlin/+spec/support-health-management-customization 13:07:46 so this etherpad is about HA design 13:08:08 https://etherpad.openstack.org/p/senlin-ha-recover 13:09:06 something is still missing 13:09:19 especially about the health policy 13:09:37 we can have a basic policy poc before summit 13:09:48 yes 13:09:57 that's a basic goal I think 13:10:16 not very difficult I guess 13:10:18 a broader topic may involve mistral 13:10:32 which should be discussed in section 4.1 13:10:39 to control the workflow of node recovery? 13:11:04 yes 13:11:29 there might be some out-of-band resources that need to be taken care of 13:11:37 I see. 13:11:43 agree 13:12:22 another item I'm still not clear is about application/service deployed inside VM 13:12:42 will senlin handle their failure as well? 13:13:00 by watching lbaas member status e.g. 13:13:20 I think so 13:13:37 we should provide listening to the event of member status change 13:13:58 it is possible to support app HA partially 13:14:16 if the lbaas service is created by senlin 13:14:32 For all different kinds of applications/services? 13:14:40 not an easy job, but we, as always, can do our best for failure detection 13:14:59 difficult for all ... I think 13:15:00 sorry, just dropped 13:15:18 detect application's failure? 13:15:19 another thought on this -- maybe a thread in parallel --- is to setup guest level Linux HA cluster 13:15:30 Is there a layer controlled by senlin which can detect their status? 13:16:00 suppose an application has been protected using Linux HA before, migrating them to Senlin cluster should be an easy job 13:16:28 zzxwill, a generic solution for application failure detection is almost impossible 13:17:11 so maybe we build a full HA story with typical use case and then figure out how senlin support it? 13:17:25 yes 13:17:58 Thanks. lixinhui and Qiming. 13:18:03 if there are some gap, we know what part is missing in Senlin's HA mechanism 13:18:04 for example, in some domains, openais has been planted into the application code, so that reliable failure detection becomes a possibility 13:18:19 Qiming, yes 13:19:14 actually, I think whether senlin can support the fail over of an app also depends on how app expose their user interface 13:19:22 we can start with some wireframe diagram 13:20:52 yes. we need more detailed design and complete the proposal which is just a draft now 13:21:01 based on such a draft we can identify where it makes sense for senlin to play a role, and where it makes sense to reuse/integrate with existing technology/solutions 13:21:13 agree 13:21:23 to decide our scope 13:21:37 we have quite some piece technologies to integrate, to massage ... 13:21:40 and then we can start more concrete work 13:21:50 and decide their priority 13:22:01 yes, actually I just try to add zaqar driver to senlin 13:22:06 but met lots of problem 13:22:31 will talk about this issue a bit more later 13:22:34 right, so ... comment on the etherpad would be very helpful: https://etherpad.openstack.org/p/senlin-ha-recover 13:22:47 ok 13:23:17 Qiming, will you be free for a call this week or next week before tuesday? 13:23:29 yes, I think so 13:23:49 great, then I will arrange a call for some discussion about this topic 13:24:28 pls involve everyone who feels interested in this 13:24:29 ok, anything else about HA? 13:24:33 sure 13:24:39 will send out the invitation 13:24:41 Great. 13:24:58 the listener implementation has some flaws 13:25:00 #action yanyanhu send out invitation about the call for HA topic discussion 13:25:16 ok, lets move on 13:25:17 I was seeing some exceptions thrown from oslo.messaging occasionally 13:25:26 will dig more into it 13:25:30 #topic profile for docker 13:25:58 hi, haiwei_ , any new progress about it? 13:26:08 I have no progress on container jobs, this week 13:26:21 big network latency here... 13:26:29 honestly, I am a little confused what to do the next 13:26:31 Qiming, great, thanks :) 13:26:40 haiwei_, yes? 13:27:33 the network is not good to me, either 13:27:47 pour your confusion here, haiwei_ 13:28:48 I have added container create/delete functions to Senlin, maybe container_list , and after that what should we do? 13:28:58 to support storage and network? 13:29:23 think from a user's perspective 13:30:18 or support cluster actions for container first? 13:30:53 either way 13:31:29 This is really a big topic I think. Maybe we first finish the work we have done as PoC in last summit? 13:31:36 supporting cluster actions doesn't sound a great challenge, IIUC 13:31:49 including the docker driver/profile and a basic placement policy for it 13:32:20 however, if we are running some workloads, we will find that we will need to provision storage/network for some basic services 13:33:09 right, "scheduling" is another topic to work on 13:34:01 I am still concern about the container's host when thinking about cluster actions, because things are different when containers are in one host or different hosts 13:34:28 that is not senlin's scope 13:34:45 we will just leverage whatever existing technology to provision the network 13:35:40 oh, BTW, in latest docker release, more and more functionality has been integrated into the docker-engine, including the orchestration 13:35:44 the difference should be masked from users 13:35:49 because ... they don't care 13:35:59 this is not a bad news I think 13:36:27 that means relying on less 3rd service/tools to support container cluster 13:36:32 imo 13:37:04 s/3rd/3rd part 13:37:14 docker will use docker swarm to support container cluster I think 13:37:32 haiwei_, in latest 1.12, swarm has been the history :) 13:37:40 it is part of docker engine now I think 13:37:40 Yes, I heard it from your WeChat. 13:37:44 so ... my general feeling is ... sometimes we are too easy to be brain washed 13:37:47 ok, really 13:38:01 I am not a docket kong 13:38:02 haiwei_, yes, they call it docker orchestration 13:38:11 to understand the reality, we have to try it out 13:38:14 but feel interested at network part 13:38:18 Qiming, yes 13:38:29 will you leverge kuryr for that? 13:38:44 so my feeling is their are in fast progress 13:39:05 setup swarm and use it, and see if it works as expected 13:39:07 lixinhui_, you mean senlin? 13:39:15 yes 13:39:19 yanyanhu 13:39:33 orchestration part is very complicated 13:39:37 Qiming, announcement is different from reality some times :P 13:39:39 everyone is moving fast, because this is a brutal, crazy world 13:39:52 I am trying to understand the driver part firstly 13:40:04 some companies are very good at "generalizing" things 13:40:19 for example 13:40:30 lixinhui_, based on my understanding, docker network is still in preliminary mode, compared with SDN support in openstack 13:40:33 deploy a lbaas in container 13:40:45 better test and verify it using your own machine 13:40:49 can senlin automatically fo that? 13:40:50 its overlay network or some 3rd part tools support 13:41:39 lixinhui_, lb in side container? 13:41:44 or lb for container cluster 13:42:26 the the one 13:42:35 the later one 13:42:48 I'm not sure about the former one. for the latter one, it is a basic requirement I guess? 13:43:01 do not know 13:43:04 at least 13:43:04 yes, will definitely need it 13:43:12 network need to be done 13:43:14 they need to support it, otherwise, it is useless :P 13:43:59 so we may need more investigation here 13:44:15 for what docker-engine can support now 13:44:24 and what it doesn't 13:44:27 I do not think so 13:44:40 yes, yanyanhu 13:44:41 since floating ip is created by neutron in this way 13:44:57 how for applications or container know that? 13:45:05 after we better understand it, we can make the correct decision :) 13:45:48 lixinhui_, if you mean network integration with openstack for docker, I'm not clear about the progress... 13:46:03 but I feel it is not an easy job 13:46:15 for networking stuff, we already have kuryr working on that, so we don't have to reinvent a thing (hopefully) 13:46:40 if we don't know kuryr, then we should learn it, instead of just imagining it 13:46:41 at least not as easy as integrating nova with neutron :P 13:46:48 Qiming, yes 13:47:02 this is a homework have to do 13:47:08 yes 13:47:31 forgive me if driver layer is not ready 13:47:35 when we have found something missing from kuryr, we propose a work item for that team/project 13:47:36 I'm trying to build a docker1.12 env and try it 13:47:36 I think it's almost about container cluster, not about the IaaS cluster for container, we can proceed it and assume that network is done. 13:47:43 orchestration is more complicated to discuss 13:47:48 hope can understand its latest release better 13:48:03 We should focus how do we use senlin to deploy docker application on senlin cluster cluster. 13:48:39 right, 'orchestration' may mean a thousand different things from different group, to different people 13:48:40 openstack have magnum to deploy the IaaS cluster for cluster. 13:49:04 for docker cluster 13:49:20 kinda agree with elynn 13:49:49 em ... magnum is not about clustering 13:50:03 yes, it isn't 13:50:16 IIUC, it is always about glueing a COE onto openstack 13:50:39 just feel maybe focusing on docker app cluster deployment is a better idea? 13:50:48 and more useful 13:51:02 for end user 13:51:19 from end user's perspective, that is true, I think 13:51:40 That's what I mean yanyanhu :) 13:51:41 yanyanhu, not really understand you, build a docker app cluster is different from building a docker cluster? 13:51:48 yes, since pure docker instance(s) is not that useful I feel 13:52:16 haiwei_, it is the same if you treat docker instance deployment as app deployment 13:52:33 when you start a container, you use an docker image, and in the image the app is installed, right? 13:52:34 since the reason user deploy docker is for deploying app I think 13:52:41 haiwei_, yes 13:52:50 yes? what is the difference? 13:52:52 so a docker cluster is actually an app cluster 13:53:01 so, I think it is the same thing 13:53:17 so docker orchestration is actually service orchestration I feel? 13:53:19 haiwei_, yes 13:53:27 I think so 13:53:29 why are we deploying pure docker instances? 13:53:39 just want to clarify which layer we are focusing on :) 13:53:53 it is only about wording 13:53:59 Qiming, yes :) 13:54:01 any technology difference? 13:54:31 I am not a big fan of strict layering 13:54:36 it means nothing to end users 13:54:40 they don't care 13:54:52 yea 13:54:55 so I think we are on the same page now for this topic :) 13:54:57 is this service a paas, or a iaas? 13:55:04 who cares? 13:55:46 ok, last 5 minutes 13:55:55 since we import containers to senlin, we need to think about how to make containers fit for senlin's architecture 13:56:07 haiwei_, +1 13:56:19 after we finish our homework :) 13:56:36 yes, we can transparently adapt senlin's architecture for this purpose, provided that we are not breaking existing users 13:56:41 will learn more about in coming weeks 13:56:49 and lets have further discussion about it later 13:57:24 ok, anything else about this issue? 13:57:25 cool 13:57:30 ok, I will express my thoughts by code 13:57:32 +2, got to jump into the pool an learn swimming that way 13:57:46 oops, big latency again... 13:57:53 instead of reading a thousand books on swimming, staying away from the pool 13:58:08 Qiming, +2 13:58:20 trying it is the best way to learn it in most cases 13:58:22 I went to pool last sunday, and got burnt 13:58:24 it's always a controversial topic, better done then always discuss. Just do IT ;) 13:58:28 haiwei_, :P 13:58:32 LOL 13:58:34 yea 13:58:40 ok, last 2 minutes 13:58:43 https://etherpad.openstack.org/p/senlin-proposal-barcelona-summit 13:58:46 btw, forwarded you (yanyanhu) an email from a user 13:58:55 Qiming, ok 13:58:56 they are seeing some NULL respose from senlin API 13:59:05 which is impossible in theory 13:59:06 ok, will check it 13:59:11 thx 13:59:17 no problem 13:59:18 https://etherpad.openstack.org/p/senlin-proposal-barcelona-summit 13:59:26 plz add concrete idea to this etherpad 13:59:37 it is for our topic proposal for coming summit 13:59:53 we may have to finish the proposal before this weekend 14:00:02 ok 14:00:16 since next Wednesday is the deadline 14:00:24 ok, time is over 14:00:36 lets move back to senlin channel 14:00:40 thanks you guys for joining 14:00:48 #endmeeting