Stig, it's possible but we don't run test master branch with JDK 7, so we all don't know the tests in storm-hdfs will succeed with JDK 7. It is occurred consistently in only JDK 7, and I'm just expecting no patch was targeted to fix that issue. (If it was, it must be pushed to 1.x version lines.)
Taylor, yes it's too flaky, not only in my experiment but also build results on Github pull requests page. It would be really tough to see build succeeds 10 times (maybe 5 times?) in a row, even 1.x version line with JDK 7 would break more often. IMHO, once we support JDK 7 in 1.x version line, tests should follow the environment: it must succeed with JDK 7 most of the time. Hopefully the flaky tests I filed for older versions and forgot are not seen recently, so closed such issues and filed new issues. I hope they would be easy to fix but it may not, so investigation from multiple contributors are definitely needed. 2018년 2월 2일 (금) 오전 5:12, P. Taylor Goetz <ptgo...@gmail.com>님이 작성: > I can confirm everything you’ve experienced. The maven release process > runs all tests 3 (current SNAPSHOT, release version, post-release SNAPSHOT) > times and they must pass for the build to succeed. That tends to highlight > flaky tests. > > In the last 6 release candidates, only one made it through on the first > try. IIRC it was 1.0.x. Sometimes switching JDKs helps, sometimes rebooting > helps, sometimes I turn to voodoo… ;) > > For the last round of RCs, I’ve been using the following JDK: > > java version "1.8.0_152" > Java(TM) SE Runtime Environment (build 1.8.0_152-b16) > Java HotSpot(TM) 64-Bit Server VM (build 25.152-b16, mixed mode) > > -Taylor > > > On Feb 1, 2018, at 6:31 AM, Jungtaek Lim <kabh...@gmail.com> wrote: > > > > Another update: I've tracked first page of pull requests and filed build > > failure issues under STORM-915. > > https://issues.apache.org/jira/browse/STORM-915 > > > > 2018년 2월 1일 (목) 오후 6:29, Jungtaek Lim <kabh...@gmail.com>님이 작성: > > > >> UPDATE: I've run build 3 times per branch from Travis CI, which ensures > >> the test result is no longer related to my (possible) messed up local > dev. > >> > >> 1. 1.x-branch (https://github.com/HeartSaVioR/storm/tree/1.x-branch) > >> > >> * storm-core, JDK7 : OK, FAIL, OK (66.6%) > >> * storm-core, JDK8 : OK, OK, FAIL (66.6%) > >> * !storm-core, JDK7 : ERROR, ERROR, ERROR (0%) > >> * !storm-core, JDK8 : OK, OK, OK (100%) > >> > >> 2. 1.1.x-branch (https://github.com/HeartSaVioR/storm/tree/1.1.x-branch > ) > >> > >> * storm-core, JDK7 : OK, OK, OK (100%) > >> * storm-core, JDK8 : OK, FAIL, OK (66.6%) > >> * !storm-core, JDK7 : ERROR, ERROR, ERROR (0%) > >> * !storm-core, JDK8 : OK, OK, OK (100%) > >> > >> 3. 1.0.x-branch (https://github.com/HeartSaVioR/storm/tree/1.0.x-branch > ) > >> > >> * storm-core, JDK7 : FAIL, FAIL, OK (33.3%) > >> * storm-core, JDK8 : OK, FAIL, OK (66.6%) > >> * !storm-core, JDK7 : ERROR, ERROR, ERROR (0%) > >> * !storm-core, JDK8 : OK, OK, OK (100%) > >> > >> I didn't see consistent test failure except storm-hdfs test stuck in > JDK 7 > >> which you may be already aware of (It must be fixed unless this cannot > be > >> fixed), but the chance of build failure is still high even except > >> integration-test. > >> > >> Test failures occur from below tests (all failures are occurring both > JDK > >> 7 and JDK 8): > >> > >> * org.apache.storm.metrics-test > >> * org.apache.storm.nimbus-test > >> * integration.org.apache.storm.integration-test > >> > >> I'm relieved that consistent failure in my local was my env. issue > >> (actually I found iPhone USB connect assigned private IP to my macOS > and it > >> completely messed up other things like manual test as well) but the test > >> failures still need to be investigated since we don't have any plan to > drop > >> supporting 1.x version lines. > >> > >> - Jungtaek Lim (HeartSaVioR) > >> > >> 2018년 2월 1일 (목) 오후 3:13, Jungtaek Lim <kabh...@gmail.com>님이 작성: > >> > >>> Hi devs, > >>> > >>> I have been verifying release candidates for 1.2.0/1.1.2/1.0.6 and > >>> encounter high chance of build failures across all the versions. > >>> I thought the issue is only with JDK 7, but sometimes also with JDK 8. > >>> > >>> Since it affects my verification of release candidates ongoing, I would > >>> like to see whether it is the issue from my environment (OSX 10.12.6, > >>> Oracle Java SE 1.7.0_79-b15), or there is something we should fix. > >>> > >>> I'm now trying out rebuilding branches with Travis CI. > >>> > >>> 1.x : https://travis-ci.org/HeartSaVioR/storm/builds/335962894 > >>> 1.1.x : https://travis-ci.org/HeartSaVioR/storm/builds/335962725 > >>> 1.0.x : https://travis-ci.org/HeartSaVioR/storm/builds/335962792 > >>> > >>> Please try out building branches with JDK7/8 (multiple times would be > >>> really appreciated) and share whether the build passes smoothly or not. > >>> > >>> Thanks, > >>> Jungtaek Lim (HeartSaVioR) > >>> > >> > >
