13:00:38 #startmeeting senlin 13:00:39 Meeting started Tue Apr 11 13:00:38 2017 UTC and is due to finish in 60 minutes. The chair is Qiming. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:00:40 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:00:42 The meeting name has been set to 'senlin' 13:01:11 hi,QiMing 13:01:23 evening :) 13:01:27 hi, evening 13:01:38 evening, guys 13:01:39 evening:) 13:02:17 pls add items to the agenda if you have things to talk about 13:02:19 #link https://wiki.openstack.org/wiki/Meetings/SenlinAgenda#Weekly_Senlin_.28Clustering.29_meeting 13:02:55 #topic pike work items 13:03:04 #link https://etherpad.openstack.org/p/senlin-pike-workitems 13:03:58 API errors check, no seeing any patches 13:05:13 I will do that 13:05:35 VDU profile, not seeing progress from ethan 13:06:00 CLUSTER_CHECK improvement, still there for review 13:07:12 I was hoping a revision that makes some actions self destroyable 13:07:28 Qiming 13:07:47 self destroyable in db layer or in senlin ? 13:08:27 in db layer 13:08:38 or else the logic propagates to many places 13:08:50 ok 13:09:07 a NODE_CHECK can be maked transient 13:09:27 and it will remove itself from the dependency table after sucessfully completed 13:09:54 Node adoption: two patches for review 13:10:06 still stuck at API endpoint 13:10:28 basically, node_adopt_preview and node_adopt are two different operations 13:10:53 because the first one should return the (complete) spec of the profile to be created, while 13:11:07 the second will return the node properties after a successful adoption 13:11:53 so ... the first api may eventually land at /v1/profiles/adopt, and the second may eventually land at /v1/nodes/adopt 13:12:07 neither one looks decent 13:12:15 would like to know your opinions 13:12:48 ok, spend time to review 13:13:09 the question I raised is not yet there for review 13:13:22 the two patches are some basic mechanisms for node adoption 13:13:40 next thing is about scaling improvement 13:13:49 rui jie is onto it 13:14:30 yes, its in good progress now :) 13:14:32 the plan is to do an optional health check (by invoking node.do_check() directly) before a scale-in operation 13:15:14 it makes good senses to me because this is solving a practical problem without introducing a lot of disruptive changes 13:16:04 next thing to work on is to do an optional converge to the (last) desired_capacity 13:16:16 it is also a wanted feature 13:16:35 about RDO prep, XueFengLiu any update? 13:16:58 In process 13:17:24 alright 13:17:37 pls let team know when there are new progress 13:17:47 health management 13:17:50 Ok, I will 13:18:06 mistral workflow work is stagnated, we need hands on it 13:18:27 yes 13:18:29 I don't think the neutron LB problem has a quick fix 13:18:44 maybe we should put fencing support back to FEATURES.rst 13:20:05 fencing physical nodes are desirable, but we may need some IPMI logics to get that done 13:20:05 fencing virtual machine is already there, i.e. a forced delete 13:20:05 Tempest tests 13:20:05 XueFengLiu is on it 13:20:26 however, my perceiption is that there were some misunderstandings about cloud_backend 13:20:33 yes, start recently 13:20:37 yes, seems so :) 13:20:41 en 13:20:48 but the logics proposed are still very valuable for sovling integration test problems 13:20:49 I guess we don't need this https://review.openstack.org/#/c/455331/ 13:20:53 I read the readme.rst 13:21:20 feel free to propose clarifications or documentation anywhere 13:21:20 creating of new key and network is only required for integration test I feel 13:21:44 It did not mention cloud_backend 13:21:56 don't hesitate don't feel shy about english, team can help you polish it 13:22:30 or you can report a bug and let me know, I can help fix the doc 13:22:35 XueFengLiu, that is our fault to forget to mention it :) 13:22:54 :) 13:23:21 Some where we have add the cloud_backend conf 13:23:28 I believe it has been documented: https://docs.openstack.org/developer/senlin/developer/testing.html 13:23:33 But not in tempest readme.rst 13:23:44 yes 13:24:28 feel free to add a pointer then 13:24:46 ok. will add 13:25:11 about CLUSTER_RECOVER and LB policy, Ruijie_ still have proposals? 13:25:30 np Qiming, will revise it when doing current bp 13:25:47 okay, so will leave the item there 13:25:52 I thought it too simple before 13:26:30 :) nothing is simple in real life 13:26:58 the node.status=WARNING need to be handled in several places.. 13:27:08 yep 13:27:26 better think it through before introducing piecemeal changes 13:27:42 as for NFV work, haiwei has been working on it 13:28:22 here is the link: https://review.openstack.org/#/c/449471/ 13:28:58 when he worked on this he also found a concurrency bug 13:29:03 I believe he is solving it 13:29:47 that's all I have on the etherpad items 13:29:53 anything to add? 13:30:19 no from me 13:30:41 okay, moving on 13:30:48 #topic pike-1 release 13:30:48 ok 13:31:00 I'm planning to cut a p-1 release tomorrow 13:31:07 however, bad things happened 13:31:12 our gate is broken 13:31:24 we have just stood up py35 gate jobs 13:31:40 yes, seems problem with py35 13:31:50 the gate is failing us, which means no patch can be validated for merge 13:32:04 I spent some time just now looking into the logs 13:32:20 it seems that we have glance-api failing during devestack installation 13:32:22 the change in gate side has been merged? 13:32:28 yes 13:32:32 yanyanhu, 13:32:32 the patch for enabling py35 test 13:32:55 could be devstack's problem? 13:33:07 the glance store by default is using swift as its backend 13:33:16 and ... swift is not yet py35 ready 13:33:22 sigh... 13:33:45 so ... need some fix to disable glance install 13:33:57 for api and functional tests, we need nothing 13:34:05 can we remove py35 in gate side? 13:34:08 there must be a way to disable all them 13:34:16 we can set it to non-voting 13:35:02 if you are interested, you can check the log here: http://logs.openstack.org/75/455575/1/gate/gate-senlin-dsvm-tempest-py35-api/66595c2/logs/screen-g-api.txt.gz 13:35:44 glance is trying to invoke swift ... 13:35:53 yes 13:36:01 I saw it 13:36:11 one possible way is to define which services we want to install specificly in job template, rather then using default list. But this is not recommened 13:36:15 So why we support py35 recently? 13:36:24 Can we do this work later? 13:36:38 it is a pike cycle goal for the whole community 13:36:51 you need to carefully add every depedencies one by one 13:37:11 so it could be easy to make mistake and then fail the devstack installation... 13:37:12 another cross-community goal is to have senlin-api deployed as a wsgi module 13:37:31 I have checked our gate job script 13:37:49 we didn't do it very "carefully" 13:38:26 currently, we just use default list plus extra services we need 13:38:43 e.g. http://git.openstack.org/cgit/openstack-infra/project-config/tree/jenkins/jobs/senlin.yaml#n37 13:38:59 this line may have glance installed, and possibly swift 13:39:21 or swift wasn't installed ... so glance is getting connection refused error 13:39:42 but without it, heat installation will fail 13:39:51 since it is not in default project_list 13:40:01 and we were installing nova components using this line, right? 13:40:25 so git_clone won't work... 13:40:29 no 13:40:38 this is only for heat 13:40:49 Yes, this need a repeated attempts to conf local.conf 13:40:52 since heat is installed as a plugin 13:41:17 then how nova gets installed? 13:41:18 but nova/neutron/glance/cinder/keystone are not 13:41:28 those services are in default list 13:41:36 they will get installed by default 13:41:44 in default project_list 13:41:46 yes 13:41:51 so git_clone works with them 13:42:00 nova/cinder/glance/neutron/keystone 13:42:02 even you don't add them to project_list manually 13:42:15 but heat is not... 13:42:22 right, project list is the problem I was suspecting 13:42:24 en 13:42:27 that's why we enable it explicitly and add it to project_list 13:42:29 if you compare this to heat.yaml 13:42:46 http://git.openstack.org/cgit/openstack-infra/project-config/tree/jenkins/jobs/heat.yaml#n50 13:43:04 they enable each and every service explicitly 13:43:08 Qiming, heat's job template is special :) they define all required services specificly using OVERRIDE_ENABLED_SERVICES 13:43:41 okay, that is a good trick 13:43:43 we tried this way, but per Andreas's suggestion, we'd better using default list + extra ones required 13:43:53 so we changed it back :) 13:44:10 I was really curious why glance store was set to swift 13:44:12 and where 13:44:47 could be in glance devstack plugin? 13:44:59 not sure... but that sounds weird... 13:45:13 yes, but I am not seeing any projects gate job explicitly overwriting it 13:45:47 only when s-api is enabled will devstack set the backend to swift 13:45:58 that brings me back to the services set up 13:46:19 because swift is known to be bad wrt py35 13:46:25 we need to exclude it 13:46:51 http://git.openstack.org/cgit/openstack-infra/project-config/tree/jenkins/jobs/heat.yaml#n64 13:47:11 honestly, I do expect a service_exlcude support in devstack gate... 13:47:43 just found there isn't such feature after searching... 13:47:48 that is beyond our control and I don't think we have extra bandwidth on fixing it that way 13:47:59 any way 13:48:06 they only support "services+=", but no "services-="... 13:48:21 so the only way is what heat is using now... 13:48:31 using OVERRIDE_ENABLED_SERVICES 13:49:01 right, we have returned to the same problem now 13:49:12 yes... 13:49:18 will try set the py35 gate nv tomorrow 13:49:26 then cut p-1 release 13:49:32 ok 13:49:43 are you aware of any high priority bugs to fix? 13:49:49 except for this: https://bugs.launchpad.net/senlin/+bug/1681620 13:49:51 Launchpad bug 1681620 in senlin "ACTION failed when can't get the cluster lock" [Critical,New] 13:50:01 nope from me... 13:51:02 Let me see the bug list 13:51:06 okay 13:51:21 #link https://bugs.launchpad.net/senlin/ 13:51:30 #topic open discussions 13:51:43 since neither xinhui or elynn is in 13:52:02 I'm not sure anyone else have summit proposals to discuss 13:52:50 I and haiwei in prepare for the proposal 13:52:52 ok, maybe next week. Only 3 weeks left? 13:53:11 yep 13:53:32 if you need help on polish it, pls share with the team 13:53:48 sure, we will 13:54:04 cool 13:54:08 anything else? 13:54:38 nope 13:54:39 Depends on adoption patch in senlin and the patch in tacker 13:54:58 okay 13:55:30 your work has a hard dependency on node adoption? 13:55:49 not hard dependency 13:56:12 :) 13:56:29 If we have the feature, we can show it better 13:56:51 * Qiming feeling a little bit relieved 13:57:14 you will have it before the summit 13:57:15 :) 13:57:28 ok, greate 13:57:36 so, please review, :D 13:57:50 sure, I will 13:58:02 Thanks Qiming for the hard work 13:58:04 guess that's all for today? 13:58:12 seems so 13:58:14 :) 13:58:22 no more from me 13:58:30 thank you all for joining, take back your two minutes 13:58:32 good night 13:58:34 #endmeeting