09:02:03 <oanson> #startmeeting Dragonflow
09:02:03 <openstack> Meeting started Mon Jun 26 09:02:03 2017 UTC and is due to finish in 60 minutes.  The chair is oanson. Information about MeetBot at http://wiki.debian.org/MeetBot.
09:02:04 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
09:02:06 <openstack> The meeting name has been set to 'dragonflow'
09:02:34 <oanson> All right. Let's wait another minute to see if anyone is coming in late (and let me update the agenda)
09:02:42 <lihi> Hi
09:03:21 <oanson> All right. Agenda is updated ( https://wiki.openstack.org/wiki/Meetings/Dragonflow )
09:04:39 <oanson> All right. Let's start
09:04:43 <oanson> #topic Roadmap
09:04:53 <oanson> I'll starting by announcing that our gate is broken.
09:04:55 <oanson> Again.
09:04:59 <lihi> :(
09:05:24 <oanson> Fix is available here. It's in Neutron. I hope it'll pass review tonight. https://review.openstack.org/#/c/476511/
09:05:58 <oanson> Many thanks to Kevin and Armando (not using nicks not to alert them for no reason) for their great help.
09:06:21 <oanson> And this one: https://review.openstack.org/#/c/477206/ for the nova issue
09:06:49 <oanson> I'd recommend adding the neutron fix as a Depends-On, and rebasing all new patches on top of dimak's fix
09:07:03 <oanson> Roadmap! (Really, this time)
09:07:33 <dimak> I can add depends-on on my patch and see if we get a verified+1
09:07:34 <oanson> SFC - I didn't see many updates this week. But dimak, you were tackling the dnat issue, right?
09:07:48 <dimak> I worked on both
09:07:57 <oanson> Won't hurt
09:08:02 <dimak> Managed to get SFC working on multinode
09:08:11 <dimak> I'll upload my patches today
09:08:31 <dimak> I algo got rid of that split l2 table patch
09:08:33 <oanson> Cool.
09:08:45 <oanson> Great. Happy to hear it
09:08:52 <dimak> And I won't need to touch tunnel/provider apps
09:09:19 <dimak> I also put in some hours on DNAT effort
09:09:23 <oanson> You mean for SFC
09:09:31 <dimak> Yes
09:09:35 <oanson> I'm guessing for dnat you will need to touch provider
09:09:41 <dimak> yes
09:09:59 <dimak> Found several things that needed fixing
09:10:41 <dimak> I'll break them out of the dnat patch chain today or tomorrow
09:10:50 <dimak> To make the merging easier :)
09:11:36 <oanson> Yes. That would be great.
09:11:36 <dimak> itamaro, I also had some changes in provider app, will be happy if you could take a look
09:11:47 <itamaro> sure
09:12:03 <dimak> thanks
09:12:43 <oanson> Anything else for SFC?
09:12:59 <itamaro> I have an updated change in the gen_mapping which I would like u to have  a look.
09:13:13 <oanson> itamaro, linky?
09:13:34 <itamaro> it will uploaded after tests today.
09:13:34 <dimak> https://review.openstack.org/#/c/475166/
09:13:45 <dimak> I have posted a comment there
09:14:11 <oanson> All right. Let's get back to this one in the open discussion section
09:14:17 <dimak> I think having 2 patches is preferable in case of snat
09:14:23 <dimak> sure
09:15:23 <oanson> All right. LBaaS
09:15:53 <oanson> I'm uploading this instance a spec
09:16:03 <oanson> It is *very* much WIP
09:16:08 <oanson> But I wanted to show some progress
09:16:14 <oanson> https://review.openstack.org/477463
09:16:53 <oanson> There are a bunch of things I am not happy about yet, and I hope for next week I can resolve them
09:17:24 <oanson> L3 flavour
09:17:26 <oanson> https://review.openstack.org/#/c/475174/
09:17:45 <dimak> I'll take a look today
09:17:50 <oanson> The spec breaks the gate... So it won't be merged
09:17:54 <dimak> (the lbaas)
09:17:55 <oanson> dimak, I wouldn't bother.
09:18:06 <dimak> ok then
09:18:13 <oanson> It's there to show that I am not ignoring it. But it's very WIP
09:18:31 <lihi> I'll take a look
09:18:43 <dimak> the l3 flavor spec didn't get much attention
09:18:59 <oanson> Yes. Sorry about that
09:19:23 <oanson> I'll try to get to it this week
09:19:36 <oanson> ETCD publisher
09:19:39 <oanson> lihi, ?
09:20:24 <lihi> Not much progress. Only yesterday (EOD) I managed to get a working env with etcd
09:20:55 <oanson> All right. Please also review https://review.openstack.org/#/c/476401/
09:21:02 <lihi> sure
09:21:21 <oanson> RPM packaging - dimak I know we haven't mentioned this in a while
09:21:33 <dimak> Yes I havent touched it since :(
09:21:50 <oanson> But I seem to remember you volunteered, and if I recall, it isn't a big undertaking
09:21:54 <oanson> (Unless I'm wrong)
09:21:55 <dimak> juggling too many things
09:21:59 <oanson> All right.
09:22:14 <dimak> I don't think its a huge endeavor
09:22:24 <dimak> btw we should also take a look at df kolla images
09:22:27 <oanson> Anyone wants to take RPM packaging? I don't think it should take long, and I think it will have a good impact on PR
09:22:34 <dimak> and update them if needed
09:22:40 <oanson> Yes
09:22:48 <oanson> But we're getting spread a bit thin
09:23:08 <dimak> I think we need to add bgp service there (iirc)
09:23:10 <oanson> Why don't I take the RPM package, and lihi, can you look at the kolla deployment in parallel?
09:23:24 <oanson> Let's start with setting it up and testing it.
09:23:54 <oanson> I'll take this time to add that I think we have 3 gate jobs that are important to add: OSA, kolla, and devstack/multinode
09:24:19 <dimak> Do we need both OSA and Kolla?
09:24:28 <oanson> OSA gate already exists, but it will probably remain unstable until lihi finishes the etcd publisher work (which is why I think it's so important)
09:24:37 <lihi> Sure
09:24:42 <oanson> dimak, because both deployments are already written, and we want to support as much as possible
09:24:55 <dimak> More gates, more breakage ;)
09:25:06 <oanson> non-voting gates, obviously
09:25:18 <oanson> and so far the etcd gate seems to be worth its bread
09:25:24 <dimak> How long does it take to get through DF in OSA?
09:25:58 <dimak> to gate*
09:26:49 <oanson> Dunno. The gate currently fails after 4-5 minutes
09:27:04 <dimak> Oh
09:27:06 <dimak> a shame
09:27:37 <oanson> Yeah.
09:28:19 <oanson> Once the etcd publisher is finished, I think we can make it much more stable. Most of the issues were that the ZMQ publisher wasn't able to guess its container's IP.
09:28:57 <dimak> We plan to drop zmq after we have etcd publisher?
09:29:12 <oanson> No
09:29:30 <oanson> We plan to support both. ZMQ might be better suiting in some scenarios
09:29:42 <oanson> Since we're driver-based
09:30:07 <oanson> A deployment can choose what fits them best
09:30:08 <dimak> +1
09:31:08 <oanson> All right. In that case, dimak, I'm guessing we're also waiting with Skydive?
09:31:17 <dimak> yeah
09:31:48 <oanson> Anything else for roadmap?
09:32:19 <oanson> #topic Bugs
09:32:30 <oanson> I am looking at bug 1690775
09:32:31 <openstack> bug 1690775 in DragonFlow "Remove special handling for lport/ofport in local controller" [High,In progress] https://launchpad.net/bugs/1690775 - Assigned to Omer Anson (omer-anson)
09:33:02 <oanson> I am looking at getting the NbAPI/controller to fire events for all dependent instances.
09:33:35 <oanson> This came up because in classifier, I need both lport and ovs port. So I need to write a mechanism to make classifier app wait for both events
09:33:40 <oanson> But this is something reusable.
09:33:47 <dimak> Yes, it could save us .get_object() is not None checks
09:33:50 <oanson> So I'm thinking to let the framework/IS do it
09:34:16 <oanson> I have an early draft written. Once it passes unit tests I'll upload it to see what the gate thinks of it
09:34:29 <oanson> (Assuming it will be non-broken for an hour)
09:34:41 <oanson> Anything else for bugs?
09:35:08 <dimak> We don't have a bug for that
09:35:21 <oanson> Sorry?
09:35:29 <dimak> but we need to put some effort into the topology/refresher/consistency mechanism
09:35:39 <dimak> it doens't cover all models
09:35:51 <dimak> only ones that come with topic (and version for consistency)
09:36:13 <dimak> so topicless objects like listeners/chassis are never refreshed
09:37:13 <oanson> Yes
09:37:21 <oanson> This is a long-standing problem
09:37:44 <dimak> I also opened another bug a few weeks ago
09:37:52 <dimak> about db drivers ignoring topic
09:37:56 <dimak> (except for redis)
09:38:09 <oanson> And redis is doing it badly
09:38:10 <dimak> which makes selective topology distribution not so selective
09:38:20 <dimak> yes redis is far from optimal
09:38:24 <oanson> dimak, are there bugs for all these issues?
09:39:00 <oanson> Let's open bugs on these. The redis mis-implementation and db drivers ignoring topics is a medium, since its a performance issue.
09:39:11 <dimak> I have a bug on the latter
09:39:23 <oanson> The topology thing is a High. It would also do well to organise all our thoughts there
09:39:54 <dimak> https://bugs.launchpad.net/dragonflow/+bug/1696142
09:39:55 <openstack> Launchpad bug 1696142 in DragonFlow "Topic is ignored in NbApi.get_all" [Undecided,New]
09:40:42 <oanson> All right. Could you also open bugs for the other two?
09:40:44 <dimak> BTW just came across https://bugs.launchpad.net/dragonflow/+bug/1697439
09:40:44 <openstack> Launchpad bug 1697439 in DragonFlow "router_port_rarp_cache and floatingip_rarp_cache dictionaries consider just mac address as key " [High,New]
09:40:51 <dimak> I'll take as I already fixed it
09:40:55 <dimak> in the DNAT patches
09:41:04 <dimak> (the l3 part is not uploaded yet but done)
09:41:16 <oanson> Sounds good
09:41:39 <dimak> I'll open bugs for the sync/topology
09:42:02 <oanson> Great. Thanks!
09:42:05 <oanson> Anything else for bugs?
09:42:45 <oanson> #topic Open Discussion
09:42:51 <oanson> itamaro, you had something to say here?
09:43:46 <itamaro> there is  a newly discoved bug related to bum flows
09:43:57 <oanson> Yes
09:44:01 <itamaro> https://bugs.launchpad.net/dragonflow/+bug/1700473
09:44:03 <openstack> Launchpad bug 1700473 in DragonFlow "bum flow on tunnel vtp is limited to 989 replications" [Undecided,New]
09:44:19 <itamaro> which need some thinking
09:44:29 <dimak> You can use groups to overcome it I think
09:44:59 <dimak> Check the max number of buckets OVS will treat
09:45:42 <itamaro> ok. lets take it offline...
09:45:55 <dimak> 👍
09:45:58 <oanson> That's actually a cool solution. I like it!
09:46:40 <oanson> The floor is free?
09:46:54 <lihi> I think https://bugs.launchpad.net/dragonflow/+bug/1647362 now deserves a lower priority. Does it blocks anyone?
09:46:55 <openstack> Launchpad bug 1647362 in DragonFlow "Fullstack tests does not simulate actual networks settings" [High,In progress] - Assigned to Lihi Wishnitzer (lihiwish)
09:47:29 <oanson> Bumped down to medium. It blocked me with trunk ports, but that's solved now
09:48:11 <oanson> Anything else?
09:48:55 <oanson> All right. Thanks everyone
09:49:02 <oanson> Come again next week.
09:49:09 <oanson> #endmeeting