17:15:06 #startmeeting ovn_community_development_meeting 17:15:07 Meeting started Thu Mar 18 17:15:06 2021 UTC and is due to finish in 60 minutes. The chair is mmichelson. Information about MeetBot at http://wiki.debian.org/MeetBot. 17:15:08 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 17:15:11 The meeting name has been set to 'ovn_community_development_meeting' 17:15:23 It's time once again for the weekly OVN meeting. 17:15:57 Thanks to everyone who helped out, we released OVN 21.03.0 last week. I have a PR up for the ovn-website that adds the release and release notes. I don't think it's been merged yet though. 17:16:49 I've mentioned in here a couple of times work I was doing regarding an oddball floating IP scenario that was not working for OpenStack. numans and blp gave reviews to my two-patch series. numans pointed out there's a failing test, and I'm working on getting that fixed now. 17:17:13 I hope to have a new version of that up before the end of the day but might have to wait until tomorrow if it's tougher to fix than I had expected. 17:17:18 Most of my reviews these days are about ddlog though. 17:17:31 blp, yeah, in my case you just ensured the C and DDLog were equivalent. 17:17:36 Right... 17:18:19 We have a bit of a problem with tests being inconsistent in whether they pass or fail. blp has done a good job during DDLog development of fixing up some of them. 17:18:33 I've got some more of those coming up. 17:18:56 I'm planning to get some issues filed on redhat's bugzilla for each of the "flaky" tests so we can get to the bottom of them. We can determine if the tests are flaky or if OVN is flaky. 17:18:59 blp, oh nice. 17:19:04 blp, there may be some overlap there then. 17:19:18 If we can't trust CI to be correct, then it's doing us no good. 17:19:43 So we need to ensure that a red run actually means there is a problem, and not get in the habit of constantly rechecking or ignoring failures. 17:19:58 Thanks to imaximets for fixing the weirdness with flake8 checking 17:20:05 That's a big help 17:20:16 And I think that's all I had wanted to mention. 17:20:27 Whoever wants to go next, feel free. 17:20:42 I know we're unlikely to hear from dceara since this is a bad time for him, and _lore_ is on PTO today. 17:21:15 I have a small update. Hi. 17:22:11 I tried to review ovsdb-idl fixes from dceara. It's on v3 now. I have couple of comments to this patch-set. Will send them shortly. 17:22:52 Otherwise, worked a bit on GHA with its wierd python configurations. 17:23:16 that's it from me. 17:23:46 I have an update. 17:24:28 I received some assistnace from Leonid on optimizing the ddlog implementaiton of northd. 17:24:42 We looked at the benchmark from Numan. 17:24:49 I did some work on the benchmark itself 17:24:52 and some work on ovn-sbctl 17:25:09 And managed to make the benchmark run several times faster without ddlog 17:25:20 and something like 3x faster than that with ddlog 17:25:41 Now the timing on my laptop is something like 115 seconds without ddlog and 35 seconds with it. 17:25:59 blp, what sort of changes did you make in ovn-sbctl? 17:26:03 I haven't managed to fit together the full patch series yet, but that will be coming up when I have some actual time outside of meetings. 17:26:10 I added the same daemon feature that ovn-nbctl has. 17:26:25 and made the benchmark use the daemon for both utilities. 17:27:08 blp, interesting. I thought that since ovn-sbctl was selective with which tables/columns it was interested in, it didn't have the same scaling issues as ovn-nbctl. Or at least the scaling issues wouldn't be nearly as dire. It's interesting to learn this. 17:27:21 It made a huge difference. 17:27:45 Every iteration in the benchmark runs ovn-nbctl once and ovn-sbctl twice. The cost of ovn-sbctl overwhlemed the rest of the test. 17:28:08 Well, in the fomr that Numan sent it, the benchmark ran ovn-nbctl dozens of times, but it didn't need to, so I merged into just one. 17:29:43 blp, that is some good news. You mentioned during my update that you had more test fixes coming up. Are those related to your ddlog optimizations or are they a separate patch series? 17:30:11 They're in my series. They aren't huge things, just minor improvements to tests. 17:30:41 blp, ack 17:30:46 I get lots of test failures while checking ddlog, and I try to eliminate them when they're races. 17:30:53 (usually, they are races) 17:32:08 blp, I suspect that's the case for most of the tests. 17:33:12 I would welcome more challenge benchmarks if anyone wants to send me them. 17:33:36 I guess, daemon mode for sbctl was not implemented just because users normally doesn't need it. 17:33:47 It doesn't cost us much. 17:33:56 It's used way less frequently than nbctl 17:34:16 My most typical use of sbctl is to inspect logical flows 17:34:58 Anyone else present? Anyone else want to give a report? 17:35:07 Currently my daemon mode is cut-and-paste from ovn-nbctl, but I might work to factor it into common code. 17:35:18 blp, that would probably be for the best 17:35:25 (factoring, I mean) 17:35:40 * blp nods 17:36:03 That's my whole report. 17:37:25 OK, last chance for someone to speak up. Otherwise this meeting is OVER 17:37:36 I certainly don't mind a short meeting :) 17:37:55 SHORT MEETING FTW 17:38:15 Hi folks! I have a --kind of a-- question for the IP gurus (like zhouhan). :) 17:38:15 I seem to have hit a bug where rate in the meter band gets updated in the NB+SB, but 17:38:15 ovn-controller is failing to act on it, even when calling "ovn-appctl -t ovn-controller recompute". 17:38:15 ref link: https://bugzilla.redhat.com/show_bug.cgi?id=1939524 17:38:17 Is there anything special when it comes to meter tables for the incremental processing logic? 17:38:17 No worries if you don't know from the top of your head. Just something to be worked on I 17:38:17 bugzilla.redhat.com bug 1939524 in ovn2.13 "Changing rate on meter does not take effect unless ovn-controller is restarted" [High,New] - Assigned to ovnteam 17:38:17 wanted to make you aware of. 17:39:03 flaviof: I can look at it 17:39:10 flaviof: Does it work if you restart ovn-controller? I mean, is it a general bug as opposed to a I-changed-something-and-it-didn't-update bug? 17:39:32 zhouhan++ yes, it works fine if I restart the ovn-controller 17:40:27 flaviof: not sure what's wrong. It is either a missing dependency or something more subtle 17:40:42 zhouhan: ack. 17:40:57 flaviof: let met check more details and come back 17:41:44 blp: thanks for the daemon mode work for sbctl, I think it is quite useful, especially for some operational tasks 17:42:47 I don't have anything to update. I will get more time on the review next week 17:43:43 zhouhan: Glad someone has a positive view :-) 17:44:18 blp no lack of appreciation for what you do, here! 17:44:52 All right everyone. I guess that's all for today. Have a nice day/weekend 17:44:56 #endmeeting