17:17:08 <mmichelson> #startmeeting ovn_community_development_discussion 17:17:09 <openstack> Meeting started Thu Oct 8 17:17:08 2020 UTC and is due to finish in 60 minutes. The chair is mmichelson. Information about MeetBot at http://wiki.debian.org/MeetBot. 17:17:10 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 17:17:13 <openstack> The meeting name has been set to 'ovn_community_development_discussion' 17:17:25 <numans> Hello 17:17:40 <mmichelson> Hi numans 17:18:18 <mmichelson> I figure I can go first. 17:18:34 <dceara> Hi 17:19:32 <mmichelson> I've been working on adding some unit test infrastructure to OVN. This way it's easier to test smaller units (e.g. functions). I've added a couple of small example unit tests and I plan to submit an RFC patch series before the end of the week. 17:19:54 <mmichelson> Other than that, I've been doing reviews. And that's all from me. 17:21:42 <numans> I guess I can go next. 17:22:09 <_lore_> hi all 17:22:27 <numans> I worked on the reject action and submitted the patches for review 17:22:47 <numans> this reduces a lot of logical flows now. 17:23:06 <numans> I have started working on reducing the load balancer hairpin lflows. 17:23:24 <numans> I'm hoping I'll submit the patches for review early next week. 17:23:43 <numans> I couldn't do much review this week. 17:23:47 <numans> That's it from me. 17:24:01 <dceara> numans: Thanks for taking care of the hairpin lflows, I wanted to have a go at it for a while but didn't get a chance to. 17:24:16 <dceara> numans: Looking forward to the patches :) 17:24:20 <numans> dceara, welcome 17:24:40 <numans> I was able to reduce 50% of the patches. 17:25:04 <numans> If ovs could support something like - ct(commit, nat(dst=reg0)) that would have been great. 17:25:10 <numans> but looks like it doesn't support. 17:25:23 <numans> because of which we still need to add an lflow to do ct_snat for each vip. 17:25:27 <zhouhan> numans: 50% of the patches? or you meant flows :) 17:25:36 <numans> zhouhan, :) 17:25:38 <numans> s/patches/lflows. 17:25:50 <numans> my bad :) 17:25:51 <zhouhan> although 50% patches would be awesome, too 17:26:13 <numans> zhouhan, I was talking about the hairpin load balancer lflows. 17:26:16 <mmichelson> numans, why doesn't that ct() action work? Does nat not accept a register as an rvalue? 17:26:34 <numans> zhouhan, FYI - I've submitted the reject action patch series. 17:26:44 <numans> mmichelson, I didn't support that way. 17:27:05 <zhouhan> numans: thanks! It is in my review list 17:27:07 <numans> May be we can enhance ovs for that. I'm planing to give a shot at that, if its possible. 17:27:43 <numans> not sure if imaximets is here.. do you think ovs supports something like - ct(nat(dst=reg0)) 17:27:53 <numans> since it supports the zone value from a register. 17:28:13 <imaximets> numans, I'm here, and I'm not sure. :) 17:28:26 <numans> imaximets, I tried it isn't supported. 17:28:42 <numans> imaximets, it does support ct(zone=NXM_NX_REG0,...) 17:29:03 <numans> so I wonder ovs could support something similar for nat too 17:29:26 <numans> Ok. that's it from me. If someone wants to go next. 17:29:50 <zhouhan> I can go quickly 17:31:03 <zhouhan> I did some reviews. Why reviewing dceara's patch for "persistent conflict flows", I figured out 2 problems of my code and I submitted the fixes. 17:31:34 <zhouhan> I am waiting for the review and also waiting for next version from dceara. 17:31:44 <zhouhan> That's it from me. 17:32:00 <dceara> zhouhan: I just reviewed the 2 patches today. I acked the first one but for the second one I'm a bit confused. 17:32:25 <dceara> zhouhan: It seems to me we're not changing behavior at all. I replied on the ML though. 17:32:53 <dceara> zhouhan: Re the "predictable conflict resolution", I'll try to send a patch soon. 17:32:59 <zhouhan> dceara: ok, thanks. I haven't checked ML yet. 17:33:02 <dceara> s/a patch/a new revision 17:33:17 <imaximets> numans, according to docs, nat src and dst only supports addresses and ports as a value. 17:33:54 <numans> imaximets, ack. 17:34:39 <imaximets> zhouhan, I looked through old patches once again and found this: https://patchwork.ozlabs.org/project/openvswitch/patch/1564100775-80657-1-git-send-email-hzhou8@ebay.com/ 17:36:57 <zhouhan> imaximets: oh, this was never applied? 17:37:04 <imaximets> zhouhan, nope 17:38:06 <zhouhan> imaximets: oh, that's why I kept wondering how would the bug reported by Mark happened again. I remembered that I fixed it long ago ... 17:38:48 <zhouhan> Mark = Mark Gray 17:38:59 <mmichelson> zhouhan, btw, I got word from Mark that he's planning to do a v2 next week. He couldn't make this meeting 17:39:36 <imaximets> zhouhan, I think you need to discuss with him, which approach is better. 17:40:08 <zhouhan> mmichelson: imaximets Ok, thanks for this information. I will check it again with him 17:42:13 <imaximets> btw, I do not have anything else for this meeting today. 17:42:21 <dceara> I can go next if that's OK. 17:43:00 <zhouhan> dceara: Mark Gray talked about "There is a failure on one ovn UT with this change. It gets resolved by the fix that Dumitru suggests." Do you know what's that? 17:43:38 <dceara> zhouhan: It's related to chassis/chassis_private. The simplest way to reproduce is to start an OVN sandbox and "ovn-sbctl destroy chassis ." 17:44:16 <dceara> zhouhan: With Mark's latest patch applied ^ 17:44:37 <zhouhan> dceara: ok, thanks, I will try it. What's the suggested fix then? 17:45:43 <zhouhan> I still doesn't understand "Essentially *_is_new() will never return true for rows inserted locally.", and how destroying a chassis would trigger the _is_new() call. 17:46:19 <zhouhan> Maybe you could reply in the email? 17:46:30 <dceara> zhouhan: Sure, I'll try to detail it there. 17:46:51 <zhouhan> thanks a lot dceara. (please continue your update) 17:48:41 <dceara> We had a bug reported by OpenStack related to IGMP Snoop and localnet. I sent the fix for it (http://patchwork.ozlabs.org/project/ovn/patch/1602164887-24755-1-git-send-email-dceara@redhat.com/) but while testing I realized there's another problem with IGMP Relay and localnet causing multicast packets to be duplicated sometimes. I don't know how to fix it yet or if it's possible to but I added a TODO entry and am looking forward to 17:48:41 <dceara> suggestions. :) 17:49:22 * numans waves and disappears 17:50:12 <dceara> And more of an OVS issue but spotted with an OVN deployment, there's a bug in the OVS kernel module when doing DNAT. I sent a patch for it on the netdev mailing list. The issue was seen very often with OVN-k8s. 17:50:23 <dceara> That's it from my side today, thanks! 17:51:15 <dceara> Oh, link for the conntrack issue: https://bugzilla.redhat.com/1877128 17:51:18 <openstack> bugzilla.redhat.com bug 1877128 in openvswitch2.13 "OVS complains Invalid Argument on TCP packets going into conntrack" [Urgent,Post] - Assigned to dceara 17:52:50 <zhouhan> dceara: thanks, could you share the link to the patch as well? 17:53:35 <dceara> zhouhan: Sure: https://patchwork.kernel.org/patch/11820857/ 17:55:00 <zhouhan> thanks dceara 17:57:42 <mmichelson> Anybody else going to give an update? 17:58:13 <_lore_> I can update very fast 17:58:37 <_lore_> I am working on adding renew/rebind msg support to IPv6 Prefix Delegation 17:59:01 <_lore_> I guess I will post v1 by the end of the week 18:00:42 <_lore_> that's all from my side 18:01:40 <mmichelson> OK, anyone else? 18:02:59 <mmichelson> All right everyone, thanks! Have a good day/afternoon/night! 18:03:31 <dceara> Bye! 18:03:34 <imaximets> Bye! 18:04:17 <mmichelson> #endmeeting