15:01:14 <regXboi> #startmeeting neutron_northbound 15:01:14 <odl_meetbot> Meeting started Fri Jun 26 15:01:14 2015 UTC. The chair is regXboi. Information about MeetBot at http://ci.openstack.org/meetbot.html. 15:01:14 <odl_meetbot> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 15:01:14 <odl_meetbot> The meeting name has been set to 'neutron_northbound' 15:01:20 <regXboi> #chair flaviof 15:01:20 <odl_meetbot> Current chairs: flaviof regXboi 15:01:20 <flaviof> #info flaviof 15:01:28 <regXboi> #topic roll call and agenda 15:01:32 <regXboi> #info regXboi 15:02:08 <regXboi> #link https://wiki.opendaylight.org/view/NeutronNorthbound:Meetings agenda in the usual place 15:02:18 <regXboi> anybody want to add anything to the agenda? 15:02:57 <edwarnicke> #info edwarnicke 15:03:04 <regXboi> ok, if not 15:03:09 <regXboi> #topic action item update 15:03:38 <regXboi> #info 1. regXboi to write up wiki page on what information is needed when a bug is filed 15:03:52 <regXboi> #link https://wiki.opendaylight.org/view/NeutronNorthbound:Bugs:_Help_Us_Help_You wiki page 15:03:55 <regXboi> so that one is done 15:04:04 <flaviof> nice 15:04:19 <regXboi> armax to work with regXboi and edwarnicke on fixing bug 3812 15:05:02 <regXboi> #link https://git.opendaylight.org/gerrit/#/c/23180/ patch set handling bug 3812 15:05:15 <regXboi> #info that patchset will be merged to stable/lithium today 15:05:28 <edwarnicke> regXboi: Is stable/lithium unblocked for us? 15:05:42 <regXboi> edwarnicke: yes - I've been merging patches all morning 15:05:48 <regXboi> edwarnicke: where I can 15:06:16 <edwarnicke> mestery: Welcome :) 15:06:17 <regXboi> I think we have 5 patches left to merge 15:06:27 <regXboi> and stable/lithium will be caught up 15:06:30 <flaviof> regXboi: if you think that is useful, I caould tweak the wiki page to include a example way of cappturing the nn packets 15:06:41 <regXboi> flaviof: please 15:06:50 <AnkurGupta> please. Thank you 15:07:05 <flaviof> regXboiL AnkurGupta: ack. 15:07:12 <regXboi> #info 3. yamahata to triage bug 3824 15:07:22 <regXboi> yamahata: welcome - any progress? 15:07:25 <yamahata> I put a comment on it. 15:07:44 <yamahata> I agree with flaviof for lowering the priority 15:08:06 <yamahata> Given those two are ignored by ODL, neutron side works fine 15:08:22 <regXboi> so, we have a bunch of model updates to do for Be 15:08:29 <regXboi> I think this goes on the list 15:08:33 <yamahata> I'll continue to work on it as background task. 15:08:38 <regXboi> o 15:08:42 <regXboi> er ok - thanks 15:09:04 <regXboi> #info 4. edwarnicke to work with alagalah on verifying 3368 and 3304 15:09:12 <regXboi> edwarnicke: you around for an update? 15:09:36 <regXboi> I see that 3368 is closed and 3304 is *better* but still not perfect 15:09:46 <regXboi> because we have a bunch of cruft to clean up in Be related to it :( 15:09:58 <flaviof> yamahata: thanks for jumping in! 15:10:25 <regXboi> #info 3368 is closed and 3304 is *better* but still not perfect as there is more cruft to clean up in Be related to it 15:10:45 <regXboi> #info 5. regXboi to send email to publicize the future release work list 15:10:53 <regXboi> #link https://lists.opendaylight.org/pipermail/neutron-dev/2015-June/000196.html future release wish list announcement 15:10:57 <regXboi> I hope folks read it :) 15:11:09 <regXboi> because that was item numbre 6 15:11:22 <flaviof> #action flaviof to elaborate a little on getting the info mentioned in https://wiki.opendaylight.org/view/NeutronNorthbound:Bugs:_Help_Us_Help_You 15:11:31 <regXboi> but we'll talk more about it later 15:11:35 <regXboi> #topic Lithium 15:11:50 <regXboi> #info Lithium is out, the stable/lithium branch is unfrozen 15:12:05 <regXboi> #action regXboi to finish merging patches from master into it 15:12:17 <regXboi> any other Lithium items? 15:12:31 <regXboi> #info regXboi believes there are 5 more patches to merge to catch up 15:12:55 <regXboi> #info there is a current problem with the IT in VM running correctly, but regXboi has an idea for how to address that 15:13:15 <regXboi> going once on other Li items? 15:13:26 * regXboi waits 15 seconds 15:13:34 <regXboi> going twice on other Li items? 15:13:50 * regXboi waits another 15 seconds 15:14:01 <regXboi> #topic planning for Be and beyond 15:14:19 <regXboi> #link https://wiki.opendaylight.org/view/NeutronNorthbound:FutureReleaseWishList wiki page of future tasks 15:14:42 <regXboi> If folks want to go to that page, that is what I'm going to be talking from 15:14:43 <AnkurGupta> ' 15:14:55 <regXboi> AnkurGupta: yes? 15:15:05 <AnkurGupta> accident sorry 15:15:20 <regXboi> no worries :) 15:15:34 <regXboi> so... the main things I think we need to do are at the top of the Be list 15:15:55 <edwarnicke> regXboi: I think we also need to prioritize HA and scaling for NN and our OS driver. 15:16:19 <regXboi> edwarnicke: note: that list isn't in an order as of yet 15:16:26 <flaviof> regXboi: I see HA being a big part of Be. In openstack terms, I expect that to percolate into neutron ml2. 15:16:53 <edwarnicke> flaviof: HA *and* scalability... you can do both together, or you can do HA in a way that scales poorly... I think we'd like to do both :) 15:17:06 <flaviof> edwarnicke: 100% agree. 15:17:11 <regXboi> folks, please do the unconference thing and edit the page and add +1's under topics that you are willing to work on 15:17:12 <edwarnicke> flaviof: I strongly suspect from a preliminary look at the problem that it breaks into these pieces: 15:17:17 <edwarnicke> 1) HA within NN 15:17:31 <edwarnicke> 2) Enabling HA for the Providers 15:17:52 <edwarnicke> 3) Improved mechanisms and algorithms for communicating from OS to us 15:18:26 <edwarnicke> I think #1 mostly comes down to the MD-SAL migration 15:18:35 <regXboi> edwarnicke,flaviof,others: ^^^^^^^ on marking items you are willing to work on 15:18:41 <edwarnicke> I think for #2 we need to figure out our asks to clustering in partnership with our consumers 15:18:48 <regXboi> or better yet 15:18:56 <edwarnicke> #3 is going to involve a lot of collaboration upstream 15:19:02 <regXboi> can somebody take the action item of making trello cards for all of these? 15:19:05 <regXboi> and we can vote there? 15:19:11 <edwarnicke> flaviof: Do you have any ideas on #3 yet (or armax ? ) 15:20:28 <yamahata> Regrading to #3, it basically comes down to synchronize two databases. 15:20:29 <flaviof> edwarnicke: i do not yet. but you are right in saying that it involves a lot of collaboration with the neutron community. 15:20:39 <odp-gerritbot> Ryan Moats proposed a change to neutron: Run jacoco in UT VM and report to sonar https://git.opendaylight.org/gerrit/23428 15:20:46 <edwarnicke> yamahata: You are correct 15:21:01 <edwarnicke> yamahata: The tricky part is that sinking two DBs in an HA world gets more complicated 15:21:01 <yamahata> I suspect modification to db schema is inevitable 15:21:11 <edwarnicke> yamahata: Especially when events keep coming while you try to sync :) 15:21:21 <edwarnicke> yamahata: Please say more, I'd love to hear your ideas :) 15:22:23 <yamahata> one possibility is to throw away one of two db 15:22:34 <yamahata> I'm not sure it's feasible 15:22:36 <flaviof> yamahata: if odl is not behind a load-balancer; I also think ml2 metworking-odl will need to accommodate more than one OLD_IP_MGR. 15:22:49 <flaviof> s/OLD/ODL 15:23:20 <odp-gerritbot> Ryan Moats proposed a change to neutron: Clean up IT and address Bug 3812 https://git.opendaylight.org/gerrit/23433 15:23:50 <edwarnicke> flaviof: I agree, we need to accomodate more than one ODL_IP_MGR for HA 15:24:25 <edwarnicke> yamahata: I don't think we will get reasonable performance and scalability if we throw away one of the two dbs 15:24:31 <regXboi> so, that's a definite change to the ODL ML2 agent 15:24:53 <edwarnicke> regXboi: Yes, and then we need to decide what to *do* with multiple IPs 15:25:09 <edwarnicke> Do we round robin, have a way to figure out where the shards are mastered and talk to that? 15:25:09 <regXboi> edwarnicke: I'd probably vote for RR 15:25:19 <regXboi> as a first step 15:25:25 <flaviof> regXboi: +1 15:25:31 <edwarnicke> How do we handle join/leave from the cluster... if someone adds a new ODL cluster node, then clearly static IPs are not going to be enough 15:25:34 <regXboi> if we can figure out shard masters, great 15:26:04 <edwarnicke> regXboi: flaviof Agreed... its just been my experience if you *think* a little about where you are going, you can leave archetectual 'Whitespace' for the things you know are coming 15:26:06 <regXboi> and now you've introduced something that is borderline NN (imho) 15:26:12 <edwarnicke> (as opposed to painting yourself into a corner ;) ) 15:26:49 <regXboi> so, honestly, telling upstream systems about ODL cluster IP addresses is something for controller, I would think 15:26:53 <regXboi> we should piggyback on it 15:28:07 <edwarnicke> regXboi: Yes, but if we want to have clustering expose some of that via restconf, we need to ask them to do so :) 15:28:20 <edwarnicke> regXboi: Basically, ask them to keep an operational model folks can access :) 15:28:30 <regXboi> so... you *KNOW* my opinion on restconf for anything northbound 15:29:04 <regXboi> but 15:29:05 <edwarnicke> regXboi: One way or the other... we would need to tunnel info from ODL to OS to find out about our cluster nodes 15:29:14 <edwarnicke> regXboi: If we are using rest, that means we need a rest API 15:29:14 <regXboi> can we come back out of the weeds for a minute? 15:29:25 * edwarnicke likes the weeds, but assents 15:29:48 * alagalah surfaces now folks emerge from weed induced frenzy :D 15:29:56 <regXboi> so... (1) do we have a volunteer to create trello cards for the tasks in the wiki page? and to create new ones for other tasks? 15:30:18 <regXboi> (2) will people please put their IRC handles and a +1 under the cards so that we can see what folks are willing to work on ? 15:30:33 <regXboi> (3) then we can come back in a week or two and see what Be looks like? 15:30:41 * edwarnicke is very bad at trello (as shague and flaviof they can attest ;) ) 15:31:10 * regXboi looks towards alagalah and flaviof/shague as having most experience with 1 :) 15:31:51 <alagalah> regXboi: I'll make cards 15:31:57 <alagalah> action me oh keeper of the chair 15:32:13 <alagalah> Due: Wed next week 15:32:14 * flaviof is probably worse than edwarnicke on trello cards :( 15:32:25 <flaviof> #chair alagalah 15:32:25 <odl_meetbot> Current chairs: alagalah flaviof regXboi 15:32:26 * edwarnicke is not sure how that is physically possible ;) 15:32:34 <alagalah> flaviof: niiiiiiice 15:32:36 <regXboi> #action alagalah to kindle create trello cards for the Be items with a deadline of Wed next week 15:32:42 <regXboi> #undo 15:32:42 <odl_meetbot> Removing item from minutes: <MeetBot.ircmeeting.items.Action object at 0x1942cd0> 15:32:56 <regXboi> #action alagalah to kindly create trello cards for the Be items with a deadline of Wed next week 15:33:10 * regXboi doesn't want to see flaming trello cards :) 15:33:14 * alagalah not sure I can do anything kindly, since I mash at my keyboard at best but yes.... 15:33:32 <regXboi> #action everybody to +1 cards they are willing to work on 15:33:56 <regXboi> #action regXboi to add agenda item to check status of interest in Be items 15:34:06 <regXboi> ok... now that we've got *that* out of the way 15:34:14 <regXboi> edwarnicke is welcome to partake of weeds again 15:34:45 * edwarnicke *is* in Colorado after all ;) 15:35:10 <alagalah> edwarnicke: flaviof So with HA, are we looking at a mechanism as well for DB sync between ODL and OS ? 15:35:22 <alagalah> edwarnicke: flaviof i.e. What did YOU do whilst I was gone ? 15:35:47 <flaviof> edwarnicke: lol 15:35:54 <alagalah> Who made changes to neutron, and who made changes to ODL ? (note "changes" could be operational as well as configuration) 15:36:10 <edwarnicke> alagalah: For HA and Scalability we need a way of both keeping in sync with the OS DB *and* receiving events in the correct order 15:36:24 <edwarnicke> alagalah: No major shift was made yet in how we are doing it 15:36:27 <alagalah> Thats one direction, no thought as to the other ? 15:36:45 <edwarnicke> alagalah: Not opposed to the other, just hadn't thought about it as much 15:36:55 <edwarnicke> alagalah: The thing we currently face I believe is this: 15:36:58 <alagalah> flaviof: Do you? 15:37:04 <flaviof> in regards to ml2 specifically, I'd like to take a step back and learn from our neutron gurus how ha is intented to be consumed. 15:37:11 <edwarnicke> 1) Our OS driver talks to only one ODL IP (ie, no HA) 15:37:13 <flaviof> maybe existing working models we could leverage. 15:37:13 <alagalah> flaviof: yeah 15:37:16 <odp-gerritbot> Ryan Moats proposed a change to neutron: Clean up IT and address Bug 3812 https://git.opendaylight.org/gerrit/23433 15:37:19 <alagalah> flaviof: Well more align to 15:37:33 <edwarnicke> 2) If there are multiple control nodes, each will try to do a *full* dump of the neutron DB to ODL on startup 15:38:02 <edwarnicke> #2 doesn't scale :( 15:38:24 <edwarnicke> 3) Each control node will try to tell ODL about each event (ie, event happens, we receive *N* events, one from each control node) 15:38:34 <edwarnicke> Which also won't scale well and is likely to be racy as shit 15:38:59 <alagalah> flaviof: regXboi edwarnicke Doesn't it make sense to step back and see where OS is going with all this? And then we can say "We align with the L release or M release of HA" ??? 15:39:17 <edwarnicke> alagalah: If they are going somewhere with this, absolutely 15:39:21 <edwarnicke> (or have gone somewhere) 15:39:27 <edwarnicke> But browsing through the public ML2 drivers 15:39:31 <alagalah> edwarnicke: I don't know... thats what I mean... step back, survey the ground... 15:39:35 <edwarnicke> It looks like almost all of them have the same defects 15:39:36 <regXboi> alagalah: when it comes to HA, the story with respect to the mechanisum drivers is a bit muddy 15:39:46 <regXboi> alagalah: as in non-existent 15:39:47 * edwarnicke *has* looked at other peoples code ;) 15:40:09 <yamahata> the current ML2 design is already racy. and there are an effort with taskflow on-going 15:40:29 <edwarnicke> armax: Any word on where HA is going with ML2 drivers? We are feeling the need for stuff, would love to learn from others :) 15:40:43 <regXboi> yamahata: +1 (in all sorts of connotations of the word "racy" (smile)) 15:40:53 <edwarnicke> yamahata: Could you give some examples of where its racy? :) 15:41:27 <yamahata> the step to update is to commit neutron db, then call mechanism driver. 15:41:42 <yamahata> The order of db commit and calling mechanism driver can be reordered. 15:42:16 <odp-gerritbot> A change was merged to neutron: Clean up IT and address Bug 3812 https://git.opendaylight.org/gerrit/23433 15:42:19 <yamahata> So even what neutrno db has can be differ from what ODL has 15:42:35 <edwarnicke> :( 15:42:58 <yamahata> Usually it's quite rare to update on same resource. i.e. same network, same neutron port. 15:43:09 <yamahata> at same time 15:43:12 <alagalah> yamahata: So its semi-synchronous 15:43:37 <yamahata> Yes. So ML2 are working on synchronizing mechanism with taskflow 15:43:39 <alagalah> yamahata: The local neutron DB is committed before the ML2 message to ODL is ack'd 15:43:57 <alagalah> yamahata: nice... but it does bring up scale... but nice. 15:44:37 <alagalah> yamahata: let me restate 15:44:40 <alagalah> yamahata: So its semi-synchronous? 15:44:43 <alagalah> yamahata: The local neutron DB is committed before the ML2 message to ODL is ack'd ? 15:44:50 <alagalah> I forgot punctuation... :) 15:44:51 <yamahata> Yes. 15:45:00 <alagalah> yamahata: thank you 15:45:23 <alagalah> yamahata: So these efforts, do you have pointers to them ? 15:45:31 <yamahata> The reason why ML2 doesn't call mechanism driver during db transaction is timeout 15:45:34 <alagalah> yamahata: The ones to improve consistency of the workflow 15:45:45 * yamahata looking for the link... 15:45:49 <alagalah> yamahata: thank you. 15:46:53 <odp-gerritbot> Ryan Moats proposed a change to neutron: Run jacoco in UT VM and report to sonar https://git.opendaylight.org/gerrit/23428 15:47:12 <alagalah> yamahata: I have to step away for another meeting, but I promise I will look when I return. I assume there is a "think" for HA on the wiki I'm to turn into a card, so I shall put your link to the card this weekend. 15:47:22 <alagalah> * "thing" 15:47:59 <regXboi> folks: are we done? 15:49:53 <yamahata> https://wiki.openstack.org/wiki/Meetings/ML2 15:49:59 <yamahata> at the agenda of June 17 15:50:12 <yamahata> But right now I couldn't find good one. 15:50:17 <flaviof> #link https://wiki.openstack.org/wiki/Meetings/ML2 Neutron ML2 meeting 15:50:18 <yamahata> Try it later 15:50:25 <flaviof> yamahata: thanks! 15:50:31 <regXboi> #topic cookies#linfo lots of discussion about HA - see the logs for the full text 15:50:35 <regXboi> #undo 15:50:35 <odl_meetbot> Removing item from minutes: <MeetBot.ircmeeting.items.Topic object at 0x1a9d890> 15:50:41 <regXboi> #info lots of discussion about HA - see the logs for the full text 15:50:53 <regXboi> again - are we done? 15:50:57 <regXboi> going once? 15:51:19 <regXboi> going twice? 15:51:31 <regXboi> #action regXboi to update meeting page for next meeting 15:51:43 <regXboi> edwarnicke: you will miss next week, yes? 15:51:53 <regXboi> if so, please info that :) 15:52:45 <regXboi> oh well 15:52:49 <regXboi> #topic cookies 15:52:55 <regXboi> thanks all! 15:52:57 <regXboi> #endmeeting