03:04:11 <Sundar> #startmeeting openstack-cyborg 03:04:12 <openstack> Meeting started Wed Jul 3 03:04:11 2019 UTC and is due to finish in 60 minutes. The chair is Sundar. Information about MeetBot at http://wiki.debian.org/MeetBot. 03:04:13 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 03:04:15 <openstack> The meeting name has been set to 'openstack_cyborg' 03:04:23 <Sundar> #topic Roll call 03:04:28 <Sundar> #info SUndar 03:04:29 <Yumeng> #info Yumeng 03:04:36 <Biwei> #info Biwei 03:05:04 <Sundar> Hi Yumeng, Biwei, xinranwang 03:05:27 <Sundar> Let us wait a min for others 03:05:48 <Yumeng> Hi sundar Biwei and xinran 03:05:55 <xinranwang> Hi all 03:06:04 <xinranwang> welcome back Yumeng 03:06:48 <Yumeng> ^ ^ thanks xinran 03:06:52 <Sundar> #topic Spec review 03:07:02 <s_shogo> Hi I joined this team as successor of Ikuo Otani from NTT Corp, thank you. 03:07:19 <Sundar> Hi s_shogo 03:07:24 <Sundar> Welcome to Cyborg 03:07:58 <Yumeng> welcome s_shogo 03:08:26 <Sundar> Nothing much on the agenda today except for completing the spec reviews: https://review.opendev.org/#/q/status:open+project:openstack/cyborg-specs+owner:Sundar+after:2019-04-01 03:08:42 <yikun> #info yikun 03:09:09 <Sundar> Please review and provide comments and/or a +1/+2 03:10:38 <Sundar> s_shogo: Hope you have got the information from Ikuo on the activities that NTT took up: Python 3 migration and Cyborg client 03:12:22 <Sundar> #topic AoB 03:12:40 <Sundar> Please continue the spec review. Any thing else to discuss today? 03:15:27 <wangzhh> Hi all. 03:15:40 <Yumeng> I just reviewed the tempest patch https://review.opendev.org/#/c/667231/ we may need add a zuul job to run it. 03:15:52 <Sundar> Hi wangzhh 03:16:04 <Yumeng> I can add it today. 03:16:47 <yikun> A question on the example on fake driver slot: https://review.opendev.org/#/c/665318/3/cyborg/accelerator/drivers/fake.py@42 03:17:53 <Sundar> yikun: Please see https://review.opendev.org/gitweb?p=openstack/cyborg.git;f=nova-integ/demo_setup.sh;hb=refs/changes/20/626420/24#l80 03:18:00 <Sundar> Yumeng: Thanks 03:18:13 <yikun> For now, we already use the string style slot like "0000:0c:00.0" in exist GPU/FPGA patch, so do you think we should change these to json something like "{domain:xx, slot:xx}" 03:19:05 <Sundar> yikun:Yes, that is what we concluded in our last IRC meeting: http://eavesdrop.openstack.org/irclogs/%23openstack-cyborg/%23openstack-cyborg.2019-06-19.log.html#t2019-06-19T03:32:48 03:20:52 <yikun> ok, sorry to miss that, then we should change it in GPU/FPGA and all other related patch, right? 03:20:52 <xinranwang> wangzhh: do we need this pci_addr string to name deployables in gpu driver 03:21:51 <yikun> and then we do NOT support string style slot anymore? only json format will be supported, right? 03:22:01 <Sundar> yikun: Yes 03:22:35 <Sundar> xinranwang:: The PCI string is for controlpath_id and attach_handle, not deployables. 03:22:50 <yikun> OK, It's okay to me, Let's see other idea, xinranwang, wangzhh, 03:22:53 <wangzhh> xintanwang: Yes. Checked it just now. 03:23:04 <wangzhh> *xinranwang 03:23:10 <yikun> : ) 03:23:23 <xinranwang> Sundar: I mean “this string is used to name the deployable” 03:23:40 <wangzhh> Yes. 03:23:49 <xinranwang> wangzhh: ok, thanks. 03:23:59 <wangzhh> Just name of deployable. 03:24:03 <yikun> wangzhh: 'r' 't' is very close I know 03:24:31 <yikun> https://review.opendev.org/#/c/665318/3/cyborg/accelerator/drivers/fake.py@34 03:25:10 <Sundar> yikun: I'll reply in that patch too. 03:25:54 <yikun> Yes, so "driver_dep.name = pci.get('device', '') + '_' + pci["slot"]" the deploy name still should be "device_strstylepciaddr" like "device_0000_00_000" 03:27:29 <Sundar> A deployable may have more than one VF. Then which PCI address would we give it? 03:27:41 <Sundar> I mean, what name would be give it? 03:28:37 <yikun> not sure, wangzhh do you have any idea for GPU deployable name? 03:29:21 <xinranwang> it's gpu's case, in FPGA driver, currently we do not name deployable using pci addr. 03:29:48 <xinranwang> But I am not sure how OPAE will report VFs' name 03:31:03 <yikun> As I know, the deployable logic model is useless in GPU, so I guess the device name can be changed to other. 03:31:36 <yikun> anyway, we need wangzhh to confirm. 03:32:16 <Sundar> We may use the deployable name as part of the Placement RP name. So, should we have some uniform naming convention across devices? 03:32:19 <wangzhh> <ComputeNodeName>_<NumaNodeName>_ <Device_name>_<Device_address> 03:32:49 <wangzhh> yikun, Can we named it like this? 03:33:02 <yikun> deployable name? 03:33:11 <wangzhh> Yes. 03:34:35 <Sundar> By device address, do you mean ctrl path id? 03:34:50 <yikun> whatif we do not have the numa? what the name looks like? 03:35:04 <yikun> Sundar: I guess it is 03:37:05 <wangzhh> yikun, we can set a default numa node value, such as 'default_numa'. 03:37:16 <Sundar> I think we had this discussion a long time ago. Must be in some IRC log 03:38:38 <Sundar> If there is more than one deployable but only one ctrlpath_id, I think we considered some solution for that. 03:39:03 <Sundar> Let's try to find the old discussion. 03:39:42 <Sundar> Meanwhile, can we all make it a priority to get the specs reviewed? 03:39:51 <Sundar> Milestone 2 is approaching fast 03:40:35 <yikun> anyway the name is not my concern, so back to original question, json style is ok to me if no one is against, I will change it to json style 03:41:02 <yikun> Sundar: thanks for reminder, I will review the spec today 03:41:58 <xinranwang> yikun Sundar I'm ok with json, I can change it in FPGA driver too 03:42:18 <Sundar> Thanks, yikun, xinranwang 03:42:26 <wangzhh> Me too. 03:42:27 <yikun> wangzhh: so could change the GPU one, and I will address fake and ascend patch 03:42:57 <yikun> and next topic, :) 03:44:09 <Sundar> Thanks, all 03:44:11 <Sundar> Anything else for today? 03:44:26 <xinranwang> I have one question about test related code 03:45:15 <Sundar> xinranwang: sure 03:45:33 <xinranwang> should we form a class like TestDeployableClass with test_list, test_get inside? 03:46:30 <yikun> it's for test Deployable obj? 03:46:44 <Sundar> I'd say yes. We are doing that for v2 (thanks to Biwei): like https://review.opendev.org/#/c/626057/12/cyborg/tests/unit/api/controllers/v2/test_device_profiles.py 03:47:08 <Sundar> But v1 is not following that pattern. Not sure if we should spend the time to fix v1 tests. 03:47:18 <yikun> ok, it's the api test 03:47:37 <xinranwang> I have an impression that I see some thing like TestList() class somewhere 03:48:06 <yikun> xinranwang: what's your expected? it's okay tome 03:49:29 <Sundar> xinranwang: Yes, some v1 API tests do that: https://review.opendev.org/gitweb?p=openstack/cyborg.git;f=cyborg/tests/unit/api/controllers/v1/_test_accelerators.py;hb=refs/changes/57/626057/12#l68 03:49:42 <xinranwang> Ah, v1 api test uses TestList() class. I think it's better to do like https://review.opendev.org/#/c/626057/12/cyborg/tests/unit/api/controllers/v2/test_device_profiles.py 03:50:07 <xinranwang> should we change v1, or just let it be : 03:51:25 <Sundar> IMHO, no point fixing v1 code, let alone v1 tests 03:52:41 <xinranwang> ok, sure. Just want make sure about this. 03:52:50 <xinranwang> I have no other questions 03:53:15 <Sundar> Anything else for today? 03:53:16 <yikun> https://review.opendev.org/#/c/660548/ 03:53:16 <yikun> one more topic from me, I will add the "aichip" type in this patch as we discussed in PTG before. 03:53:53 <Sundar> Sure, yikun. 03:54:11 <yikun> OK, thanks~ 03:54:28 <Sundar> I can give a +2, if others don't need more time 03:54:55 <yikun> OK, sure 03:55:10 <yikun> I have no other topic then 03:56:37 <Sundar> All done for today, then? 03:57:09 <Sundar> Thank you all. Let's complete the specs by next week. Have a good day! 03:57:14 <Sundar> #endmeeting