17:01:46 #startmeeting Octavia 17:01:47 Meeting started Wed Jul 19 17:01:46 2017 UTC and is due to finish in 60 minutes. The chair is johnsom. Information about MeetBot at http://wiki.debian.org/MeetBot. 17:01:48 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 17:01:49 o/ 17:01:51 The meeting name has been set to 'octavia' 17:01:54 Pradeep 17:02:01 Hi folks! 17:02:04 Hi 17:02:10 Hi 17:02:24 #topic Announcements 17:02:35 PTL elections are coming up 17:02:41 #link https://governance.openstack.org/election/ 17:03:17 Please consider running for Octavia PTL. The link above has information on criteria, etc. 17:03:52 I cannot say at this time whether I will run for another release.... 17:04:00 :( 17:04:06 :-( 17:04:14 Plus, it's always good to have new folks involved. 17:04:18 :( 17:04:23 johnsom: u are awesome :) 17:04:29 +1 17:04:34 Yeah, I'm not sure if my employer will want me to run again 17:04:38 +1 17:04:44 +1 17:05:01 hopefully we don’t go the way of designate 17:05:05 Thanks folks! grin 17:05:30 Yeah, I have faith in our community (couldn't do it without your support) 17:05:41 xgerman_: what happened with designate? 17:06:11 may be not important 17:06:33 they imploded because of lack of participants/leadership 17:06:40 Also, the PTG is September 11-15th. I have an Octavia room reserved Wednesday through Friday, though I will be there all week. 17:06:44 xgerman_: ohh 17:06:50 #link https://www.openstack.org/ptg 17:07:51 The next summit is in Sydney, Australia November 6-8. I am not planning to attend this one. 17:07:58 #link https://www.openstack.org/summit/ 17:08:04 rm_work and I submitted a talk 17:08:17 Yes, I think some folks are planning to attend. 17:08:31 well, I wouldn’t call that planning 17:08:55 Well, I'm pretty sure at least one other person is.... 17:09:01 but if somebody has budget and the talk is approved we can add people 17:09:28 Yeah, our talks are usually a community collaboration. 17:09:41 Which I think is cool. 17:10:09 Final announcement I have: Feature freeze and Pike-3 milestone is next week, July 24th!!!! 17:10:28 #link https://releases.openstack.org/pike/schedule.html 17:11:12 Next week we will be locking down the features for Pike. Please plan accordingly. If you have something critical to get in, please add it to the priority bug list and let me know. 17:11:22 #link https://etherpad.openstack.org/p/Octavia-Pike-priority-patches 17:11:41 Any other announcements I am forgetting about? 17:12:27 #topic Brief progress reports / bugs needing review 17:12:36 https://review.openstack.org/#/c/418530/ - needs review 17:12:39 johnsom, just joined in. I think you'll get our votes even if you won't run. 17:12:43 Ok, so I just mentioned a reminder about the priority bug list. 17:12:50 yep added 17:13:11 Cool 17:13:21 but we should probably discuss if/how much we swotch to V2 API 17:13:26 later 17:13:28 nmagnezi also wanted input on: 17:13:34 #link https://bugs.launchpad.net/octavia/+bug/1705259 17:13:35 Launchpad bug 1705259 in octavia "Octavia health-manager and worker should work better with any given interface_driver " [Undecided,New] - Assigned to Nir Magnezi (nmagnezi) 17:13:47 Ok 17:14:01 yup, please have a look and comment i'd like to work on that one 17:14:21 Recently I have started to pick up the python-openstacksdk patches that the folks from Intel started 17:14:33 #link https://review.openstack.org/#/c/454410/ 17:14:40 #link https://review.openstack.org/#/c/451574/ 17:15:12 #link https://review.openstack.org/449264 17:15:27 This is a python library for OpenStack that will include Octavia. 17:15:39 I plan to use it for the octavia-dashboard 17:16:34 I think I can get most of API done this week. 17:16:55 Any other progress updates or bugs to note? 17:17:39 #topic Proposed mascot 17:17:54 The foundation has come back with a proposed mascot rendering 17:18:01 #link https://usercontent.irccloud-cdn.com/file/KdpSnmb0/OctaviaMascot.jpg 17:18:24 I think rm_work had some comments/feedback. Any other thoughts from the team? 17:18:26 looks nice! 17:18:29 o/ 17:18:30 +1 17:18:33 ship it 17:18:42 +1 17:18:45 looks good 17:18:46 I did have some feedback 17:18:48 it looks nice 17:18:51 Can you link it again? 17:18:59 rm_mobile, https://usercontent.irccloud-cdn.com/file/KdpSnmb0/OctaviaMascot.jpg 17:19:07 looks good 17:19:07 I thought it looked really busy 17:19:16 And why wasn't it 8 feathers 17:19:33 That'd help with the busyness 17:19:34 its our national birs 17:19:37 bird 17:19:39 And make it a pun 17:19:58 Nice 17:19:58 great :) 17:20:06 Oct-avia 17:20:11 So, proposing dropping the two bottom feathers? 17:20:21 sounds good 17:20:32 +1 17:20:41 They could do that or resize them all 17:20:54 rm_mobile: avia? 17:21:09 I actually was hoping for the resize but both work 17:21:18 Pksingh "oct" 17:21:25 Means 8 :P 17:21:32 What do others think about dropping the two bottom feathers to make it eight (octa)? 17:22:07 Just dropping those two might make it look awkward because of the lines 17:22:13 then we can increase the width of feathers little bit, to look it full 17:22:13 Personally I'm fine with it as-is, but looking for feedback 17:22:19 if that's important for Adam I'd vote in favor of it :P 17:22:53 To give context, the foundation is putting them here: 17:22:54 yeah, for me it’s more important we get a mascot/sticker after what happened the last time ;-) 17:22:57 #link https://www.openstack.org/software/project-navigator/ 17:23:22 They will also make up stickers for the summit/PTG 17:23:41 I already reserved a space on my laptop ;-) 17:24:01 * nmagnezi needs a mascot shipped :D 17:24:29 nmagnezi We can probably make that happen 17:24:34 +1 17:24:37 please ship to us ;) 17:24:45 that would be really nice :-) 17:25:30 We can figure out a way to get any active contributor a sticker 17:25:33 Yeah just note how simple most are 17:25:45 On that page 17:25:48 Ok, so should we vote on this? More discussion? 17:25:59 To many lines makes them look really bad as a low res icon 17:26:47 When I zoom it small in chrome the dark and light sections show up pretty well actually 17:26:59 Yeah it's close 17:27:22 https://usercontent.irccloud-cdn.com/file/FwOSBaQh/image.png 17:27:39 Eugh 17:27:56 Yeah less plz 17:28:07 i think less feathers would help that 17:28:49 8! 17:29:25 #startvote Should we change the proposed mascot? Yes, No 17:29:26 Begin voting on: Should we change the proposed mascot? Valid vote options are Yes, No. 17:29:27 Vote using '#vote OPTION'. Only your last vote counts. 17:29:39 Just to keep our voting heritage alive.... 17:29:45 ol 17:29:51 #vote yes 17:30:03 #vote No 17:30:04 #vote Yes 17:30:20 #vote No 17:30:26 #vote yes 17:30:33 #vote yes 17:30:42 #vote yes 17:31:24 Ok, going once... 17:31:32 going twice.... 17:31:42 #endvote 17:31:43 Voted on "Should we change the proposed mascot?" Results are 17:31:44 Yes (4): sanfern, nmagnezi, rm_mobile, jniesz 17:31:45 No (2): johnsom, pksingh 17:32:10 8 feathers wins the day! Huzzah 17:32:18 Ok, so help me with instructions for the artist. We want eight feathers, how do we do that? 17:32:33 Reduce the number of feathers by 2 17:32:36 :P 17:32:40 * johnsom glares 17:32:51 Ideally scale the others around to match the current layout 17:32:53 reduce by 2 resize the rest 17:33:31 So we still want two dropped features, just remove two, resize and some white space between feathers? 17:33:59 Help me, to help you, help the artist.... grin 17:34:11 Err I don't know about white space 17:34:18 Just drop two and resize imo 17:34:29 So, you are thinking making the feathers bigger? 17:34:33 yea, no whitespace 17:34:34 Yes 17:34:44 Ok 17:35:15 Everyone good with that? Additional comments? 17:35:37 Don't make me call another vote... grin 17:36:03 All right, moving on 17:36:11 #topic Discuss L3 active-active spec confusion (xgerman) 17:36:58 xgerman and I were talking this week about the active/active patch he is working on and the distributor driver 17:37:12 I think there is some confusion about the L3 spec 17:37:43 Specifically, what actions the L3 act/act distrubutor driver will take against the top of rack switch? 17:38:04 and what part the bgp speaker plays 17:38:20 the ToR will be the bgp peer 17:38:20 What are the mechanics if an amphora needs to be removed from rotation? (distributor-> ToR switch?) 17:38:37 And what exactly the BGP speaker will be sending out. 17:38:48 xgerman_ Does that capture the questions? 17:38:55 the bgp speaker will essentially be sending out route announcements 17:39:02 to the bgp peer of the ToR 17:39:41 ok, johnsom thought the speaker would just annoucne and cold be configured when starting an amp and didn’t need to be reconfigured 17:40:22 we would want the ability to cleanly withdraw the route as well 17:40:28 so when we remove it from the distributor 17:40:35 it will pull the route 17:40:44 otherwise have to wait for timeout 17:40:53 which is dictated by the hold time 17:41:05 So, give us a little detail on how that would work. Is that a distributor driver call to ToR? 17:41:19 no calls directly to the ToR 17:41:30 all the calls would go through the bgp speaker on the amphora 17:41:57 Ok, so to remove an amphora, it needs to stop speaking. Correct? 17:42:20 to remove it would have to withdraw the /32 route and then terminate the peering session 17:42:42 that is for shutdown 17:42:49 Because essentially, in your model the "distributor" is the ToR doing the ECMP right? 17:43:15 the ToR is just bgp neighbor that is taking the route and injecting it into the L3 fabric 17:43:23 the neighbor could be a route reflector as well 17:43:31 doesn't have to be the ToR 17:43:36 in our model it will be ToR 17:43:57 but the design should work with different network topologies based on L3 fabric 17:44:36 when the amphora boots up we can automatically create the peer session 17:44:41 So, to setup the ECMP hashing, your distributor driver would talk to what? 17:44:44 and then announce the route when everything is ready 17:45:15 for the ECMP hashing, we would need multiple amphora launched 17:45:30 Or are you planning to just accept any multi-advertised route as an ECMP hashed 17:45:30 and they would inject the same anycast VIP route with different next-hops 17:46:14 each amphora injects the same route with different next hop 17:46:24 ECMP happens when you have 2+ 17:46:39 as the network then has multiple best paths of equal eight 17:46:41 weight 17:47:21 * johnsom Thinks he had this all wrong and German was on the right path. 17:47:39 I think we need to step back and re-think the distributor driver. 17:47:49 well, we still should note that in the spec 17:47:59 so others can read & understand 17:48:18 jniesz Can we update the spec to call out who is talking to what components for each of the actions? I think that would help clarify. 17:48:36 +1 Yeah, that was what I was typing. 17:48:37 it can also just be an example and we can go from there 17:48:56 yes, it would be an example, as a different operator might have different devices 17:48:58 Bonus points for a lifecycle 17:48:59 that they BGP peer with 17:49:05 Right 17:49:18 yes, but it would help us to understand ;-) 17:49:57 ok, I can add some details in the spec with our example 17:50:03 Thanks! 17:50:29 I still feel we need to whiteboard ACTIVE-ACTIVE and talk that through at the PTG 17:50:32 also for the distributor driver talk if you want to discuss after on a call I can set up 17:50:53 ok, so now the discussion is really about how we abstract the act/act to support the "service vm distributor", "appliance distributor", and 'L3 fabric distributor" models. 17:51:03 indeed 17:51:20 and where we put the abstraction: flows for each or hide it in the driver 17:51:52 yea, I remember the flow question came up 17:52:26 I can't do a call today, I have some other meetings, but yeah, whiteboard and/or proposals would be good. 17:52:42 I need to think about that abstraction a bit. 17:53:16 they are pretty different, as some you just call api from driver, others you have to orchestrate the service vm distributor 17:53:17 same here and we probably should loop in rm_work to see if it fits "our architecture” 17:53:35 Right 17:54:15 #action Think about how we abstract the act/act to support the "service vm distributor", "appliance distributor", and 'L3 fabric distributor" models. 17:54:26 #topic Open discussion 17:54:37 Any other topics for the last 5-6 minutes today? 17:54:46 My main things is how much do we want to switch to V2 API? 17:54:53 https://review.openstack.org/#/c/484325/ 17:55:04 Are we planning to change the default in devstack. etc. 17:55:45 unrelated I asked infra and they can build diskiamges… 17:56:17 Hmmm, so changing the service "octavia" to not require neutron-lbaas? 17:56:33 and making the proxy default driver in Neutron 17:56:48 We need to merge that first... 17:56:53 indeed 17:56:54 (needs some reviews) 17:57:20 but depending on what we decide that might inform priorities 17:57:27 Yeah 17:57:50 It might be a bit late to pull that switch on other projects. I'm thinking heat for example 17:58:16 well, I think we can punt that until next week…if we do proxy it shoudl eb transparent 17:58:26 I would propose making sure the proxy gets into Pike, then at the start of queens making the switch 17:58:53 yeah, just wanted us to have a decision… works for me 17:59:15 Ok, let's do that. 17:59:27 pksingh 17:59:41 johnsom: i replied to your comments 17:59:44 Reading your comments. I will comment there as we are now out of time 17:59:52 johnsom: sure, thanks :) 17:59:55 Other please chime in 18:00:09 Ok, thanks folks! 18:00:15 thanks 18:00:20 #endmeeting