13:00:44 <baoli> #startmeeting PCI Passthrough
13:00:45 <openstack> Meeting started Tue Oct 21 13:00:44 2014 UTC and is due to finish in 60 minutes.  The chair is baoli. Information about MeetBot at http://wiki.debian.org/MeetBot.
13:00:46 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
13:00:48 <openstack> The meeting name has been set to 'pci_passthrough'
13:00:59 <baoli> Hi there
13:01:05 <irenab> hi
13:01:55 <beagles> hi
13:01:58 <rpothier> hi
13:01:58 <sadasu> Hello!
13:02:14 <irenab> beagles: Hi, long time...
13:02:38 <beagles> irenab, I've been lurking lots, but seem to miss the first 10 minutes or so ;)
13:02:53 <baoli> beagles: hi, nice to have you back
13:02:55 <heyongli> hi
13:03:10 <heyongli> i'm back
13:03:12 <baoli> heyongli, hi, hope you had a nice vacation
13:03:19 <heyongli> yeah
13:03:34 <baoli> heyongli, how about traffic over there during the holidays
13:03:57 <heyongli> en, ... bad
13:04:11 <sadasu> heyongli, hope you had a good break and celebration
13:04:14 <heyongli> some guy alway on the way
13:04:55 <heyongli> s/way/road/
13:04:59 <irenab> good to see all are back
13:05:21 <heyongli> i have reading the meeting log, notice there are several patches need review
13:05:38 <heyongli> and i will take a look soon
13:06:02 <irenab> yes, there is one for macvtap vlan assignment, currently macvtap is broken
13:06:23 <baoli> https://review.openstack.org/#/c/126357/
13:07:47 <baoli> Folks, I normally don't have a specific agenda. Do you guys prefer to have a general agenda for each meeting, and last open discussion?
13:08:03 <irenab> baoli
13:08:06 <heyongli> i see there are plug/unplug, does this mean it works now?
13:08:10 <irenab> sounds like good idea
13:09:04 <baoli> So any suggestions?
13:09:15 <irenab> heyongli: without patch, macvtap is not working
13:09:47 <heyongli> irenab, what the plug/unplug mean there?
13:09:58 <irenab> I think we should focus on specs to push for kilo and bugs reported for SR-IOV
13:10:00 <baoli> heyongli, it's not for the hot plug, it's the vif plug/unplug
13:10:15 <heyongli> got it, thank boli
13:10:17 <heyongli> baoli
13:10:54 <baoli> irenab, sounds good. So we go over the bugs, review, and then blueprints, followed by issues/open discussion
13:11:11 <irenab> baoli: +1
13:11:21 <heyongli> for the resize bug, i been drag to other thing weeks. i i investigating it now, and wil continue work on that
13:12:01 <baoli> heyongli, sounds good.
13:12:29 <heyongli> for hotplug/unplug, i had a rough search, and
13:12:36 <heyongli> the interface hotplug is there
13:12:57 <heyongli> problem is buggy and lack of pci and sriov supporting
13:13:25 <heyongli> for this feature nova interface attach/detach is a good start point to work on
13:13:37 <baoli> heyongli, that's right.
13:14:52 <heyongli> the pci and sriov plug/unplug is diffrent
13:15:22 <heyongli> this interface meet the sriov request but not for pci
13:15:22 <irenab> heyongli: generic pci and netwroking sriov?
13:15:27 <heyongli> yeah
13:15:38 <baoli> there may be common area for resize, hot plug and live migration, especially in the area of dealing with old/new PCI requests
13:15:54 <baoli> or additional requests
13:16:07 <heyongli> sound like this situation
13:16:24 <heyongli> hotplug is somehow stand alone but resize, migration share most of code
13:16:44 <irenab> for my understanding interface-attach is only for networking case
13:17:21 <heyongli> irenab, maybe, sriov can be make when a interface attached to vm, if the port is not "normal"
13:17:29 <irenab> so it should somehow leverage tyhe same solution as with VM creation (pci_request, resource allocation,...)
13:18:18 <baoli> irenab, that's right. resize will deal with general pci passthrough, but interface attatch/deattach deals with resizing (adding/removing) interfaces.
13:18:24 <heyongli> ireab, yes, eveh now the resize is broken and buggy, the common process is there
13:19:20 <irenab> there are also some worst case scenarios, that no more VFs are available and actually interfac-attach can fail
13:20:03 <baoli> irenab, that's true.
13:20:28 <heyongli> sure, but it should be allow to fail, right? lack of resource is kind of normal 'fail' case , does it problem?
13:20:45 <irenab> heyongli: is this something you plan to deal with for kilo? or mainly focus on generic pci hot-plug?
13:21:16 <irenab> heyongli: I agree
13:21:17 <heyongli> i think the sriov is easy and straitfoward,
13:21:28 <heyongli> adn i will focus on sriov first
13:21:40 <heyongli> and sametime push generic hotplug
13:22:02 <irenab> heyongli: are you planning some spec/blueprint?
13:22:17 <heyongli> yeah, but not draft yet
13:22:20 <irenab> for kilo not every development require spec
13:22:42 <baoli> heyongli, that sounds good approach.
13:22:47 <heyongli> what ? is there detail?
13:23:04 <heyongli> dont' need bp?
13:23:29 <irenab> http://docs.openstack.org/developer/nova/devref/kilo.blueprints.html#project-priorities
13:23:56 <baoli> irenab, heard it's not finalized, but still under discussion
13:24:18 <heyongli> thanks irenab, i will check it very carefully, that's seems new trend
13:24:42 <irenab> heyongli: as baoli said, it  is still evolving
13:25:03 <heyongli> got, irenab. we should keep track that
13:25:37 <baoli> Another review: https://review.openstack.org/#/c/125482/
13:26:23 <heyongli> thanks head up
13:27:17 <irenab> thanks, this just refactoring, right?
13:27:35 <baoli> irenab, yes.
13:27:57 <irenab> I have opened two bugs for documentation, didn't have time to follow up yet: https://bugs.launchpad.net/openstack-manuals/+bugs?search=Search&field.bug_reporter=irenab
13:28:47 <heyongli> irenab, do you have contributor?
13:29:33 <irenab> heyongli: what do you mean?
13:29:52 <heyongli> any one will work on the doc?
13:30:53 <irenab> heyongli: I assume someone of us who has few available cycles should take it
13:31:36 <heyongli> i'd like talk to our planner to check if there anyone will warm up with this, it's a good start point
13:32:05 <irenab> heyongli: This is great
13:32:53 <heyongli> any topic today?
13:33:04 <heyongli> i'd like a short meeting -:)
13:33:19 <irenab> baoli: do you want to discuss live migration for macvtap?
13:33:34 <baoli> irenab, yes
13:33:40 <beagles> there was a bug reported https://bugs.launchpad.net/nova/+bug/1383345
13:33:42 <uvirtbot> Launchpad bug 1383345 in nova "PCI-Passthrough : TypeError: pop() takes at most 1 argument (2 given" [Medium,Confirmed]
13:33:48 <beagles> that had to do with whitelist configuration
13:33:53 <heyongli> great i like been update with a idea
13:34:22 <baoli> beagles, dan asked me to take it up
13:34:33 <beagles> baoli, cool
13:34:39 <heyongli> this will not a tough one baoli, -:)
13:35:28 <baoli> heyongli, I haven't looked at closely, it's on my today's to-do list
13:35:34 <irenab> from now on, seems every fix we should backport to stable/Juno
13:36:05 <heyongli> baoli, great
13:36:28 <irenab> maybe work to add links to open bugs on meeting wiki page
13:37:10 <baoli> irenab, right, we haven't updated that link for a while. I will do that before next meeting.
13:37:33 <baoli> Ok, back to live migration. I am working on its nova spec.
13:37:56 <heyongli> any rough idea to update us?
13:38:25 <baoli> We discussed three solutions long back: network XML for each physical net, network XML for each SR-IOV interface, and interface renaming
13:39:30 <baoli> The first one has a major drawback, which is that libvirt allocates device, and doesn't seems to have a facility to tell nova which one is allocated.
13:39:32 <heyongli> will this work for both macvtap and direct?
13:40:03 <baoli> heyongli, libvirt doesn't support live migration for direct sr-iov yet.
13:40:11 <heyongli> so first one is almos not work here
13:40:42 <baoli> heyongli, so we'll initially support live migration on macvtap sr-iov
13:41:00 <heyongli> reasonable
13:41:16 <rook> beagles: there is a new bz i opened yesterday, seems a bit more urgent to fix.
13:41:26 <irenab> heyongli: the first one not alligned in current nova/scheduler solution
13:41:27 <rook> beagles: https://bugs.launchpad.net/nova/+bug/1383465
13:41:28 <uvirtbot> Launchpad bug 1383465 in nova "[pci-passthrough] nova-compute fails to start" [Critical,Confirmed]
13:41:47 <baoli> The second approach would work, but requires the support of netowrk xml in Nova.
13:42:11 <baoli> The third approach looks like the simplest, so I'm leaning towards that
13:42:21 <beagles> rook: ouch
13:43:43 <rook> beagles: yup, i can get around it... but this shouldn't be happening. dansmith is aware of the bz.
13:43:46 <irenab> rook: was it VM with pci function as vnic or as generic pci device?
13:44:09 <rook> passed through a 10G intel nic
13:44:28 <heyongli> seems the pci device is not been update to pci db when it back
13:45:52 <heyongli> discovery pci when rebooting seems broken on first glance
13:46:20 <baoli> rook, with nova-compute disabled, should the VM be destroyed?
13:46:38 <rook> baoli: so, you reboot a compute node...
13:46:49 <rook> or a power outage.
13:47:17 <irenab> baoli: I think VM can run even though nova-compute maybe stopped
13:47:39 <dansmith> yeah, restarting the compute service for any reason *has* to work
13:47:54 <dansmith> for upgrades, for general maintenance, and because sometimes it falls over and needs to be restarted
13:48:25 <baoli> dansmith, that's the area that needs more work.
13:48:32 <baoli> for pci passthrough
13:48:39 <dansmith> well, it needs to be fixed ASAP
13:48:43 <heyongli> need think this carefully, pci discovery stage is little late on such a situation
13:48:47 <dansmith> it's a major problem
13:50:26 <baoli> dansmith: will take a look at it today.
13:50:32 <dansmith> thanks
13:50:39 <rook> thanks baoli
13:51:04 <heyongli> i can continue work on that when baoli offline
13:51:40 <baoli> heyongli, thanks, let's work it out together.
13:52:04 <heyongli> yeah, keep tune, mail me if any clue or progress
13:52:19 <baoli> heyongli, I will comment on the bug once I found something
13:52:31 <heyongli> that's better than mail
13:53:19 <irenab> so back to live migration, are all ok with interface renaming approach?
13:53:41 <heyongli> seems suitable for now, irenab
13:54:27 <irenab> we had previously discussed with baoli some issues with release flow, but seems it can be worked out
13:55:31 <heyongli> does libvirt support network xml be problem?
13:56:02 <baoli> heyongli,  we need to add its support in Nova.
13:56:35 <baoli> I don't think it's being supported now.
13:56:51 <irenab> heyongli: network xml per VF?
13:57:04 <heyongli> i have no idea, add it to nova should not be a problem, if this approch had some thing we want, does it?
13:57:18 <irenab> since fore network xml per physicla network we need somehow integrate with scheduler
13:58:39 <heyongli> i mean the second one, xml per sriov interface
13:59:23 <baoli> ok, let's think about it a little more.
13:59:45 <baoli> We'll continue the discussion next time
13:59:46 <irenab> ok
13:59:56 <heyongli> ok we run of time
14:00:04 <baoli> thanks everyone
14:00:11 <heyongli> thanks, 88
14:00:14 <irenab> baoli: let me know if need help to verify bugs
14:00:24 <irenab> thanks
14:00:29 <baoli> irenab, sure, thanks.
14:00:34 <baoli> #endmeeting