2020-10-27 13:49:15 UTC - Lari Hotari: The flakiness problem is really bad. It's hard to get a pass. It takes 5-6 attempts in many cases to get a PR to pass. ---- 2020-10-27 13:49:41 UTC - Lari Hotari: I created an issue for InactiveTopicDeleteTest: <https://github.com/apache/pulsar/issues/8390> . It's flaky. ---- 2020-10-27 13:53:28 UTC - Lari Hotari: After <https://github.com/apache/pulsar/pull/8386> gets merged (which should fix the maven download issue), it would make sense to start addressing the most common test failures asap. A lot of the common failures can be found by simply going to <https://github.com/apache/pulsar/actions?query=is%3Afailure> . I would say that over 50% of failures across all PR builds are currently caused by flaky tests, so it's easy to find things to fix if one wants to contribute in the effort. Please contribute! ---- 2020-10-27 14:07:32 UTC - Lari Hotari: there's a build queue building up. 53 items atm in <https://github.com/apache/pulsar/actions?query=is%3Aqueued> ---- 2020-10-27 14:11:15 UTC - Lari Hotari: when someone pushes a lot of changes to a PR, it overloads the CI pipeline. I can see lots of in progress items, <https://github.com/apache/pulsar/actions?query=is%3Ain_progress> . 698 items. 181 items are from a single user: <https://github.com/apache/pulsar/actions?query=is%3Ain_progress+actor%3Acongbobo184> . @CongBo is it from you? ---- 2020-10-27 14:13:12 UTC - Lari Hotari: @CongBo Since the Pulsar CI is very flaky at the moment, a lot of builds might fail because of flaky tests. Would it be possible to minimize the amount of changes pushed to the PR until the CI is in a better shape? Every time a commit is pushed to a PR branch, new build jobs will spin off. That's the reason to minimize the number of times you make changes. ---- 2020-10-27 14:14:02 UTC - Penghui Li: @Lari Hotari <https://github.com/apache/pulsar/pull/8391> I have pushed a PR to fix the flaky test in the InactiveTopicTest ---- 2020-10-27 14:14:23 UTC - Lari Hotari: Cool, thanks @Penghui Li ---- 2020-10-27 14:22:41 UTC - Alexandre DUVAL: Hi, for <https://github.com/apache/pulsar/pull/7523#discussion_r483377081> I don't know how to retrieve tenantInfo and check isTenantAdmin as default. WDYT? ---- 2020-10-27 16:24:38 UTC - Sijie Guo: @jia zhai ^ ---- 2020-10-27 16:50:39 UTC - Lari Hotari: one thing that I was wondering about the Github workflows is that the fetch-depth is set to 0 which "indicates all history for all branches and tags". This disables a "shallow clone". Why is that setting used? It would be more efficient to use a small number. The default for the Github checkout action is 1 (docs at <https://github.com/marketplace/actions/checkout>). If a few recent commits are needed, I'd suggest setting the number to something limited, for example 50. If it's fine to have just the most recent commit, 1 should be fine. ---- 2020-10-27 16:55:20 UTC - Lari Hotari: It looks like <https://github.com/apache/pulsar-test-infra/blob/master/diff-only/entrypoint.sh> needs at least the commits of the PR. Setting fetch depth to some reasonable limit, let's say 25 should be fine? ---- 2020-10-27 17:08:08 UTC - Lari Hotari: it might be possible to use `${{ github.event.pull_request.commits }}` as the value for fetch-depth ---- 2020-10-27 17:12:24 UTC - Lari Hotari: Actually, fetch depth would have to be at least github.event.pull_request.commits + 1. I'm not sure if it's possible to add 1 to the returned value or not. ---- 2020-10-27 17:13:03 UTC - Lari Hotari: It might be easier to just use a number that is the maximum for any PR. Let's say 25 or 50. ---- 2020-10-27 22:24:48 UTC - Lari Hotari: I have created PR <https://github.com/apache/pulsar/pull/8393> to mitigate the issue the CI gets overloaded when a single PR gets frequent changes. Currently this leads to a new build being queued for each change. This PR will cancel the previous jobs when a new change is introduced on the PR branch. +1 : Sijie Guo ---- 2020-10-27 22:53:24 UTC - Lari Hotari: Shallow clone is 15-35 *times* faster than a full clone in Github actions for the Pulsar repository. I pushed a PR for this <https://github.com/apache/pulsar/pull/8394> ---- 2020-10-27 23:11:50 UTC - Addison Higham: those both look great, we will get some eyes on them soon, @Penghui Li will be online in a little bit +1 : Lari Hotari ---- 2020-10-28 01:53:06 UTC - jia zhai: we could consider that in the future. current way is to align with the old manner. usually user will provide their own security jars, such as BouncyCastle, and non-fips version maybe used in a more wide range. ---- 2020-10-28 06:39:59 UTC - Shahar Fermon: @Shahar Fermon has left the channel ----
