03:01:14 #startmeeting openstack-cyborg 03:01:15 Meeting started Thu Nov 21 03:01:14 2019 UTC and is due to finish in 60 minutes. The chair is Sundar. Information about MeetBot at http://wiki.debian.org/MeetBot. 03:01:16 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 03:01:18 The meeting name has been set to 'openstack_cyborg' 03:01:26 Hello all 03:01:28 #info Li_LIu 03:01:33 Hi Sundar 03:01:43 Hi Li, how are you? 03:01:44 #Yumeng 03:01:49 #info s_shogo 03:01:51 Hi all 03:01:52 good~ 03:01:54 #info chenke 03:01:54 hi all 03:01:58 hi all~ 03:02:11 Hi all 03:02:18 Good to have a quorum. 03:02:20 #info xinranwang 03:02:46 Agenda: https://wiki.openstack.org/wiki/Meetings/CyborgTeamMeeting#Agenda 03:02:53 #topic Status update 03:03:37 Glad to see lots of movement and progress in Cyborg 03:04:25 On the Nova side, we got another +1 on the spec: https://review.opendev.org/#/c/684151/ 03:04:37 We re trying to get this over the edge 03:04:57 I updated the patch series: https://review.opendev.org/#/q/status:open+project:openstack/nova+bp/nova-cyborg-interaction 03:05:58 Added API versioning for the notification event (https://review.opendev.org/692707), which turned out out be a bit tough, because lots of the requirements are not documented anywhere 03:06:08 We just have to read previous code and history 03:06:33 Added functional tests for Cyborg in Nova --- this was another recent requirement 03:06:59 I am not aware of any blocker now, but we are still working on getting reviews 03:07:25 Please do review this patch series and offer comments or +1 03:07:36 sure, will do 03:08:00 One prerequisite for tis patch series is this Cyborg patch: https://review.opendev.org/694856 03:08:17 Please review and help to merge it 03:09:15 should microversion be configurable? 03:09:47 I just made it 'latest'. Is there a reason why we would want an older microversion? 03:09:54 I want to know, we need to always set the version to latest? 03:10:38 Not sure if we need to do it, but it is convenient, because it will work with the Nova patches or without 03:11:06 Otherwise, tempest tests with the Nova patches will fail 03:11:49 not know why nove need a microversion science there is no reason to request an older microversion 03:11:50 The Nova patches move the microversion to '2.82' because we defined the Cyborg ARQ notification event 03:13:00 Any objections to using 'latest' microversion? 03:13:09 I think maybe some production environment needs to use the old api. 03:13:33 For the current state, agree. 03:13:38 there maybe some compatible problems in a real env. 03:13:49 Ok 03:14:52 If the operator is deploying older Nova, he will not get Cyborg functionality -- that needs '2.82' 03:15:54 no objections for using 'latest'. Just wonder should it be configurable. 03:16:09 OK. We can enhance this if there is a need. 03:16:18 Agree. 03:16:20 yes ,aggree with shaohe, it should be configurable 03:16:22 OK 03:16:32 On another note: the devstack patch for using Python 3 has merged: (https://review.opendev.org/#/c/649097 03:17:18 The Zuul gate was broken because Nova expected Py3 by devstack was using Py2. Now that should work. 03:17:25 *but 03:18:16 FYI: The Cyborg meeting info was out of date: I am updating it -- https://review.opendev.org/#/c/695343/ 03:18:56 Esp. as we expect more interaction with neutron 03:19:56 #topic Programming APIs 03:20:52 We have been talking about the programming requirements. I created this etherpad to continue the discussion from the PTG: https://etherpad.openstack.org/p/cyborg-ussuri-programming-apis 03:21:06 Please review it offline when you can 03:22:20 We can take this up next week. 03:23:06 #topic Who's doing what 03:23:49 All of us have been busy with various things. Plus, many of us signed up for various areas (last part of https://etherpad.openstack.org/p/shanghai-ptg-cyborg) 03:24:17 Just trying to get a sense of what folks are busy with. 03:25:21 Sundar: Continuing to work on Nova patches, Cyborg tempest plugin, 3rd party CI for FPGAs (with another colleague), starting on support for networked devices (with other colleagues in Intel) 03:25:54 Plus assorted patches as needed 03:25:54 will we support SmartNic? 03:26:06 Shaohe: Yes, different types 03:26:07 I currently got nothing on my plate 03:26:51 Li_Liu: please feel free to look at the etherpad and take something of interest. For example, the entire area of functional tests for Cyborg is open. 03:27:15 That is absilutely needed to make Cyborg into a production-quality project 03:27:16 sure, I will take a look 03:27:23 Thanks, Li_Liu 03:27:42 Do I need any hardware to do the functional test? 03:27:54 No, it can all be with the fake driver for now 03:27:57 Sundar if you have a smartnic env, please let me try. Thanks. 03:28:01 great 03:28:33 shaohe_feng: Let's talk offline 03:28:40 Shogo: 03:29:59 BTW, I may get another colleague to support on openstacksdk, so chenke can focus on other things too :) 03:31:11 ok. 03:31:39 When I have free time. I will help to review it. 03:32:00 s_shogo: I think you have up the Pythn client plus some APIs. Please let us know if you need help, or if you'd like to spread the work more 03:33:42 maybe we can put the tasks on the storyboard 03:33:46 xinranwang: Similarly, you have the Intel driver plus /v2/devices GET, including microversion support, I think. Do you need help? 03:33:47 Sundar: That's right. 03:34:16 ( And I intend to send reminder to openstacksdk folks to review the patch > https://review.opendev.org/#/c/679914/) 03:34:31 xinranwang: Yes, Storyboard is a good place. We already have many entries: https://storyboard.openstack.org/#!/project/openstack/cyborg 03:34:49 I am ok, s_shogo will work with me device API 03:35:17 But it is useful only if all of us use it regularly -- otherwise we'll still not know what tasks are remaining, and who is doing what 03:35:33 we'd better to list the tasks we discussed during ptg ther 03:36:09 Cool. I know Yumeng has been active with the docs update. Can others please LMK what they are working on? 03:36:11 Sundar: yes, I agree 03:36:48 Do we all vote to use Storyboard regularly for tasks? 03:36:54 +1 03:37:22 +1. and I would start UT improvement in the next. 03:37:42 Yumeng: nice, thanks :) 03:37:43 +1 03:37:50 or tempest 03:37:56 +1 03:38:14 Li_Liu: ^ 03:38:32 shaohe_feng: ^ 03:38:51 tempest and UT will be the next thing I will do. 03:38:58 Sundar: just notice you mentioned intel driver, which one you mean? 03:39:09 Intel FPGA driver 03:39:28 +1 03:39:30 Sundar: BTW, do we have plan about shelve and unshelve in this cycle? I'm intrested. 03:42:56 Got disconnected :( 03:42:58 +1 03:43:06 Thanks, all 03:43:25 #agreed We'll use storyboard regularly for tasks going fwd 03:43:40 Yumeng: Can you say more about the use case for un/shelve? 03:45:10 Sundar: not very specific use case for now. we can discuss more maybe next time. 03:45:49 Ok. I don't mind adding it. But hope to get the basic cases merged first. 03:46:15 #topic Patches 03:46:46 Special :+1: to Yumeng for the thorough doc patcches 03:47:13 I will update today. I saw your comments. thanks for review. 03:47:50 Cool. I think it will be nice to have some introductory doc on the object model and Cyborg-Nova workflow 03:47:57 I'll add it 03:48:21 We have a story for this: https://storyboard.openstack.org/#!/story/2006384 03:48:41 All: please add any missing doc improvements here 03:49:10 Yumeng: you used sql benchwork for cyborg doc ?? ;) 03:49:33 xinranwang: haha not for doc 03:50:14 just generated a ER diagram of cyborg-db automatically 03:50:39 amazed by it productivity 03:50:50 lol 03:50:55 *its 03:51:48 Cool. Does any patch need specific discussion? 03:52:08 https://review.opendev.org/#/c/693784/ 03:52:16 Is this OK? 03:52:33 Did you test on multiple VMs? with fake driver? 03:52:50 I have list the configure options in the etherpad 03:53:13 seems only Sundar and xinranwang give feedback. 03:54:10 If no problem about the config options, I will retest it again to let it go on. ' 03:54:49 #link https://etherpad.openstack.org/p/cyborg-ptg-ussuri 03:55:27 Yea, I gave some comments. The agent needs to reach glance, so it would need keystone 03:55:54 We need to set up an env with multiple physical nodes with FPGAs, to test programming paths 03:56:31 Or can we somehow extend the fake driver to simulate programming? 03:57:06 yes, we can extend. 03:57:11 Latter will be quite useful for UT/functional tests 03:57:38 such as write a fake program script 03:57:48 Shaohe_feng: would you like to take that up? That will give more confidence in the multi-node setup. Which is needed for real-world operations. 03:58:25 Yes, the fake device could advertise a region/function ID, and the fake driver coul always return success for programming calls 03:58:51 when agent download a image, we let the fake program script do some simple check and then return error or OK 03:59:15 Sure 03:59:31 maybe the fake program script can output some information to tmp/file 03:59:50 not sure it can output to the cyborg log 04:00:02 We don't need a program script per se, right? Just the program() API in the driver 04:00:30 what is per se? 04:00:52 I mean, we don;t really need a separate script. 04:01:35 The program API can just look at something in the image file and decide to return success or failure 04:01:46 That way, we can use different images to force failure orsuccess 04:01:59 maybe just deploy a fake script by devstack when we use fake driver 04:02:23 yes. let program API do it is OK. 04:03:08 OK. Let's talk offline if needed. We are past the end time 04:03:12 #topic AoB 04:03:16 Anything else, folks? 04:03:18 OK. 04:03:41 will we setup a multi-node gate? 04:03:48 any plan for it? 04:03:53 Do we need to? 04:03:59 not sure. 04:04:30 Yumeng, chenke, all: do other projects have it? 04:04:48 OK. maybe all in one is enough. 04:05:12 seem others are going to have lunch :) 04:05:36 Ah ok 04:05:50 Goodbye, everybody 04:05:55 #endmeeting