15:03:13 #startmeeting distributed-virtual-router 15:03:14 Meeting started Wed Jul 23 15:03:13 2014 UTC and is due to finish in 60 minutes. The chair is Swami_. Information about MeetBot at http://wiki.debian.org/MeetBot. 15:03:15 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 15:03:17 The meeting name has been set to 'distributed_virtual_router' 15:03:27 hi dvr folks 15:04:08 #info 15:04:14 hi swami 15:04:35 As you all know the DVR have been moved to Juno 3 milestone. 15:05:05 Even though we are constantly pushing the patches upstream, all our patches may not land during the next day or so. 15:05:32 So we will continue to work on it and push the patches upstream 15:05:58 That would be our high priority. 15:07:15 hello? 15:07:52 Current patches that have merged. 15:07:57 #link https://review.openstack.org/#/q/status:open+project:openstack/neutron+branch:master+topic:bp/neutron-ovs-dvr,n,z 15:08:10 mrsmith: hi 15:08:51 I think the latest addition to the patches that have merged is the L3 Agent 15:09:17 Swami_ I believe you meant https://review.openstack.org/#/q/status:merged+project:openstack/neutron+branch:master+topic:bp/neutron-ovs-dvr,n,z 15:09:47 Rajeev_: thanks, the first list is the non-merged patch but with the exception of L3 agent. 15:10:04 The second link that rajeev provided gives the list of merged patches so far. 15:10:42 L3 Agent merged. Congratulations .. 15:10:56 Swami_: your first link missed the devstack change 15:11:07 Apart from the above patches, there was also a python-neutronclient patch that was merged. 15:11:47 haleyb: thanks for bringing it up, there was another patch for the devstack enablement for DVR 15:12:10 haleyb: can you post that link here. 15:12:22 https://review.openstack.org/103180 15:12:38 #link https://review.openstack.org/#/c/106147/ 15:14:24 There is an updated DVR Howto wiki for people who wanted to test the code. 15:14:28 #link https://wiki.openstack.org/wiki/Neutron/DVR/HowTo 15:15:04 Swami_: hi. My internet connection is really spotty today. Hope I can stay on for a little while. 15:15:18 Folks please feel to test the code and let us know if you have any issues. 15:15:23 (well, I know what I'm doing the rest of the week...) 15:15:43 carl_baldwin: thanks for joining, sure and I understand your problems. 15:16:23 #topic L3 plugin patch for router conversion 15:16:50 Folks there is patch that address the router conversion and I am currently working on that patch and posted it for review. 15:17:24 #link https://review.openstack.org/#/c/105855/ 15:18:35 This patch should help the scheduler identify the transition state of the router, for any router conversion and should be able to schedule the router to the right agent based on the type of the router. 15:18:47 Swami_ - how much testing have you done with agents ? 15:19:06 wrt conversion? 15:19:44 mrsmith: As I mentioned above this patch addresses the changes on the plugin side and not on the scheduler side. 15:19:52 ok 15:20:11 mrsmith: we need to start working on the scheduler side and then test the agent part. 15:21:22 I spoke to murali and he mentioned, right now if the router is already scheduled on an agent and when router update happens, the routers are not rescheduled. 15:21:22 Swami_: I think this patch should address conversion across all components. I don’t think we should split it into multiple patches. 15:22:16 carl_baldwin: sure, then we might have to include the other components such as the scheduler and agent. 15:22:59 mrsmith: does that answer your question 15:23:15 Swami_: Yes, I think you’re right. Hopefully the L3 agent code will merge in the next hours and we can rebase it to the scheduler patch. 15:23:22 Swami_ - yes... sounds like we will all work on this patch 15:23:51 mrsmith: yes that would be plan 15:24:08 sure, sounds like a good plan 15:24:24 so we remove the NN side router first and then intiate conversion all the way from plugin to agents 15:24:28 as part of this patch> 15:24:32 #topic L2 DVR 15:24:49 Swami_: Also, if you’re authoring the bulk of the patch you can either reset the author or add yourself as co-author. 15:25:06 viveknarasimhan: can you provide an update on the L2 sidw 15:25:07 nit: transition will be offline, i.e. there will be a short interruption to traffic 15:25:32 L2 Agent changes for DVR merged 15:26:06 tested GRE with L2 Agent and it works OK when I tried for East-west traffic 15:26:52 With regard to unit testing, am following up on Scheduler side tests and parallely focusing on ml2/db.py and ml2/plugin.py lines 15:27:24 addressing review comments with armax on ML2 Additions for dvr #102398 15:27:36 that is gist of my update 15:28:10 viveknarasimhan: thanks for the update 15:28:40 viveknarasimhan: do you expect any change on your side for the router conversion 15:29:01 carl_baldwin: thanks 15:29:11 i want to understand this conversion better before seeing any change required 15:29:28 we should not be adding DVR while NN router interface ports exist 15:29:35 as that will fail in the current code scenarios 15:29:45 due to overlap of subnets 15:30:54 This should not be any different than how a router can be moved from one L3 agent to another L3 agent in a different node. 15:31:00 viveknarasimhan: would like to know more about it 15:31:09 viveknarasimhan: I’m not sure I understand yet. 15:31:32 today we donot allow a subnet of dvr to be made available on centralized NN router and vice versa 15:31:51 a subnet interface is either on DVR or on NN and not on both 15:32:23 so when we do conversion, we should first remove the NN hosted router's interface subnets and then re-apply them on a DVR router with the same name and properties 15:33:48 viveknarasimhan: yes I agree with the above statement. 15:34:07 viveknarasimhan: That makes sense. 15:34:22 so there is going to be drop in traffic 15:34:41 and that could be from minimal to more depending on how many VMs are 15:34:50 spread out in the cloud for such subnets 15:34:56 viveknarasimhan:agreed, traffic interruption will be there 15:35:02 viveknarasimhan: yes that's what rajeev mentioned and I think that is acceptable. 15:35:24 this will be done in a maintenance window (or) live? 15:36:07 viveknarasimhan: Cloud admin call ? 15:36:14 viveknarasimhan: It depends, since there would be a network glitch, admins may want to do it in a maintenance window or they can choose to just do on a case by case basis. 15:36:22 viveknarasimhan: We need to learn more about the downtime and how significant it will be. If we can characterize that the the deployer can decide. 15:36:37 agree. but we may need to test live case 15:37:10 viveknarasimhan: Sure once we have the conversion working we can test, make sure how much time it takes and then document it. 15:37:19 i don't foresee change in L2 for this 15:37:37 if the process is as I mentioned above, sequential removal of NN ports followed by DVR ports creation 15:37:56 viveknarasimhan: thanks for your update. 15:38:02 things should work without change from L2 side , except for configuration updates 15:38:11 Swami_: ok to priortize legacy --> dvr Over dvr --> legacy -- right ? 15:38:15 like l2-pop and enable_distributed-routing must be ON on compute nodes 15:38:46 Rajeev_ : We are not targeting DVR-to-Legacy at this time... 15:39:08 Swami_: thx, sounds good for now 15:39:15 Swami_: Rajeev_: Agreed 15:39:36 #topic L3 Agent 15:40:00 mrsmith: the stage is yours 15:40:24 looks like the patch is about to land 15:40:35 viveknarasimhan: Do we really need things to be sequenced? I’m wondering if the plugin side does its thing and the L3/L2 agents do their thing on the various nodes will things fall in to place? 15:40:42 we have a few follow up issues/patches to work on 15:40:55 conversion, agent-restart/namespace cleanup 15:41:12 also, we still have a conflict with other code with SNAT 15:41:16 and VLAN tagging 15:41:34 Rajeev_ can you comment on that issue? 15:41:45 carl_baldwin: right now the agent namespace cleanup is a different patch, can we merge this along with the migration patch or should it be a different one. 15:41:48 mrsmith: yes 15:42:07 we noticed the SNAT ports were not getting tagged 15:42:14 Swami_ migration or conversion ? :) 15:42:24 Swami_: What would be the advantage of merging them? 15:42:33 There a fix on the way or maybe already in 15:42:59 armax is integrating it into l3_rpc_base 15:43:02 We need to get away from large un-focused patches that try to do too much. We need to think leaner. They will merge more easily that way. 15:43:06 Rajeev_: already did 15:43:22 thanks armax! 15:43:34 carl_baldwin: Since mrsmith will be also working on the conversion part, he can also fix the namespace cleanup issues, rather than having another patch for it. 15:44:06 carl_baldwin: agreed 15:44:29 i mentioned sequencing in algorithmic point of view 15:45:00 Swami: mrsmith: I can also help with agent restart and namespace cleanup. I’ll be back full time starting tomorrow. 15:45:14 the code can execute concurrently, but we need to ensure racing is avoided, be in dvr scheduling in controller, l2 rule handling by l2 agent and l3 namespace handling by l3 agent 15:45:18 k 15:45:26 carl_baldwin: thanks 15:45:28 carl_baldmin - maybe I can focus on conversion then? 15:45:41 mrsmith: Sounds good. 15:46:02 mrsmith: carl_baldwin: sounds like a good plan 15:46:59 #topic DVR scheduler 15:47:00 i will focus on CLI enhancement to show replicated DVR bindings 15:47:13 would that be ok? 15:47:31 also i can focus on concurrency of flow additions issue with ovs_lib 15:47:44 please feel free to let me know where I could help 15:48:11 all: what about the SNAT add/remove/scheduling ? 15:48:17 viveknarasimhan: sure we will work with you on this, once you finish up the current work, or else you can also help on the scheduler side. 15:49:08 viveknarasimhan: armax: murali: any updates on the scheduler 15:49:44 armax , are you there? 15:49:57 let me give update then: 15:50:08 viveknarasimhan: go ahead 15:50:36 armax refactoring some part of scheduler 15:50:38 Swami_, viveknarasimhan the patch has been updated 15:50:45 and it is current 15:50:51 predominantly it was inheriting both L3 DB and was getting inherited to ML2 Plugin 15:50:58 which puts it into cyclic issues 15:51:12 so we removed scheduler direct dependence on L3 DB and all calls 15:51:12 viveknarasimhan: yes, that’s being taken care of 15:51:16 are throuh l3-plugin 15:51:23 from the dvr scheduler 15:51:31 armax, pls go ahead 15:52:01 viveknarasimhan: I don’t have anything else to add on the l3 sched 15:52:21 we have proceed with scheduler UT, like 30% complete 15:52:27 armax: viveknarasimhan: thanks for the updates 15:52:33 remaining 70% coverage is ongoing 15:52:59 So for this week, let us focus on the "router conversion" patch 15:53:40 Also address any upstream review comments for the current non-merged patches. 15:54:25 #topic Open Discussions 15:54:38 Any other topic that we wanted to discuss here. 15:55:14 Swami_: DVR testing 15:55:26 including Tempest and Gate 15:55:35 Rajeev_: yes 15:55:55 can someone give an update please ? 15:56:05 Once all our patches have landed into the upstream, they will open up the gate testing for the dvr 15:56:26 armax: was working on it. 15:56:47 Swami_: there are a number of patches I pushed to add an experimental check job 15:56:52 armax: do you have any update on the process on what is test it tempest, gate and others. 15:57:04 waiting on patches to merge 15:57:06 hang on 15:57:10 lemme find the links 15:57:19 https://review.openstack.org/#/c/98235/ 15:57:29 this is passing against dvr currently 15:57:31 so that’s great 15:57:45 https://review.openstack.org/#/c/106902/ 15:57:54 this one is the experimental job that tests dvr 15:57:54 armax: thank you 15:58:03 this depends on: 15:58:03 https://review.openstack.org/#/c/106901/ 15:58:19 and: 15:58:20 https://review.openstack.org/#/c/103180/ 15:58:31 so we need people helping review/approve these ones 15:58:43 once we do and DVR is all in trunk 15:58:51 armax: thanks for the update 15:58:55 we can kick a job doing ‘recheck experimental' 15:58:59 on the Gerrit review 15:59:12 and we’ll kick off a DVR-based tempest run 15:59:21 Yes we will take a look at those patches. 15:59:42 i am seeing a problem using that dev devstack patch to get it up and running with mode=dvr_snat, stephen thinks it might just be something missing from that patch as his hand-created devstack install works 15:59:49 armax: What is "recheck experimental" - my ignorance 16:00:06 let’s take this offline 16:00:08 Time check 16:00:11 Swami_: if that’s ok 16:00:21 Sukhdev: we’re out of the way :_) 16:00:22 armax: thanks for the update 16:00:26 we are at the end of the hour 16:00:27 bye folks 16:00:39 Thanks everyone for joing the meeting. 16:00:42 bye 16:00:49 #endmeeting