09:30:55 <BobBall> #startmeeting XenAPI
09:30:56 <openstack> Meeting started Wed Nov 25 09:30:55 2015 UTC and is due to finish in 60 minutes.  The chair is BobBall. Information about MeetBot at http://wiki.debian.org/MeetBot.
09:30:57 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
09:31:00 <openstack> The meeting name has been set to 'xenapi'
09:31:03 <BobBall> huanxie jianghua ping
09:31:08 <jianghua> hello verybody.
09:31:09 <huanxie> ping all
09:31:20 <BobBall> also I don't see John Hua here? unless I'm missing him :)
09:31:25 <BobBall> Could you poke him please
09:31:58 <huanxie> john will join soon, he is on call
09:32:27 <BobBall> Hi huazhihao
09:32:34 <huazhihao> hi bob
09:32:35 <BobBall> OK - so - let's talk about bugs first
09:32:37 <BobBall> #topic bugs
09:32:48 <BobBall> huanxie and jianghua have uploaded a couple of Neutron bug fixes (in Nova)
09:32:53 <BobBall> Could you walk through these?
09:33:18 <jianghua> yes.
09:33:26 <jianghua> for the ovs port configuration issue:
09:34:10 <jianghua> we have a solution by adding an interim bridge between the vm and the integration bridge.
09:34:32 <jianghua> please refer to this review:
09:34:36 <jianghua> link: https://review.openstack.org/#/c/242846/
09:35:10 <jianghua> it's finished the sub team review and is ready for core team review.
09:35:14 <BobBall> So this change was particularly for HVM guests
09:35:27 <BobBall> where we have two ports - VIF and TAP - and Neutron was sometimes picking the wrong one
09:35:43 <jianghua> yes.
09:36:08 <jianghua> it depend on if PV driver exists.
09:36:31 <BobBall> Some reviews on that would be great johnthetubaguy - particularly because it's a conceptual change
09:36:45 <BobBall> and huan - could you explain your change?
09:36:48 <johnthetubaguy> yeah, suck in spec land right now I am afraid
09:36:51 <jianghua> with our solution, it will only expose one fixed port to neutron.
09:37:14 <BobBall> Could you give a very quick high level check to see if you're happy for it to go through as a bugfix rather than a spec?
09:37:19 <huanxie> Mine is to fix nova and neutron race condition when booting a new instance, the detailed is to let nova wait for neutron notification event when port from down to active
09:37:29 <huanxie> https://review.openstack.org/#/c/241127/
09:38:25 <huanxie> I have tested this fix in my env and it's waiting for review now
09:38:27 <BobBall> huanxie's is 100% a bugfix - porting the libvirt event changes to XenAPI
09:39:04 <johnthetubaguy> BobBall: it does feel a bit like this is adding neutron support, and feels a bit like a blueprint
09:39:12 <BobBall> Neutron support is already there
09:39:20 <BobBall> It works fine but with a couple of races
09:39:24 <johnthetubaguy> so this is just fixing HVM guests?
09:39:41 <BobBall> In some cases.  Most of the time HVM guests work fine because they use VIF rather than TAP
09:39:44 <johnthetubaguy> BobBall: do we have a blueprint up to add the events support like libvirt has, or are we not doing that?
09:40:05 <BobBall> The issue is in the case is that clearly if you don't have the PV drivers then it doesn't work
09:40:15 <johnthetubaguy> BobBall: OK
09:40:33 <BobBall> Regarding the events support - I think we talked previously and agreed that it was a bugfix because it was porting the libvirt change (actually I think that was through a bugfix too)
09:41:35 <huanxie> currently we treat the race condition as a bug
09:42:10 <BobBall> and it is actually a bug fix since the use of events is just fixing a race in nova
09:42:22 <BobBall> events support in neutron was added with a BP I think, to support fixing this race in Nova.
09:43:06 <BobBall> Do you agree that both of these are bug fixes then?
09:43:39 <johnthetubaguy> unsure, they could be spec-less blueprints, but I could by either way really
09:44:26 <BobBall> To understand, what's the benefit of having a BP without a spec here?  I thought the BPs were useful to collate multiple changes - but in both cases here they are single fixes?
09:45:56 <johnthetubaguy> BobBall: its more advertising a new feature
09:46:05 <johnthetubaguy> BobBall: like I say its all boarderline
09:46:14 <johnthetubaguy> BobBall: many blueprints are single patches
09:46:22 <BobBall> OK - so if we create a BP saying "Fix XenAPI neutron races" or similar then that would be useful to have?
09:46:27 <johnthetubaguy> BobBall: doesn't matter
09:46:45 <johnthetubaguy> I didn't finish that sentence, oops, I forget what it was
09:46:54 <BobBall> haha
09:46:57 <BobBall> :)
09:47:08 <johnthetubaguy> Sorry, we are stuck in a hole, so lets dig out
09:47:23 <johnthetubaguy> bug fix seems fine to me, just highlighting it feels a little boarderline
09:47:51 <BobBall> *nod* understood.  OK - we'll progress as we are for now then
09:47:55 <BobBall> thanks.
09:48:15 <BobBall> #topic BluePrints
09:48:42 <BobBall> Just to be clear on this topic, the VGPU spec has been -1'ed by jaypipes for the resource tracking, so we'll be abandoning the spec
09:49:19 <BobBall> There may be an alternative that we can  investigate (e.g. faking up the VGPU devices as PCI devices).  Quite hacky but restricted to the driver.
09:49:24 <BobBall> We can talk more about that another time
09:49:28 <BobBall> #topic AOB
09:49:42 <BobBall> Last - but by no means least - huazhihao - can you give a MOS update?
09:50:00 <huazhihao> Sure
09:50:26 <huazhihao> We have finally reached a release for MOS 7.0
09:50:40 <huazhihao> But still with nova-network
09:51:30 <huazhihao> There is only one small issue with file injection
09:51:47 <BobBall> By 'reached a release', John means that we've sent a version of the plugin to Mirantis for certification
09:52:01 <huazhihao> Sorry, yes
09:52:09 <BobBall> This plugin has been verified internally so can be provided to people wanting to do early validation of the 7.0 integration
09:52:12 <huazhihao> That is what I want to mean.
09:52:48 <huazhihao> Maybe that is all for MOS 7.0
09:53:23 <BobBall> Awesome
09:53:26 <BobBall> Any other business?
09:53:44 <huazhihao> Nope
09:54:04 <johnthetubaguy> sounds like good progress
09:54:16 <johnthetubaguy> about the VGPU, I should look at jay's worries on there
09:54:30 <johnthetubaguy> but resource tracking is in massive flux right now
09:54:59 <johnthetubaguy> really happy to see XenAPI into a "distro" sort of solution
09:56:04 <BobBall> Would be appreciated john - but to us it seems quite a strong veto of the VGPU spec
09:56:16 <BobBall> If you have alternatives to suggest we would be very open to them
09:58:10 * johnthetubaguy nod
10:00:28 <johnthetubaguy> BobBall: why can't you track vGPU as PCI devices?
10:05:52 <johnthetubaguy> BobBall: did you mean to end the meeting?
10:42:57 <BobBall> #endmeeting