[11:20:08] lunch [12:29:52] o/ dcausse: my test-flink.sh fails because of Caused by: java.lang.NoSuchMethodError: 'org.apache.flink.streaming.api.datastream.DataStream org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator.getSideOutput(org.apache.flink.util.OutputTag)' at org.wikimedia.discovery.cirrus.updater.producer.graph.ProducerGraphFactory.createStreamGraph(ProducerGraphFactory.java:191) - I tried to reproduce that locally by [12:29:52] using flink 1.16.0 (the version provided by docker-registry.wikimedia.org/flink:1.16.0-wmf5-20230326 but that does not seem to be the reason [12:30:22] Does that ring a bell? [12:34:11] Another problem I run into is, that the kafka brokers cannot be resolved from within docker buildx/blubber/gitlab ci. So I don’t know how much more effort I should put into that kind of integration test. [12:41:23] pfischer: I don't think we should do much testing at image building, just a quick smoke test to just see that flink can start [12:42:11] the org.apache.flink.streaming.api.datastream.SingleOutputStreamOperator.getSideOutput is interesting... [12:42:34] sounds like a mixed version (compile time vs runtime) [12:44:16] or perhaps we embark flink-streaming-java in the fat jar? [12:44:36] flink-streaming-java should be provided I believe [12:51:55] the test-flink.sh I wrote does not even start flink really, it just builds the stream graph and exit [12:52:24] but I guess that's sufficient enough to capture small dependencies issues [13:10:20] flink-streaming-java is not part of either of the jar(s)-with-dependencies [13:11:02] I’ll add the build-only-the-graph flag and see what happens. [13:18:01] o/ [13:20:06] inflatador: did you publish an image with flink 1.16.1? [13:23:15] dcausse I thought I did? Did it not show up? [13:23:38] inflatador: can't find it in https://docker-registry.wikimedia.org/flink/tags/ [13:25:09] build logs says: `2023-04-12 16:22:37,129 [docker-pkg-build] INFO - Successfully tagged docker-registry.discovery.wmnet/flink:1.16.0-wmf6` [13:25:21] Maybe I should've changed a tag somewhere? [13:25:24] ah ok [13:25:38] yes should have been flink:1.16.1-wmf0 [13:25:42] cf the changelog [13:26:41] https://gerrit.wikimedia.org/r/c/operations/docker-images/production-images/+/908256/1/images/flink/flink/changelog [13:27:09] I can push a new patch if you like [13:27:57] this is this patch I think: https://gerrit.wikimedia.org/r/c/operations/docker-images/production-images/+/908486 [13:29:19] Interesting, the git log does not have that entry. Let me try that again [13:32:10] perhaps this repo, unlike deployment-prep, needs an explicit fetch+rebase? [13:32:24] It looks like one of those git directory errors [13:33:01] it's refusing to update git on the build server [13:33:08] fatal: detected dubious ownership in repository at '/srv/images/production-images' [13:34:03] this issue rings a bell [13:34:24] ottomata: ^ [13:34:35] yeah, it was causing problems before...I can still do a `git pull` as root though...one sec [13:35:39] OK, looks like it's building. The directions at https://wikitech.wikimedia.org/wiki/Kubernetes/Images#Production_images don't work; you have to be root as opposed to using sudo [13:43:04] dcausse OK, the image is built. The docker-registry web UI doesn't show it yet though. let's give it a few minutes [13:43:24] inflatador: thanks! [13:53:18] errand [14:55:06] \o [14:56:08] o/ [15:57:43] inflatador: ya sudo -i, if instructions don't say that, can you fix? [16:21:04] ottomata will do, was waiting to see if it worked first [16:24:43] working out first though ;) . Back in ~40 [16:48:44] \o [16:54:01] back [18:56:04] errand, back in ~30 [19:35:43] back [20:36:23] huh, unexpected...it works. 41 passed, 1 skipped, 42 total (100% completed) in 00:09:46 [20:36:41] (it actually takes 14m, including setting up the environment) [20:43:21] Good news!