03:05:10 <Yumeng> #startmeeting openstack-cyborg
03:05:10 <openstack> Meeting started Thu Jul  2 03:05:10 2020 UTC and is due to finish in 60 minutes.  The chair is Yumeng. Information about MeetBot at http://wiki.debian.org/MeetBot.
03:05:11 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
03:05:13 <openstack> The meeting name has been set to 'openstack_cyborg'
03:05:59 <Yumeng> songwenping__: that's can be done automatically by gerrit. you just need to add story:story-id, task:task-id to commit message of your patch.
03:07:35 <Yumeng> here is a reference: https://wiki.openstack.org/wiki/Cyborg/CyborgStoryboard
03:08:40 <songwenping__> So what's the meaning of doing this? I donot think we should record it in the storyboard again.
03:10:12 <Yumeng> you can track either in cyborg-spec storyboard or the feature storyboard, but need somewhere to track.
03:10:59 <Yumeng> we had a storyboard cyborg-spec, which I personally agree if we don't need, we can remove.
03:11:53 <brinzhang_> Any topic need to discuss? today
03:12:03 <Yumeng> but for the specification, I think need a place to record together with the feature code.
03:12:55 <brinzhang_> In QAT, SSD, FPGA, and add device attribute API, I had left some comment, I have nothing to say before get the respone.
03:13:04 <Yumeng> I have a question, what do you think of third-party CI about the driver that your are going to submit?
03:13:13 <brinzhang_> In the specs of QAT, SSD, FPGA, and add device attribute API,...
03:13:25 <songwenping__> Thanks Yumeng.
03:13:57 <brinzhang_> Yuemng: We have not the hard device to support the 3rd CI, would you like to provide?
03:14:03 <Yumeng> shaohe and I had a discussion with sean about mdev, he suggested a good way of mdev first-party CI:http://lists.openstack.org/pipermail/openstack-discuss/2020-June/015696.html
03:15:53 <brinzhang_> I had seen this ML, I have not looked into, will see later
03:16:02 <Yumeng> I mean, the Inspur FPGA driver and Inspur NVME SSD which you are plannig to submit. If you don't have ,I'm more impossible to have. :(
03:16:16 <brinzhang_> just saw you response in the ML
03:16:34 <Yumeng> http://lists.openstack.org/pipermail/openstack-discuss/2020-June/subject.html
03:16:40 <Yumeng> here is more.
03:17:17 <brinzhang_> yeah, we dont have server to support the CI evn in project planning, hope you can understand.
03:17:35 <Yumeng> maybe that will inspirs you any new ideas.
03:18:02 <brinzhang_> Yumeng: but we can show the test results on the implements patch, should not limit the spec go
03:20:03 <Yumeng> I understand. We need thinking the third-party CI or any other workarounds, since that's a topic asked by other projects several times.
03:20:15 <brinzhang_> Yumeng: I can understand this, but I feel helpless to provide 3rd-party CI.
03:20:49 <brinzhang_> Yes, this requires hardware support. I think this is not just a problem we encountered at Cyborg.
03:21:34 <Yumeng> don't have to be a very serious 3rd-party CI. I think the main concern from others should be if that driver really works? or things like that
03:22:27 <brinzhang_> If the contributor can provide a complete test report, I think it is acceptable, what do you think?
03:23:03 <brinzhang_> maybe we shuold mark the driver lack of it's CI to run
03:23:26 <Yumeng> brinzhang_: makes sense to me!
03:24:18 <brinzhang_> Yuemng: thanks, this is a better way I can think of.
03:24:19 <Yumeng> we publish these kind of report to potential customers (either in documentation or other ways)
03:25:53 <shaohe_feng> Hi all
03:26:07 <brinzhang_> we publish these kind of report to potential customers (either in documentation or other ways)
03:26:11 <songwenping__> His shaohe.
03:26:11 <Yumeng> hi shaohe_feng
03:26:35 <shaohe_feng> conflict with another meeting, sorry for late
03:26:41 <brinzhang_> we can left comment under the under the code to implement SPEC.
03:27:28 <brinzhang_> Two ways: one is provide the 3rd-party CI; another is proivder the complete test report for the target driver.
03:28:22 <songwenping__> brinzhang_: What is the complete test report like?
03:29:07 <Yumeng> shaohe_feng: no worries. we were discussing if we need third-party CI for new drivers in this release.
03:29:47 <shaohe_feng> got it, thanks.
03:30:43 <brinzhang_> The report should show the driver can run successfully, may contain each api requst/response test result
03:31:54 <Yumeng> agree. That makes sense to me now. I cannot think of any better idea for now.
03:32:02 <songwenping__> Can u provide a template?
03:32:05 <brinzhang_> or others result you can provide, should let the reviewer can know it's a better driver
03:32:56 <brinzhang_> Each driver is implemented differently and cannot provide templates. Provide a test list yourself if possible.
03:33:13 <Yumeng> I think nova operations like creation, deletion are very important
03:34:36 <brinzhang_> nova-cyborg interaction is not a mandatory requirement, and may not be called through nova.
03:35:06 <brinzhang_> But if used, it is best to provide.
03:36:14 <Yumeng> if we have  third-party CI, that's a mandatory I suppose
03:37:08 <Yumeng> current tempest only test fake driver. third-party CI test real driver.
03:37:31 <brinzhang_> yeah, can as soon as possiable to provider the nova operaction
03:37:54 <Yumeng> so from my understanding, test report will be a make up for third-party CI.
03:38:44 <Yumeng> I can bring up a ML thread to discuss this.
03:39:22 <brinzhang_> Yuemng: You can
03:39:33 <Yumeng> Is there any other things want to bring up for today?
03:40:17 <Yumeng> I was testing rbac policy API, will soon submit the patches.
03:40:39 <brinzhang_> Yuemng: cool, ths
03:40:47 <brinzhang_> Yumeng, songwenping_, shaohe_feng: I have to go now, bye~
03:41:02 <Yumeng> if nothing else, let's wrap up today's meeting
03:41:04 <shaohe_feng> brinzhang_ bye, have a good day
03:41:12 <songwenping__> bye
03:41:15 <Yumeng> bye
03:41:21 <shaohe_feng> Yumeng, I'd like to talk about program
03:41:29 <Yumeng> ok. sure.
03:41:33 <Yumeng> please go ahead.
03:41:41 <shaohe_feng> Now it will take about one hour for us to program FPGA.
03:42:29 <shaohe_feng> and I find the program API will report timeout
03:42:29 <Yumeng> sounds like a long time.
03:42:42 <shaohe_feng> yes, a long time
03:43:15 <s_shogo> shaohe_feng : I would like to know the FPGA spec , e.g. Arria10 GX
03:43:26 <shaohe_feng> the client wait for the API response and receive the timeout
03:43:37 <shaohe_feng> s_shogo: yes it is A10
03:43:48 <shaohe_feng> s_shogo: but we do not program the PR
03:44:09 <shaohe_feng> we program the whole flash.
03:44:35 <s_shogo> ok, I got it. It seems to be similar to N3000 case.
03:44:40 <shaohe_feng> for PR takes so many logic units
03:44:53 <shaohe_feng> s_shogo: yes, it is N3000
03:45:36 <shaohe_feng> It really take too long time:')
03:45:51 <shaohe_feng> can we support async program?
03:46:07 <s_shogo> I know, that takes 25-40min :)
03:46:52 <s_shogo> IMO, supporting for async program needs modification for driver, agent, and so on . That is big task.
03:46:57 <Yumeng> shaohe_feng: havewe already supported async program ?https://review.opendev.org/#/c/681005/
03:47:00 <shaohe_feng> the program API start to program and return immediately
03:47:12 <shaohe_feng> s_shogo: yes, I know
03:47:36 <shaohe_feng> we need a new API to show the program status
03:47:45 <shaohe_feng> then user know the program process
03:47:59 <shaohe_feng> and notify mechanical
03:48:08 <shaohe_feng> for driver, agent.
03:48:24 <shaohe_feng> really big task
03:48:45 <shaohe_feng> like async bind, it is also a big task
03:49:12 <Yumeng> aha. sorry. that's async bing.
03:49:22 <Yumeng> bind
03:49:55 <shaohe_feng> s_shogo, Yumeng, any ideas on it?
03:52:11 <s_shogo> shaohe_feng I approve your idea. IMO, the async architecture should be same to async bind, for uniformity.
03:52:47 <shaohe_feng> yes, agree
03:53:18 <Yumeng> I'm not quite familar with programing. sounds like waiting such a long time is horrible.
03:53:39 <Yumeng> from this point, I would support async program
03:54:58 <Yumeng> how will that affect agent?
03:55:03 <Yumeng> agent workflow?
03:56:07 <shaohe_feng> that may be a complex case
03:56:26 <Yumeng> haha. no worries. too big question.
03:56:51 <shaohe_feng> such as during this program time
03:57:48 <shaohe_feng> a new program API is issued, the cyborg-API should know there's already one is in progress, and should reject the request
03:58:45 <Yumeng> ok.
03:59:35 <shaohe_feng> and as s_shogo in his patch say: the driver should report the program info to agent, and at last, to tell the end user
04:00:02 <shaohe_feng> such as it is successful, or failed with some reason
04:01:04 <Yumeng> alright. thanks shaohe.
04:01:27 <shaohe_feng> Yumeng, and maybe a new field in DB(or other ways) to show the program status to user.
04:01:51 <shaohe_feng> Yumeng: we should think about it.
04:02:19 <Yumeng> sounds like a huge task.
04:02:52 <shaohe_feng> yes, async means complex.
04:03:06 <Yumeng> emmm do you think  we start next release?
04:03:31 <shaohe_feng> we are waiting for this program API
04:03:45 <shaohe_feng> hopeful it can work as soon as possible.
04:04:02 <shaohe_feng> -_-
04:04:10 <Yumeng> aha, is it part of this ?https://review.opendev.org/#/c/698190/
04:04:47 <shaohe_feng> yes
04:04:58 <s_shogo> IMO, that is too huge task for the first program patch. That should be better to implement in another one.
04:05:24 <Yumeng> yes, agree. I will find some time to review this.
04:05:43 <shaohe_feng> OK, we can split it into some small patch sets
04:06:01 <Yumeng> yes.
04:06:07 <s_shogo> agree, thanks
04:06:19 <Yumeng> thanks shaohe and shogo. we've run out of time.
04:06:34 <Yumeng> let's continue discussion in wechat.
04:06:35 <shaohe_feng> sorry, that's all
04:06:53 <Yumeng> I will wrap up today's meeting. see you next time!
04:07:04 <Yumeng> bye guys. Thank you.
04:07:19 <shaohe_feng> bye, thanks everyone
04:07:22 <Yumeng> #endmeeting