Tuesday, 2016-03-01

*** Qiming has quit IRC00:17
openstackgerritMerged openstack/senlin: Add unit test for cluster check action  https://review.openstack.org/28589601:00
*** Qiming has joined #senlin01:13
*** zzxwill has joined #senlin01:13
zzxwillMorning Qiming.01:17
openstackgerritzzxwill proposed openstack/senlin: separate two different commands in two lines  https://review.openstack.org/28634801:23
openstackgerritQiming Teng proposed openstack/senlin: Simplify action initialization  https://review.openstack.org/28635101:37
*** lixinhui_ has joined #senlin01:38
lixinhui_Qiming, there01:38
lixinhui_last night, I submit this https://review.openstack.org/#/c/286131/01:38
lixinhui_on hypervisor resource01:39
lixinhui_it will be nice if you can help review01:39
lixinhui_hope it can catch up the timeline01:39
Qimingcan we find a hypervisor by name?01:41
lixinhui_I thinl01:43
Qimingcan you help try?01:44
lixinhui_Will confirm it then return to you01:44
*** Yanyanhu has joined #senlin02:05
*** elynn has joined #senlin02:11
zzxwillMorning Yanyan and Elynn.02:14
Yanyanhumorning :)02:28
openstackgerritQiming Teng proposed openstack/senlin: Simplify action initialization  https://review.openstack.org/28635102:39
*** dixiaoli has quit IRC02:41
*** idonotknow_ has joined #senlin02:55
openstackgerritzzxwill proposed openstack/senlin: separate two different commands in two lines  https://review.openstack.org/28634803:13
*** yuanying has joined #senlin03:21
*** yuanying_ has quit IRC03:23
*** yuanying has quit IRC03:27
*** zzxwill has quit IRC03:36
*** zzxwill has joined #senlin03:44
*** elynn has quit IRC03:59
*** zzxwill has quit IRC04:03
*** yuanying has joined #senlin04:10
*** zzxwill has joined #senlin04:14
*** elynn has joined #senlin04:26
*** elynn has quit IRC04:30
*** elynn has joined #senlin04:31
*** zzxwill has quit IRC04:32
*** zzxwill has joined #senlin04:37
*** zzxwill has quit IRC05:20
openstackgerritCindia-blue proposed openstack/senlin: Fix Health Manager Problems  https://review.openstack.org/28638705:28
*** zzxwill has joined #senlin05:44
openstackgerritCindia-blue proposed openstack/senlin: Fix Health Manager Problems  https://review.openstack.org/28638805:46
*** dixiaoli has joined #senlin05:49
openstackgerritCindia-blue proposed openstack/senlin: Fix Health Manager Problems  https://review.openstack.org/28638805:50
openstackgerritEthan Lynn proposed openstack/python-senlinclient: Fix node update receive a assertion error  https://review.openstack.org/28639105:53
openstackgerritMerged openstack/senlin: separate two different commands in two lines  https://review.openstack.org/28634805:59
*** shu-mutou-AWAY is now known as shu-mutou06:10
Qiminghi, guys06:11
Qimingtoday we will have another weekly meeting06:12
QimingI have put an agenda item to the meeting agenda06:12
Qimingwe need to make a plan for the design summit06:13
openstackgerritOpenStack Proposal Bot proposed openstack/senlin-dashboard: Imported Translations from Zanata  https://review.openstack.org/28639706:14
Yanyanhuok, will think about it06:14
Qimingttx is asking each team to present an estimation of space needs06:15
Qimingwe will feed him with some numbers: # fishbowl slots, # workroom slots,  and contributor meetup06:16
Yanyanhumaybe we can list topics we want to discuss in etherpad06:16
Yanyanhulike what we did for midcycle meetup06:17
Qimingit would be a pretty easy math if we have topics at hand06:17
Yanyanhuyes, maybe we can have a discussion and give a list in tonight weekly meeting06:18
Qimingthat's the reason I'm communicating with you beforehand06:18
Qiming1 hour is too short for brainstorming and discussion06:18
QimingI'm creating a etherpad06:18
Qimingshould we call it senlin-mitaka-summit or senlin-newton-summit?06:19
Yanyanhumaybe austin-summit?06:20
Yanyanhuthis summit is for mitaka release, right?06:20
Yanyanhuso mitaka-summit is also ok I think06:20
haiwei_should be for newton cycle06:20
Qimingwhen searching around, I found this: https://wiki.openstack.org/wiki/Design_Summit/Mitaka/Etherpads06:21
Yanyanhuok, so the coming one should be newton06:23
*** zzxwill_ has joined #senlin06:33
*** zzxwill_ has quit IRC06:33
*** lixinhui_ has quit IRC06:34
openstackgerritYanyan Hu proposed openstack/senlin: Rework to_dict method of sdk resource in functional test  https://review.openstack.org/28640306:41
openstackgerritYanyan Hu proposed openstack/senlin: Rework to_dict method of sdk resource in functional test  https://review.openstack.org/28640306:51
*** lixinhui_ has joined #senlin06:54
openstackgerritEthan Lynn proposed openstack/python-senlinclient: Fix node update receive a assertion error  https://review.openstack.org/28639106:58
*** lixinhui_ has quit IRC07:10
openstackgerritYanyan Hu proposed openstack/senlin: Add functional test for node get details  https://review.openstack.org/28641707:14
*** zzxwill has quit IRC07:15
*** zzxwill has joined #senlin07:20
Qiminghi, guys07:27
*** lixinhui_ has joined #senlin07:28
Qimingjust got a question from i18n team07:29
Qimingthey were asking whether we can have a liaison on translation07:29
Qimingit is okay if we don't have one07:30
Qimingjust please speak up if you want to assume such a role07:30
zzxwillI translated some strings for senlin and i18n core said senlin cores will approve the translation merge.07:37
zzxwillBut I don't know the following process.07:37
Qimingzzxwill, so it means we will need one core member from senlin to review the translations?07:39
Qimingzzxwill, do you have a link to the patch?07:40
Qiminglixinhui, there?07:40
zzxwillHold on for a second.07:40
lixinhuiYes, Qiming07:40
Qiminga question about the vSphereDRS policy07:41
QimingI'm working on generalizing it to a generic affinity policy07:41
Qimingadding a property named 'enable_drs_extension' which defaults to False07:41
Yanyanhuhi, Qiming, the translation is between english and chinese?07:41
openstackgerritCindia-blue proposed openstack/senlin: Fix Health Manager Problems  https://review.openstack.org/28638807:42
Qimingyes, Yanyanhu07:42
zzxwillQiMing, there is no patch for the translation. https://translate.openstack.org/project/view/senlin07:43
zzxwillOne can translate senlin as an i18n translator in Zanta.07:43
Qimingah, looks like a very nice website07:44
openstackgerritCindia-blue proposed openstack/senlin: Fix Health Manager Problems  https://review.openstack.org/28638807:45
zzxwillSo please confirm the i18n process with i18n team:) (I am not familiar with it and I'm afraid I might mislead you guys.)07:47
Qimingzzxwill, the translations will be proposed back to senlin when it is (half) done07:47
Qimingzzxwill, a million thanks for starting this07:50
zzxwillHappy to hear that as a small potato:)07:53
lixinhuiQiming, I am okay with your change on vSphereDRS07:53
*** lixinhui_ has quit IRC07:54
lixinhuiIf things goes on smoothly, we can customise the general policy for specific user case07:55
lixinhuiExisting vSphereDRSplacement does not expose DRS specific rules actually besides affinity and unti-affinity.07:57
Qiminglixinhui, but it does search for host names07:59
lixinhuiThat is reason we discussed once to name it vSphereDRS07:59
Qimingwhich is not necessary for all cases08:00
Qimingwhat I want to confirm is ...08:00
Qimingin current implementation, the policy is inserting a 'zone' key08:00
Qimingand the 'zone' value would default to 'nova'08:00
QimingI want to know whether this is very specific to vsphere08:00
lixinhuiThe logic can be enabled with your add condition08:01
lixinhuithat is common08:01
lixinhuieven for vsphere nova08:01
Qimingyou mean, when I'm specifying 'group' as part of the 'scheduler_hints', I HAVE TO provide a 'zone' key?08:02
QimingI am not seeing that a requirement from nova api08:02
Qimingit sounds to me that only vSphere DRS requires a zone key, and that zone value has to be in format like '<zone_name>:<host_name>'08:04
lixinhuithat is common08:04
lixinhuiyou mean can we remove hostname?08:04
Qimingokay, let's start all over again08:05
Qimingquestion 1: to do an affinity based placement, we don't have to specify an availability zone, this is not required by nova api, right?08:05
lixinhuino. it is not enforced08:06
Qimingquestion 2: we can support an availability zone name as optional key. it means the affinity/anti-affinity will happen within an availability zone, right?08:08
Qimingthis question is not specific to vSphere either08:08
lixinhuiit can cooperate with zone08:09
Qimingokay, now point 308:09
Qimingvsphere (as a special kind of 'hypervisor') will need the availability zone to be specified in format '<zone_name>:<drs_host>'08:10
lixinhuiyes, that is way to find the vSphere cluster as a hypervisor backend08:11
Qimingbecause the 'drs_host' usually contain more than one physical machine --- some vsphere internals08:11
Qimingso, I will make the affinity policy to support an optional availability zone name, but I'm not defaulting that to 'nova'08:11
Qiminghowever, if 'enable_drs_extension' is set to True, we will check the zone name, and default that to 'nova'08:12
Qimingis this a viable solution?08:12
lixinhui'nova' is the default zone name if no special creation actions with zone08:13
Qimingyes, but in production environment, I don't know what happens08:13
Qimingin devstack, that is true08:13
lixinhuiI do not quiet sure I understand this question08:16
lixinhuinova is default set zone name by common nova08:16
lixinhuibut user can generate their own zone08:16
Qimingthat is the exactly the point08:16
Qimingif 'nova' is the implict az08:17
Qimingwe don't need to make that assumption in our policy, let nova do it08:17
Qimingif we make an assumption that an az is 'nova' and pass that value to nova08:17
Qimingthe implication is different08:17
Qimingit means we are CHOOSING nova, instead of letting nova to make decision08:18
Qimingthanks, now gained a better understanding of this08:19
lixinhuiglad to help.08:19
Qimingvsphere won't accept something like ':host1' as an availability zone name ... I guess08:19
Qimingso there has to be an assumption08:19
Qimingin other words, the guessing is still needed for vsphere, in order to form a correctly formated 'zone name'08:20
lixinhuiyes, we need suggest user who want enable it to check correct zone08:21
openstackgerritMerged openstack/senlin: Rework to_dict method of sdk resource in functional test  https://review.openstack.org/28640308:21
Qimingthey can still use the general 'availability_zone' key for customization08:21
lixinhuizone is higher concept than hypervisor, vsphere cluster is one kind of backend to hypevisor08:22
lixinhuiso generalized zone is the ideal zone08:23
*** dixiaoli_ has joined #senlin08:24
lixinhuivSphere nova plugin will follow rules of community nova on naming of AZ08:25
openstackgerritCindia-blue proposed openstack/senlin: Fix Health Manager Problems  https://review.openstack.org/28638808:28
Yanyanhuhi, Qiming, I have a question about cluster update API behavior08:46
Yanyanhushould cluster status be marked as 'UPDATING' when the cluster update API returns08:47
*** openstackgerrit has quit IRC08:48
Yanyanhuactually, we have made some related discussion with ethan before08:48
*** openstackgerrit has joined #senlin08:49
Yanyanhuabout how to check status changing when updating cluster resource in heat08:49
Yanyanhuhe mentioned the cluster status will keep staying at 'ACTIVE' before update action is scheduled08:51
Yanyanhutherefore, the status check will give incorrect result if it is done very soon after cluster update API request returns08:51
Yanyanhuguess restful API should have some standard or discipline about this, will check it08:52
openstackgerritEthan Lynn proposed openstack/senlin: Make map schema resolve json serialized map value  https://review.openstack.org/28644708:53
elynnYanyanhu, I think now cluster update will return an action id and we can check the status of action ID.08:55
Yanyanhuyes, elynn, I know we can get action_id from resp. Just curious whether resource should have been marked as 'UPDATING' when update API returns08:56
elynnit might not be good to change cluster status to update since cluster might not actually go into 'update' status...08:57
Yanyanhuyea, that's our current understanding08:57
elynnThat's what Qiming told me before...08:57
Yanyanhunot sure whether RESTful has some discipline about this question08:57
Yanyanhusince it is all about resource operation :)08:58
elynne.g. a scale-out action is on-going, and update action might need to wait for it to complete.08:58
elynnIf only consider restful API guidelines, I'm not so sure...08:59
Yanyanhuelynn, yes, this understanding makes sense.08:59
QimingI don't think the status change of resources is business of APIs08:59
Yanyanhujust also not sure whether there is guideline :)08:59
Qimingthere's no guideline on this I believe09:00
Yanyanhusince it is not in the scope of 'API interface'?09:01
Qimingthe only 'status' you get is the HTTP status code09:01
YanyanhuYes. that is the status of http resp09:02
Qimingthat code only implies whether the request has been processed (200), accepted (202)09:02
Yanyanhuhmm, so RESTful doesn't provide constaint on what the resource looks like, e.g. what its status is or even whether it has a property named as status09:03
Yanyanhunot its responsibility09:04
Yanyanhuseems I thought too much :)09:05
elynnHi Qiming , wanna know your thoughts about https://review.openstack.org/#/c/286447/ .09:08
Qimingelynn, acceptable09:12
Qimingoh, no09:12
Qimingif the value provided is a string, it has to be able to be parsed into a map09:12
Qimingline 354 is guarding against invalid inputs09:14
elynnif a string can be parsed to a map, will raise error.09:14
elynncan not09:14
Qiminggot it09:14
elynnIf it's acceptable, I will abandon the bug in heat.09:17
openstackgerritEthan Lynn proposed openstack/senlin: Make map schema resolve json serialized map value  https://review.openstack.org/28644709:19
elynnSeems still not work as expected, sigh...09:24
*** Yanyanhu has quit IRC09:28
elynnoh, it worked as expected, forget to restart senlin-engine...09:28
openstackgerritQiming Teng proposed openstack/senlin: (WIP)Revise DRS policy to be a generic affinity policy  https://review.openstack.org/28646109:31
Qiminglixinhui, please help review the above patch ^09:33
Qimingstill work needed on reworking the unit tests09:33
*** idonotknow_ has quit IRC10:01
*** idonotknow_ has joined #senlin10:02
*** elynn has quit IRC10:16
openstackgerritBéla Vancsics proposed openstack/senlin: Use assert(Not)Equal/Less/Greater  https://review.openstack.org/28647710:22
*** dixiaoli_ has quit IRC10:28
*** Qiming has quit IRC10:30
-openstackstatus- NOTICE: Gerrit is going to be restarted due to poor performance10:38
*** ChanServ changes topic to "Gerrit is going to be restarted due to poor performance"10:38
*** ChanServ changes topic to "IRCLog: http://eavesdrop.openstack.org/irclogs/%23senlin/ | Bugs: bugs.launchpad.net/senlin | Review: https://review.openstack.org/#/q/project:openstack/senlin,n,z"10:45
-openstackstatus- NOTICE: gerrit finished restartign10:45
*** idonotknow_ has quit IRC11:28
*** zzxwill_ has joined #senlin11:32
*** zzxwill has quit IRC11:33
*** zzxwill_ has quit IRC11:34
*** zzxwill has joined #senlin11:34
*** Qiming has joined #senlin11:36
*** shu-mutou is now known as shu-mutou-AFK11:54
*** idonotknow_ has joined #senlin11:59
openstackgerritMerged openstack/senlin: Simplify action initialization  https://review.openstack.org/28635112:03
*** xuhaiwei has joined #senlin12:53
*** yanyanhu has joined #senlin12:54
openstackgerritCindia-blue proposed openstack/senlin: Fix Health Manager Problems  https://review.openstack.org/28638812:58
*** idonotknow_ has quit IRC13:01
openstackgerritCindia-blue proposed openstack/senlin: Revise DB API to Ensure Delete Success  https://review.openstack.org/28657513:16
Qiminglixinhui, meeting?13:31
Qimingyanyanhu, continue the discussion here14:01
Qimingwon't have a conclusion today14:01
yanyanhuhonestly, I understand what you mean about the different between IaaS and PaaS14:01
QimingI'm not anticipating a conclusion this week, this month14:01
yanyanhuagree that's not that important actually14:01
Qimingeventually, there will be containers running on baremetal14:02
Qimingthe big security gap will be closed14:02
yanyanhubut I think one problem is the design of Senlin should be different if we want to act as some services like k8s or Mesos14:02
Qimingalthough the tech is not really that ready today14:02
yanyanhuyes, that's right14:02
yanyanhuwe miss some important conception in senlin as of now14:03
Qimingthe value of senlin lies in its unification of cluster management14:03
yanyanhuif we want to provide that support14:03
Qimingare you talking about pod? service? ...14:03
yanyanhuyes, service, not pod14:04
Qimingor 'offers'?14:04
yanyanhuor tasks14:04
yanyanhuyes, I think we miss them currently14:04
xuhaiweiso what is the mostly different things of Senlin's container cluster and K8s ?14:04
yanyanhuservice is an element supported by k8s I think14:05
Qimingright, from users perspective, they think about micro-service, for example14:05
yanyanhupod is one thing to undertake it14:05
lixinhuiI think who do it fistly then who will catch the eyeball14:05
Qimingand there will be a lot developers educated by their terms14:06
yanyanhubased on my understanding, Senlin Now is a generic cluster management service14:06
Qimingso ... to deploy a cluster of 'task's or 'service's14:06
yanyanhubut we haven't touched stuff that deployed upon cluster that much14:06
yanyanhutill now14:06
Qimingwhat are the artifacts you will feed to the service (be it senlin or not)14:07
yanyanhuQiming, it's possible14:07
yanyanhuactually, it doesn't have to be connected with Container cluster14:07
Qimingtoday, it is just a docker image or some other image, right?14:07
yanyanhuapp/service can be deployed in cluster of any kind of compute resource14:07
yanyanhuQiming, I think we need something like 'DEPLOYMENT'14:08
yanyanhuto finish this gap14:08
QimingI see the direction you are going14:08
yanyanhuit's kinda like service deployment in k8s14:08
yanyanhuor task framework in Mesos14:09
Qiminghowever, 'deployment' is still a little be too abstract14:09
yanyanhuQiming, yes, but that's our advantage, right14:09
QimingI prefer we aiming high while working low14:09
yanyanhuwe don't want limit the cluster type to Container14:09
yanyanhuQiming, agree14:09
yanyanhubut the design should be generic14:09
yanyanhuactually just like what we have done on cluster management :)14:10
Qimingso, speaking of deployment, can it be a jar, a war, an ISO, a vmdk?14:10
yanyanhucluster can be composed of any kind of resources14:10
yanyanhuif we want to introduce this conception to Senlin, maybe we'd better make it generic14:11
yanyanhuas generic as cluster14:11
Qimingi was imagining that 'custom actions' can play a big role here14:11
Qimingor something like ansible14:11
Qimingor just mistral?14:11
yanyanhuem, not sure, maybe it varies cross the resource type at the back of node?14:12
Qimingyou mean perform some operation on a cluster as a whole? where the 'some operation' here can be14:12
Qimingdeploy a container image?14:12
yanyanhuno, I mean deploy an App/service14:12
Qimingright, it will vary a lot14:13
Qimingso ... i was asking14:13
Qimingwhat are you referring to by 'App/service'14:13
yanyanhuah, I means something like a webserver14:13
Qiminghow is that 'App/service' packaged14:13
Qimingwhat is the artifact to be deployed14:13
yanyanhuor db. or a spark services:)14:14
Qimingstill don't get the point14:14
yanyanhudepends on what kind service it is14:14
yanyanhumaybe a docker image14:14
yanyanhumaybe just a script14:14
Qiminga service is the function provided by some software, right?14:14
Qimingan app is a software14:15
Qimingyou package your software into a distributable format14:15
Qimingso that people can deploy again and again14:15
Qimingthat 'distributable format' is the artifact I am talking about14:15
yanyanhuok, I see.14:16
Qimingtoday, the most popular one is docker image14:16
Qimingyes today, it was jar/war14:16
Qimingtomorrow, it could be some xxxkernel/xxxos package14:17
yanyanhuok, for docker image, maybe we can deploy it in a cluster of VM14:17
yanyanhuwhere docker service is enabled in all VMs14:17
Qimingif we can deploy docker image14:17
Qimingcan we say we are deploying the 'app/svc' you talking about?14:17
yanyanhuit's kind of app/svc14:18
Qimingthere are other kind of distributions14:18
Qimingjust like TOSCA has been trying to abstract14:18
yanyanhuso what I want to express is we should make it a generic conception14:19
Qimingthey are even talking about web service bundles14:19
yanyanhunot just focus on Container14:19
Qimingtotally makes sense14:19
yanyanhuat least when we design the conception14:19
Qimingwe can call it a pdu14:19
yanyanhumaybe we have special 'driver' or something for container cluster14:19
Qimingor dup14:19
Qimingdeployable unit package14:20
Qimingit is also 'dup'lication ... we are reinventhing things, but for a good reason14:21
Qimingif the package type is 'docker' we call docker api to deploy14:21
Qimingif the du type is a script, we ssh into a vm and run it14:22
yanyanhuyes, any compute resource can run script14:22
Qimingalright, you are leading us to the software config/software deployment trap14:22
Qimingwe'll need to monitor their progress, their log, their return value14:23
Qimingwe'll need to teach users to find logs for debugging ...14:23
yanyanhuhmm, after thorough think, I really did...14:23
Qimingbut anyway, that is some great vision14:23
yanyanhubut if you want to manage a deployment or mabye lifecycle of an App/Service14:24
yanyanhuwe may have to think about these issues14:24
yanyanhuactually, there are also other things need to think about14:24
Qimingcome on, lifecycle is murano's scope14:24
yanyanhue.g. for container cluster, we need to consider the network connection inside cluster14:24
Qimingthe top concern for me ... is scheduling14:24
yanyanhuthe loadbalancer, the proxy14:25
Qimingwe need a basic, very naive scheduler14:25
Qiminganything that works for container networking, we can use14:25
yanyanhuhmm, can swarm help to do this?14:25
Qimingwe are consumer of those technologies14:25
yanyanhuI mean the basic scheduling14:25
Qimingdon't think so14:26
Qimingscheduling problem can be decomposed to several sub-domains14:26
yanyanhuk8s is also not what we want to talk with14:27
Qimingfirst you will need to know what you need for each UOW (unit of work)14:27
Qimingthen you will need to know what you have at hand (resource provisioning, a thousand dimensions here)14:27
xuhaiweiSo Qiming, for container cluster, you are willing to have almost the same design as vm cluster?14:27
Qimingfinally, you need to pick some location to put your UOW14:28
yanyanhuQiming, this is exactly what a Mesos scheduler does14:28
yanyanhuxuhaiwei, really anxious to have more talk with you guys about this topic :)14:29
Qimingall schedulers do the same type of jobs14:29
yanyanhureally need more thinking on it14:29
Qimingxuhaiwei, yes, that is what we call a generic clustering service14:29
Qimingit's very late14:30
xuhaiweiyanyanhu, yes, need to learn more about it14:31
QimingI don't mind if you want to continue dig more14:31
yanyanhuxuhaiwei, yes, and hope we can figure out the correct way to do so14:31
yanyanhumaybe what I said was all wrong :)14:31
xuhaiweiI will go to bed soon14:31
yanyanhuyes, also really need to go14:31
yanyanhuotherwise, have life danger14:31
Qimingyes, don't waste time doing something useless, at least14:31
yanyanhutalk to you guys later14:32
yanyanhuhave a good night14:32
xuhaiweisee you tomorrow14:32
Qimingto me, (secretly), magnum is useless, :)14:32
Qiminggood night14:32
*** yanyanhu has quit IRC14:32
*** xuhaiwei has quit IRC14:32
*** zzxwill has quit IRC14:44
openstackgerritLiuqing Jing proposed openstack/senlin-dashboard: Delete useless return statement  https://review.openstack.org/28664515:08
*** jdandrea_ has joined #senlin15:23
*** zzxwill has joined #senlin15:24
*** jdandrea has quit IRC15:26
*** zzxwill has quit IRC15:47
*** Qiming has quit IRC16:17
*** openstackgerrit has quit IRC18:18
*** openstackgerrit has joined #senlin18:19
*** openstack has joined #senlin19:18
*** openstackstatus has joined #senlin19:18
*** ChanServ sets mode: +v openstackstatus19:18
*** openstackgerrit has joined #senlin19:20
*** Qiming has joined #senlin23:41
*** zzxwill has joined #senlin23:58

Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!