Earlier  
Posted Nick Remark
#openstack-cyborg - 2019-05-29
03:03:36 Sundar__ #startmeeting openstack-cyborg
03:03:37 openstack Meeting started Wed May 29 03:03:36 2019 UTC and is due to finish in 60 minutes. The chair is Sundar__. Information about MeetBot at http://wiki.debian.org/MeetBot.
03:03:38 openstack Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
03:03:40 openstack The meeting name has been set to 'openstack_cyborg'
03:03:46 Sundar__ #topic Roll call
03:04:17 Sundar__ #info Sundar
03:04:20 wangzhh #info wangzhh
03:04:36 ikuo_ #info ikuo_
03:04:49 Sundar__ #topic Agenda
03:04:52 Sundar__ https://wiki.openstack.org/wiki/Meetings/CyborgTeamMeeting
03:04:53 Coco_gao #info Coco_gao
03:05:05 Sundar__ Anything to add to this agenda?
03:05:36 Sundar__ #topic Nova spec update
03:05:41 ikuo_ It's good
03:06:29 Sundar__ There was a feedback to move the Cyborg binding to Nova compute manager. I implemented it and made some measurements on 1 2-node system, but that doesn't predict scaling impact
03:07:11 Sundar__ Meanwhile, several folks gave that feedback. SO, I have changed the spec to match that
03:07:22 Sundar__ Does anybody have any concerns or comments?
03:08:17 Sundar__ The spec is close to finish, so please provide your comments soon
03:08:36 ikuo_ Where do you mean in the spec?
03:09:36 Sundar__ https://review.opendev.org/#/c/603955/13/specs/train/approved/nova-cyborg-interaction.rst,unified Line 240
03:10:01 wangzhh Fine with me.
03:10:03 Sundar__ Lines 240-263
03:10:29 Sundar__ Good
03:10:44 Sundar__ #topic python cyborg client
03:10:54 Sundar__ Thanks to ikuo_ for starting http://lists.openstack.org/pipermail/openstack-discuss/2019-May/006543.html
03:10:55 yikun #info yikun
03:11:02 Sundar__ Hope you all had a chance to catch up
03:11:27 Coco_gao still need to some time to catch up, sorry.
03:11:50 Sundar__ Coico_gao, NP. I'll try to summarize.
03:12:26 Sundar__ python-cyborgclient should be rewritten as an OSC plugin or a new repo should be created for that: http://lists.openstack.org/pipermail/openstack-discuss/2019-May/006574.html
03:13:07 Sundar__ Dean tTroyer (http://lists.openstack.org/pipermail/openstack-discuss/2019-May/006577.html) suggests we could write a plugin to OSC, as part of OSC repo itself.
03:13:39 Li_Liu #info Li_Liu
03:13:56 Coco_gao Thanks for the summay~
03:14:00 ikuo_ So we should rewrite the whole cyborgclient.. In fact, I cannot imagine the impact.
03:14:05 Sundar__ Either way, it is good for Cyborg to implement OpenStack SDK, so that the client can use it to call the API
03:14:45 Sundar__ First, do we all agree that we should use the OpenStack SDK (independent of other options)?
03:15:26 xinranwang #info xinranwang
03:15:35 xinranwang Hi all, sorry for late
03:15:56 ikuo_ Is it other thing but OSC plugin?
03:16:24 Sundar__ ikuo_ : It seems like a large impact. In practice, current client is mostly v1. So its objects either don't exist (like accelerators) or have changed a lot (like deployables). Plus, there are new pones like device profiles
03:16:43 Sundar__ So, I suspect anyway it will be a big change
03:16:43 wangzhh Sundar, if we rewrite cyborgclient as an OSC plugin, who guys are responsible for the milestone?
03:17:00 Coco_gao Agree to use openstack sdk first.
03:17:21 Li_Liu is this a train target?
03:17:35 Sundar__ wangzhh: Good question, I don't know :). It means we are not in control, as Dean says. Or, we could write our own plugin from scratch -- more work but more control.
03:17:42 Sundar__ Li_Liu: Yes
03:18:38 Sundar__ ikuo_: "Is it other thing but OSC plugin?" <-- OpenStack SDK and OSC plugin are 2 different things
03:18:50 Sundar__ SDK is a way for other services (or cyborg client) to call CYborg API
03:19:01 ikuo_ OK, I agree Use openstack sdk
03:19:10 Sundar__ Whereas OSC plugin is a way to implement the client using standard mechanisms
03:19:54 Sundar__ Coco_gao, ikuo_: Thanks for voting on SDK. Anybody disagrees?
03:20:31 wangzhh Agree SDK +1
03:20:37 Li_Liu sounds good
03:20:52 Sundar__ Great :) I like consensus :)
03:21:03 xinranwang Agree
03:22:06 ikuo_ But rewriting for OSC plugin by only NTT, it is difficult, I think.
03:22:50 Sundar__ ikuo_: I understand
03:23:11 Sundar__ Now, for the difficult question -- should we write a separate osc plugin (like many other projects have done) or try to make this a part of osc repo? Dean says: "if Cyborg were completely backed by the SDK we would consider adding its commands directly to the OSC repo."
03:23:37 Sundar__ Once we adopt the SDK, may be it gets easier? Plus, I agree others should support NTT
03:24:16 wangzhh ikuo, can u help to split the task on storyboard?
03:24:45 Sundar__ I suspect there will be less UT and functional tests if we use the OSC repo. If there is concern about dependence and timelines, I could ask Dean.
03:24:56 Sundar__ (Or any of you could respond on the ML.)
03:26:43 ikuo_ Wangzhh: so we should split our cyborgclient task to 1) use openstacksdk and 2) OSC plugin /repo?
03:27:07 Coco_gao Maybe more smaller tasks.
03:28:19 wangzhh Coco, Yep. So that others can take part of them to support NTT.
03:28:35 ikuo_ I got it.
03:29:10 Coco_gao I will first finish my part and then join you to finish other tasks.
03:29:22 wangzhh Thx. ikuo_.
03:29:39 ikuo_ All right. thanks.
03:30:03 Sundar__ Sounds like a plan. Should I ask on the ML what is the risk to timeline from using osc repo?
03:30:21 ikuo_ Yes, please!
03:30:31 Sundar__ Will do
03:30:49 Coco_gao Yes
03:31:01 Sundar__ #agreed Cyborg will implement OpenStack SDK. The client will call the API through that
03:31:26 Sundar__ #action Sundar: gets pros and cons of using the osc repo, vs. writing a plugin
03:31:49 Coco_gao BTW, does engineering meeting still exist?
03:31:51 Sundar__ #action: ikuo_ will create storyboard tasks
03:32:41 Sundar__ Coco_gao: We can convene it any time if there is a need. IMHO, it is good to keep it as a placeholder. The networked FPGA stuff is still open
03:33:00 Sundar__ What do you all think?
03:33:12 Coco_gao Sounds good for me.
03:33:17 ikuo_ Yes, we should talk when we need.
03:33:49 Sundar__ OK, thanks
03:33:57 Sundar__ #topic Communication updates
03:34:26 Li_Liu I am still working on the nova's external_event_api patch
03:34:41 Sundar__ As you may know, there are some restrictions on communication, and some of us have been asked not to use WeChat by default
03:35:17 Coco_gao What do you suggest?
03:35:34 Li_Liu lol
03:35:37 ikuo_ Yes, my colleague cannot use WeChat. Thank you for proposing the topic.
03:35:41 Sundar__ It would be good to use IRC and ML by default. For IRC, does it make sense to have a window of time where we all login? For example, it could be morning 9-12 am by CHina time?
03:35:46 Li_Liu I think irc is our only option now
03:36:07 Sundar__ Li_Liu and all: yes
03:36:19 Li_Liu can we still talk in zoom meetings?
03:36:41 xinranwang Sundar__: Hi Sundar, I have pull your pilot branch in local, and find some bugs, for example, we cannot delete extended_arq when deleting vm, and some other. I have done some bug-fix. Do you want me to leave comments where bug occurs , or I can co-author with you if you want. Which way you prefer?
03:36:59 Sundar__ Li_Liu: My guidelines are that public meetings are ok. I have put the Zoom URL in wiki along with the times. SO it is open to all
03:37:14 ikuo_ That's sounds good staying 9-12 am.
03:37:44 Sundar__ ikuo_: Thanks. Others ok with that time?
03:38:09 Li_Liu yup, I will be on irc all the time now
03:38:20 Sundar__ Cool.
03:39:46 Sundar__ xinranwang_: The Cyborg code to delete ARQs exists. Bu the Nova code to call the DELETE for ARQs is not done yet. If you want to take that up, NP. But please be aware that this will be on the critical path.
03:40:27 Sundar__ If you want to leave comments, that is also fine

Earlier   Later