| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-cyborg - 2017-10-11 | |||
| 15:17:35 | jkilpatr | looks good, but I think we need to focus on drilling down on these specs. we have high level stuff (nova interaction, spdk) but next is spec'ing standard traits/flavors in nova placement. | |
| 15:17:45 | jkilpatr | like format and such. | |
| 15:19:31 | zhipeng | agree | |
| 15:20:25 | zhipeng | but spdk/dpdk is less of a concern on traits, since they are not directly related to the compute resource | |
| 15:20:46 | jkilpatr | not saying it's a concern for that spec specifically, but in general both of these need follow up specs to flesh out details. | |
| 15:21:03 | crushil | What is the timeline for the implementation of the Cyborg Nova interaction spec? | |
| 15:21:06 | zhipeng | oh definitely | |
| 15:21:35 | zhipeng | Q-2 if we are fast enough ? | |
| 15:22:02 | jkilpatr | it's just calling existing placement api stuff, hard part is getting the format for the tags we want to use then keeping track of all of them | |
| 15:22:26 | crushil | And don't we need to have a patch in Nova as well? | |
| 15:22:53 | jkilpatr | the live whitelist addition? | |
| 15:22:57 | zhipeng | we agreed in denver that we should experiment with nova extension first | |
| 15:23:08 | zhipeng | to get a sense of what we want to add | |
| 15:23:30 | zhipeng | attach/detach for hotplug to start with | |
| 15:23:57 | zhipeng | then we could have a formal proposal to the nova team | |
| 15:24:02 | Li_ | basically issue virsh attach in libvirt right? | |
| 15:24:10 | zhipeng | this is what we think nova api should be extended and why | |
| 15:24:31 | zhipeng | Li_ that would be the nova part | |
| 15:24:35 | crushil | If we are planning to have a patch in Nova, then we should be conservative as the Nova queue is already overloaded with patches and the review time will be months if not weeks | |
| 15:24:46 | jkilpatr | didn't think of that. | |
| 15:25:03 | jkilpatr | at the same time we don't want to make a nova patch first and then figure out we don't like it's workflow once we write the other code... | |
| 15:25:15 | zhipeng | yes exactly | |
| 15:25:18 | crushil | +1 | |
| 15:25:33 | zhipeng | so what we want to have out of Queens is to function with placement | |
| 15:25:48 | zhipeng | and an unofficial hacky extension of nova | |
| 15:25:53 | zhipeng | to make it e2e working | |
| 15:26:00 | crushil | And have a semblance of a driver so that vendors have something tangible to play with | |
| 15:26:10 | zhipeng | then we could discuss with nova team about the api extension in Rocky | |
| 15:26:19 | zhipeng | yes crushil | |
| 15:26:51 | crushil | We basically want to end the vendor dependence on Nova for accelerators by the end of this release | |
| 15:27:09 | zhipeng | provide a leeway | |
| 15:27:12 | zhipeng | so to speak | |
| 15:27:27 | zhipeng | #link https://review.openstack.org/508572 | |
| 15:27:48 | zhipeng | btw the conversation so far is about this magnificent beast | |
| 15:27:57 | jkilpatr | lol | |
| 15:30:37 | zhipeng | i will work with zhuli to add more resource provider related info to the spec | |
| 15:32:07 | zhipeng | jkilpatr for the spdk spec if it is ok plz green light it so that helloway could start code implementation | |
| 15:32:18 | jkilpatr | sounds good. I'm not sure we want the placement stuff in the same spec as high level nova cyborg interaction. | |
| 15:32:43 | zhipeng | and #action zhuli and howard will provide additional info on resource provider/trait design for the nova interaction spec | |
| 15:32:47 | jkilpatr | ok +3 | |
| 15:33:26 | zhipeng | jkilpatr let's try and figure out then | |
| 15:33:34 | crushil | +3? | |
| 15:33:54 | jkilpatr | +2 cr +1 wf | |
| 15:33:59 | zhipeng | beast privilege | |
| 15:34:45 | zhipeng | XD | |
| 15:35:19 | zhipeng | #topic AoB | |
| 15:35:53 | zhipeng | okey any other topics/ideas/what have you | |
| 15:36:40 | jkilpatr | nope, I'll start poking at adding devices to the nova whitelist live, see how hacky I can get away with at first. | |
| 15:36:58 | jkilpatr | if you have placement and crushil has the driver I think that's what's left uncovered? | |
| 15:37:18 | zhipeng | yep | |
| 15:37:19 | jkilpatr | until it comes time to implement all of the placement calls from the conductor/agent then I can switch to that. | |
| 15:38:39 | crushil | Is the Nova team aware of all the Nova-Cyborg interaction? I see that they have commented on the patch | |
| 15:39:01 | crushil | \me doesn't remember | |
| 15:39:12 | zhipeng | yes they are | |
| 15:41:36 | crushil | zhipeng, Cool. I haven't been following the Nova-Cyborg interaction much. Nova team has a lot on their plate. Just making sure they are aware that this is coming their way | |
| 15:42:00 | zhipeng | no problem :) | |
| 15:42:13 | zhipeng | btw how's the driver going crushil ? | |
| 15:42:29 | zhipeng | we will start with GPU ? | |
| 15:43:10 | crushil | It's going. I have been closing on a lot of my downstream work. We are implementing an in house cloud with accelerators embedded in them | |
| 15:43:54 | crushil | We actually demoed to a customer about SPDK integration POC in our lab. They seemed interested in Cyborg | |
| 15:44:06 | zhipeng | ah cool | |
| 15:45:02 | crushil | Btw zhipeng do you know whether you'll be at the Sydney summit? | |
| 15:45:17 | zhipeng | chances are i will not be there :( | |
| 15:45:32 | zhipeng | we've cut a lot of people for sydney attendance | |
| 15:45:39 | zhipeng | budgeting issue | |
| 15:45:50 | zhipeng | i will still try, but not likely | |
| 15:46:02 | zhipeng | I will prepare the slide tho, for our lightening talk | |
| 15:46:34 | crushil | I'm not going either | |
| 15:46:41 | crushil | 2 of my coworkers are | |
| 15:46:51 | crushil | I'm going to make them cover our lightning talk | |
| 15:47:10 | jkilpatr | Sounds like Sydney isn't going to have attendance from the team then :( | |
| 15:47:58 | zhipeng | yep :( | |
| 15:49:07 | jkilpatr | too bad we can't attend via telepresence robots and be real cyborgs :P | |
| 15:49:55 | zhipeng | that would be the goal for S release | |
| 15:50:19 | zhipeng | scare the S out of everybody | |
| 15:51:16 | crushil | haha | |
| 15:52:14 | crushil | I'll make one of my coworkers cover the talk | |
| 15:52:21 | jkilpatr | https://shop.suitabletech.com/products/beam-smart-presence-system?variant=28643990416 | |
| 15:52:55 | jkilpatr | what they make you pay a subscription to use it forever? | |
| 15:53:12 | jkilpatr | great now we need FOSS robots. | |
| 15:54:05 | zhipeng | ok i'm staring the girl in the picture for too long | |
| 15:54:20 | jkilpatr | you're tired, go to sleep. It's what midnight over there? | |
| 15:55:25 | zhipeng | 5 min to midnight | |
| 15:56:24 | zhipeng | okey folks great meeting today | |
| 15:56:42 | zhipeng | i'm gonna close it on time, like any official projects do | |
| 15:56:47 | zhipeng | lol | |
| 15:56:53 | zhipeng | #endmeeting | |
| 15:56:55 | openstack | Meeting ended Wed Oct 11 15:56:53 2017 UTC. Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4) | |
| 15:56:57 | openstack | Minutes: http://eavesdrop.openstack.org/meetings/openstack_cyborg/2017/openstack_cyborg.2017-10-11-15.01.html | |
| 15:56:58 | openstack | Minutes (text): http://eavesdrop.openstack.org/meetings/openstack_cyborg/2017/openstack_cyborg.2017-10-11-15.01.txt | |
| 15:56:59 | openstack | Log: http://eavesdrop.openstack.org/meetings/openstack_cyborg/2017/openstack_cyborg.2017-10-11-15.01.log.html | |
| #openstack-cyborg - 2017-10-18 | |||
| 15:10:12 | jkilpatr | morning 0/ | |
| 15:19:18 | Guest73850 | #info Li Liu | |
| 15:19:51 | jkilpatr | #info Justin Kilpatrick | |
| 15:19:59 | jkilpatr | zhipengh[m], are we not meeting this week? | |
| 16:55:52 | zhipengh[m] | Yes I'm on travel, sent an email to the ml earlier | |
| 17:38:39 | crushil | zhipengh[m], Next time, you don't have to cancel the meeting. jkilpatr or I can run it. :) | |
| 17:39:46 | zhipengh[m] | I suppose that would be better :) | |
| 18:44:18 | jkilpatr | man how did I miss that? I trash too much mailing list email | |
| #openstack-cyborg - 2017-10-23 | |||
| 13:50:14 | jkilpatr | https://git.openstack.org/cgit/openstack/os-traits/tree/ is looks like we should just hook into this for trait name standardization | |