13:00:25 <baoli> #startmeeting PCI Passthrough 13:00:26 <openstack> Meeting started Thu Feb 13 13:00:25 2014 UTC and is due to finish in 60 minutes. The chair is baoli. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:00:27 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:00:29 <openstack> The meeting name has been set to 'pci_passthrough' 13:00:36 <rkukura> hi 13:00:39 <baoli> Hi 13:00:47 <BrianB_> hi 13:01:19 <irenab> hi 13:02:04 <baoli> Let's wait a couple of minutes for Yongli 13:02:07 <b3nt_pin> hi 13:02:23 <beagles> (sorry my nick was flipped) 13:02:30 <baoli> Hi beagles 13:02:35 <irenab> hi 13:03:48 <irenab> short update on nova bug I raised before, Itzik pushed the fix https://review.openstack.org/#/c/59093/ 13:04:11 <irenab> baoli: there is a code that gets provider network physical_netowrk attributes, take a look 13:04:24 <baoli> irenab, sure 13:05:19 <irenab> rkukura: any objections on vnic_type patch? 13:05:44 <rkukura> irenab: I never got back to my review of that yesterday, but I will today 13:06:42 <irenab> I'll probably push slight fixes to mainly fix alembic migration failure 13:09:11 <baoli> I think that we should get started. 13:10:02 <baoli> I guess that everyone has seen the emails on the Nova BPs pending for approval 13:11:35 <sadasu> yes 13:11:44 <baoli> rkukura, do you see any issues for the neutron sriov related BPs being approved? 13:11:58 <irenab> baoli: yes, John sent the email 13:12:29 <rkukura> We are still waiting for markmclain to approve the vnic_type BP, right? 13:12:41 <heyongli> baoli, sorry my network can not reach this meeting for a while. 13:12:55 <baoli> Hi heyongli 13:13:05 <irenab> rkukura: I think mestery approved and need salvatore to look at this too. 13:14:05 <rkukura> irenab: Its all set 13:14:33 <irenab> rkukura: Cool, so just need you time for review :-) 13:15:29 <rkukura> Are there separate BPs for each SR-IOV MD? 13:15:51 <rkukura> I see https://blueprints.launchpad.net/neutron/+spec/ml2-sriov-nic-switch 13:16:24 <irenab> rkukura: yes, I posted it 13:17:11 <irenab> rkukura: will start to bring it once vnic_type in good shape 13:17:23 <rkukura> Is there still work to do for https://blueprints.launchpad.net/neutron/+spec/pci-passthrough-sriov? 13:18:21 <baoli> rkukura, good question. I opened it for neutron related sriov works 13:18:45 <sadasu> I think this work is covered in Irena's BP 13:19:03 <irenab> rkukura: it comes to common parts for every SRIOV plugin 13:19:33 <sadasu> but we might keep it around, to support Neutron changes when the full blown Nova changes come in 13:19:36 <irenab> I think sadasu and I need to see if/how common parts can be extracted in order to eliminate code duplication 13:19:50 <sadasu> irenab: agreed 13:20:20 <irenab> still need to see if there is SRIOVMDBase or some DBMixin 13:20:33 <baoli> So seems that we are in good shape from neutron side 13:20:36 <sadasu> yeah, I looked at it a little bit 13:20:58 <baoli> Now that yongli has joined, let's talk about nova side of things 13:21:07 <irenab> sadsu: I think you are ahead of me here, so any inputs you have, please share 13:21:07 <rkukura> We'll need to get at least the 1st of those BPs approved, and some code in review by Tuesday to make icehouse 13:21:41 <irenab> rkukura: How we can do it without binding:profile? 13:21:42 <baoli> rkukura, which BP specifically? 13:22:49 <rkukura> The binding-profile BP is approved,and will be in review today I expect 13:23:34 <baoli> rkukura, that sounds great 13:23:49 <rkukura> Whichever BP(s) actually define the keys used in binding:vif_details and binding:profile, and implement the MD(s) 13:24:26 <baoli> sadasu, what about your BP? is it approved? 13:24:52 <sadasu> baoli: yes, its approved ....putting up code shortly for review 13:25:00 <irenab> rkukura: I'll update the above BP to mention the attributes baoli summarized on wiki 13:25:13 <rkukura> sadasu: Link to your BP? 13:25:15 <baoli> sadasu, cool 13:26:22 <irenab> baoli: switch to nova? 13:26:37 <baoli> irenab, is there a BP for your MD? You said that you are going to bring it up once the vnic bp is in good shape? 13:27:02 <sadasu> rkukura: https://blueprints.launchpad.net/neutron/+spec/ml2-ucs-manager-mechanism-driver 13:27:09 <irenab> baoli: yes, https://blueprints.launchpad.net/neutron/+spec/ml2-sriov-nic-switch is mine. 13:27:15 <rkukura> sadasu:: found https://blueprints.launchpad.net/neutron/+spec/ml2-ucs-manager-mechanism-driver, and that it is approved 13:27:24 <baoli> I'll put all of them in the meeting wiki 13:27:34 <baoli> cool. let's switch to nova 13:27:46 <irenab> Still need the old Mellanox Plugin as MD due to InfiniBand support, so had to push it with no regards to surrent design 13:28:12 <irenab> s/surrent/current 13:29:47 <baoli> With regard to the nova BPs, we need to decide what to do if they won't be approved for icehouse. Any comments? 13:31:34 <irenab> baoli: not sure what we can do 13:31:43 <heyongli> it's hard part 13:31:57 <sadasu> rkukura: how has this type of scenario worked in the past? 13:32:51 <heyongli> i have no idea, yunhong talked with john in meetup, there still some concert should be resolved, 13:33:49 <rkukura> We need to find supportive nova core reviewers, and deal with any legitimate objections/concerns 13:34:24 <irenab> maybe if we continue with the code and even its accepted in J-1, it can be backported 13:35:00 <heyongli> code will continue, definitely 13:35:03 <rkukura> Right now, it seems that one nova core has not been supportive, but is he raising valid concerns other than just core reviewing workload? 13:35:43 <baoli> I thought Russell would look at this. But his email seems to indicate that his hands are full 13:35:49 <heyongli> aggregate is the one of problem 13:37:33 <sadasu> heyongli: are u planning to rework your design to include host-aggregates? 13:37:57 <heyongli> at leas i should considerate it carefully 13:38:10 <heyongli> and post some idea about it 13:38:45 <baoli> I put together an agenda for today earlier: https://wiki.openstack.org/wiki/Meetings/Passthrough#Agenda_on_Feb._13th.2C_2014 13:38:54 <sadasu> I think you'll likely end up getting the same comment from other cores 13:38:57 <baoli> It talks about what we need from nova generic support 13:40:31 <heyongli> yunhong also try to split the basic support to another bp, but can not make it 13:41:04 <baoli> I think that we discussed host aggregates extensively, and thought in its current shape it won't support our requeirements 13:41:44 <heyongli> baoli, this the major gap 13:41:49 <heyongli> An API to retrieve a PCI device that is linked to the original PCI request. 13:42:32 <baoli> heyongli, any comments on that? 13:43:08 <heyongli> for this, is it can be done by pci_manager.get_instance_pci_devs and plus a function find the device by specs, what do you thinks about such approach? 13:44:01 <sadasu> baoli: can we put a section in your wiki, just to capture that reasoning? 13:44:26 <baoli> heyongli, what do you mean by specs? 13:44:27 <sadasu> I think it is spread out in emails and IRCs 13:44:38 <heyongli> { key: v, ...} 13:44:57 <baoli> sadasu, about the aggregate? 13:44:58 <heyongli> which wiki, did i miss it? 13:45:13 <sadasu> baoli: yes 13:45:35 <heyongli> our reason is not strong enough for john, i think 13:45:57 <sadasu> baoli: don't want to add more to your plate...but trying to help 13:46:40 <baoli> heyongli, it won't work in this case: if you have similar requests in both the nova flavor and the --nic request 13:47:00 <heyongli> baoli , can you make a example? 13:47:04 <baoli> heyongli, I sent you an email about my proposal, you didn't get back to me 13:47:15 <heyongli> baoli, i might miss it? 13:47:45 <heyongli> i will check it later. ooh, i take one day leave at tomorrow 13:48:19 <baoli> heyongli, it was a few weeks back, and also other emails that talk about it. 13:48:25 <heyongli> baoli, even similer, but you can make the request's all spec to the function 13:49:03 <heyongli> baoli, sorry for this, i did take a long vacation. 13:50:06 <heyongli> if we had get_pci_device_by_specs(pci_devs, specs) 13:50:18 <heyongli> the specs can translate from a request 13:50:28 <baoli> heyongli, taking the existing pci alias for example, you can have the alias specified in the flavor, at the same time, the alias can be used in the --nic option. 13:50:30 <heyongli> get_specs_from_request(request) 13:50:56 <heyongli> only if they are have any diffrent , this can work 13:51:20 <heyongli> if same, no need to get recognized, i think. 13:52:37 <heyongli> only problem might be, if network and regular pci share same alias, we should find a 'free' one in libvirt config layer 13:52:48 <baoli> heyongli, so get_pci_device_by_specs() would return a list of pci devices. Which one to pick? 13:53:13 <baoli> heyongli, let me forward you my email again. It basically uses the cookie approach, and it's very simple 13:54:17 <heyongli> baoli, i try to make a cookie version ever, but seems more like hack one, i drop it. anyway, this is code detail, and we can discuss offline in mail. 13:56:20 <heyongli> baoli, forward that to yunhong also and to mail list, yunhong might quick response when i take my short vacation 13:57:03 <heyongli> any way you list base support of nova is what i think, this cool 13:57:10 <baoli> heyongli, he had it already. You were on the mailing list, I beleive 13:57:11 <irenab> any change regarding daily meetings? 13:57:51 <heyongli> baoli, sorry again, once we find the way, code is very soon 13:58:47 <heyongli> irenab, 2 meeting might enough? 13:58:59 <baoli> heyongli, we need to agree on what is needed, and the way to do it, 13:59:10 <heyongli> sure. 13:59:30 <irenab> heyongli: I think we may try Mon and Wed + updates on mails if needed 14:00:09 <heyongli> irenab, i'm ok with it 14:00:31 <sadasu> 2 times a week is good 14:00:43 <baoli> Ok, let's say the next meeting is Monday. And If needed, we can have one on Wednesday. 14:00:53 <irenab> agree 14:00:56 <heyongli> cool 14:01:04 <rkukura> +1 14:01:30 <rkukura> I'm likely to miss Monday's meeting, but not sure 14:01:51 <baoli> The channel is still open, can we go on for a little bit longer? 14:01:51 <rkukura> If so, I'll read the logs. 14:02:22 <markwash> is the pci-passthrough meeting over? 14:02:56 <baoli> markwash, you need the channel? Sorry, I'll end the meeting now 14:03:30 <baoli> #endmeeting