14:11:47 <Li_Liu> #startmeeting openstack-cyborg
14:11:48 <openstack> Meeting started Wed Aug 15 14:11:47 2018 UTC and is due to finish in 60 minutes.  The chair is Li_Liu. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:11:49 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
14:11:52 <openstack> The meeting name has been set to 'openstack_cyborg'
14:12:01 <Li_Liu> #topic Roll Call
14:12:02 <Sundar> #info Sundar
14:12:04 <Coco_gao> #info Coco_gao
14:12:08 <Li_Liu> #info Li_Liu
14:12:15 <wangzhh> #info wangzhh
14:12:39 <Li_Liu> #Topic Status Update
14:12:42 <xinran> #info xinran
14:12:54 <Li_Liu> https://review.openstack.org/#/q/status:open%20project:openstack/cyborg
14:13:53 <Li_Liu> I see xinran just updated https://review.openstack.org/#/c/564968/
14:14:18 <Li_Liu> Please help reviewing this
14:14:29 <Coco_gao> I will test my object code with agent part tomorrow.
14:14:42 <xinran> Yes I just updated it including to the new api design
14:14:53 <Li_Liu> There is no commit on that yet right Coco_gao?
14:15:01 <Sundar> Are we aiming code changes for Stein release, since the Rocky cycle is past code freeze?
14:15:15 <Li_Liu> thanks a lot xinran
14:15:23 <Coco_gao> past?
14:15:44 <xinran> But I am not sure how we handle the quota granularity
14:16:00 <shaohe_feng_> #info shaohe_feng_
14:16:44 <shaohe_feng_> deployable is OK.
14:16:54 <Sundar> Coco_gao: #link https://releases.openstack.org/rocky/schedule.html July 27 was feature freeze. We are past the RC1 milestone
14:16:58 <shaohe_feng_> for granularity at present
14:17:35 <Sundar> However we do quotas now, it should be upgradable to the per-RC quota that is coming in Nova.
14:19:19 <shaohe_feng_> It is OK for placement do quota in openstack
14:20:09 <Li_Liu> Sundar, I will double check on that, but there is another window Aug20-24 for Final RCs
14:21:05 <Li_Liu> We hope we can squeeze CoCo's stuff in, as these code are not affecting external customers/users/components anyways
14:22:51 <Coco_gao> And zhenghao's part is quite related to my part, the agent code need to update either.
14:24:49 <Coco_gao> If we commit before August 20, is that possible to catch up the Final RCs?
14:24:56 <Li_Liu> ok, I can see the only major pieces that are missing are 1. Coco_gao and wangzhh's code 2. merging xinran's code
14:25:20 <Li_Liu> I think you can, but I need to double check for you
14:25:37 <shaohe_feng_> Li_Liu: Sundar: how is the new FPGA driver?
14:26:16 <Sundar> Will Coco-gao's code include driver-API interaction via OVOs, and reflect the os-acc-related aspects too?
14:26:55 <shaohe_feng_> it is OVOs. :)
14:27:00 <Li_Liu> shahe_feng_, are you talking about https://review.openstack.org/#/c/584170/?
14:27:18 <Sundar> shaohe: The OPAE-based driver needs to be tested well before committing.
14:27:36 <shaohe_feng_> Li_Liu: yes.
14:28:02 <Coco_gao> I and zhenghao will try our best.
14:28:35 <Li_Liu> I think it still needs to be tested against CI at least
14:28:54 <shaohe_feng_> Sundar: Is it necessary for we to update the old drivers?
14:29:05 <Coco_gao> Yeah, I think so
14:29:20 <shaohe_feng_> Coco_gao: thanks. hard work.
14:29:28 <wangzhh> Coco_gao: Yes. ლ(╹◡╹ლ)
14:29:53 <shaohe_feng_> Sundar: the agent depend on the driver.
14:30:18 <Sundar> Just curious: Why do we want to get code into Rocky at this stage? Aren't we better off getting agreement on the overall flow and doing it properly?
14:30:25 <Coco_gao> In the agent , we need the code to be more general rather than FPGA specific.
14:30:56 <Coco_gao> Sundar, that's an option if there are no hurry.
14:31:04 <Coco_gao> I don't know.
14:32:00 <shaohe_feng_> Sundar: do you have better design for the new refactor?
14:32:18 <Sundar> Coco_gao: agreed that the agent is not FPGA-specific. The drive-agent API should not be device-specific either. Also, there are requests to let the driver tell the agent about custom traits for its device.
14:33:20 <Sundar> shaohe: I have some ideas on the driver API along the lines above, and addressing the diff update etc. I am writing it as a new spec, to be out this week (for Stein PTG).
14:33:27 <Sundar> Hope to get input from all of you!
14:38:00 <Sundar> Also, is anybody thinking about the test framework that we have today? How do we get Zuul to work with real devices?
14:41:07 <Coco_gao> Should Zuul test real devices?
14:41:47 <efried> Procedurally, trying to get something of this size into Rocky is a no-go.
14:43:20 <efried> I think to test real devices you would need a third party CI. You can't count on the nodes in the pool managed by zuul for tox- and devstack-based tests to have devices.
14:44:37 <Sundar> May be we should create fake devices in the testing framework for every supported device type. Like shaohe's fake FPGAs.
14:44:43 <Li_Liu> Can you fake the device information for CI?
14:46:40 <shaohe_feng_> we had better do some fake
14:46:42 <wangzhh> Li_liu: Now we fake them for UT.
14:46:55 <Coco_gao> Yeah, we need to make sure the fake information is same with the what we get from devices.
14:47:13 <Sundar> If we do create fake devices, they even need to simulate timeouts and unresponsiveness, to test various code paths.
14:47:30 <Li_Liu> efried, I get your point, we are just keep committing, if we can make to Rocky, that's cool, if we can not, it will just slide into S
14:48:36 <shaohe_feng_> Sundar: for hardware devices a third party CI is better
14:48:39 <Li_Liu> unlike other projects, Cyborg is still not in its usable stage yet. Keep having code to make it mature is the key right now
14:49:00 <efried> The project has already forked, so you would have to explicitly backport changes to stable/rocky. That is the thing I'm saying you shouldn't do.
14:49:12 <efried> Keep committing to master (which is Stein at this point)
14:49:37 <Coco_gao> my consideration is, if there are no urgent, I should work with Sundar first to get the spec will recognized by others.
14:49:49 <Coco_gao> well
14:50:07 <efried> ++ to creating fixtures for functional tox-based tests.
14:50:45 <shaohe_feng_> but 3 part CI need more resource/effort to maintain
14:51:10 <efried> yup
14:52:09 <Coco_gao> Li_Liu, What should I do next?
14:52:34 <Li_Liu> 3rd party CI should be maintained by device vendors IMO, openstack CI should just keep the fake devices
14:53:19 <Li_Liu> Coco_gao, if you guys think the time is too tight until Aug 20. just slide the feature into S.
14:53:35 <Li_Liu> so that you have the time to refine your documentation and code
14:57:28 <Li_Liu> ok, it's time to wrap up
14:57:48 <Li_Liu> #todo merge xinran's patch
14:58:35 <Li_Liu> #todo Coco_gao and wangzhh keep working on their patch, but does not have to make into R
14:59:05 <Coco_gao> Bye
14:59:39 <Li_Liu> #todo OPAE PFGA patch needs to be refined for CI and fake the devices if necessary
15:00:48 <Li_Liu> Sundar, are you going to have a new spec to replace 561849 ?
15:03:50 <Li_Liu> I will end the meeting for now
15:03:55 <Li_Liu> #endmeeting