Thursday, 2016-01-07

*** Qiming has quit IRC00:19
*** pratikmallya has quit IRC00:22
*** Qiming has joined #senlin01:12
Qimingjust got some complaints about upgrading senlin01:21
Qimingthe report is from IBM internal01:21
Qimingpeople upgraded senlin to the latest code, they had to recreate database, due to latest revision to db schema01:22
Qimingand they have lost references to all objects they have created01:22
Qimingso, I'm thinking maybe we should freeze verion 001 database schema starting mitaka-201:23
Qimingany future revisions to db should go into 002, 003 etc01:23
xuhaiweiup to now, we hardly conside backward capability things01:28
xuhaiweithat is a problem01:28
Qimingyes, this is the right thing to do01:28
Qimingthe only question was when to start enforcing such a rule/practice01:29
Qimingconsidering that there are users start using it01:29
QimingI'm raising awareness for core reviewers01:29
Qimingplease be stricter when reviewing DB changes01:30
xuhaiweiabout the meetup agenda, I want to talk about tempest-external-plugin, it's about function test, currently it is not in the schedule, Qiming01:32
xuhaiwei20 mins would be ok I think01:33
xuhaiweican you reschedule it?01:33
Qimingis it an experience sharing?01:33
xuhaiweiyou know in tempest, some projects will make them function tests ran in tempest, but there are two many projects now, tempest only want to support core projects now01:35
Qimingso it would be a senlin related topic, which means day 2 is better?01:35
xuhaiweiso they bring into this method, make other projects' function test work as a plugin01:36
xuhaiweiit's ok I think, not very big topic01:36
xuhaiweiafter this plugin worked, you can run senlin function with tempest command01:37
Qimingcheck etherpad, editing01:38
Qimingxuhaiwei, we have to define our scope01:57
xuhaiweiif we create a stack and leave it there, why should we create it?01:57
Qimingbe very very careful when walking into anoter service's domain01:57
Qimingsenlin is about a cluster of stacks01:57
xuhaiweiI think I am understanding the original design for stack node01:57
xuhaiweiI am not01:58
Qimingeach stack can be a wrapper of a nova server + a cinder volume + a floating IP ...01:58
Qimingxuhaiwei, I don't think you understand the design01:58
Qimingyou are dragging us to a very dangerous zone01:58
xuhaiweiyes, I mean I dont understand01:58
Qimingstack is an abstraction or a wrapper of many different kinds of resources01:59
xuhaiweiso why senlin create a stack node?01:59
Qimingin Heat's design, a stack can contain an embeded stack, which contains yet another level of child stacks01:59
Qimingxuhaiwei, to avoid reinventing all resource types02:00
Qimingthink about array and struct metaphore02:00
QimingHeat provides you a struct, Senlin provides you an array02:00
Qimingcluster of heat stacks <==> array of structs02:00
Qimingwe are NEVER touching the internals of a struct02:01
xuhaiweiso we only create it?02:01
Qimingcreate, update, delete, scale ....02:01
Qimingwe scale a stack as a whole02:02
Qimingsenlin will never care how many instances, how many volumes inside a stack02:02
Qimingif your heat template doesn't have an output of the IPs of the servers created, we will never get a chance to do load-balancing02:03
xuhaiweibut if your stack work together with other instances(not in this stack), there will confict02:03
Qimingwe cannot force users to write heat templates this way and that way02:03
Qimingxuhaiwei, that is true02:04
Qimingthat will be true forever02:04
Qimingjust like you are running openstack02:04
Qimingand you are also launching KVMs using qemu-kvm command line02:04
xuhaiweiin that case, the stack auto-scaling will fail in a high opportunity02:04
Qimingyou are using Heat to do orchestration, while you are also using nova to create VMs02:04
Qimingxuhaiwei, I don't agree02:05
Qimingif architected properly, you can still get auto-scaling work properly02:05
Qimingthere are 1000 ways to fail02:05
Qiming999 are not our business02:05
xuhaiweijust ask an common question02:05
Qimingsay you a cluster of heat stacks and you want that cluster to be autoscaled02:06
xuhaiweiif we dont care about the instance inside a stack, we auto-scale this stack, which az will the instance go?02:06
Qimingtell me the metrics you want to trigger this "auto" scaling02:06
Qimingcross-az policy won't deal with heat stack02:07
Qimingyou cannot put heat stack into an AZ02:07
Qimingcinder, if I'm remembering correctly, has its own definition of AZ02:08
Qimingthere are also questions how to do cross AZ references from one Nova server to a Cinder volume02:08
Qimingbefore we have some basic functional support from nova, cinder and other services, you won't be able to put a Heat stack as a whole into an AZ02:09
xuhaiweiI still think stack autoscaleing doesn't make sense, because you have no policy to evaluate a whole stack02:10
Qimingwell it can be done02:10
Qimingpretty easily02:10
Qimingyou can not speak for the users02:10
Qimingcloud monitoring is another domain02:11
Qiminga huge one02:11
Qimingeven today, using ceilometer, you can already do a decent metric collection, aggregation, filtering and then trigger an alarm from that02:11
xuhaiweiok, if you want to scale out a stack, you will create an exactly same stack?02:12
Qimingsame collection of resources02:12
Qimingjust like you are calling Heat twice02:12
*** elynn has joined #senlin02:13
xuhaiweiif the stack has only one node, it seems ok, but it has many nodes, it's dangeous02:13
Qimingyes, if you are scaling a stack of many nodes02:13
Qimingyou will take care of the nodes you created from Heat02:13
Qimingone way of doing that is to add metadata to your nova server resource in the stack02:14
Qimingso that ceilometer can filter out which resource comes from which stack02:14
Qimingthen ceilometer will do the correct aggregation and evaluate whether an alarm is needed02:14
Qimingif is not Senlin's business02:15
xuhaiweiif a cluster contain some nova nodes and some stack nodes, the policy attached to it will still work?02:15
Qimingwe don't allow a cluster to contain nodes of different profile type02:16
xuhaiweisorry, I forget that02:16
xuhaiweia stack cluster can be attached some policies?02:17
Qimingcheck the policy definition02:18
Qimingeacy policy is supposed to tell what profile types it can support02:18
Qimingnot all policies will be applicable to all profiles02:18
Qimingxuhaiwei, I'm wondering if we should take some actions about this:
Qimingor it should be a patch to SDK02:39
Qimingdon't have bandwidth to following at the moment02:39
xuhaiweiOk, I will read this patch02:40
xuhaiweiwe will discuss this in the API  topic at the meetup02:41
Qimingwe don't have an API sesssion02:41
xuhaiwei1-11 project status: API , just share some information?02:43
xuhaiweidon't discuss something?02:43
Qimingyes, it is a share02:43
Qimingday one is about meeting with users02:43
xuhaiweigot it02:43
Qimingadded one on day 202:44
Qimingputting you as the lead02:44
xuhaiweiok, thanks02:45
openstackgerritCindia-blue proposed openstack/senlin: Implement node_recover in Profile
*** gongysh_ has quit IRC03:57
*** elynn has quit IRC04:00
*** gongysh_ has joined #senlin04:10
openstackgerritCindia-blue proposed openstack/senlin: Add Recover into Node Actions and Node Model
openstackgerritCindia-blue proposed openstack/senlin: Add Recover into Node Actions and Node Model
openstackgerritCindia-blue proposed openstack/senlin: Add Recover into Node Actions and Node Model
*** elynn has joined #senlin05:12
*** dixiaoli has joined #senlin05:14
*** gongysh_ has quit IRC05:16
*** gongysh_ has joined #senlin05:35
*** Yanyanhu has joined #senlin05:36
Qimingplease jump onto this:
Yanyanhusure, will check it06:35
Qimingit smells dangerous06:35
Yanyanhuwill confirm it06:36
YanyanhuQiming, I think this change is safe for us. Actually only loadbalancer has two status related properties: provisioning_status and operating_status06:40
Yanyanhuother lb related resources don't have 'status' attr based on Neutron API interface06:41
Yanyanhuand we are not using this 'status' attr in our lb driver06:41
Yanyanhuso I'm gonna +1 to that patch :)06:44
QimingYanyanhu, okay06:46
Qimingjust want to make sure we are not checking 'status' fields06:46
Yanyanhuyes, understand06:47
QimingIf I'm remembering correctly, we WERE checking status when ever doing some operations06:47
Yanyanhuit could influence our lb driver if so06:47
Yanyanhuyes, everytime we do lb related operation, we will check loadbalancer's provisioning_status and operating_status06:47
QimingI see06:48
Yanyanhuto decide whether the operation is done06:48
Qimingthose are the two status fields we care about06:48
QimingYanyanhu, have you verified if Brian's patch to metadata works?07:09
Yanyanhuhaven't yet07:09
Yanyanhuhaven't read his latest change07:09
Yanyanhulet me have a look07:09
Qimingokay, he just proposed a new patchset07:09
Yanyanhuyes, the old one looks good I think07:10
Yanyanhunot sure image part is also ready. But I noticed he didn't -workflow this time07:10
Qimingyep, I think he has tested it07:11
*** elynn has quit IRC07:11
*** elynn has joined #senlin07:11
*** elynn has quit IRC07:22
*** elynn has joined #senlin07:22
*** elynn has quit IRC07:27
*** elynn has joined #senlin07:27
*** elynn has quit IRC07:32
*** elynn has joined #senlin07:33
*** dixiaoli_ has joined #senlin07:50
*** dixiaoli_ has quit IRC07:51
*** dixiaoli has quit IRC07:54
openstackgerritQiming Teng proposed openstack/python-senlinclient: Remove references to oslo incubator modules
openstackgerritYanyan Hu proposed openstack/senlin: Rework get_details of nova server profile
openstackgerritYanyan Hu proposed openstack/senlin: [WIP]Rework policy check for built-in policy types
Yanyanhuhi, Qiming, just propose a patch for policy_check rework. It is still in progress, but any comment is very helpful08:54
YanyanhuI haven't revised test cases for it08:55
YanyanhuBTW, this problem in sdk is still there08:55
Yanyanhuneutron will throw exception when accepting an unexpected argument in request body08:56
Yanyanhuthis behavior is different from Nova. Nova just ignore this kind of arguments silently08:57
Qimingcan we revise neutron resource directly?08:57
Qimingif it is generic enough, we can propose a revision at resource level08:58
Qiminglixinhui, there?08:58
Yanyanhuyes, I think Brian want this kind of revision. But he also didn't have very good idea for this issue08:58
YanyanhuI think08:58
Yanyanhurevising lb related resource is a possible way to address this issue08:59
Qimingif it is only a neutron problem, we can revise itself08:59
Yanyanhuwithout influence other resources08:59
Yanyanhuunderstand, will try it08:59
Qimingelynn is pinging me about austin summit proposal09:00
Qimingdo you have any ideas? about senlin, what can we propose?09:00
Yanyanhuoh, right, I received his mail, almost forgot this...09:00
elynnyes, Internal review will start from 1/11, if we need to catch up with internal review, need to submit them befoer 1/1109:03
QimingI'm thinking if the health management thing can be a feature to show case09:07
xuhaiweiI think we have not introduced Senlin to openstack community formally yet09:08
xuhaiweiexcept Heat team09:09
Qimingwe have proposed senlin deep dive to vancouver, no luck, to tokyo, no luck09:12
Qimingwhy don't we try a 3rd time09:12
Yanyanhuwe should09:13
*** dixiaoli has joined #senlin09:13
elynnVM HA is a good use case to show in summit, just don't know how tc choose topic...09:13
Qimingso we propose two?09:14
Qimingone is about a deep dive about senlin09:14
Qimingthe other is about a real use case09:14
Qimingit would be controversial if we propose a topic like "Clustering containers in an OpenStack native way."09:15
xuhaiweiwe should make a cool topic name anyway, it can add the possibility to be chosen09:18
xuhaiweishould we still talk about container?09:18
Qimingout of magnum's context09:19
Qimingwe can do that09:19
xuhaiweiwe can put that in the future job of Senlin introduction session09:19
openstackgerritMerged openstack/senlin: Fix KeyError caused by getting 'security_groups' from server_data
openstackgerritMerged openstack/senlin: Updated from global requirements
*** lixinhui_ has joined #senlin09:42
Qiminglixinhui_ is back09:44
Qimingso ...09:44
Qimingabout submitting someting to austin summit09:44
QimingIBM internal deadline is next Monday09:44
xuhaiweiSenlin introduction again, about the background, the structure, where are we, future jobs09:46
lixinhui_I like Elynn's mention to VM HA use case09:50
lixinhui_We should china mobile to vote for senlin09:50
lixinhui_we should ask china mobile to vote for senlin09:51
lixinhui_Seems container is very popular in openstack also09:52
Yanyanhunew gerrit UI is terrible, sigh...09:54
Qimingit has been popular for two years now09:54
Qimingdevelopers are never grouped into human09:55
Yanyanhuwe are people...09:55
lixinhui_I am thinking if it is proper tp provide a talk/demo like AWS10:03
Qiminglike AWS ?10:04
lixinhui_as in the blood of senlin, better support to scaling, lb and health management10:04
Qimingnot yet, :)10:04
Qimingthat is the vision10:05
Qimingback in Tokyo, Heat team was asking a question: what should Heat do with the AWS autoscaling API if autoscaling is offloaded to Senlin10:06
QimingI told them this API should be taken over by Senlin as well10:06
Yanyanhuso an AWS-compatible API might be needed in future10:06
QimingAngus and Steve Baker was cheering10:06
Qimingapplauding I think10:06
Yanyanhuthey sufferred from this kind of thing I think :)10:07
Yanyanhuto keep compatible with another API interface10:07
openstackgerritQiming Teng proposed openstack/senlin: Rework DB sorting functions
elynnSo we directly provide an API at senlin side or to organize senlin resources in heat side?10:08
lixinhui_it is a headache before senlin10:08
Qimingthey were trapped10:08
Qimingtrapped by something they woud rather stay away from10:08
xuhaiweiso can't we drop AWS autoscaling API in senlin?10:09
xuhaiweiwhy should we receive them?10:09
Qimingxuhaiwei, it is a community wide requirement10:10
Qimingthere are requirements for OpenStack to support AWS APIs10:10
*** elynn has quit IRC10:11
xuhaiweiec2 api has been kicked out from Nova10:11
xuhaiweiHeat can't kick AWS autoscaling API?10:11
QimingI think there is still a work group doing that10:11
xuhaiweimanage them somewhere else10:12
Qimingnova team ...10:12
Qimingno comment10:12
lixinhui_I can not say other users, VMware users really like AWS alike things10:12
lixinhui_I mean the VIO users, VIO is openstack product10:12
lixinhui_running on vSphere10:12
Qimingnova cores believe they are Gods10:13
lixinhui_it has AT&T and Nike ... as customers10:13
Qimingyes, I do hear a lot requirements on mailinglist, people need AWS APIs sometimes10:14
Qimingbut anyway, for senlin, I don't think it is a high priority10:14
Qimingjust put it in FEATURES.rst at the moment10:14
Qimingabout the proposals10:15
Qimingmy suggestion10:15
lixinhui_maybe nowdays, we could mentions these advantages as a part of senlin introduction10:15
Qimingwe propose two10:15
Qimingone is about a generic intro to Senlin10:15
*** Yanyanhu has quit IRC10:16
Qiminganother is about a use case, say VM health management10:16
xuhaiweiok, the chance of being chosen is doubled10:17
lixinhui_Just try our best, and collect name list to vote then...10:20
*** Qiming has quit IRC10:22
*** dixiaoli has quit IRC11:11
*** Qiming has joined #senlin11:14
openstackgerritCindia-blue proposed openstack/senlin: Add Description about Overriding Profile
openstackgerritCindia-blue proposed openstack/senlin: Add Description about Recover of Profile
openstackgerritCindia-blue proposed openstack/senlin: Add Description about Recover of Profile
openstackgerritCindia-blue proposed openstack/senlin: Add Description about Recover Function in Profile
*** lixinhui_ has quit IRC13:15
openstackgerritMerged openstack/senlin: Add 'wait_for_delete' call for functional tests
openstackgerritQiming Teng proposed openstack/senlin: Fix sample placement policies
openstackgerritMerged openstack/senlin: Rework DB sorting functions
openstackgerritQiming Teng proposed openstack/senlin: Add utility function for parsing sort parameter
*** Qiming has quit IRC16:12
*** rebase has joined #senlin17:55
*** bdrich has joined #senlin18:18
*** rebase has quit IRC19:06
*** junxu has quit IRC19:29
*** openstackgerrit has quit IRC20:23
*** openstackgerrit has joined #senlin20:25
*** dixiaoli has joined #senlin22:45
*** dixiaoli_ has joined #senlin22:55
*** dixiaoli has quit IRC23:01
*** dixiaoli_ has quit IRC23:01
*** dixiaoli has joined #senlin23:03
*** Qiming has joined #senlin23:32
*** bdrich has quit IRC23:34

Generated by 2.14.0 by Marius Gedminas - find it at!