16:04:28 #startmeeting networking_ml2 16:04:28 Meeting started Wed Jul 1 16:04:28 2015 UTC and is due to finish in 60 minutes. The chair is rkukura. Information about MeetBot at http://wiki.debian.org/MeetBot. 16:04:30 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 16:04:32 The meeting name has been set to 'networking_ml2' 16:04:45 #topic Agenda 16:04:55 #link https://wiki.openstack.org/wiki/Meetings/ML2#Meeting_July_1.2C_2015 16:05:23 Would anyone like to add anything to today’s agenda? 16:05:47 #topic Announcements 16:06:03 * rkukura doesn’t have any announcements - does anyone else? 16:07:01 #topic Action Items 16:07:32 I started an eitherpad on the ML2 coding sprint, which we’ll get to in the later in the agenda 16:07:53 banix: Any update on the modular agent proposal? 16:08:18 i sent the email out 16:08:26 the response was overall positive 16:08:49 with Kyle saying its probably late for this cycle but there were support for the work 16:09:18 sean opened an rfe 16:09:23 sean collins 16:09:41 but there hasnt been much discussion there last i checked 16:10:07 banix: Thanks! Is an incremental approach possible? 16:10:43 rkukura: that’s what is being suggested. have to take a fresh look again 16:10:59 banix: did you looked at agent extension manager stuff? 16:11:23 yamamoto: no i haven’t 16:12:00 it seems like a step to similar direction 16:12:21 yamamoto: any link? 16:12:41 #link https://review.openstack.org/#/c/195439/ 16:12:54 and its parent/child patches 16:14:34 silly MacOS just killed all my apps trying to install updates :( 16:14:39 #chairs Sukhdev 16:14:59 rkukura: Opps - 16:15:22 yamamoto: Thanks for the link 16:15:29 Was able to cancel colloquy from shutting down, so I think I’m OK 16:15:57 yamamoto: did a quick browsing - looks interesting 16:16:06 #chairs rkukura 16:16:12 rkukura: back to you 16:16:15 OK 16:16:34 Sukhdev: Just wanted to make sure you could take over if I dropped out 16:16:47 rkukura: sure I will 16:16:53 no worries 16:17:05 Should we continue to track/discuss modular agent on the ML2 meeting agendas going forward? 16:17:48 I think so - banix ? 16:18:17 Sukhdev: sure; i hope to catch up this weekend 16:18:31 OK 16:19:03 I don’t see asomya or shivharis 16:19:21 They have an action regarding physical topology 16:20:04 We’ll get back to that next week 16:20:08 rkukura: in the last meeting asoumya mentioned they will be ready for phisical topology work in ML2 sprint 16:20:18 I wonder if that is still the case 16:20:39 Sukhdev: good question - lets get back to that later in the agenda 16:20:58 #topic Neutron Mid-cycle sprint recap 16:21:21 I tried but couldn’t make this, but I know Sukhdev did 16:21:44 Yes - I attended for two days 16:21:47 Sukhdev: Can you summarize what happened there regarding ML2? 16:22:05 I went there for the ML2 plugin decomposition effort 16:22:35 We discussed about it - and I asked my favorite question - i.e. ROI for the effort 16:22:55 i.e. why do we want to do it and what do gain out of it, etc. 16:23:12 and I also raised concern that we may end up breaking lots of things 16:23:41 folks agreed that there was a possibility of lots of breakages - so, we decided to take two phase approach 16:24:10 Phase 1 - is to keep everything in the tree, but, move everything into separate directories 16:24:31 we decided to call Phase 1 as pre-decomposition phase 16:24:53 we wanted to make sure nothing smokes with code restructure 16:25:23 if all works well with phase 1, then we will decide with phase - 2 - i.e. actually move all the restructured code out 16:25:33 or we may decide not to do phase 2 16:25:42 so, all the effort was on phase 1 16:25:59 Kyle was working on the spec for phase 1 16:26:25 BTW, this means all the L2 agents as well as core plugin 16:26:35 That is all about this 16:26:44 Would phase 1 have any impact on patches under development or in review, other than needing to rebase? It seems phase 2 would be an issue for these, but not phase 1. 16:27:06 rkukura: correct 16:28:40 anything else from the neutron mid-cycle? Were any others there? 16:29:00 there were bunch of people - 16:29:37 see here - https://etherpad.openstack.org/p/neutron-liberty-mid-cycle 16:30:17 Do you see ML2 impact from any of the other efforts? 16:31:54 Sukhdev: How about “get me a network” and “routed networks”? 16:32:00 not really - there are two specs 16:32:27 Sounds like this sprint involved more planning than coding 16:32:29 yes - those two may have some impact - but, I do not believe there will be anything significant - 16:32:50 most likely, they will use Core plugin as opposed to making too many changes to it 16:33:12 Is there still concensus that mid-cycle coding sprints are productive? 16:33:52 I believe they are :-) 16:34:17 Thanks Sukhdev! Anything else on the neutron mid-cycle before we move on to the ML2 mid-cycle? 16:34:39 at the end of second day mestery did a round table about the achievements - everybody seem to believe it was very productve 16:35:13 lets move on 16:35:26 #topic ML2 Mid-cycle Sprint 16:36:01 I started an etherpad for this at https://etherpad.openstack.org/p/Neutron_ML2_Mid-Cycle_Sprint 16:36:32 It looks like July 13 and 14 is possible, or else early September 16:36:56 If we are going to do July, we need to finallize this right away so people can arrange travel 16:37:39 rkukura: july 13 is week after next 16:37:50 I still don’t see asomya, and he was one of the people planning to travel for it 16:38:07 I can do it - but. for travelers it may be a short notice 16:38:47 Week after next is getting kind of tight. I have the OK to do it, but would want to book it today before heading out for a bit of PTO. 16:38:59 I have not seen anything on physical topology 16:39:31 asoumya wanted to add it to the sprint 16:40:10 I lined up a guy at Arista who could help as well - but. we need a firm plan 16:41:05 rkukura: if we did it in sept, it gives everybody more time to regroup 16:41:43 considering this is a short week, most people are on PTO - hence, asoumya and shivharis missing as well :-) 16:41:55 Sukhdev: Agreed, but it also makes it less likely to get significant patches into liberty. 16:42:32 rkukura: correct - we risk pushing it to M-cycle 16:42:48 Maybe we can make significant progress between now and September, and finish up then? 16:43:13 rkukura: I think we can push the RFE now and get the approval for it 16:43:37 manishg: are you here? 16:43:57 Sukhdev: It sounds like your preference would be to do the mid-cycle in September. 16:43:58 Just a note that doign it in September is really late-cycle, and we'll be in feature freeze then. 16:44:16 FYI in case that affects what you're planning on working on 16:44:25 mestery: Agreed 16:44:41 mestery: thanks - good point 16:45:13 rkukura: I am OK with july 13 as well - but, folks are not here to finalize - 16:46:14 sorry got disconnected for a couple of minutes... 16:46:29 manishg: ah ha - was looking for you :-) 16:46:39 manishg: We are debating 7/13 vs September for the sprint 16:47:15 yeah, I saw that discussion. anything is fine with me. 16:47:45 Ignoring the topology effort for the moment, is there anyone who would really want to participate on the synchronization efforts, but can’t be in San Jose 7/13 and 7/14? 16:47:46 manishg: will you be able make arrangements to host on july 13th? 16:47:55 looks like asoumya hasn't responded and he wanted to be part of it (based on your previous comment). right? 16:48:23 Sukhdev: yes I can make the arrangements. if it's not too tight for folks. looks like it might be tight for rkukura ... 16:49:08 kevinbenton: From your mail, looks like you’d attend but 6/13 could be problematic. Are you here? 16:49:19 I mean 7/13 of course 16:51:19 OK, looks like 7/13 and 7/14 are a possibility. Anyone who is interested in participating, please update the etherpad. 16:51:42 We can continue via email among those planning to attend. 16:51:53 rkukura: sounds good. 16:52:02 #topic Discussion on binding_profile_type to facilitate filtering of Ironic ports or using device_owner 16:52:19 Sukhdev: go ahead 16:52:27 thanks 16:52:33 rkukura: hey 16:52:40 rkukura: 7/13? 16:52:48 So, as you know we are working on Ironic/neutron integration 16:53:08 kevinbenton rkukura : I will let you finish that conversation 16:53:13 kevinbenton: Are you in the San Jose area and available 7/13 and 7/14? 16:53:25 kevinbenton: rkukura was asking if you can make it on 7/13, 7/14 for ml2 sprint (at yahoo office in sunnyvale) 16:53:29 rkukura: sorry, i can't make 7/13 because i'm flying to montana for two weeks on 7/10 16:54:08 Lets continue trying to schedule that via email, and let Sukhdev discuss his topic 16:54:15 kevinbenton: would sept. be preferable to you (and would like to push this out => you want to attent)? 16:54:18 rkukura: thanks 16:54:25 rkukura: sounds good, sorry for interupting Sukhdev! 16:54:31 rkukura: ok, let's proceed via email. 16:54:47 so, continuing - there is a requirement from the neutron/ML2 side to be able to filter on Ironic ports 16:55:02 we had considered few possibilities - 16:55:32 using "host" field in the ml2-biniding 16:55:44 we also considered using device_owner: 16:56:10 Using device_owner kind of makes sense to me 16:56:13 currently nova sets it to "compute:" - we considered using it to say "compute:ironic" 16:56:27 Sukhdev: the issue is that binding:profile is a json blob in the db so it can't be filtered? 16:56:31 but, that sound problematic 16:56:50 kevinbenton:correct - hence, we are looking for possible solutions 16:57:19 Sukhdev: ack, what is the issue with using a different device owner? 16:57:19 kevinbenton rkukura : so, one possibility is to create a new field under ml2_port_binding 16:57:54 and call it something like binding_profile_type and set it to "ironic" 16:58:10 kevinbenton: the way Nova uses the zones - there may be a conflict 16:58:15 Would setting device_owner to something like “compute/ironic:” be any less problematic? 16:58:54 nova has no notion of ironic or virt stuff 16:58:59 it is all compute to nova 16:59:26 there is absolutely no distinction in mova - it is all compute (BM or VM) 16:59:43 Sukhdev: I wouldn’t rule out the binding_profile_type idea, but I kind of think of the binding:profile as being open-ended, able to hold different kinds of info, so a single type might be constraining 17:00:40 rkukura: the idea is to use binding-profile for the blob of info, but, have a separate field to say that this binding-profile is of type Ironic 17:00:42 we could also manually filter in ML2 as a temporary workaround 17:01:16 Sukhdev: There is also a binding:vnic_type attirubute that might make some sense to flag ironic 17:01:20 kevinbenton: yes, but, it will be better to have cleaner solution if possibel 17:01:45 rkukura: I looked at that field - it is used by vmware only 17:02:12 In a sense, ironic is a special VNIC type in that its not a virtual NIC, but a physical NIC 17:02:21 rkukura: but, was not sure who else might be using - so, thought if we added a new field, there will be no conflicts - 17:02:39 The ML2 plugin implements binding:vnic_type and triggers rebinding if its changed. I think SR-IOV uses it. 17:02:45 rkukura: so, you think that will be a good idea? 17:02:54 We are overtime 17:03:20 rkukura kevinbenton : can you switch to the neutron channel to finish this discussion 17:03:40 Sukhdev: I think you should consider binding:vnic_type for this. But we need to finish up for the next meeting. 17:03:55 Anything else before we wrap up? 17:04:04 rkukura: yes, lets move over to neutron channel, please 17:04:15 free up this channel 17:04:15 Thanks everyone! 17:04:17 #endmeeting