20:00:03 #startmeeting Octavia 20:00:04 Meeting started Wed Aug 26 20:00:03 2015 UTC and is due to finish in 60 minutes. The chair is xgerman. Information about MeetBot at http://wiki.debian.org/MeetBot. 20:00:05 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 20:00:08 The meeting name has been set to 'octavia' 20:00:13 o/ 20:00:17 #chair blogan 20:00:18 o/ 20:00:19 o/ 20:00:19 hi 20:00:23 Current chairs: blogan xgerman 20:00:23 xgerman you were three seconds late 20:00:24 hi 20:00:25 hi 20:00:37 hello 20:00:44 Howdy, howdy! 20:00:45 #topic Announcements 20:00:47 o/ 20:01:05 o/ 20:01:07 Accepted Summit talks 20:01:08 Load Balancing as a Service, Liberty and Beyond 20:01:08 Hands on Lab: Install and Configure OpenStack Octavia 20:01:32 how many people are signed up for the lab? 20:01:54 it’s huge — now we need to see who actually gets funded to go 20:01:54 I'd be happy to help out with that. 20:02:04 I'm about 95% certain I'm going. 20:02:22 #action german add sbalukoff to lab talk 20:02:28 I ... don't think they are planning to send me at the moment 20:02:36 i don't know if i adequately expressed that they SHOULD be 20:02:36 you have a talk... 20:02:39 I'm funding my self. Due to health concerns 20:02:41 we'll see 20:02:59 we honestly weren't sure if they would send anyone 20:02:59 I'm not sure if anyone from rax is going on company dime 20:03:05 I'll have to talk to our guy. 20:03:08 Cuz I'd love to 20:03:25 Wow, that sucks, eh. 20:03:40 Have I mentioned that IBM is sending a whole bunch of people? Also, we're hiring. XD 20:03:41 maby we'll get funded due to the intel boost. 20:03:59 Hopefully! 20:04:09 Tis possible, we'll see I guess :D 20:04:19 #link https://www.openstack.org/summit/tokyo-2015/schedule/main-conference 20:04:24 ^^ blogan 20:04:34 first day as always 20:05:05 yeah, 8 people are in the lab 20:05:23 or we have 8 presenters 20:05:26 thats a big meatball 20:05:44 i think blogan should do all of the lbaas talk. it's his favorite thing. 20:05:50 HAHA! 20:06:08 Seriously though-- I'm happy to help out with whatever there, eh. 20:06:18 awesome 20:06:29 ill be sure to fake a cold 20:06:30 i'll be there, also happy to help. equally happy to let people that need to be speakers to attend be the speakers. 20:06:59 being a speaker isn't as powerful as it was for the paris summit, in regards to rax sending us 20:07:05 low on the totem poll now 20:07:12 so if someone else needs to speak... 20:07:16 what about keynote speaker? 20:07:18 * blogan_ feels a cold coming on 20:07:32 dougwig: +1 as well (same sentiment) 20:07:33 o/ 20:07:35 well ill be doing that for sure 20:07:41 openstack on shingled magnetic recordings? 20:07:46 ??? 20:07:46 Since I'm going regardless of whether I'm speaking at anything. 20:07:48 #chair blogan_ 20:07:49 Current chairs: blogan blogan_ xgerman 20:07:58 2 blogans > 1 xgerman 20:08:10 k, so next is GSLB 20:08:14 johnsom? 20:09:17 I think they had a short meeting and nothing of substance got decided ;-) 20:09:23 there he is 20:09:29 Sorry. 20:09:51 i did show this cool pic of why i wasn't there: http://i.imgur.com/HTp5Wlu.jpg 20:09:53 GSLB Short meeting, we are reviewing updated diagrams that are going into gerrit 20:10:29 xgerman: I would like to discuss L7 tasks if there is time 20:10:37 yes it’s on the agenda 20:10:39 evgenyf: there's an agenda item for it 20:11:13 next, L3 deadline — dougwig is release master since mestery is on vacation 20:11:25 dougwig: Fun! 20:11:32 dougwig changed that to armax 20:11:33 actually it's armax. myself and carl_baldwin are the backups. 20:11:50 smart!! 20:12:21 so deadline is still 8/31 20:12:34 does this mean code complete by then? 20:13:16 in gerrit by then, yes. 20:13:24 ok, cool 20:13:48 other announcements? 20:14:07 #topic Brief progress reports 20:14:10 Are we going into depth on that last point about what needs to be in gerrit? 20:14:20 can we do liberty first? 20:14:27 i have to bail at 2:30 for dentist. 20:14:36 i have a dentist appt too! 20:14:41 k, let’s do L7 and then Liberty 20:14:51 #topic L7 20:15:04 xgerman clearly didn't hit refresh on the agenda. :) 20:15:19 what's left to get l7 in? do we have people on all the pieces? 20:15:28 #topic Liberty deadline stuff 20:15:32 * xgerman hit refresh 20:15:35 lol 20:15:35 Let's talk about Liberfy 20:15:40 Liberty 20:15:46 liberfinity 20:15:51 eew. 20:15:56 we have 3 liberty features. l7, horizon, octavia as ref. 20:16:01 pick one to start. 20:16:01 Seriously, you just made me shudder. 20:16:14 Octavia as ref is most important. 20:16:24 octavia as ref, where are we? 20:16:45 heartbeats, failover are coming 20:16:54 seems like its close 20:16:59 +! 20:17:02 +1 20:17:02 I think I may have just finished failover. 20:17:06 need to get scenario tests working with the octavia driver 20:17:19 gate is in but fails 20:17:31 heartbeats is done, just updating unit tests. 20:17:39 if we fire it up in devstack, does it work better than the current ref? is it performant at all? 20:17:51 ptoohill: tls? 20:18:16 dougwig: its performant if you're got enough space for nova instances 20:18:32 +1 20:18:45 yeah, if you use a small machine it will be slower 20:18:51 Has anyone done any kind of load testing or anything? Not scale, but load? 20:18:59 +1 20:18:59 and nova scheduler will fail 20:19:00 TrevorV no 20:19:09 if not much capacity 20:19:11 do we have any docs or deployment guides? 20:19:25 not yet — we decided we can deliver them after L3 20:19:27 johnsom_: did you commit the main part of your udp changes yet? 20:19:27 that was set up to be done after deadline 20:19:29 I didn't think load testing (or really, any kind of significant performance numbers) were a part of it? 20:19:48 they were but we skirted 20:19:58 i don't think they need to be, but we should get them done at some point 20:20:02 rm_work main part yes, unit test update part we talked about earlier, no, just about donw. 20:20:11 are we ready for the octavia api to be partial baked into stone? 20:20:15 johnsom_: not sure if you actually did a review? 20:20:44 dougwig: hmm there was the goal to make octavia api exactly like neutron-lbaas, its not there 20:20:47 dougwig — do we really need to bake in stone since we use LBaaS V2? 20:20:52 does that need to be baked in stone though? 20:20:54 dougwig: i think there's a couple of changes that need to be ported by blogan_ about the tls containers 20:21:06 but yeah, octavia isn't used by end-users yet, just the driver 20:21:11 so as long as we update the driver... 20:21:28 well, i said partially. once we release the service, we'll likely need to be careful about breaking api changes. especially to the operator api, right? 20:21:29 The tentant-facing API for both Octavia and Neutron LBaaS v2 should be the same. 20:21:50 dougwig: really only a user api at this point, though its noauth 20:21:56 dougwig: There is no operator API right now. 20:22:00 +1 20:22:02 dougwig: no operator style commands 20:22:11 well you knwo what i mean 20:22:24 We will be adding one, but it wasn't seen as necessary in order to replace the current refrence implementation. 20:22:25 maintenance is via nova commands, then? 20:22:41 can be 20:22:52 dougwig: When the system isn't self-correcting, sure. 20:22:52 its something we can add in later 20:22:53 ls -alF 20:22:56 danmit 20:22:57 Right. 20:23:07 ok, this sounds doable, then. does anyone think it's not? 20:23:09 but its not something the current ref impl has anyway 20:23:20 I think it's totally doable. 20:23:23 `1 20:23:31 +1 20:23:35 +1 20:23:38 though still experimental 20:23:48 dougwig: i am concerned a bit by me not being able to create more than one load balancer at the moment in devstack bc nova fills up fast with that one 20:24:04 we can create more than one 20:24:09 if hte ref impl is experimental does that mean neutorn lbaas v2 remains in experimental? 20:24:16 blogan_: one per nova, you mean? 20:24:33 well, is neutron-lbaas not still experimental? 20:24:39 we can create like 4 before our 8 GB VMware devstack dies 20:24:51 I thought neutron-lbaas v2 API was still experimental. 20:25:05 it is, at least until we get horizon support (later in the agenda) 20:25:08 yeah, but we like it to be stable so we can deprecate bass v1 20:25:12 k then no worries yet 20:25:14 dougwig: yes, but im sure its a local issue though 20:25:18 Right. 20:25:21 well, for devstack, they can still enable the old driver if needed. 20:25:49 agreed 20:26:25 what's the minimum VM size, btw? 20:26:42 Yes... so, the namespace driver becomes deprecated, Octavia becomes recommended, and then in "M" release, namespace driver goes away (hopefully). 20:27:09 dougwig, I think ptoohill saw performance degradation under 1GB... 20:27:14 sounds like a plan to me 20:27:23 +1 20:27:43 dougwig I run tests on m1.small 20:28:34 johnsom_: int that 1 GB? 20:28:36 isnt 20:29:09 i can see how multiple LB's per amphora would be a useful optimization later. 20:29:26 ok, let's get this on the L-3 docket unless anyone objects? 20:29:33 or we can make a compute driver that just spins up a process on the controller worker in a namepsace! 20:29:35 dougwig: Yep! But first things first. 20:29:38 we could call it the namespace compute driver 20:29:47 lol 20:29:58 Haha 20:29:59 let's move on to L7/liberty (rushing, have to leave shortly) 20:30:11 dougwig, problem with multiple lb's per amphora (in my eyes) is bad neighbor effect 20:30:18 so is L7 useful at all without being able to point to another pool? 20:30:21 evgenyf: You wanted to say something ablut L7, right? 20:30:21 TrevorV: same tenant though, no problem. 20:30:35 blogan_: Yes-- you can set a rule to block. 20:30:40 Maybe, still one LB could tank another, tenant causes own problem, still "bad neighbor" right? 20:30:53 sbalukoff: so only useful for one rule? 20:31:06 Sorry, I meant policy type. 20:31:07 or redirest to URL 20:31:09 but yes. 20:31:10 BTW, L7 has not been added to Octavia yet either 20:31:16 Two policy types. 20:31:20 That's correct. 20:31:37 well, I think it’s doable but... 20:31:39 I think we're likely to miss the L3 deadline because of too many missing "support stuff" to make L7 more useful. 20:31:40 But! 20:31:54 The work evgeny has done does expand the API in useful ways... 20:32:04 dougwig was talking extension 20:32:05 And in ways that won't need changing later on, right? 20:32:19 how many weeks away are we? 20:32:29 There are 3 patches for review for L7, extension, CLI and ref.implementation, tempest is missing 20:32:30 Probably 3. 20:32:36 but if there's no driver using it, it shouldn't go in 20:32:58 i mean if the ref impl isn't, and if octavia bcomes the ref impl then that'll be odd 20:33:04 We're also missing functionality to create / manipulate pools without being associated with a listener (essential for redirect_to_pool L7 policies) 20:33:09 ref impl doesn't support it, but deprecated driver does 20:33:17 sbalukoff: how many work is there for ref.implementation you think? 20:33:18 I think we're probably going to miss the deadline for this this time. 20:33:42 evgenyf: I'm pretty close on the namespace driver. It's mostly fleshing out tests that I needed to do. 20:34:04 i'm wondering if it's kosher to merge the extension and then not publish the feature, to enable iteration or decomp'ed drivers to use it (which includes the ref in M.) 20:34:13 and then publish in M. 20:34:25 But again, the wind is really stolen from our sails here if we can't do redirect_to_pool rules, since that's what like 95% of our customers use in production with L7 stuff. 20:34:32 it'd be nice to have api versioning for this 20:34:41 Right. 20:34:45 I personally don 20:34:55 't think it's the end of the world if we miss this deadline. 20:35:13 I'd rather we make sure that Octavia is a solid replacement for the ref. impl.... 20:35:14 well, Sam *really* wants this 20:35:16 blogan_: that work is in the main service plugin, right? 20:35:17 well what has neutron-lbaas accomplished for L? 20:35:19 Because if it's not, our project loses credibility. 20:35:33 dougwig: independent pools? 20:35:41 blogan_: yes 20:36:00 dougwig: yeah but maybe some work with the backends as well 20:36:05 one more driver 20:36:06 drivers i mean 20:36:17 yeah, it might be too rushed 20:36:22 but i think every driver probably allows independent pools 20:36:32 except octavia at this point 20:36:39 but that'd be worked on in parallel 20:36:44 Right. 20:37:02 well, we also need to finish all the other work 20:37:10 sbalukoff/evgenyf - can you guys make a list of what's left and an expected timeline, and then we can circle with the L-3 folks? 20:37:11 but exposing to API is just service plugin work 20:37:47 If we merge evgeny's patch (which is honestly pretty much ready to be merged), then in order to deliver this for Liberty we need to solve the independent pools problem for Neutron LBaaS and Octavia, do the L7 stuff for Octavia, and also do it for the namespace driver for good measure (which, again, is mostly done) 20:37:52 oh and db work 20:37:59 right. 20:38:11 dougwig: sure, how abiut to have it on the ML 20:38:17 sbalukoff: ha, i didn't mean right now. :) 20:38:18 sounds good 20:38:21 And tempest test 20:38:23 So... it's a lot of medium-to-small work kind of stuff. 20:38:28 Oh yes, tempest test. 20:38:33 let's get an etherpad going with a plan, then we can ML that. 20:38:37 Which I have no idea how difficult is to do. 20:38:38 +1 20:38:42 Ok. 20:38:45 i personally dont see it making it in 20:38:55 evgenyf: Do you want to do that etherpad, or shall I? 20:39:02 blogan_: +1 20:39:04 that depends on how motivated evgenyf is. i'm willing to give him a shot. 20:39:10 Ok! 20:39:28 if independent pools is a must, which seems like it is, its too much work for too little time 20:39:28 * xgerman shipd redbull to evgenyf_ 20:39:44 sbalukoff. I will 20:39:48 priority is likely octavia first, l7 second. 20:39:55 horizon? 20:39:55 can we do horizon quick now? 20:39:55 sbalukoff: I will 20:40:07 we've got the repo and initial code. has there been any activity since then? 20:40:19 haven’t heard from vivek_ebay in a while so 20:40:26 i dont even know who's working on it, i havent even tested it out bc im a horrible dev 20:40:33 blogan_: I'd be willing to call the independent pools problem a "bug" which affects the ability to utilize some features of the new L7 patch... but I realize that might be stretching the truth a bit. ;) 20:40:52 probably safe to move horizon to M 20:41:10 #action evgenyf to do etherpad, start ML discussion of remaining L7 tasks. 20:41:14 i think its more like a cockroach that has been mutated by radiation 20:41:16 can all of the neutron-lbaas cores take a look at the horizon patch (and anyone else, of course.) but neutron-lbaas cores are also cores in that repo. 20:41:20 (I don't know if I can add action items, eh.) 20:41:50 #action evgenyf to do etherpad, start ML discussion of remaining L7 tasks. 20:41:57 for good measure 20:42:21 dougwig: +1 20:42:22 dougwig will look 20:42:24 I tried vivek's horizon patch, merged it, restarted, the new section did not show up, left it there 20:42:28 thanks ajmiller 20:42:40 evgenyf good to know 20:43:02 evgenyf: how does that get loaded into horizon? 20:43:27 it might be a new panel that needs to get added to /etc/openstack_dashboard, but i haven't looked 20:43:30 * dougwig runs off to dentist. 20:43:41 I merged the code into horizon dashboard like vivek said 20:44:20 bunch of new files and new import, that;s it 20:44:49 okay 20:44:50 hmm 20:45:02 mmh, let’s try and update on the ML 20:45:04 ive gotta head to the dentist soon as well 20:45:14 airport soon for me 20:45:27 blogan_ you and dougwig use the same dentist? 20:45:32 same teeth? 20:45:33 So, I was in a meeting yesterday with IBM bluemix people giving me a demo of the GUI work they did around neutron LBaaS v1... it was prettty good. I think I also made a really strong case for them moving to v2 as soon as possible. If there's interest, I can ping them about the possibility of getting engineering talent on an upstream LBaaS v2 horizon UI. :/ 20:45:35 xgerman: will do 20:45:45 Haha! 20:45:59 sbalukoff the more the merrier 20:46:05 yeah i'm seeing dougwig's dentist, he's doing a remote root canal 20:46:07 on me 20:46:09 xgerman: Okeedokee! 20:46:12 with a robotic arm 20:46:16 Haha! 20:46:20 I hope they hit a nerve 20:46:23 dougwig is probably driving that 20:47:00 #topic Brief progress reports 20:47:10 really brief 20:47:17 so for the neutorn lbaas talk for Liberty, if horizon and L7 dont get in, we only have one thing accomplished, octavia as ref impl 20:47:31 yep, and that is huge!! 20:47:39 health manager is fully functional (DB is updated for status and stats). Working on finishing up updating unit tests 20:47:45 its huge but takes up like 1 minute of talk time 20:47:46 yeah but ... not sure it will be perceived as quite as momentus as we see it :) 20:47:56 Heh! 20:48:06 rm_work we also have active-passive 20:48:12 That's true. 20:48:14 huge as I said 20:48:17 yeah that is true 20:48:30 and ... TLS? 20:48:31 thats a huge review to merge before the deadline 20:48:38 TLS is in actually, and wasn't, right? 20:48:40 and its not a requirement for ref impl 20:48:45 it was in 20:48:50 And it's not the only one that would need to be reviewed / merged. 20:48:57 TLS was enabled in lbaas in K? 20:48:58 I will try to strongram dougwig for an extension 20:49:02 yes 20:49:06 T_T k 20:49:08 just found out there were no docs about it 20:49:10 but it was there 20:49:15 yep 20:49:16 I need some eyes again on https://review.openstack.org/#/c/202336/ 20:49:18 no idea how it would have worked 20:49:19 #link https://review.openstack.org/#/c/202336/ 20:49:23 johnsom demoed it on stage 20:49:25 since we JUST managed to merge the final commits to make it work 20:49:26 lol 20:49:30 I'll work hard on getting my stuff updated and stable. I apologize I've not gotten y'all an updated patch yet. 20:49:30 I think its ready for merging in its most basic failover form 20:49:32 I just tested it. 20:50:00 Cool 20:50:10 grande 20:50:18 ok dentist time 20:50:21 talk to yall later 20:50:26 Enjoy 20:50:27 for TLS tests, looks like adam is working on setting up barbican with devstack and that test can be tested 20:50:29 good luck 20:50:33 johnsom_, captures the routes, adds the haproxy config, and I tested with a LB with a listener, pool, member, and health monitor. 20:50:57 I also tested failing over a failed-over amphora 20:51:03 madhu_ak: yeah, soon barbican will support the new plugin method so we can actually use it correctly in our devstack gates 20:51:05 To make sure it couldn't only fail over once :D 20:51:15 Nice 20:51:24 testing and will continue 20:51:25 rm-work nice 20:51:26 which involves me fiddling with THEIR gate right before freeze <_< 20:51:32 nice ram_work 20:51:33 thanks bana_k 20:51:36 rm_Work 20:52:18 I like ram_work 20:52:23 heh 20:52:24 heh 20:52:37 HAHA 20:52:37 time to change his nick 20:52:41 ok, I guess we can move to 20:52:42 T_T 20:52:53 #topic Open Discussion 20:52:56 time to head to the airport, maybe i'll wave at the Seattle HP people on Friday 20:53:16 yeah, stop by I am sure they can hook you up with red bull and stuff 20:53:18 rm_work ping us if you have a chance... 20:53:21 heh 20:53:33 if there is any open task needs to be done, I can try to work on it 20:53:38 any other tasks to be done ? 20:53:44 i'll prolly be occupied most of the day, but i can wave in passing at least :P 20:53:45 we'll see 20:53:58 the convention center will be BONKERS, so have fun with that 20:54:10 well, In saw a potential L7 tempest test made_ak 20:54:13 mshu_ak 20:54:15 ridiculous number of people will be there 20:54:17 madhu_Ak 20:54:21 yep 20:54:35 sure, can take a look into it. best person to reach? 20:54:50 sync with sbalukoff and evgeny 20:54:51 bana_k: I don't think anyone has started on the L7 work for Octavia directly yet. I can give you pointers on that if you'd like. 20:54:59 +1 20:55:12 awesome, I can sync with them 20:55:21 +1 20:55:24 also there was an ask for performance tests 20:55:40 sbalukoff, that do be great. 20:55:44 xgerman, who asked that? 20:55:45 Sweet! 20:55:48 dougwig 20:55:51 dougwig was asking earlier 20:55:55 kk 20:55:58 just curious 20:55:58 TrevorV you guys have a lb? 20:56:00 However, I would consider a performance problem to be a "bug" 20:56:01 lab? 20:56:18 So... work on it after this week. In the short term, let's get that Octavia as a reference driver done. 20:56:26 +1000 20:56:32