17:21:08 <blp> #startmeeting ovn_community_development_discussion
17:21:09 <imaximets> blp, I think so.
17:21:09 <openstack> Meeting started Thu Sep 24 17:21:08 2020 UTC and is due to finish in 60 minutes.  The chair is blp. Information about MeetBot at http://wiki.debian.org/MeetBot.
17:21:10 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
17:21:12 <openstack> The meeting name has been set to 'ovn_community_development_discussion'
17:21:25 <blp> I'll start with my progress report then.
17:21:36 <imaximets> ack
17:21:37 <blp> I am preparing to post the DDlog patches for OVN.
17:21:48 <blp> They will add ddlog support without removing the existing northd implementation.
17:22:15 <blp> It is valuable to have them in ovn because that makes it easier to maintain
17:22:26 <blp> and easier to start doing performance benchmarks and tuning.
17:22:34 <blp> and some deployments might find them useful.
17:22:52 <blp> It's taking me some time to do it because I need to crossport ~200 patches that have been committed in the meantime.
17:23:06 <blp> Most of them are trivial or require no work at all, but some are tricky and I'm spending most of the time on that.
17:23:19 <blp> When I'm done, and tests pass, I'll start posting patches for review.
17:23:36 <blp> That's my report. I can take questions or comments.
17:24:13 <imaximets> Good news.  Thanks!
17:24:48 <blp> Who is next? Ilya?
17:24:59 <imaximets> will make check test both implementations at once?
17:25:38 <blp> Yes, tests are paired; if you don't build ddlog, then the ddlog versions are skipped.
17:25:49 <imaximets> ok.
17:26:27 <imaximets> I have one one small update
17:26:35 <imaximets> *only one
17:27:21 <imaximets> It's kind of OVN related.  I'm trying to finish v2 of stream record/replay patch set and going to post it this week.
17:27:54 <blp> I might not know about this. What kind of streams are being recorded?
17:28:15 <imaximets> All streams.  tcp, unix, ssl.
17:28:39 <blp> Oh, for debugging purposes?
17:29:24 <imaximets> Yep. The idea is to record incoming ovsdb operations and replay them locally to reproduce issues or test performance of production usecases.
17:29:46 <blp> OK
17:29:59 <imaximets> The same could be applicable to C version of northd. or ovn-controller in the future.
17:30:51 <imaximets> will ddlog use C implementation of streams? or it will have some different implementation?
17:31:04 <blp> It's using the C jsonrpc code.
17:31:11 <blp> (which is built on C streams)
17:31:53 <imaximets> ack.  So it should be possible to use from ddlog version of northd too.
17:32:27 <imaximets> Here is the link for v1, if you're interested
17:32:31 <imaximets> #link https://patchwork.ozlabs.org/project/openvswitch/cover/20200630012435.205590-1-i.maximets@ovn.org/
17:34:02 <imaximets> for now it requires some tricks, i.e. disabling some randomness and also doesn't support clustered dbs, but these are points for future improvements.
17:34:21 <imaximets> That's it from my side.
17:34:33 <blp> OK!
17:34:48 <blp> Is anyone else here?
17:35:04 <imaximets> _lore_, are you still here?
17:35:17 <_lore_> yes
17:35:52 <_lore_> I have a very brief update, this week I worked mainly on a QoS regression introduced by I-P processing
17:35:58 <_lore_> posted v4 upstream
17:36:21 <_lore_> now I am working adding new option for lr-policy-add/delete
17:36:34 <_lore_> like --may-exist/--if-exist
17:37:50 <_lore_> I will post it today or tomorrow
17:38:24 <_lore_> btw imaximets: any new about the raft patch I posted some time ago?
17:39:02 <imaximets> _lore_, you mean extra information for cluster/show command?
17:39:10 <_lore_> imaximets: yes
17:39:32 <blp> That's good...
17:39:38 <blp> Anyone else today?
17:39:39 <imaximets> _lore_, sorry, I didn't look at it closely yet.
17:40:13 <imaximets> Maybe I could say a couple of words about what numans worked on.
17:40:36 <_lore_> https://patchwork.ozlabs.org/project/openvswitch/patch/95b2a271bc6f04b91bed76847b70359b4234eeef.1599127790.git.lorenzo.bianconi@redhat.com/
17:40:45 <_lore_> imaximets: ok, no worries
17:41:07 <imaximets> numans just posted a patch to avoid addition of invalid lflows for reject ACLs.
17:41:38 <imaximets> northd adds 4 flows unconditionally for each reject ACL
17:41:52 <imaximets> one for ip4, ip6, tcp and icmp
17:42:45 <imaximets> and if 'match' already contains some protocol e.g. ip4, half of these flows will be invalid as they will match on ip4 and ip6 at the same time
17:43:13 <imaximets> there are deployments where such flows takes significant part of SB db
17:43:36 <imaximets> e.g. 300-400 MB of invalid flows of total 1GB SB db.
17:44:29 <imaximets> Solution is that CMS could specify which protocols it targets, so northd will generate only relevant flows.
17:44:49 <imaximets> #link https://patchwork.ozlabs.org/project/ovn/patch/20200924164955.2733564-1-numans@ovn.org/
17:45:17 <blp> Sounds like it solves a real problem.
17:45:55 <zhouhan> imaximets: hi sorry I was in a conflict meeting
17:46:05 <imaximets> And also numans accepted a pile of refactorig patches for northd, to split up big functions.
17:46:18 <imaximets> That's all I know from his side.
17:47:13 <imaximets> zhouhan, do you have any updates?
17:47:37 <zhouhan> imaximets: not really much, but just code reviews. (I am done :)
17:48:20 <imaximets> zhouhan, any ddlog related questions while blp is here? :)
17:51:27 <zhouhan> imaximets: no questions. Sounds like a great progress. Thanks blp!
17:54:18 <imaximets> It seems like we could end this meeting now.
17:55:00 <blp> OK.
17:55:02 <blp> #endmeeting