15:03:38 <zhipeng_> #startmeeting openstack-cyborg
15:03:38 <openstack> Meeting started Wed Apr  5 15:03:38 2017 UTC and is due to finish in 60 minutes.  The chair is zhipeng_. Information about MeetBot at http://wiki.debian.org/MeetBot.
15:03:39 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
15:03:41 <openstack> The meeting name has been set to 'openstack_cyborg'
15:03:55 <zhipeng_> #info zhipeng crushil
15:04:02 <zhipeng_> who else do we have at the meeting ?
15:10:34 <jkilpatr> o/
15:10:36 <jkilpatr> sorry I'm late
15:12:08 <zhipeng_> no problem,just saw your patch
15:12:16 <zhipeng_> #topic BP discussion
15:12:54 <zhipeng_> I must apologize that due to the cloudnativecon week, I have not been attentive to the api BP
15:13:09 <zhipeng_> will get to it starting tmr I hope
15:13:29 <zhipeng_> crushil jkilpatr could you guys walk through the updates ?
15:13:44 <crushil> Sure
15:14:08 <jkilpatr> crushil, you first?
15:14:15 <crushil> yup
15:14:28 <crushil> Updated the bp for the generic driver implementation and trying to incorporate all the changes that were requested
15:15:08 <crushil> Did some reviews on other BPs as well and waiting for the BP owners to come back with a response
15:15:53 <zhipeng_> crushil cool man, any outstanding issues at the moment ?
15:15:58 <crushil> The generic driver implementation bp is at https://review.openstack.org/#/c/447257/
15:16:19 <crushil> zhipeng_, Nope, just need more review comments atm
15:16:34 <zhipeng_> great :)
15:18:02 <zhipeng_> jkilpatr how about your updates ?
15:18:34 <jkilpatr> So I still need to address a review from a Friday. It's not going to be the conductor's responsibility to attach accelerators, so I need to remove it from that part of the loop
15:18:51 <jkilpatr> then I think I'll be taking the suggestion to move setup responsiblities out of the conductor.
15:19:15 <jkilpatr> for the time being I guess we can maintain a directory of setup playbooks and document how to use them.
15:19:50 <jkilpatr> I also need to actually write the caching stuff we talked about in the last meeting. Might try and get through that today
15:19:52 <jkilpatr> that's all.
15:20:19 <zhipeng_> i don't remember we had discussion on moving out setup functionalities
15:20:34 <zhipeng_> is it from the review ?
15:20:48 <jkilpatr> zhipeng_, it's in the review comments.
15:21:15 <zhipeng_> my network is dreadfully slow now ... so comments from _gryf ?
15:21:33 <jkilpatr> Roman doesn't like the idea of having a program run off and install it's own dependencies. Which I can understand.
15:21:42 <zhipeng_> i thought we had a understanding that agent could do the ansible stuff
15:23:17 <zhipeng_> okey i will dig into Roman's comment later
15:24:17 <jkilpatr> There are arguments from both sides, on one hand ease of setup, on the other the potential conseqences of cyborg running wild with root.
15:24:48 <jkilpatr> in the end we're planning on making the same playbooks and just having a human press the button instead of the agent.
15:24:52 <zhipeng_> so if we do the setup as a out-of-band operation
15:25:07 <zhipeng_> okey
15:26:29 <zhipeng_> but I do have some scenarios that maybe still the agent should do the setup, despite of the security concerns
15:27:24 <zhipeng_> for example for some dataplane virtualization techniques, you could compose a set up virtual functions into one bigger function
15:27:47 <zhipeng_> a case wouldbe to compose a virtual FW on a intelligent NIC card
15:28:25 <zhipeng_> using different dedicated small virtual networking functions which will do q-in-q, switching respectively
15:29:02 <zhipeng_> jkilpatrt I think the overall question is that , could we have a compromise between the two sides ?
15:31:01 <jkilpatr> zhipeng_, I think we could find some sort of compromise I'm just not sure where to draw the line
15:32:00 <zhipeng_> I was thinking the line could be where the root privilege is required
15:32:24 <zhipeng_> if the root is required on the host side, then maybe it is good, as Roman suggested, to leave to the humans
15:32:39 <jkilpatr> that sounds reasonable enough.
15:32:40 <zhipeng_> but if it is root privilege on the smart device
15:33:06 <zhipeng_> then we could, at the moment, assume that the conductor could perform the setup with root privilege
15:33:31 <zhipeng_> it has security concerns as well, no doubt, but i think we could deal with it later on
15:34:30 <jkilpatr> you're talking about permissions levels on the accelerator itself? What is the "smart device" ?
15:35:51 <zhipeng_> yes i was. smart device like intelligent NICs I referred to
15:36:13 <zhipeng_> there are FPGA enabled devices that you could program the "micro" virtual functions
15:36:27 <zhipeng_> deploy them in different ways to compose different functionalities
15:36:49 <zhipeng_> it is not widely used yet, but I think it would be a major use case later on
15:37:08 <jkilpatr> I know FPGA's can be deployed like that but I never thought about user levels on them.
15:37:08 <zhipeng_> for example in edge computing, where you could dynamically compose your network dataplane
15:37:46 <zhipeng_> that was just my thinking
15:38:09 <jkilpatr> it makes sense, just never considered it that way.
15:38:56 <zhipeng_> if we could have cyoborg agent setup an ansible playbook on a rasberry pie
15:39:03 <zhipeng_> that would be supper awesome :P
15:39:22 <jkilpatr> but why?
15:39:49 <zhipeng_> (just thinking out loud)
15:40:12 <zhipeng_> never mind the pie example
15:40:34 <zhipeng_> so I think I will also comment on the patch, to see what's _gryf's take on this
15:41:17 <zhipeng_> if in the end , the dynamic composition FPGA use case is still too far fetch, then we could just have the regular playbook procedure
15:43:04 <jkilpatr> I think that's fine for now. Maybe add some notifications to the operator that things need to be setup.
15:43:14 <jkilpatr> I think we could detect new accelerators without root
15:43:39 <zhipeng_> yes that is ture
15:43:43 <zhipeng_> true ...
15:43:49 <zhipeng_> fat fingers
15:44:00 <zhipeng_> #topic AoB
15:44:09 <zhipeng_> okey any other buisness or topics
15:46:13 <crushil> Not from me
15:48:40 <jkilpatr> not from me either
15:48:45 <zhipeng_> okey thank you guys for the great work on the bp patches, let's conclude the meeting earlier today :)
15:48:53 <zhipeng_> see you guys on gerrit :P
15:48:58 <zhipeng_> #endmeeting