15:01:36 #startmeeting distributed-virtual-router 15:01:37 Meeting started Wed Mar 12 15:01:36 2014 UTC and is due to finish in 60 minutes. The chair is Swami. Information about MeetBot at http://wiki.debian.org/MeetBot. 15:01:39 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 15:01:41 The meeting name has been set to 'distributed_virtual_router' 15:02:00 #topic Time-change 15:02:55 Just to inform people in the West, because of the time change, our meetings will be held from 8.00a.m to 9.00a.m PST and still the UTC time does not change. 15:03:19 UTC time for this meeting is still at 15.00 UTC. 15:03:27 hi carl 15:03:31 hi Swami 15:04:17 #topic DVR L3 Agent design doc 15:04:45 I am working on the L3 Agent doc and will be posting it as soon as it is complete. 15:04:56 Once it is complete I will post it out for review. 15:05:32 Looking forward to it. 15:05:49 #topic L2 Agent Design doc 15:06:17 The L2 Agent Design doc is out there for review and for people who have not reviewed please review the doc and provide me your feedback. 15:06:36 Carl: I think Vivek had responded to your queries on the L2 Agent doc. 15:07:07 I've seen some replies. I added one more comment yesterday. I'll look at it again. 15:07:34 carl: Yes I will take a look at it and if not addressed, I will address your concerns 15:07:49 Swami: Great, thanks. 15:08:23 #topic Distributed-DHCP 15:08:51 Hi folks, I would like to add an additional topic this week, that addresses the Distributed DHCP. 15:10:02 Going forward for the Juno summit, I would like to take up the discussion of distributing the exisiting DHCP service to the compute nodes so that we can get rid of one other service from the centralized node and distribute. 15:10:27 Will this be based on the multi host bp? 15:11:01 carl:Similar to that. 15:11:04 Swami, I have a question about this which may sounds stupid :-) What will be different between this DHCP service on compute node VS current DHCP agent by scheduler? 15:11:35 my understand is currently you can also have multiple DHCP agents, right? 15:12:03 xuhanp: Having multiple DHCP agents is different than distributing the DHCP to each and every node. 15:12:32 xuhanp: Distributed DHCP if in place, need not deal with the HA for that service. 15:13:06 so multiple DHCP agent still means only one DHCP is serving a certain VM? 15:13:19 xuhanp: Yes 15:13:23 they are just go to different places. I see. 15:13:37 xuhanp: yes 15:13:52 But I think they share some common host files, If I remember correctly 15:14:07 can another one help when one goes down? 15:14:07 xuhanp: It will be similar to the Distributed virtual router, were we have Enhanced L3 agent on each and every compute node. 15:14:46 xuhanp: We need to investigate if another agent can service if an agent in a specific compute node goes down. 15:15:09 Also note that DNS is served from the DHCP server. 15:15:34 carl: Do you see any issue there with the DNS 15:16:18 carl_baldwin, I think I saw one code review to make two host file sync on two dhcp agent so DNS can be served from another agent when one is down. 15:16:27 not sure if that can address the HA problem. 15:17:04 I still have a question on distributed dhcp, what are the asvantages? 15:17:09 advantages? 15:17:28 dhcp traffic is not much compared to data traffic 15:17:54 shivharis: Yes I agree with you that dhcp traffic is not much traffic when compared to data traffic. 15:18:20 dvr made sense to me, i am not convinced on dhcp 15:18:44 There are lot of solutions currently worked on the HA for the current Network Node. But there is no HA solution for the DHCP. 15:19:02 We do have L3-HA. 15:19:33 When we distribute the services the need for HA will be reduced for that particular service. 15:20:17 The idea is just have centralized service ( may be snat, vpn) in a single node and distribute all other services. 15:20:25 distrubuted and more connotations, having a 2nd reduntant dhcp may be sufficient? 15:20:35 Swami, so we will only need L3 HA for external network translation when DVR is enabled? 15:20:36 the multi-host dhcp was a topic that was discussed even before the DVR. 15:21:23 swami: we can discuss this later, dont want to change the main theme of the meeting.. lets move on. 15:22:05 shivharis: I appreciate your feedback, I am just brain storming ideas for the next juno release and will make sense with the Distributed model. 15:22:45 Is safchain in here 15:23:36 Will do a write up on this and will post it out for review. If you have any thoughts we can discuss in the next meeting. 15:23:54 #topic DVR+HA 15:24:05 Is amuller in here 15:24:20 Hi Swami sorry to be late, currently in another meeting 15:24:36 not sure to be able to speak a lot 15:24:41 carl: Do you know if the L3-VRRP HA got FFE for the Icehouse release. 15:25:26 safchain: thanks 15:25:32 I don't think it did. 15:25:51 In the meeting on Monday, Mark said that only bps that were stuck in the gate got an exception. 15:25:56 I did see some +1 on it, but I don't know the status 15:26:23 Yes I did hear that but did not see any list out of the meeting. 15:26:45 safchain: Did you get an exception for the L3-VRRP HA 15:27:17 Swami, I don't think, no new status on it 15:27:50 safchain: carl: thanks on the information. 15:28:27 safchain and amuller sent an email regarding having a session at the Juno summit to talk about the DVR+HA 15:29:47 with respect to the HA solution, we might need to make sure that the L3-VRRP code can be re-used to provide the HA solution for the centralized DVR SNAT. 15:31:07 I will try to chat with safchain and amuller during this week and probably give an update to the team in the next meeting. 15:31:42 I don't have any other topic for this week. 15:31:49 #topic General discussions 15:32:07 Do anyone have other questions or topics to discuss. 15:32:28 Swami, I have a question about the L3 DVR on compute node 15:32:34 if you have a minute 15:32:38 xuhanp: Yes go ahead. 15:32:53 why do we need to create a namespace for floating IP? 15:33:34 Previously in the non distributed model, everything was handled in the router namespace. 15:34:12 In this case, we do have one "floatingip" namespace per compute node, where all the floating ip association will happen. This will be shared by all the tenants in the compute node. 15:35:33 so I am still confused why we cannot have the floating IP association happen in the qr namespace of compute node 15:36:10 xuhanp: If need more information on this we can chat offline or send you a mail with the information. 15:36:25 Sure. That sounds great. Thanks a lot 15:37:00 xuhanp: can you send me an email with this request and I will reply to you. I don't think I have your email id. 15:37:16 I can surely do that. will do! 15:37:24 xuhanp: The problem is that if the qr namespace is attached directly to ext-net then it will consume an IP address per router/compute node. 15:38:26 xuhanp: Note that each qr namespace now consumes an ip address to use as its primary ip address. That is before adding floating ips. 15:38:36 carl: thanks for the clarification, yes the IP consumption is a major driving fator. 15:38:56 When you distribute the router, you can't use the same ip for each DVR instance. You'd need a new ip for each one. 15:39:31 Swami: Happy to help. 15:40:15 If there are no further questions or topic to discuss, then I will end this meeting. 15:40:34 Thanks everyone for joining this meeting. 15:40:37 See you all next week. 15:40:40 thanks Swami and carl_baldwin for the clarification. 15:40:49 #endmeeting