18:16:14 <mmichelson> #startmeeting ovn_community_development_discussion
18:16:15 <openstack> Meeting started Thu Jan  7 18:16:14 2021 UTC and is due to finish in 60 minutes.  The chair is mmichelson. Information about MeetBot at http://wiki.debian.org/MeetBot.
18:16:16 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
18:16:19 <openstack> The meeting name has been set to 'ovn_community_development_discussion'
18:16:57 <mmichelson> So is anyone else here for the OVN meeting?
18:17:05 <blp> hi
18:17:09 <mmichelson> hi blp
18:17:25 <mmichelson> blp, I just started the meeting. Not sure who all is actually here this week
18:17:45 <blp> I've been cocooned in staycation mode for a couple of weeks. Now I have emerged as a beautiful butterfly and ready to do some work.
18:17:52 <mmichelson> yay!
18:18:17 <blp> I found the linking bug that some people were experiencing trying to build ovn+ddlog. That will be fixed in the next version.
18:18:38 <blp> I need to do a new rev of that. I'll post it when it's ready.
18:18:42 <mmichelson> blp, what's the status of the IDL patches for OVS that DDLog is dependent on that?
18:18:50 <mmichelson> s/that//
18:19:08 <blp> mmichelson: I posted v4. Haven't seen any reviews.
18:19:12 <_lore_> hi all
18:19:25 <blp> mmichelson: I'll rebase it against master to make sure that no changes are needed.
18:19:31 <mmichelson> blp, OK cool.
18:19:44 <blp> and if any are, I'll post a v5.
18:19:49 <mmichelson> blp, When the new version of the ddlog patches are posted, I'm tempted to say we should get them merged as soon as we can
18:20:08 <mmichelson> But if the IDL split patches are also needed, then it seems silly to merge the DDLog patches too
18:20:17 <blp> The IDL split is a prereq.
18:20:24 <mmichelson> ok, that's what I thought
18:20:56 <blp> (It wasn't originally, but I got feedback that there was too much code duplication without the split.)
18:21:06 <mmichelson> blp, okie-dokie
18:21:07 <blp> (which was fair)
18:21:14 <imaximets> blp, mmichelson: I'll take another look at IDL split v4 tomorrow or early next week.
18:21:20 <mmichelson> imaximets, excellent
18:21:24 <blp> imaximets: thanks!
18:21:36 <blp> I'm a little shaken today with political events from yesterday.
18:21:40 <mmichelson> We're about 4 weeks away from the proposed OVN soft freeze date for 21.03
18:21:50 <mmichelson> So getting more time with DDLog baked in is a good thing
18:22:25 <blp> mmichelson: yes
18:23:27 <mmichelson> On my side of things, I've been working up a document that discusses pain points and possible solutions for ovn-controller code.
18:23:40 <mmichelson> At this point, the document isn't ready to be shared publicly
18:23:56 <blp> mmichelson: OK.
18:24:10 <blp> mmichelson: I've found the ovn-controller code pretty impenetrable lately.
18:24:12 <mmichelson> But the gist of the doc is that ovn-controller has reached a point where it is becoming more and more common that we try to add something new but end up breaking something as a result
18:24:19 <blp> (partly my fault of course)
18:24:37 <mmichelson> blp, it's all our faults
18:25:28 <mmichelson> I was kind of hoping zhouhan, blp, and others might be able to chime in with quick high-level ideas for what y'all think are the sticking points in ovn-controller, and if you have ideas for how to fix them.
18:25:47 <mmichelson> I'd like to ensure that whatever document I publish takes into account the viewpoints of everyone and not just Red Hat people.
18:26:24 <blp> Well, ultimately I think ddlog might help? It's very good at making incremental computation easily understood.
18:27:47 <zhouhan> mmichelson: The lately problem was mostly related to the changes related to change-handler of local interface and port-bindings. There were some temporary code which numans planned to rework. numans also mentioned that's the next thing he will work on.
18:28:24 <zhouhan> blp: in general I believe ddlog is the better approach
18:29:51 <zhouhan> blp: I wonder if it is even better for ovn-controller to directly consume the NB DB, skipping the logical flow stage. Would that result in simpler implementation?
18:30:28 <mmichelson> Hm
18:30:43 <zhouhan> (of course this is a bigger architecture change)
18:31:24 <zhouhan> but in reality maybe it is not harder than converting existing ovn-controller to ddlog.
18:31:42 <blp> zhouhan: It's an option! I would like to think that doing the nb->sb transformation in one place does a lot of work in one place that otherwise would have to happen in many places.
18:31:56 <blp> zhouhan: I don't have measurements to back that up.
18:31:58 <mmichelson> My initial thought was that a lot of the work would be duplicated if ovn-controller is handling NB directly
18:32:15 <blp> mmichelson: that's what i mean, yes
18:32:16 <mmichelson> But the thing is, if logical flows aren't being generated
18:32:26 <mmichelson> Then that could mean that the duplicated work isn't happening
18:32:53 <blp> yes, that's possible too
18:35:18 <mmichelson> Anyway, nice to hear some input
18:35:18 <zhouhan> I would like to try this with ddlog, but I will still need to learn ddlog first :)
18:35:32 <mmichelson> I should have a document ready for publication next week
18:35:43 <blp> mmichelson: great!
18:35:52 <zhouhan> I guess most part of the current northd ddlog would be reused for that.
18:35:54 <mmichelson> I probably won't be suggesting DDLog as the solution, BUT, what I suggest may pave the way for DDLog to slot in more easily at some point?
18:36:38 <blp> mmichelson: makes sense
18:36:48 <mmichelson> Anyway that's all from me
18:38:03 <_lore_> can I go next? very fast
18:38:47 <_lore_> this week I worked on v8 of bfd support for ovn (posted upstream)
18:39:02 <_lore_> moreover I posted some ovn-trace fixes
18:41:59 <blp> _lore_: all done?
18:42:57 <blp> Anyone else?
18:43:14 <imaximets> Small update from me
18:43:14 <_lore_> blp: yes sorry
18:43:22 <_lore_> thx
18:43:25 <blp> _lore_: np
18:44:16 <imaximets> For the 'diff's as a file transactions in ovsdb:  I'm going to add "downgrade" related documentation and send v2 somewhere soon.
18:45:33 <blp> imaximets: great! This change has great potential.
18:46:20 <imaximets> And OVS is in a soft-freeze now, but I think we still could accept IDL split and ovsdb "diff"s since those were discussed and reviewed mostly.
18:46:40 <imaximets> That's it from me.
18:46:48 <blp> I think IDL split is ready; it just needs a final look from someone.
18:47:03 <imaximets> blp, sure.
18:51:12 <blp> Are we done for today?
18:51:50 <mmichelson> I suppose so?
18:52:00 <mmichelson> All right thanks everyone
18:52:04 <mmichelson> #endmeeting