17:22:49 <mmichelson> #startmeeting ovn_community_development_discussion
17:22:50 <openstack> Meeting started Thu Jul 23 17:22:49 2020 UTC and is due to finish in 60 minutes.  The chair is mmichelson. Information about MeetBot at http://wiki.debian.org/MeetBot.
17:22:51 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
17:22:53 <openstack> The meeting name has been set to 'ovn_community_development_discussion'
17:23:09 <mmichelson> Sorry for the late start. Things are suddenly a bit crazy today
17:23:20 <mmichelson> ovnkube CI is having problems with OVN 20.06
17:23:31 <mmichelson> Still gathering facts on the matter.
17:23:39 <mmichelson> But as for my activity the past week
17:24:06 <mmichelson> I put up a patch for lb_force_snat_ip and dnat_force_snat_ip to work for dual stack
17:24:17 <mmichelson> And I put up a patch for ECMP symmetric replies
17:24:33 <mmichelson> That ECMP patch needs fixing because there are some test failures that numans pointed out
17:25:17 <mmichelson> blp, I know you're mostly lurking, but something to bring to your attention is that OVN committers should probably be made maintainers in patchwork
17:25:56 <mmichelson> blp, if you have the proper contact info for getting that done, then OVN maintainers can update patch states when desired.
17:26:11 <blp> mmichelson: I don't think there is any self-service administration for patchwork, that is, I don't think that I can do that myself.
17:26:44 <mmichelson> blp, ok, imaximets said that you might know who to get in contact with though
17:26:50 <blp> mmichelson: In the past, I've always emailed one of the owners of the patchwork instance. My guess is that they'd take requests from others too, or I could send an email recommending that they take it from you.
17:27:21 <mmichelson> blp, OK. I'm willing to send the message myself. But not sure who to send to.
17:27:46 <mmichelson> (I'd CC the OVN committers)
17:27:49 <blp> Jeremy Kerr <jk@ozlabs.org> is who I
17:27:53 <blp> ve emailed in the past
17:28:06 <mmichelson> blp, OK thanks.
17:28:11 <mmichelson> And that's all I wanted to bring up.
17:28:35 <blp> OK. Do you want to start by contacting Jeremy? I'll back you up as needed. Or we can do it another way if you prefer.
17:30:20 <zhouhan> I can go next.
17:31:10 <zhouhan> I am catching up after the vacation. I just submitted the formal patch series for avoid ARP related flow explosion: #link https://patchwork.ozlabs.org/project/openvswitch/list/?series=191714
17:31:45 <zhouhan> As I saw the test report from Girish for the RFC was effective
17:32:21 <zhouhan> I will catch up with some reviews, first one would be the ECMP patch from mmichelson
17:32:29 <zhouhan> That's my update
17:34:02 <flaviof> Nice. Avoiding flow explosion is always a good thing. Let me jump next...
17:34:26 <flaviof> I have just one small announcement to make to folks who may know or use ovn-org/ovn-fake-multinode.
17:34:46 <flaviof> Thanks to Eelco, we now have the ability to build ovs/kernel and run test traffic. If that is something that interests you, take a look at:
17:34:54 <flaviof> #link https://github.com/ovn-org/ovn-fake-multinode/blob/master/README_DEVELOPMENT.md  Using ovn-fake-multinode for development and testing
17:35:19 <flaviof> Huge thanks to chaudron!
17:35:24 <flaviof> That is it from me.
17:37:28 <zhouhan> thanks flaviof. That's helpful, but I don't use vagrant. Does it make big difference?
17:38:04 <flaviof> oh not at all. you can use the scripts on any env. Vagrant just help getting the host to run the cluster.
17:39:22 <zhouhan> I have tried it before, which worked well. But I didn't try the "change and retest" process on that yet. Does it require rebuilding the docker images?
17:39:54 <flaviof> nope, which is a huge time saver. once you have the images you can restart very easily/fast
17:40:13 <zhouhan> flaviof: oh that's great!
17:42:37 <panda> oneline from me, I'm a but swamped at the moment with the rest, so even my OVN day was put on hold, not sure If I can do something tomorrow, but I'll try. I have two bugs on my list and one has a poste solution but needs improvement.
17:42:50 <panda> :( sorry
17:43:42 <mmichelson> panda, we've all been there.
17:44:07 <blp> panda: Have we met before? I know at least one Panda.
17:44:49 * numans_ joins now and says Hi to everyone
17:44:50 <panda> blp: there are not many pandas out there, but probably we haven't met.
17:45:19 <panda> numans: you can say you're early for next meeting.
17:45:21 <blp> I guess you are probably not Aurojit Panda formerly at Berkeley.
17:45:40 <zhouhan> (most pandas in the world are in my hometown :)
17:45:56 <blp> zhouhan: :-)
17:46:11 <panda> blp: nope, sorry to crush your expectation. I'm Gabriele Cerami when I'm not panda. I introduced myself in the mailing list a whil ago
17:46:20 <flaviof> zhouhan++
17:46:22 <blp> panda: OK, pleased to meet you
17:46:41 <panda> I was able to clear out some bugs from bugzilla, but they are activities that can go unnoticed sometimes
17:46:59 <panda> especially if they don't require a patch
17:47:02 <panda> blp: :)
17:47:24 <panda> anyway, that's it for my update.
17:47:39 <numans> I can go real quick
17:47:51 <flaviof> I second what mmichelson said, panda. no worries. Is the patch you are talking about this one:
17:47:55 <flaviof> #link https://patchwork.ozlabs.org/project/openvswitch/patch/20200710235515.GA998029@localhost/ ?
17:48:01 * flaviof sorry numans !
17:48:10 <numans> flaviof, no worries. I can go after  you
17:48:31 <mmichelson> numans, flaviof already went, actually :)
17:48:41 <flaviof> numans: go ahead, pls.
17:48:45 <panda> flaviof: yep, it is.
17:48:50 <numans> Ok. I submitted this small patch series to improve conntrack usage in OVN
17:48:53 <numans> #link https://patchwork.ozlabs.org/project/openvswitch/list/?series=191630
17:49:28 <numans> There were some performance comparisons between OVN kubernetes and the default openshift SDN with openshift deployment
17:49:40 <numans> and ovn deployment seemed to fair bad.
17:50:17 <numans> While debugging noticed that we are sending the packet to conntrack many times when a logical switch has ACLs with allow-related and load balancers
17:50:34 <numans> This patch series should help a bit.
17:50:56 <numans> I didn't get much time this week on reviews. I have started looking into mmichelson ECMP patches.
17:51:02 <numans> That's it from me.
17:51:39 <mmichelson> Thanks. Anybody else?
17:51:43 <flaviof> numans++ openshift SDN uses OVS directly?
17:51:56 <numans> flaviof, yes it does.
17:52:00 <flaviof> ack. ty
17:52:10 <numans> it also uses iptables via kube proxy for load balancing.
17:52:25 <zhouhan> numans: is the patch expected to bring it at the same performance level as the other one?
17:52:34 <numans> zhouhan, it should improve
17:53:09 <numans> zhouhan, with openshift SDN I see 2 recirculations. with this patch its just 1 recirculation for Pod to Pod traffic
17:53:09 <zhouhan> numans: ok, just want to confirm there is no other known gap
17:53:23 <numans> ok.
17:55:32 <numans> mmichelson, probably I'm the last one to update.
17:55:35 <mmichelson> numans, ok
17:55:45 <mmichelson> Well in that case, everyone have a good day/afternoon/evening
17:55:48 <mmichelson> Bye!
17:55:51 <numans> Bye
17:55:53 <zhouhan> bye
17:55:54 <blp> bye!
17:55:56 <mmichelson> #endmeeting