| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-cyborg - 2017-08-29 | |||
| 13:59:04 | zhipeng | is it possible to do before the PTG ? | |
| 13:59:46 | crushil | Before the PTG? | |
| 13:59:53 | crushil | I am double booked till then | |
| 14:00:01 | crushil | I was planning on working on it after the PTG | |
| 14:00:19 | zhipeng | ok cool, just want to know your status :) | |
| 14:00:33 | zhipeng | so what is the accelerator you could play with in your lab ? | |
| 14:08:39 | crushil | zhipengh[m], I have a Tesla K80 in the lab | |
| 14:31:25 | zhipengh[m] | crushil cool | |
| #openstack-cyborg - 2017-08-30 | |||
| 15:01:23 | zhipeng | cyborgers | |
| 15:01:46 | jkilpatr | booyha | |
| 15:02:03 | jkilpatr | got a working version of the deployment patch up, need someone to test it on devstack | |
| 15:02:20 | zhipeng | great ! | |
| 15:02:21 | jkilpatr | well I could test it on devstack, but it could be quicker if someone else has one already stood up. | |
| 15:02:37 | zhipeng | #startmeeting openstack-cyborg | |
| 15:02:38 | openstack | Meeting started Wed Aug 30 15:02:37 2017 UTC and is due to finish in 60 minutes. The chair is zhipeng. Information about MeetBot at http://wiki.debian.org/MeetBot. | |
| 15:02:40 | openstack | Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. | |
| 15:02:42 | openstack | The meeting name has been set to 'openstack_cyborg' | |
| 15:02:57 | jkilpatr | one comment for zhuli you capture rpc connection errors and log them in the common service code, but you don't change the exit code, this makes automated checks difficult :( | |
| 15:03:00 | jkilpatr | I' | |
| 15:03:17 | jkilpatr | I'm not really sure what to do to fix it, since there seems to be some graceful exit code there I didn't want to break | |
| 15:04:22 | zhipeng | i will let him know if he's not here | |
| 15:04:22 | zhipeng | zhuli around ? | |
| 15:04:36 | zhipeng | #topic patch discussion | |
| 15:04:49 | zhuli | yep, I'm here | |
| 15:04:52 | crushil | jkilpatr, I can test it on devstack | |
| 15:05:03 | zhipeng | crushil that'd be great | |
| 15:05:16 | zhipeng | zhuli did you get justin's question ? | |
| 15:05:18 | crushil | I will add a devstack plugin too soon. Hopefully this week | |
| 15:06:35 | zhipeng | thx crushil, looking forward to it | |
| 15:06:49 | zhuli | I will confirm the rpc exit code issue later | |
| 15:08:00 | zhipeng | thx zhuli | |
| 15:08:16 | zhuli | jkilpatr, I will test it on devstack too if I have time | |
| 15:09:38 | zhipeng | so any questions on zhuli's three new patches ? | |
| 15:09:55 | jkilpatr | commit messages! | |
| 15:10:00 | zhipeng | other than comit msg fixing as uaual | |
| 15:10:04 | zhipeng | haha I was just saying | |
| 15:10:04 | jkilpatr | lol | |
| 15:10:24 | jkilpatr | looks good to me otherwise, I think I commented on at least two, I'll have to go and look at the third. | |
| 15:10:49 | zhuli | I think I need fulfill the commit messages before merged | |
| 15:10:56 | zhuli | yep | |
| 15:12:20 | zhipeng | zhuli just git commit --amend everytime after git commit -a -m :P | |
| 15:12:42 | zhipeng | so we got one more new patch, from a colleague of mine | |
| 15:12:47 | zhipeng | #link https://review.openstack.org/#/c/498690/ | |
| 15:13:12 | zhipeng | this is a new spec (targeting queen) for SPDK driver | |
| 15:13:23 | zhipeng | for high perf NVMe SSDs as accelerators | |
| 15:13:39 | zhuli | zhipeng, thx | |
| 15:14:08 | zhipeng | zhuli no problem | |
| 15:14:20 | zhipeng | everyone plz help review the spdk spec when you got time | |
| 15:14:32 | zhipeng | altho this is not Pike priority | |
| 15:15:12 | jkilpatr | so this involves cyborg reconfiguring other openstack services | |
| 15:15:25 | zhipeng | not really I think | |
| 15:15:34 | jkilpatr | don't we need to do some config in cinder? | |
| 15:15:36 | jkilpatr | can we do that live? | |
| 15:15:51 | zhipeng | no we only do that for spdk | |
| 15:16:05 | jkilpatr | ok then I'm less concerned about the scope getting too big then. | |
| 15:16:58 | zhipeng | so spdk is in the config anyways | |
| 15:16:58 | zhipeng | the premise of this scenario is that Cinder will use a accelerated backend | |
| 15:17:04 | zhipeng | yep | |
| 15:17:16 | jkilpatr | just don't want to obligate us to restart other services | |
| 15:17:24 | zhipeng | definitely | |
| 15:18:32 | jkilpatr | so the conductor is in charge of nova placement api interaction? | |
| 15:18:39 | jkilpatr | we should start looking into that. | |
| 15:18:56 | zhipeng | I was gonna discuss with you about that | |
| 15:19:04 | zhipeng | okey let's move on to the next topic | |
| 15:19:18 | zhipeng | #topic Denver PTG TC presentation prep | |
| 15:20:41 | zhipeng | for us to better present the project to the TC | |
| 15:20:41 | zhipeng | I did a internal check with our TC member, and get several really good suggestions | |
| 15:20:41 | zhipeng | so I will do a project update presentation to TC on Sep 12th | |
| 15:20:57 | jkilpatr | oh cool we're presenting to TC | |
| 15:21:04 | zhipeng | So one thing would be an end to end workflow demo of Nova works with Cyborg | |
| 15:21:14 | zhipeng | could be just bare minimum functionality | |
| 15:21:21 | zhipeng | but a demo of such would be great | |
| 15:21:49 | jkilpatr | well we need some sort of driver and I'm not even sure if we can whitelist and load a pci device live, there's some docs for it. | |
| 15:21:59 | jkilpatr | but it looks dusty to me. | |
| 15:22:18 | zhipeng | jkilpatr we could do it without the driver, in my opinion | |
| 15:22:58 | zhipeng | we could have the generic driver just response success everytime the request do get through there | |
| 15:24:40 | zhipeng | jaypipes mentioned to me that we could just copy that python file from nova compute | |
| 15:24:40 | zhipeng | Justin could help add the report.py to the agent | |
| 15:24:40 | zhipeng | jkilpatr crushil zhuli I think what we could do is that | |
| 15:24:40 | zhipeng | the main point is on the Nova Cyborg interaction | |
| 15:24:49 | zhipeng | with necessary modification | |
| 15:25:13 | zhipeng | and then it could interact with Placement, which could interact with Nova | |
| 15:27:13 | zhipeng | well let's just do a best effort, see what we could have before PTG :) | |
| 15:27:13 | zhipeng | zhuli could help with nova api hack, which could talk to cyborg api | |
| 15:27:28 | zhipeng | sounds good ? | |
| 15:31:32 | zhipeng | okey I will take that silence as a OK :P | |
| 15:31:36 | jkilpatr | sorry dual meetings | |
| 15:31:54 | jkilpatr | what does report.py do? | |
| 15:32:52 | zhipeng | zhuli has provided some on the api/db | |
| 15:32:52 | zhipeng | another feedback I got is that we lack of unit/function tests | |
| 15:32:52 | zhipeng | jkilpatr no worries | |
| 15:32:54 | zhipeng | jkilpatr and crushil could you guys help on the conductor/agent and driver side as well ? | |
| 15:32:55 | zhuli | you mean the cyborg api to response something like detach/attach action from nova ? | |
| 15:33:05 | zhipeng | zhuli yes | |
| 15:33:18 | zhipeng | jkilpatr it reports the resource information to the placement api | |
| 15:33:18 | zhuli | ok | |
| 15:33:19 | jkilpatr | ok I think that's doable, I need to debug some things though, zhuli when someone sends an api request does it actually go out on rpc to the conductor? We need that to work for an instance request so that I can test this. | |
| 15:34:44 | zhipeng | okey, back to the unit/function testing issue | |
| 15:35:30 | zhuli | jkilpatr, yes the api will send rpc message to conductor | |
| 15:35:52 | jkilpatr | well we can run the deployment playbook and make some requests against it for functional tests. Unit testing wise there isn't much buisness logic anywhere but the API at the moment so test will have to come with functionality | |