14:01:51 <zhipeng> #startmeeting openstack-cyborg 14:01:52 <openstack> Meeting started Wed Jul 11 14:01:51 2018 UTC and is due to finish in 60 minutes. The chair is zhipeng. Information about MeetBot at http://wiki.debian.org/MeetBot. 14:01:53 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 14:01:57 <openstack> The meeting name has been set to 'openstack_cyborg' 14:02:03 <zhipeng> #topic Roll Call 14:02:09 <zhipeng> #info Howard 14:02:14 <Li_Liu> #info Li_Liu 14:03:33 <Helloway> #info Helloway 14:03:45 <shaohe_feng> #info shaohe_feng 14:03:46 <zhipeng> everyone plz use the #info cmd to sign up 14:04:03 <wangzhh> #info wangzhh 14:04:35 <Coco_Gao> #info 14:04:59 <zhipeng> #info gaojinghua 14:05:54 <sum12> #sum12 14:06:21 <zhipeng> shaohe_feng will sundar attend ? 14:06:50 <shaohe_feng> zhipeng, I do not think so 14:07:00 <zhipeng> on vacation ? 14:07:00 <shaohe_feng> he is on vocation. 14:07:02 <shaohe_feng> yes 14:07:12 <zhipeng> until when ? 14:07:25 <shaohe_feng> seems \7/25 14:08:11 <shaohe_feng> from July 10 night through July 25, coming back on 26th. 14:08:22 <zhipeng> lol 14:08:40 <Coco_Gao> long vacation. 14:08:48 <zhipeng> okey let's get into business 14:09:00 <zhipeng> #topic Rocky Development Update 14:09:05 <zhipeng> #link https://review.openstack.org/#/q/status:open+project:openstack/cyborg 14:09:12 <zhipeng> #topic os-acc design 14:09:33 <zhipeng> https://review.openstack.org/577438 14:09:56 <zhipeng> Li_Liu shaohe_feng is there any consensus converging on the current proposal ? 14:11:26 <Li_Liu> I don't have any further comments on that one 14:11:38 <zhipeng> meaning satisfied ? 14:11:45 <zhipeng> or still waiting for answers 14:11:47 <shaohe_feng> I really want to know the scope of the os-acc. 14:12:34 <shaohe_feng> will it cover the program and instance binding? 14:14:20 <shaohe_feng> we have not define the binding interface. The binding should pass hostname and instance uuid and request spec. 14:14:32 <zhipeng> Li_Liu wangzhh Coco-Gao what's your opinion 14:14:55 <wangzhh> I think so. os-acc call cyborg to program and bind. 14:15:36 <Li_Liu> it should be able to call the program API 14:16:04 <shaohe_feng> rest full API 14:16:15 <shaohe_feng> ? 14:16:22 <Li_Liu> but how it calls it is the problem here 14:16:23 <wangzhh> Not just restful api. 14:16:32 <Li_Liu> exactly 14:16:43 <Li_Liu> rest/rpc/something else? 14:17:14 <Li_Liu> for the moment, the most intuitive way is calling the REST API I am currently writing 14:17:35 <shaohe_feng> can program also call RPC? 14:17:46 <zhipeng> wangzhh how's the os-brick handling the attach ? 14:17:58 <Coco_Gao> wangzhh sundar suggest to use rpc call? 14:18:14 <wangzhh> I've talked to Sundar about it. We agreed that os-acc should call not just api. 14:19:06 <wangzhh> In fact, call api only is another option. 14:19:38 <zhipeng> besides api what could it call ? 14:20:29 <Li_Liu> Maybe allowing os-acc directly call the vendor driver api is another option 14:20:53 <zhipeng> that would be the original design right ? 14:21:02 <zhipeng> contain it on the compute node level 14:21:08 <zhipeng> without any api interactions 14:21:19 <wangzhh> In some cases, we must call api. But sometimes, rpc call is a good code path, too. 14:21:33 <Coco_Gao> os-acc call cyborg-agent using rpc? 14:21:56 <wangzhh> CoCo: Yes 14:22:17 <wangzhh> The original design right. 14:23:01 <Coco_Gao> ok, when should we use rpc and when should we use api? 14:23:27 <wangzhh> Sorry, lost an "is". 14:24:41 <wangzhh> Attach, deatch use rpc. 14:26:44 <Li_Liu> zhipeng: "contain it on the compute node level". that should be the easiest way for now 14:27:00 <zhipeng> will attach/detach involving binding shaohe_feng mentioned ? 14:29:38 <xinran__> Binding is before attach I think？ 14:30:08 <zhipeng> bind the vm instance with the acc instance ? 14:30:18 <zhipeng> do we have similar process in other module ? 14:31:09 <Li_Liu> the binding should be taken care of before the process get to the vendor driver 14:31:13 <wangzhh> Neutron 14:31:40 <zhipeng> i think there was a comment from Sundar states that it is kinda different for neutron 14:31:46 <zhipeng> the port binding 14:31:48 <wangzhh> Nova call neutron-agent directly to allocate port. 14:31:50 <zhipeng> and the acc binding 14:33:15 <zhipeng> okey let's arrange another ZOOM call for specificities 14:33:28 <zhipeng> the big problem here is about the os-acc scope, like shaohe said 14:33:37 <zhipeng> what we want it to be for Rocky release 14:33:46 <wangzhh> In my opinion, we can imply it as the design now. For the api and os-acc are stable. We can refactor it if needed. 14:33:57 <zhipeng> bare minimum , or we want binding, programming in it as well 14:34:36 <zhipeng> we need to start implementation as soon as possible for sure 14:34:50 <Li_Liu> I agree with wangzhh, we should get the thing working first 14:35:03 <Li_Liu> ZOOM call would be great 14:35:24 <zhipeng> Li_Liu will it block FPGA programing if os-acc does not include program support ? 14:36:07 <Li_Liu> no it should not as my programming support is a stand alone feature 14:36:15 <Li_Liu> it does not depends on others 14:36:57 <zhipeng> okey then let's shoot for the minimum feature goal 14:37:26 <zhipeng> shall we have a ZOOM meeting say 10:00 am tmr ? 14:37:35 <zhipeng> discuss os-acc in detail ? 14:37:53 <Li_Liu> you mean china time? 14:38:07 <zhipeng> yes, 10:00pm for Toronto 14:38:14 <Li_Liu> should be fine 14:38:17 <zhipeng> about less than 12 hours later 14:38:22 <Li_Liu> up to you guys 14:38:33 <zhipeng> wangzhh shaohe_feng Coco-Gao okey ? 14:38:53 <wangzhh> I'm OK. 14:39:03 <Coco_Gao> ok 14:39:34 <zhipeng> ok then I assume shaohe is ok as well :) 14:39:37 <zhipeng> moving on 14:39:43 <zhipeng> #topic GPU driver 14:39:44 <sum12> could you guys post some notes about your discussion ? 14:39:53 <zhipeng> sum12 definitely 14:40:20 <zhipeng> wangzhh any timeline for the GPU driver ? 14:40:36 <zhipeng> #link https://review.openstack.org/574075 14:41:04 <wangzhh> I'll update the commit tmr. 14:41:23 <zhipeng> can you also reply to the NVIDIA guy ? 14:41:36 <zhipeng> i remember he asked about an introduction about the driver code 14:41:40 <zhipeng> in the email thread 14:41:44 <wangzhh> Of course. I have promised Mike. 14:42:13 <zhipeng> gr8t :) 14:42:16 <zhipeng> moving on 14:42:25 <zhipeng> #topic FPGA API 14:42:28 <Coco_Gao> I think Mike asked more than the driver code, but the cyborg related code as well. 14:42:39 <zhipeng> #link https://review.openstack.org/579315 14:42:48 <zhipeng> Li_Liu is this still premature ? 14:43:26 <wangzhh> Coco: Yes. I'll try my best to describe it. 14:44:07 <Li_Liu> zhipeng, it's not ready yet 14:44:26 <zhipeng> okey got it 14:44:40 <zhipeng> moving on to next 14:44:46 <zhipeng> #topic Quota support 14:44:50 <zhipeng> #link https://review.openstack.org/564968 14:44:59 <zhipeng> xinran___ 14:45:55 <zhipeng> why is there a 16KiB file change in the patch ? 14:46:03 <zhipeng> the swp file 14:46:37 <wangzhh> haha. :) 14:47:01 <zhipeng> lol so weird 14:47:49 <zhipeng> xinran__ are you around ? 14:47:53 <shaohe_feng> http://docs.ocselected.org/openstack-manuals/kilo/networking-guide/content/logical-architecture.html 14:48:38 <shaohe_feng> #link https://docs.openstack.org/install-guide/get-started-logical-architecture.html 14:49:08 <wangzhh> https://review.openstack.org/#/c/580022/2/.gitignore This commit could resolve the swp file. 14:50:44 <xinran__> I’m here 14:50:59 <xinran__> Sorry for the swp file... 14:51:08 <xinran__> Haha 14:51:11 <zhipeng> no problem 14:51:30 <zhipeng> other than that, plz address Li Liu's commenty 14:51:37 <zhipeng> and let's move forward 14:51:43 <xinran__> OK 14:52:28 <zhipeng> okey i think we reviewed most of the important current available patches 14:52:59 <Li_Liu> zhipeng, what's the currently time line? 14:53:18 <zhipeng> shooting for code completion before RC freeze 14:53:53 <Li_Liu> ok 14:54:13 <Li_Liu> but the os-acc and nova interaction is still faaar behind.. 14:55:28 <zhipeng> https://releases.openstack.org/rocky/schedule.html 14:55:39 <zhipeng> os-acc will miss the R release i'm fraid 14:55:44 <zhipeng> as non-client libraries 14:56:07 <zhipeng> but as long as we have code implementation, we could have docs guiding the user how to use it 14:56:19 <Li_Liu> ok 14:56:35 <zhipeng> the week of Aug 20 - Aug 24 is the RC deadline 14:56:48 <Li_Liu> I am just hoping we can make cyborg be in a usable state in R 14:56:54 <zhipeng> which gives us barely more than a month of time 14:57:01 <zhipeng> Li_Liu of course 14:57:34 <zhipeng> #topic Berlin Summit CFP 14:57:47 <Coco_Gao> since we will merge accelerator&deployable db table, should we modify the code after xinran's code being merged? 14:57:50 <zhipeng> Plz be aware that the Berlin Summit CFP ends on Jul 17th 14:58:08 <wangzhh> Zhipeng, Li_liu, if we can't merge os-acc in nova in time. Can we offer the code as patch to user? 14:58:20 <zhipeng> wangzhh that was what I meant 14:58:31 <zhipeng> because the repo is aleady there, thx xinran__ 14:58:40 <zhipeng> as long as we have workable code 14:58:49 <zhipeng> in or out of a release is not that big a deal 14:58:52 <Li_Liu> zhipeng, will you go to Berlin? 14:59:17 <zhipeng> yes definitely 14:59:22 <zhipeng> maybe will do a keynote 14:59:28 <zhipeng> with cyborg+AI 14:59:35 <zhipeng> but nothing settled yet :P 14:59:36 <Coco_Gao> that's great 14:59:44 <wangzhh> Wow~ 15:00:10 <Li_Liu> nice 15:00:27 <zhipeng> a big maybe, let see how Mark Collier and others works out the arrangement 15:00:31 <zhipeng> maybe nothing 15:00:56 <zhipeng> plz remember submit topic for Berlin Summit 15:01:05 <zhipeng> ok see you guys tmr on ZOOM 15:01:10 <zhipeng> 10am Beijing time 15:01:17 <Coco_Gao> tomorrow? 15:01:17 <xinran__> Coco_Gao: I think you can go ahead because quota support still need Oslo limit be ready 15:01:41 <zhipeng> yes tmr 15:02:02 <zhipeng> xinran__ let's discuss it tmr as well 15:02:31 <xinran__> OK 15:02:55 <zhipeng> thx everyone 15:02:59 <zhipeng> #endmeeting