On Fri, 15 Jul 2022 at 07:30, Olivier Lamy <ol...@apache.org> wrote: > > > On Thu, 14 Jul 2022 at 19:25, Slawomir Jaranowski <s.jaranow...@gmail.com> > wrote: > >> I don't think that it is a code problem ... in the most times. >> >> eg: >> >> https://ci-maven.apache.org/blue/organizations/jenkins/Maven%2Fmaven-box%2Fmaven-plugin-tools/detail/PR-112/2/pipeline/ >> >> Build was failed on three window nodes with error: >> >> [INFO] Building: ignore-plugin-class-realm\pom.xml >> Sending interrupt signal to process >> Terminate batch job (Y/N)? >> script returned exit code 1 >> >> in the different places >> > >> What can be the reason, how to investigate it? >> > > > interesting and not sure exactly. > At least we can/could ask infra how windows nodes are provisioned, and > eventually look at the code. > As it's supported by our infra and open source code we still have the > opportunity to do that and eventually fix something to help the community. > but frankly you guys want to get rid of Jenkins. So I do not feel the > motivation to find some bandwidth for this now. > Maybe in a few weeks, I will have some time as Jenkins will be still there > and I can find a bit of time to look at those sometimes sporadic errors. > > I created this https://issues.apache.org/jira/browse/INFRA-23485. As you are looking a lot and have figures on this happening a lot etc... Maybe there is a pattern? is it happening all the time with invoker plugin when forking anothe mvn?
> > >> >> >> śr., 13 lip 2022 o 22:56 Olivier Lamy <ol...@apache.org> napisał(a): >> >> > On Thu, 14 Jul 2022 at 04:04, Slawomir Jaranowski < >> s.jaranow...@gmail.com> >> > wrote: >> > >> > > Hi, >> > > >> > > To summarize the discussion we have many votes to go this way. >> > > >> > > My proposition is as first step remove windows nodes from jenkins >> > builds, I >> > > see many fals positive fails on such node ... >> > > >> > >> > false positive or false negative? :) >> > >> > >> > > Next step can be stop building PR by jenkins - it is only triggered >> for >> > > dependabot and for branch from repo - not triggered for build PR for >> > forked >> > > repo. >> > > >> > >> > we can simply tell Jenkins to stop notify gh pr checks. As I presume the >> > goal is to not have Jenkins results within a fully oriented gh workflow. >> > Normally with this option build will still happen but not notify to GH >> PR >> > workflow so don't worry there will not be notifications. >> > Not sure if someone tries to find the reason for windows issues (could >> be >> > some files locking during the build which is obviously a bug in our >> code). >> > Perso I prefer the Jenkins ui especially when trying to find the reason >> for >> > a failed test result as results are collected and displayed in a more >> easy >> > to find stack. >> > >> > >> > >> > > >> > > sob., 2 lip 2022 o 21:42 Tamás Cservenák <ta...@cservenak.net> >> > napisał(a): >> > > >> > > > Howdy, >> > > > >> > > > I'd like to spin a discussion about the ASF Maven Jenkins instance. >> > > > >> > > > As you know, ASF Infra operates one separate instance of Jenkins >> only >> > for >> > > > Maven related projects. Still, aside this being a chore for INFRA, >> it >> > > does >> > > > have quite some shortcomings: >> > > > - lack of all needed OS-es (has linux and windows nodes only) >> > > > - regular (lately more often) issues with windows nodea (like post >> > build >> > > > workspace cleanup and others). But really, like 1 out of 3 fails >> due to >> > > > some windows node issue. >> > > > In short, it does not give us needed coverage, plus regularly >> generates >> > > > false negatives (red X) for PRs and master builds. >> > > > >> > > > OTOH, GH Actions proved very usable, quick, has macOS and in general >> > > fast. >> > > > Still, GH Actions cannot deploy snapshots to repository.a.o, nor is >> > > > something we'd want (see last npmjs token leakage). >> > > > >> > > > Currently we have this "duality" of CI, and almost always one has to >> > > check >> > > > why a job failed, and MANY times it fails due Jenkins non-build >> related >> > > > issues (usually on Windows nodes). >> > > > >> > > > Hence, I'd propose something along those lines: >> > > > - "scale down" Jenkins, keep linux nodes only, and make it "deploy" >> > only >> > > > (preferably of master branch commits only) >> > > > - hence Jenkins would "loose" CI title, it would be just "deployer" >> to >> > > > repository.a.o >> > > > - use GH Actions for running tests on PRs and master branches and >> rely >> > on >> > > > its results on GH UI >> > > > >> > > > This would give us (and ASF INFRA) benefit of: >> > > > - the maintained instance becomes way simpler, linux only (ASF >> INFRA) >> > > > - PR and master CI results come in way faster >> > > > - no more (well, GH Actions has outages as well, but less) false >> > > negatives >> > > > >> > > > WDYT? >> > > > >> > > > T >> > > > >> > > >> > > >> > > -- >> > > Sławomir Jaranowski >> > > >> > >> >> >> -- >> Sławomir Jaranowski >> >
