13:01:24 <baoli> #startmeeting PCI Passthrough 13:01:24 <openstack> Meeting started Tue Dec 2 13:01:24 2014 UTC and is due to finish in 60 minutes. The chair is baoli. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:01:25 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:01:27 <openstack> The meeting name has been set to 'pci_passthrough' 13:01:31 <baoli> Hi there 13:01:34 <yongilhe> hi 13:01:55 <pczesno> hi 13:02:35 <riwinters> hi 13:03:28 <baoli> Irena can't attend today. 13:05:33 <baoli> https://wiki.openstack.org/wiki/Meetings/Passthrough#Agenda_on_Dec._2nd.2C_2014 13:05:40 <baoli> Let's get started. 13:06:50 <baoli> Looks like that we have a new neutron bug 13:07:58 <baoli> https://bugs.launchpad.net/neutron/+bug/1397675 13:08:00 <uvirtbot> Launchpad bug 1397675 in neutron "Updating admin_state_up for port with vnic_type doesn't have affect" [Undecided,New] 13:08:59 <baoli> itzikb: do you have 'agent_required' flag turned on? I guess that you are using mlnx which supports update of port's admin state? 13:09:24 <itzikb> baoli: I'm not using mlnx 13:09:46 <itzikb> baoli: So it's without the agent 13:10:11 <sadasu> itzikb: are you using an agent that can take action on admin_state_up False? 13:10:57 <itzikb> sadasu: No.I'm using Intel 1G NIC and there is no agent involved 13:10:58 <sadasu> itzikb: i see your answer. Without an agent, I don't think this is expected to work 13:12:53 <itzikb> It seems to me there should be a way to be able to set the admin_state_up if we can't do it we should at least document it and somehow write an error when there is no support for this option 13:13:55 <baoli> itzikb, that sounds reasonable. 13:14:26 <sadasu> itzikb: yes, we are lacking in documentation 13:14:33 <baoli> Ok, anything else for bugs. 13:15:26 <sadasu> last week a discussion surrounding "bonding" of 2 sr-iov ports was brought up...would that be a bug? 13:15:38 <sadasu> I guess that would be more like a new BP category 13:15:46 <baoli> sadasu, right 13:16:09 <baoli> #topic reviews 13:16:09 <sadasu> just wanted to highlight that we have this week to propose a new BP, if needed 13:16:43 <baoli> sadasu, yes, thanks for the reminder 13:18:21 <baoli> yonglihe, I put some comments in your PCI resize wip patch 13:18:53 <yongilhe> thanks, i had no time for it now, i will check it soon 13:18:57 <baoli> https://review.openstack.org/#/c/137530/ 13:19:09 <baoli> yonglihe, sure 13:21:25 <yongilhe> baoli, good suggestion 13:22:03 <baoli> yongilhe, thanks 13:22:25 <yongilhe> for 'devices' like resouce, resource tracker does not support that in nature, there some work to do. 13:25:50 <sadasu> I have a question on this review: what is the significance of sign in manager.py? 13:26:18 <yongilhe> it means 'alloce' reserver something. 13:26:27 <baoli> sadasu, check resource_tracker.py on how 'sign' is used over there. 13:26:56 <yongilhe> the negtive sign indicator a claim aborting or direct release the resources. 13:27:23 <baoli> it makes sense if the resource can be arithmetically calculated. 13:27:25 <sadasu> what other values can it have? 13:28:14 <sadasu> +1, 0. -1? 13:28:20 <baoli> 'sign' means + or - 13:28:22 <yongilhe> no other value, just searching the caller you will find it : -1 or default to 1 13:28:26 <sadasu> s/./, 13:28:54 <sadasu> +1 to allocate and -1 to release? 13:29:14 <yongilhe> yeah, it is. 13:29:36 <yongilhe> +1 some time means 'reserved' to pci devices. 13:29:39 <sadasu> thansk 13:30:06 <yongilhe> that why pci need know the stage of resizing. 13:31:23 <baoli> I'll try to push for approvals for several reviews that are in good shape this week. 13:31:40 <baoli> Guess that we can move on to the next topic: 13:31:50 <baoli> #topic Blueprints. 13:31:52 <yongilhe> baoli, thanks, sound cool 13:32:31 <baoli> like Sandhya has mentioned a moment ago, please submit your proposal if any by the end of this week, I guess. 13:32:50 <baoli> deadline is monday 13:32:53 <baoli> end of monday? 13:33:33 <sadasu> baoli: correct 13:35:58 <yongilhe> i'm stuck on other thing, but someone will cover me for the 'interface attaching' bp, hope we can meet the deadline. 13:36:39 <baoli> yongilhe: sounds good 13:36:48 <yongilhe> baoli, how about your migration bp? any progress? 13:37:33 <baoli> yongilhe, some blocking issue came up. Libvirt or the hypervisor doesn't seem to be happy with interface renaming. 13:38:00 <baoli> I've send a query to libvirt-users 13:40:17 <baoli> Hopefully, I can get some resolution soon. 13:41:24 <baoli> network xml doesn't help either. Internally, libvirt converts that to interface, and during live migration, the same interface is needed on the destination host. 13:42:41 <baoli> If anyone has any idea, please let me know. 13:43:23 <yongilhe> baoli, sure, maybe i can consult our hypervisor team to see is there anything can help 13:44:08 <sadasu> baoli: no ideas for a solution...but by "same interface" you mean pci device with same address? 13:44:11 <baoli> yongilhe, that sounds great. Let me forward the libvirt-users query to you. 13:44:35 <baoli> sadasu: same interface name like eth6 13:44:46 <itzikb> baoli: Please send me too 13:45:31 <baoli> irzikb, sure. 13:45:43 <itzikb> baoli: Just a quick question - the problem is with the first migration 13:45:54 <itzikb> baoli: Or back to the original host 13:45:57 <sadasu> ah! so this will probably work only in the use case where we are migrating to a new/clean new server 13:46:37 <baoli> itzikb: the vm couldn't even be booted up. 13:46:44 <sadasu> itzikb: I suppose either direction as long as the same interface is unavailable 13:47:19 <itzikb> baoli: Let's discuss this after the meeting. I think it worked for mlnx_vif 13:48:12 <baoli> itzikb, cool. I thought Irenab has done that. 13:48:33 <baoli> I need to get another intel card and try it on. 13:49:02 <baoli> I don't have any mlnx cards for sure. 13:49:37 <itzikb> baoli: Should work with Intel.. 13:50:36 <baoli> itzikb, cool. will give it a try. 13:50:37 <sadasu> is someone bringing up the bonded interfaces case? we were supposed to follow-up on the ML but I didn't see anything 13:51:01 <baoli> sadasu, beagles brought it up. 13:51:04 <baoli> last time 13:51:17 <baoli> pczesno: trying to review your spec. Will add some comments and questions soon. 13:51:33 <pczesno> baoli: great thanks 13:51:36 <sadasu> yes, thats right. 13:52:31 <sadasu> just want to make sure we spend some time to atleast define the problem 13:53:04 <baoli> well, beagles doesn't seem to be present. So let's wait to see his proposal. 13:53:28 <baoli> Ok, move on 13:53:41 <baoli> #topic CI Testing 13:53:48 <baoli> anything on this? 13:54:12 <yongilhe> i'm try to make our CI online asap. 13:55:40 <baoli> yongilhe, that's great! 13:56:04 <yongilhe> any hw CI there? 13:57:00 <itzikb> yongilhe: What tests are your running? 13:57:22 <sadasu> I have questions on how to convert my existing setup to do Nova CI 13:57:39 <sadasu> can we do both Nova and Neutron CI on the same TB? 13:58:02 <sadasu> I am investigating in the background...will hopefully have an update in the next meeting 13:58:03 <yongilhe> itzikb, to online, first stage only had baisc pci passthrough testing. and will add more and more gradully. 13:58:53 <yongilhe> sadasu, we trying to cover basic sriov after online. 13:59:03 <itzikb> yongilhe: You mean local testing or tests that are upstream? 13:59:59 <yongilhe> itzikb, i don't know yet, but upstream a HW specific test seems not very valuable. but if it does, we happy to upstream all of them. 14:01:24 <baoli> Time is running out. So let's continue next time 14:01:38 <baoli> thanks everyone 14:01:42 <yongilhe> sure, then, bye 14:01:46 <baoli> #endmeeting