03:15:39 #startmeeting openstack-cyborg 03:15:40 Meeting started Thu Jan 16 03:15:39 2020 UTC and is due to finish in 60 minutes. The chair is Sundar. Information about MeetBot at http://wiki.debian.org/MeetBot. 03:15:41 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 03:15:44 The meeting name has been set to 'openstack_cyborg' 03:15:50 #topic Who's here 03:15:54 #info Sundar 03:16:02 #info s_shogo 03:16:23 #info chenke 03:16:26 #info brnzhang 03:17:56 For the agenda today, apart from the patch reviews, I was thinking about the security policy refresh, microversions, and the spec for instance ops on accelerators. 03:18:08 Do any of you have something else to add? 03:18:28 I want to talk about http://lists.openstack.org/pipermail/openstack-discuss/2020-January/011990.html 03:18:46 Do we need to use the launchpad 03:19:15 Oh yes. brinzhang, I was planning to reply to that 03:19:41 brinzhang, we had some discussions last year, and folks decided to go with storyboard. 03:19:57 it's ok 03:20:00 We already have a lot at https://storyboard.openstack.org/#!/project_group/77 03:20:32 Either way, as long as it can be managed uniformly, iam ok 03:20:34 Is there any string advantage to launchpad over storyboard that you want to bring u? 03:20:50 *strong 03:21:03 Ok 03:21:26 I think the lauchpad was more intuitive 03:21:43 Is it signficantly faster? I find storyboard to be very slow 03:22:02 ore intuitive 03:22:08 As far as I know, more people use launchpad than storyboard 03:22:28 yeah, launchpad is faster than storyboard 03:23:02 sometime the storyboard always stuck 03:23:20 so I recommand to use launchpad 03:23:37 We don't seem to have a quorum today to decide that, or even discuss it at length 03:24:00 I suggest you can bring it up in a future IRC meeting. 03:24:25 yeah, I think this need more people to join, we talk about next topic, i think 03:25:00 Sure, I'll add this to the meeting agenda at https://wiki.openstack.org/wiki/Meetings/CyborgTeamMeeting. 03:25:21 Ye. 03:25:25 We don't have Yumeng today to discuss the policy refresh 03:25:49 I think storyborad is clearer。 03:26:00 There is a lot of discussions going to microversions. I'd just like to make sure we have a solid footing for the future 03:26:45 chenke, got it. I suspect folks will have different opinions. 03:26:49 Reference the nova microversion, use microversion is more useful 03:26:55 yumeng is out of office. 03:27:27 brinzhang: yes, microversions are necessary to add features and evolve the APis sommothly, which is important for a young project like Cyborg 03:27:28 When a new feature changed in Cyborg, we can record this with a microversion 03:27:56 Agree 03:28:26 So about this spec, what are you worry about? Sundar 03:29:34 brinzhang: I think we all agree on the need for microversions. The questions would be, are we following the best practices etc. I invited Chris Dent etc. to review it. If you see any gaps, please bring it up. It is better to take the time and do this properly to avoid gaps. For example, adopting conventions like docmetning APi changes in release 03:29:34 notes, as you pointed out in the Nova patch review 03:29:52 *documenting 03:30:19 I'm not sure if the existing Cyborg feature will be implemented after this SPEC merger, or in parallel with it? 03:31:21 The way I see it, we released v2.0 of APis in Train. Whatever comes out of Ussuri will/should be microversion-enabled and will probably be v2.1. After that, we have incremental changes. 03:31:24 yes, we should consider more, such as the unit tests, documenting, functional test etc. 03:32:36 Until we get Nova integration done, any version bump in Cyborg is not of great significance as people are not yet putting it in production. 03:32:59 Sundar, so while someone pushed a spec for Cyborg, we should require he/she to add a new microversion for that api (if the api changes) 03:33:32 Sure 03:34:14 Sundar, Even so, we also need to maintain microversion step by step 03:34:56 The device v2 APis would be 2.0, because they are totally new. But existing APIs like device profiles and ARQs would get a bump if they are changed. 03:35:38 This is my opion for this spec, I will review it again while I free, and refer the experience I was in Nova 03:35:54 Great, sounds good 03:36:03 #topic Spec for instance ops for accelerators 03:36:14 agree "The device v2 APis would be 2.0" 03:36:29 https://review.opendev.org/#/c/605237/ 03:37:14 I was reviewed this spec, I agree with matt, this spec mostly seem description the documents?is right? 03:37:25 I wrote this last year as a way to get agreement between Nova and Cyborg sides. Some people did not see it as a spec per se, because it is not specifying any changes. It is only documented behavior. 03:37:27 some operators are in nova 03:38:03 Yes, all these are instance ops in Nova, and the doc addresses what should happen if the instance has accelerators 03:38:06 Sundar, yeah, it's document 03:38:27 It may make sense to retire it as a spec and make it a document 03:38:39 yeah, it make sense 03:39:10 Ok, I'll take an action item for that 03:39:14 I think we should ensure the accuracy of the description 03:39:58 Is something not accurate? 03:40:10 So we need more nova core to review this, such as erifed, stephenfin, alex_xu, sean-k-moony to review it, make it better and better 03:40:39 See my comment in PS4 03:41:02 They have all reviewed it to some extent or the other in Train. If it is recast as a Cyborg doc, it will be tougher to get their review, but we can try. 03:41:07 about the rebuild action, I think you should re-word 03:41:54 ok, will go by the review 03:42:02 Maybe we can send this to the nova meeting 03:42:34 Next, i will review again, and check all instance ops 03:42:45 I'll ask the Nova PTL if that is feasible, after I recast it as a doc 03:43:06 Sundar Above is my opinion on this 03:43:59 Sure, anything else on this subject? 03:44:03 another way, it may be possible to add this to Nova's document library as an auxiliary document using Cyborg as an instance 03:44:39 'Using Cyborg as an instance' -- what does that mean? 03:45:18 put this doc in the Nova's library, while some want to use Cyborg, they can read this 03:45:48 of course, we also need put this in Cyborg's document library 03:46:05 just an idea 03:46:19 Nova's handling of instance ops varies with Neutron ports, Cinder volumes, etc. I doubt that Nova will agree to put a project-specific doc in their list. In fact, it may make sense to keep it in Cyborg, because different Cyborg drivers may potentially support different ops . 03:47:15 For example, hypothetically, if there is a problem with doing instance rebuilds with GPU drivers, that caveat should be in Cyborg docs, right? 03:48:18 Yeah, but I think nova also need this 03:48:55 Ok, no harm in asking them if they can do it. 03:49:04 maybe, put it in Cyborg is appropriate 03:49:13 more appropriate 03:49:20 I think cyborg is the main one and Nova can briefly introduce it. 03:49:33 Sure 03:51:02 #topic Other topics 03:51:41 The next PTG will be in Vancouver, June 8-11 03:51:51 About this patch. https://review.opendev.org/#/c/702578/ 03:52:03 Early Bird Full Access costs $299 and lasts till Mar 2, 2020. 03:52:31 chenke: I started looking at your patch. It is highly needed. Thanks 03:52:49 Ok. Thanks. 03:52:49 I'm considering a gap between cyborg API spec and sdk's premise(mainly arq delete API,with multi choice of argument), 03:52:50 Sundar I need to follow the company's decision whether to participate in this PTG 03:53:16 brinzhang: Sure, just mentioning it for awareness 03:53:29 Just need have a topic, also a candidate speeker 03:54:00 Would you like to collaborate with one of us for a speech or a project update? 03:54:28 Anyways, we can discuss it offline 03:54:42 s_shogo: Yes 03:55:08 I think we had a brief discussion. The HTTP DELETE's query parameters can be used to unify both forms, right? 03:55:17 about chenke's patch, i have a doubt, consistency of interface parameters, separated from rpcapi 03:55:31 sundar, while you review this, please take a care 03:55:52 brinzhang: ok 03:55:53 mianly about the 'context' parameter 03:56:31 context is useless for create db device profile 03:57:00 I removed it . https://review.opendev.org/#/c/702578/2/cyborg/db/sqlalchemy/api.py 03:57:22 What are your thoughts 03:57:36 chenke: I think brinzhang is saying that, when we remove context as a parameter for db api, it may be good to remove it from other places which use it only for dbapi, such as the main API 03:57:43 chenke, we can wait for Sundar reviewed this, then talk about 03:57:44 Sundar: That depends on sdk's delete function, I'll clarify the gap and ask this again. 03:58:13 s_shogo: ok 03:58:27 brinzhang ok. 03:58:58 brinzhang: Welcome again to Cyborg. Could you give us some background about yourself? Do you use accelerators like FPGAs in your company? 03:59:15 I have to go for lunch. bye~ 03:59:30 We are use GPU in Cyborg, FPGA in plan~ 04:00:19 Now we are applying the fpga device :) 04:00:39 chenke, bye 04:00:51 chenke: Bye 04:01:00 brinzhang: Cool. What important things are missing from Cyborg, apart from nova integ? 04:01:52 And could I ask if they are Intel FPGAs? ;) 04:01:55 Sundar, I was gurduate from Zhengzhou Unviersity, now I am working in Inspur, mainly work in OpenStack (Community) 04:02:48 Yes, in theory, Intel's fpga 04:03:26 brinzhang: Good to know that. Happy to have you in OpenStack and Cyborg. 04:03:38 Sundar, thanks 04:03:51 Welcome brinzhang (again) 04:04:08 Please feel free to reach out to me in WeChat or IRC if you need anything 04:04:17 s_shogo thanks~ 04:04:21 Anything else, folks? 04:04:34 Nothing for me today 04:04:53 I think the Chinese New year is coming up soon? 04:05:14 Should we hold a meeting next week? 04:05:51 Yeah, next Thursday, next meeting I will not join, on the way (go home) 04:06:25 Ah ok. Advance wishes -- Shenyun Kwai le :) 04:06:37 Have a good day, everybody. 04:06:42 There are question from nova member in this IRC, last FRIday - we had better to follow that http://eavesdrop.openstack.org/irclogs/%23openstack-cyborg/%23openstack-cyborg.2020-01-10.log.html 04:06:43 Thanks 04:07:24 s_shogo: Sean and I talked on Nova IRC this morning 04:07:33 Sundar OK,thanks. 04:07:34 But thanks for pointing it out to me 04:07:41 #endmeeting