13:03:35 <Qiming> #startmeeting senlin
13:03:36 <openstack> Meeting started Tue Sep 22 13:03:35 2015 UTC and is due to finish in 60 minutes.  The chair is Qiming. Information about MeetBot at http://wiki.debian.org/MeetBot.
13:03:37 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
13:03:39 <openstack> The meeting name has been set to 'senlin'
13:04:03 <Qiming> morning/evening ...
13:04:24 <yanyan> hello
13:04:26 <haiwei_> hi
13:04:28 <elynn> evening
13:04:50 <Qiming> feel free to add/modify meeting agenda
13:04:52 <Qiming> https://wiki.openstack.org/wiki/Meetings/SenlinAgenda#Weekly_Senlin_.28Clustering.29_meeting
13:06:24 <Qiming> okay, let's get started
13:06:29 <Qiming> #topic container clustering
13:07:15 <Qiming> just got an update from the SUR team
13:07:20 <Qiming> overall progress is good
13:07:54 <Qiming> some interns has built a client for senlin to talk to k8s, and they are testing it
13:08:40 <Qiming> some interns have tried creating k8s clusters by having magnum invoking senlin, using a heat profile
13:09:04 <Qiming> they are still working on improving the yaml files used in this context
13:09:29 <haiwei_> this is what we want to show in the summit?
13:09:33 <haiwei_> for the demo
13:09:58 <Qiming> some interns have tried collecting runtime metrics from docker using CAdvisor, triggering the scaling using alarms generated from these metrics
13:10:48 <Qiming> yes, haiwei_, this is the work we want to show case
13:11:07 <haiwei_> the second one or the third one?
13:11:23 <Qiming> the autoscaling work
13:11:23 <haiwei_> or all of these?
13:11:27 <haiwei_> ok
13:12:01 <Qiming> we are evaluating different options to glue the two layers of entities together, in the magnum context
13:12:46 <yanyan> Qiming, you mean the scaling of both VMs and containers?
13:12:48 <Qiming> there will be open questions, but I think that is okay
13:12:52 <Qiming> yes
13:13:11 <Qiming> currently, triggering VM scaling using container-layer metrics is working
13:13:25 <Qiming> we are discussing whether the other path is worth to do
13:13:32 <yanyan> ok
13:13:36 <Qiming> we are also looking into load-balancing options
13:13:51 <yanyan> about the lb part, I think they met some issues
13:14:04 <Qiming> yanyan, please help with the team and see if the version conflicts problem is solved
13:14:11 <yanyan> I guess the neutron lb version in their env is still v1
13:14:12 <yanyan> sure
13:14:20 <Qiming> there seems to be some LB configuration problem
13:14:30 <yanyan> so they mentioned that they have version problem
13:14:35 <yanyan> ok
13:14:43 <Qiming> yes, need to engage at this stage
13:14:54 <yanyan> will re their mail and discuss with them
13:15:52 <Qiming> thanks.
13:16:11 <Qiming> any questions/comments on this work?
13:16:35 <haiwei_> is there a deadline for this?
13:17:10 <Qiming> we want to see the whole thing up and running by the end of this month
13:17:47 <Qiming> anyone interested in this can leave a comment here: https://etherpad.openstack.org/p/magnum-senlin
13:18:05 <Qiming> #topic Trigger implementation
13:18:06 <yanyan> nice
13:18:29 <elynn> cool
13:18:39 <Qiming> so ... we decided a few weeks ago to provide an abstraction called 'Trigger'
13:19:15 <yanyan> yep
13:19:28 <Qiming> basically, it would be a wrapper of ceilometer/aodh alarms, monasca alarms, zaqar queues ... things that can trigger an action on senlin entities: clusters, nodes ...
13:20:06 <Qiming> A user will decide what tools they want to use
13:20:25 <Qiming> the basic CRUD logic is already committed
13:20:46 <Qiming> however, this work is blocked somehow by two things
13:21:13 <Qiming> 1. ceilometer alarming service is offcially separated from ceilometer into aodh
13:21:37 <Qiming> since senlin is not talking to any xyz service using xyzclient, we are using openstacksdk
13:22:09 <Qiming> so we need to add aodh support into openstacksdk first, then we can experiment with "ceilometer/aodh" alarms first
13:22:41 <Qiming> the problem is that openstacksdk is blocking any resource types at the moment
13:23:58 <Qiming> so ... we need a workaround, we will host these bits temporarily in senlin codebase and migrate them to openstacksdk later
13:23:58 <Qiming> or else we will be blocked for a long time ..
13:24:17 <Qiming> I'm working on that
13:24:46 <Qiming> questions?
13:25:09 <yanyan> so the first step is support trigger based on aodh alarm
13:25:20 <Qiming> yes
13:25:41 <haiwei_> aodh is using ceilometer-client, so it is almost the same to support ceilometer-client?
13:25:42 <yanyan> ok, after the framework of trigger is done, we add more trigger tyep
13:25:45 <yanyan> type
13:25:47 <Qiming> I'm not sure about monasca's status, or the usability of zaqar so far
13:26:08 <Qiming> haiwei_, if you using ceilometerclient, everything remains the same
13:26:40 <Qiming> there will be an redirection from ceilometer to aodh concerning alarm operations
13:26:42 <yanyan> zaqar is a little different I think, maybe we need a new handler type to support it
13:27:09 <Qiming> yep, it is not longer a webhook
13:27:26 <Qiming> it will be totally about message queue
13:27:54 <yanyan> yes
13:27:57 <Qiming> #topic heat resource type support
13:27:58 <haiwei_> aodhclient== ceilometer-client?
13:28:20 <Qiming> haiwei_, aodhclient?
13:28:28 <yanyan> hi, haiwei_ , I think there is just ceilometer client
13:28:36 <Qiming> don't know if there is such thing, we won't use it even there is a new client
13:29:10 <haiwei_> ok, I need to learn adoh first
13:29:26 <Qiming> so ethan has kindly offered to take senlin back to heat
13:29:26 <Qiming> https://review.openstack.org/#/c/226180/
13:29:30 <haiwei_> aodh
13:30:03 <elynn> yea, I'm working on this
13:30:15 <elynn> still need to discuss how to implement it.
13:30:31 <Qiming> we have quite a few things to work out
13:30:41 <elynn> I don't quite get Kanagaraj Manickam's point.
13:31:13 <elynn> He said senlin is part of requirements.txt
13:31:19 <elynn> But I don't see it.
13:31:31 <Qiming> above all design considerations, we'd strive to maintain compatibility with heat ResourceGroup ...
13:31:43 <Qiming> elynn, it is a typo
13:31:57 <Qiming> senlin is not listed there yet
13:32:20 <yanyan> in big tent, will it be?
13:32:54 <Qiming> among the concepts (objects) senlin exposed, the top priority, in my view, is cluster and node
13:33:02 <Qiming> it doesn't matter I think
13:33:12 <yanyan> yea
13:33:19 <elynn> Yes we need to compatible with heat RG, question is when.
13:33:38 <Qiming> we need to put 'UNSUPPORTED' as support status once the code is in
13:33:53 <elynn> yes, maybe cluster is enough for compatible with heat Rg
13:33:55 <Qiming> elynn, from the very beginning
13:34:21 <Qiming> cluster is meant to be a superset in terms of features from RG
13:34:48 <Qiming> if not, we need to improve cluster design/implementation
13:35:24 <Qiming> however, we may don't need to model senlin profile in Heat
13:35:45 <Qiming> from Heat's perspective, it is just an implementation detail
13:36:28 <Qiming> the big issue as I see it is about the 'update' operation
13:36:38 <elynn> So the resources for senlin would be like OS::Senlin::Cluster OS::Senlin::Node OS::Senlin::ResourceGroup ?
13:36:39 <Qiming> which is the only operation for you to change things in Heat
13:36:56 <Qiming> no OS::Senlin::ResourceGroup
13:37:17 <elynn> so OS::Senlin::Cluster replace Heat RG resource?
13:37:31 <Qiming> we can add an entry in the default environment file to map OS::Heat::ResourceGroup to OS::Senlin::Cluster
13:38:08 <Qiming> even that mapping would be a mid-term work
13:38:30 <elynn> ok, I got it.
13:38:39 <yanyan> so both properties and attrs of OS::Senlin::Cluster should be completely compatible with RSG
13:38:45 <elynn> so what's the problem for stack-update?
13:38:46 <Qiming> the near-term is to make OS::Senlin::Cluster work, cover all ResourceGroup features by invoking senlin APIs
13:39:06 <yanyan> or maybe superset of the one of RSG
13:39:06 <Qiming> stack-update -> resource update
13:39:16 <Qiming> resource update -> cluster ?
13:39:28 <Qiming> it could be cluster resize, cluster update
13:39:50 <elynn> and nested stack update
13:40:16 <Qiming> heat doesn't have to see the nested stacks
13:40:54 <Qiming> RG <=> cluster of heat stacks
13:41:04 <Qiming> heat stacks here is already the nested stack
13:41:14 <Qiming> and we implement it using profile
13:41:28 <elynn> en
13:41:34 <Qiming> when template used for the inner stack changes, we change the profile
13:41:55 <Qiming> when template is not changed, we may still need to do resize operations
13:42:10 <elynn> ok
13:42:18 <Qiming> we got to be very careful at every step
13:42:48 <elynn> I've see apis in senlin for cluster resize, is it enough ?
13:42:57 <Qiming> when doing this, we should think in Heat's philosophy, not the other way
13:43:23 <Qiming> yep, I'm pretty sure it is a superset of RG's capability
13:43:53 <Qiming> elynn, we need to sit down and work out the details
13:44:04 <elynn> yes...
13:44:49 <Qiming> yet another blocker is that we need to release senlinclient first
13:45:06 <Qiming> make it into the global requirements so that other projects can invoke us
13:45:25 <Qiming> I'll ask around how to do this
13:45:40 <yanyan> and I think the way invoking senlinclient is a little different from the client of other services ?
13:46:21 <Qiming> sadly I won't be avail 9am-13pm tomorrow
13:46:41 <yanyan> 9am?
13:46:53 <yanyan> not 9:30
13:46:55 <Qiming> but we can discuss this via email if f2f discussion is not possible
13:47:00 <elynn> senlinclient now didn't provider api to call directly, isn't it? we can only use shell commands to call it .
13:47:04 <Qiming> sigh, the meeting was rescheduled
13:47:14 <yanyan> ...
13:47:40 <Qiming> 9-11, meet, 11-12, all-hands, 12-13 lunch meet
13:47:44 <yanyan> elynn, yes, I think so. But we can build sdk session first and use it to creation senlinclient instance I think
13:48:03 <yanyan> right, there is an allhands tomorrow...
13:48:17 <Qiming> we can work out some pseudo code first
13:48:37 <Qiming> #topic open discussions
13:48:58 <Qiming> 11 mins
13:49:15 <Qiming> then I have another meeting, no meat
13:49:44 <yanyan> did you have dinner?
13:49:55 <Qiming> so other than the topics we touched so far
13:50:19 <Qiming> I'd encourage everyone to start signing up to mitaka work items
13:50:20 <Qiming> https://etherpad.openstack.org/p/senlin-mitaka-workitems
13:50:56 <Qiming> yanyan, I did, without being far from keyboard
13:51:14 <yanyan> sigh...
13:51:57 <Qiming> anything else, before we release this channel?
13:52:03 <elynn> oh...
13:52:08 <yanyan> no from me
13:52:13 <haiwei_> no
13:52:37 <elynn> I've sign my name on some item ;)
13:52:53 <elynn> Don't know if anyone are working on zaqar or not...
13:53:05 <Qiming> okay, thanks for joining, and your hard work, :)
13:53:15 <Qiming> until next time
13:53:19 <Qiming> #endmeeting