03:07:20 <Yumeng> #startmeeting openstack-cyborg 03:07:21 <openstack> Meeting started Thu Sep 24 03:07:20 2020 UTC and is due to finish in 60 minutes. The chair is Yumeng. Information about MeetBot at http://wiki.debian.org/MeetBot. 03:07:22 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 03:07:24 <openstack> The meeting name has been set to 'openstack_cyborg' 03:07:34 <Yumeng> #topic Roll call 03:07:39 <Yumeng> #info Yumeng 03:07:52 <swp20> #info swp20 03:07:55 <brinzhang> #brinzhang 03:08:00 <s_shogo> #info s_shogo 03:08:06 <xinranwang__> #info xinranwang__ 03:09:19 <Yumeng> I have no agenda for today. 03:09:36 <Yumeng> feel free to discuss any patchset if any 03:10:50 <chenke> Ok. I have two patches. One is tempest, another is wenping's patch. https://review.opendev.org/#/c/753299 https://review.opendev.org/#/c/752034/ xinranwang__ @s_ 03:11:10 <chenke> Ple all help to review. 03:12:12 <swp20> thanks chenke, my patch is about create more than one device profile with one request. 03:12:27 <swp20> it's a TODO things. 03:12:46 <swp20> if needn't, we should remove the TODO tag. 03:12:59 <chenke> Not a good TODO things. although Sundar had wrote it . 03:13:54 <Yumeng> yes, agree. if not necessary, we should remove. 03:14:05 <chenke> Ye. I think there is no need to support the creation of multiple dp at the same time. Nova's flavor has never supported this feature. 03:14:28 <brinzhang> chenke:why? 03:14:39 <chenke> The groups in our 1 dp support the definition of multiple resources at the same time. 03:14:45 <brinzhang> As we know, some users want to get this feature 03:15:23 <chenke> If user want. to define dp. I think he should create one by one. 03:15:47 <xinranwang__> IMO it causes the complexity in terms of cli. 03:15:49 <brinzhang> you mean he need to create server one by one 03:16:47 <chenke> There are still differences between dp and arq. dp is initiated by the user. If the user wants to create more than one, the interface can initiate multiple creation requests. 03:17:08 <chenke> No brinzhang 03:17:24 <chenke> I means flavor . 03:17:34 <brinzhang> the same 03:17:46 <chenke> the device profile in our cyborg is the same meaning as flavor. 03:18:30 <Yumeng> if we support multiple dps at once, does nova need to modify to support this? 03:18:31 <brinzhang> if the user want to create a server, he need to create a flavor, there will be many flavors in, and cannot re-use 03:19:20 <swp20> Yumeng, nova needn't to modify. 03:19:42 <chenke> Before create server. flavor should ready to go. Right? 03:21:29 <chenke> I think the post one dp API in enough. 03:22:01 <chenke> hi s_shogo yumeng xinran. what do your think? 03:22:36 <swp20> maybe it's useful in future if we have lots of devices to combine. 03:23:02 <brinzhang> I think you should think about the uses requirement, and then make a decision 03:23:31 <brinzhang> s/the uses/users/ 03:25:01 <xinranwang__> brinzhang: what use cases do you encouter where need to create multiple dp in one request, can you explain it to us? 03:25:10 <s_shogo> IMO, I agree with chenke(to keep one dp API), at now. Users can issue the multiple request in their client script and so on. To keep simple one dp is good from transactions management 's point of view. 03:26:31 <Yumeng> swp20, brinzhang: IMHO, devices are expensive. Was not very clear about the use case: lots of devices to combine. 03:26:51 <Yumeng> yes, the same question as xinran. can you explain it？ 03:26:59 <brinzhang> xinranwang__: we have two customers give us a requirement, they want to boot a server with support GPU and FPGA in a same server 03:27:47 <Yumeng> IMO, one dp can handle this issue. 1 dp can define two resources, right? 03:27:49 <chenke> That's ok now with one dp. 03:29:13 <chenke> openstack accelerator device profile create fpvgpu '[{"resources:FPGA": "2", "trait:CUSTOM_FAKE_DEVICE": "required"},{"resources:PGPU": "1","trait:CUSTOM_XXX": "required"}]' 03:29:22 <brinzhang> not correct in above: they want. the case is : [{dp1: gpu}, {dp2: gpu+fpga}] 03:30:56 <chenke> As shogo said, I think it is very easy if the web UI sends two creation device profile requests. 03:31:22 <swp20> Yumeng:one server can hold four devices commonly, and if GPU can be virtualized, we can have lots of devices. 03:32:01 <Yumeng> agree shogo +1 03:32:06 <brinzhang> this is a blueprint, and in Victoria we have freezed feature, we dont need to waste time to talk this 03:32:06 <chenke> In any case, the API should maintain its independence and self-explanation. We do not recommend making it complicated. 03:32:50 <brinzhang> swp20: please register a blueprint, and submit a spec, then talked this in PTG or in wallaby release 03:33:18 <brinzhang> we should not take this on today's meeting 03:33:23 <Yumeng> swp20: I undersrand. what I mean here is the hardware. 03:33:54 <Yumeng> alright. then let's move on. 03:34:04 <Yumeng> I also have two patchsets. device_profile validate are already to merge. https://review.opendev.org/#/q/status:open+project:openstack/cyborg+branch:master+topic:device-profile-post-validate 03:34:30 <Yumeng> pls help to review. 03:35:02 <Yumeng> and this document for policy-popup https://review.opendev.org/#/c/753233/ 03:35:24 <Yumeng> this has a tiny pep8 error, I will fix soon. 03:36:01 <Yumeng> the releasenote for policy-popup depends on the doc patch, so pls help to review https://review.opendev.org/#/c/753233/ 03:36:34 <swp20> alright. 03:37:35 <Yumeng> thanks. 03:37:41 <Yumeng> that's all from my side. 03:38:11 <Yumeng> Do you have any other patch to disscuss? 03:38:55 <Yumeng> ok. cool. 03:40:26 <Yumeng> if nothing else, let's wrap up today's meeting. we will reach Victoria rc soon! 03:40:52 <chenke> OK. 03:41:07 <Yumeng> Thank you all for coming. bye 03:41:45 <Yumeng> #endmeeting