02:00:35 #startmeeting mogan 02:00:36 Meeting started Thu Jul 13 02:00:35 2017 UTC and is due to finish in 60 minutes. The chair is zhenguo. Information about MeetBot at http://wiki.debian.org/MeetBot. 02:00:37 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 02:00:39 The meeting name has been set to 'mogan' 02:00:48 o/ 02:01:01 o/ 02:01:06 o/ 02:01:49 o/ 02:02:14 as always, the agenda: 02:02:14 #link https://wiki.openstack.org/wiki/Meetings/Mogan#Agenda_for_next_meeting 02:02:19 let's jump in 02:02:24 #topic announcements and reminders 02:02:35 OpenStack Summit Sydney CFP deadline is July 15, 2017, 14:59 Beijing Time. 02:03:34 zhangyang: how about adding a joint presentation for mogan run in HFBank? 02:04:04 o/ 02:04:28 zhenguo: i'm not sure... you can ask liudong :D 02:04:35 zhangyang: ok 02:04:48 shaohe_feng: we can add a joint one like last summit 02:05:46 wanghao, litao__: how about yours about adopting running baremetals? 02:06:19 zhenguo: OK 02:06:28 zhenguo: I have committed the first patch for getting adopting bms from driver. 02:06:53 zhenguo: and litao__ is working on the manage API. 02:07:00 wanghao: yeah, I mean the presentation for Summit, hah 02:07:13 zhenguo: :) 02:07:22 zhenguo: we can add a one. and I think mogan is good improve than the last summit. 02:07:34 shaohe_feng: yes 02:07:35 zhenguo: I have registed the presentation 02:07:42 wanghao: cool 02:07:45 zhenguo: we should keep to promote mogan. 02:08:03 zhenguo: now just litao_ and I, so if anyone is interesting in this. could join us. 02:08:20 wanghao: you can add me if there's still seat, lol 02:08:35 zhenguo: sure np 02:08:47 wanghao: thanks 02:09:17 zhenguo: :) 02:09:20 ok, let's move no 02:09:25 *on 02:09:47 We will change to let ironic set the node resource class during devstack setup 02:10:04 The project-config patch has landed 02:10:42 please remember to update your devstack local.conf when that's done 02:11:09 it will require IRONIC_USE_RESOURCE_CLASSES = True 02:12:02 cool 02:12:10 hah 02:12:29 ok, another thing is ironic will get rid of ssh driver, we should also clean that up on mogan side 02:12:35 zhenguo: thanks for reminder. 02:12:38 please help to do it if you find anything related to ssh 02:13:21 seems need to update our doc, a quick search can find some ssh* there 02:13:32 zhenguo: does the ssh driver just be deprecated or will be removed soon ? 02:13:40 liusheng: removed 02:13:46 liusheng: it will deprecated for a long time 02:13:53 s/will/has 02:14:00 zhenguo: oops! 02:14:28 zhenguo: we may also need to disable the console tempest test 02:14:59 liusheng: seems on our tempest config it use ipmitool driver 02:15:27 zhenguo: let me check 02:15:33 liusheng: some ironic guy has help to change that 02:15:51 I am here 02:15:55 zhenguo: oh, yes 02:15:57 litao__; o/ 02:16:39 ok. not much of announcement, anyone else have a thing here? 02:17:34 oh, the contribution stats 02:17:45 Contribution into mogan for the last 30 days 02:17:45 #link http://stackalytics.com/report/contribution/mogan/30 02:18:26 thanks everyone for the hard work 02:19:24 anything else? 02:19:43 plz update the tasks status 02:19:44 haha 02:19:54 wanghao: sure 02:20:02 #topic task tracking 02:20:09 #links https://etherpad.openstack.org/p/MoganWhiteBoard 02:20:42 looks like things are moving along nicely 02:21:05 yes 02:21:36 I think we can land adopting spec if there is no big issue. 02:21:38 the placement patches need more eyes on them, please help to review to make it land by this week 02:21:50 sure 02:21:59 wanghao: sure, let's make it land by this week 02:22:13 great 02:22:18 zhenguo: very nice 02:22:31 zhenguo: 02:22:38 just added an item about uwsgi support 02:22:56 liusheng: thanks, 02:23:05 liusheng: do we need to support uwsgi? 02:23:06 I found the mogan codes has been changed a lot because of placement. 02:23:35 maybe many conficts for new patches 02:23:37 wanghao: yes, seems it is tendency 02:23:42 litao__: yes, so I want more people review the patches to make it not break mogan 02:24:14 litao: so let's land it first and rebase other patches 02:24:15 zhenguo: got it 02:24:44 wanghao: seems many other projects have moved to uwsgi 02:24:58 liusheng: yeah, Zaqar also has it already, but I still not see the benefit 02:25:29 yes, now many projects use uwsgi 02:25:44 zhenguo, wanghao yes, not very sure, but seems it can well support multi process of api server 02:26:27 liusheng: do further research maybe ：) 02:26:42 litao__: sure, I am researching, hah 02:26:50 hah 02:26:53 zhenguo, liusheng, shaohe_feng okay, that's a good improvement. 02:27:05 the patch is ready to review, and I have tested ok locally 02:27:23 awesome! 02:27:24 if we support uwsgi, guru also need some changes to support it. 02:27:33 thanks :) 02:27:33 nice work 02:27:37 I did it in Zaqar 02:27:55 wanghao: maybe you can also do it as a following up patch :D 02:28:11 wanghao: oh, please go ahead. hah 02:28:23 zhenguo: yes, leave it to me 02:28:33 liusheng: sure : 02:28:39 :) 02:28:59 ok, let's talk about the aggregates and server groups 02:29:21 I updated to the node aggregates spec to leverage placement 02:29:51 we will handle aggregate metadata on mogan side, and placement aggregates will be used for grouping the rps 02:30:16 zhangyang: maybe you and liudong should take a look at it 02:30:39 zhenguo: yes, sure. 02:31:31 zhangyang: ok, and the server group one, we still cant find out the real purpose of it, need more feedback 02:32:10 zhenguo: I guess there still is many TODO works of placement 02:32:22 I can understand the use case that, users want an affinity group of nodes which in same RACK, or not same 02:32:32 liusheng:ｙｅｓ 02:32:52 zhenguo: I think Nova also need to consider that 02:33:11 liusheng: you mean aggregates? 02:33:39 zhenguo: yes, if Nova will totally switch to use placement 02:33:54 liusheng: hah, seems that will take years 02:34:00 zhenguo: :( 02:34:24 liusheng: They already clarified that nova host aggregates is different from placement aggregates 02:34:58 So Mogan also need 'host aggreates'? 02:35:15 wanghao: not host, but node 02:35:34 if considering affinity group, seems node aggregates more fit for us. 02:35:36 yes 02:35:57 zhenguo: yes, if so, just think a bit valueless of node aggregates of placement.. 02:36:30 yes, we need to define node aggregates on mogan side 02:36:58 placement is just a place to provide the group and provide the api to select it 02:37:19 wrt which group we needed, should be done by us 02:37:36 like we need to filter the aggregate ourselfs to get a agg list 02:37:54 then pass it down to placement with the query parameter to get the expected RPs 02:38:11 we don't need to store compute nodes in Mogan any more, right 02:38:16 ? 02:38:29 yes, we dont' save nodes and the aggregate_nodes 02:38:36 just same aggregate metadata 02:38:42 zhenguo: oh, I see, yes 02:38:45 zhenguo: We should call ironic api to get compute nodes 02:38:57 zhenguo: low efficiency? 02:39:07 litao__: no, we will provide that API 02:39:13 litao__: from placement 02:39:53 I plan to cache the RPs and map of aggregates from placement when doing resource update, then we can expose that to users 02:40:07 what's the difference aggregates and server group? 02:40:28 server group have policies 02:40:30 yeah, I want to ask too 02:40:36 affinity and anti-affinity 02:40:47 based on the server in the groups 02:40:51 zhenguo: OK 02:41:00 for vms, it's in the same host or not 02:41:08 but for us, it's not clear yet 02:41:43 seems user can choose that two servers could be in one rack or not 02:41:46 for RACK, ROW... seems all make sense 02:42:07 yes, that's why we want to add server group concepts 02:42:13 zhenguo: we may need to introduce that concepts firstly, hah 02:42:17 with scheduler hints a user can do that 02:42:18 so it's a logic concept just in Mogan 02:42:41 yes, but I don't want to add all that notions 02:42:52 maybe we need to improve the server groups 02:42:53 zhenguo: we are planning to use server groups during starting servres, and mark different aggregates with metadata rooms/racks/etc. 02:42:59 not like nova, we can make it more smart 02:43:03 maybe also besed on metadata 02:43:20 and we need to select the aggregates based on the group policy 02:43:27 zhangyang: yes, that's what I expected 02:44:04 yes, seems a bit complex, we need to think more, at least need a spec to show the details 02:44:13 but it's sure not like nova's server groups 02:44:35 zhenguo: yes. 02:44:47 wanghao: you mean combine server group and node aggregate？ 02:45:15 I think we need to combine those 02:45:52 aggregates can be RACK, ROW, DC 02:46:06 zhenguo: maybe a great idea 02:46:12 but it also could be ceph backed, FPGA, GPU 02:46:19 not sure how to handle it gracefully 02:46:49 for location notion, they will fit affinity and anti-affinity, but for hardware specs, not sure how to handle it 02:48:16 litao__: no, I just say the relationship 02:48:49 AZ is also a metadata of aggregates, but it's special 02:49:01 maybe we should make RACK also special metadata, hah 02:49:29 zhenguo: you mean traits in placment or what ? 02:50:14 liusheng: no 02:50:25 liusheng: we will handle the aggregate metadata ourselfves 02:51:13 I think we can not use traits until we add NIC, DISK there 02:51:23 zhenguo: oh, I misunderstood that 02:51:25 with a nested resource providers 02:51:39 zhenguo: I mean metadata for server group, hah 02:51:46 liusheng: hah 02:52:14 metadata can solve everything 02:52:18 liusheng: metadata for server group to match with aggregate metatada 02:52:25 seems can archieve this 02:53:27 zhenguo: so server group and node aggregates notions in Mogan are overlapped 02:53:28 I don't know what nova's server group metadata used for 02:53:44 liusheng: seems not 02:54:01 liusheng: aggregates are for nodes, server groups are servers 02:54:16 liusheng: nodes contains deployed and not deployed 02:54:41 server groups is only used for creating server 02:55:20 zhenguo: yes, I see 02:55:40 hah, seems we can do it 02:55:53 zhenguo: seems nodes aggregate is a urgent requirement, but server group not, right ? 02:56:06 liusheng: seems yes, 02:56:22 liusheng: but after aggregates landed, we can prepare the server groups things 02:56:32 so let's focus on that firstly 02:56:33 liusheng: I will try to finish aggregates soon 02:57:01 oh boy, it's almost time 02:57:05 hah 02:57:24 #topic open discussion 02:57:31 let's continue here 02:57:33 lol 02:58:35 zhenguo: are you already working on code of aggregates ? 02:58:40 zhenguo: or only spec 02:58:49 liusheng: yes, the db migrations and models patch 02:58:59 zhenguo: cool 02:59:02 liusheng: already under review 02:59:07 zhenguo: hah 02:59:52 liusheng: btw, please make sure our client work before Pike :D 03:00:04 zhenguo: sure 03:00:10 thanks 03:00:15 ok, it's time 03:00:22 thanks y'all 03:00:30 #endmeeting