Thanks Robert for the ideas. Some explanation: we rerun all the tests
within a test class in case there is a failing test case because surefire
does not allow running individual test cases from multiple test classes
(see bin/test-patch-20-tests). In other words: passing
-Dtest=Test1#testMethod1,Test2#testMethod2 to mvn test does not work, while
-DTest1,Test2 works at the cost of running all other tests in Test1 an
Test2.

Clean up before and after tests is a good idea. In fact, I have seen
multiple undeleted test resources (after mvn clean too):
        core/activemq-data/
        core/dist2.txt
        core/dist3.txt
        core/distcp-log4j.properties
        core/distcp-oozie-1514391729048.log
        core/dst1.txt
        core/test-invalid-workflow-app.xml
        core/test-workflow-app.xml

I filed some new JIRA-s:
- OOZIE-3145 TestDistcpMain shall remove created files after test execution
- OOZIE-3148 Rerun Failing Tests through Maven surefire
( Right now I am running again mkdistro.sh with
-Dsurefire.rerunFailingTestsCount=2. I will look into mini cluster logs if
there are failures.)

Unfortunately, there are some known flaky tests in Oozie (tracked by
OOZIE-3111 umbrella JIRA). I am not sure if we need to fix all of them
before the first 5.0.0-beta1 release candidate; just out of curiosity, I
tried to run bin/mkdistro.sh on release-4.3.0 and some tests failed too.

- Attila

On Fri, Dec 22, 2017 at 1:20 AM, Robert Kanter <rkan...@cloudera.com> wrote:

> I took a look at the latest PreCommit job
> <https://builds.apache.org/job/PreCommit-OOZIE-Build/288/consoleFull> and
> it reported 55 rerun tests.
>     Tests rerun: 55
>     Tests failed at first run:
> org.apache.oozie.action.hadoop.TestJavaActionExecutor,
>
> However, looking through the actual output, I only see 1 test that failed
> (and was rerun): TestJavaActionExecutor. testCredentialsSkip.
>
> [INFO] Running org.apache.oozie.action.hadoop.TestJavaActionExecutor
> [ERROR] Tests run: 55, Failures: 0, Errors: 1, Skipped: 0, Time
> elapsed: 126.213 s <<< FAILURE! - in
> org.apache.oozie.action.hadoop.TestJavaActionExecutor
> [ERROR] testCredentialsSkip(org.apache.oozie.action.hadoop.
> TestJavaActionExecutor)
>  Time elapsed: 0.532 s  <<< ERROR!
> org.apache.oozie.action.ActionExecutorException: JA020: Could not load
> credentials of type [abc] with name [abcname]]; perhaps it was not
> defined in oozie-site.xml?
>         at org.apache.oozie.action.hadoop.TestJavaActionExecutor.
> _testCredentialsSkip(TestJavaActionExecutor.java:1106)
>         at org.apache.oozie.action.hadoop.TestJavaActionExecutor.
> testCredentialsSkip(TestJavaActionExecutor.java:1006)
>
>
> In fact, the report only lists the one test class, not 55 of them.  So I
> think there's something wrong with our reporting here.
>
> Anyway, typically, when we see test that succeed on their own but fail when
> run all together, that means that (likely some other) test is not cleaning
> up properly.  This is unfortunately tricky to debug because it's hard to
> figure out what the other test is.  A long time ago, a big example of this
> problem was not properly shutting down the Services singleton, so we'd have
> duplicates and other issues.
>
> For these specific issues, some hints:
> # RUNNINGWITHERROR: This problem means that a yarn job in the mini cluster
> has failed.  To find out why, you should be able to dig out the app id from
> the test output, and then find it's yarn logs somewhere (there's a
> minicluster logs dir, but I forget where).  That'll hopefully make it
> obvious what's going on.
> # Credentials: There's probably an oozie-site or Configuration class
> leaking from somewhere or not properly cleaned up by a previous test or
> setup by this test.  The Credentials class is missing.
> # TestJMSAccessorService: Sounds like something didn't get cleaned up.
>
> One way that might be easier to fix this is to have the setUp methods
> ensure that things are clean just in case.  We actually have a number of
> tests that do things like this too.
>
>
> - Robert
>
>
>
> On Thu, Dec 21, 2017 at 12:40 PM, Attila Sasvari <asasv...@cloudera.com>
> wrote:
>
> >  Update:
> > - bin/mkdistro.sh fails because there are test failures.
> > [ERROR] Failures:
> >
> >
> > [ERROR]
> > TestCoordActionsKillXCommand.testActionKillCommandActionNumbers:96
> > expected:<RUNNING> but was:<RUNNINGWITHERROR>
> >
> >
> > [ERROR] Errors:
> >
> >
> > [ERROR]
> > TestJavaActionExecutor.testCredentialsSkip:1006->_
> testCredentialsSkip:1106
> > ? ActionExecutor
> >
> > [ERROR]   TestJMSAccessorService.testConnectionRetryExceptionLi
> stener:211
> > ?
> > InstanceAlreadyExists
> >
> >
> > - if I run the tests separately, they pass. Looking at the latest
> precommit
> > builds (https://builds.apache.org/job/PreCommit-OOZIE-Build), it turns
> out
> > that a lot of tests had to be re-executed to get a +1 for the TESTS part.
> > Problem is that the Oozie tests have impact on each other, and it looks
> > like the execution order matters too.
> > - uploaded work in progress (SNAPSHOT)  artifacts here:
> > http://people.apache.org/~asasvari/oozie-5.0.0-beta1-SNAPSHOT/
> > - I plan to update the "How To Release" page as it contains some errors
> > (e.g sftp shall be used to upload artifacts)
> > - agreed with Artem that OOZIE-2231 will slip to 5.0.0
> >
> > Regards,
> > - Attila
> >
> > On Mon, Dec 18, 2017 at 11:47 PM, Attila Sasvari <asasv...@cloudera.com>
> > wrote:
> >
> > > Hi everyone,
> > >
> > > branch-5.0.0-beta1 has been created.
> > >
> > > Regards,
> > > Attila
> > >
> > > On Mon, Dec 18, 2017 at 7:25 PM, Attila Sasvari <asasv...@cloudera.com
> >
> > > wrote:
> > >
> > >> Thanks gp. I will follow the steps described on
> > >> https://cwiki.apache.org/confluence/display/OOZIE/How+To+Release
> during
> > >> the process.
> > >>
> > >> Next steps:
> > >> - A new branch is about to be created from master.
> > >>
> > >> Artem, many thanks. I will review and commit that patch if everything
> is
> > >> okay. I don't see any problem with including it in 5.0.0-beta1.
> > >>
> > >> On Mon, Dec 18, 2017 at 4:32 PM, Artem Ervits <artemerv...@gmail.com>
> > >> wrote:
> > >>
> > >>> just uploaded patch for OOZIE-2231.
> > >>>
> > >>> On Mon, Dec 18, 2017 at 10:04 AM, Peter Cseh <gezap...@cloudera.com>
> > >>> wrote:
> > >>>
> > >>> > Hey Attila,
> > >>> >
> > >>> > I won't be able to work on the release for a couple weeks now.
> > >>> > Thanks for getting the release rolling!
> > >>> >
> > >>> > Cheers,
> > >>> > gp
> > >>> >
> > >>> >
> > >>> > On Mon, Dec 18, 2017 at 4:00 PM, Attila Sasvari <
> > asasv...@cloudera.com
> > >>> >
> > >>> > wrote:
> > >>> >
> > >>> > > Hi everyone,
> > >>> > >
> > >>> > > I would like to create the release branch, branch-5.0.0-beta1
> > >>> (following
> > >>> > > Hadoop release versioning), earlier.
> > >>> > >
> > >>> > > Looking at https://issues.apache.org/
> jira/projects/OOZIE/versions/
> > >>> > 12342048
> > >>> > > there are 3 issues in progress (OOZIE-2231, OOZIE-2942,
> OOZIE-2974)
> > >>> and 3
> > >>> > > issues to do (OOZIE-2600, OOZIE-3093, OOZIE-1987). I will push
> > those
> > >>> out
> > >>> > to
> > >>> > > 5.0.0 if there are no objections.
> > >>> > >
> > >>> > > At the same time, I am volunteering to be the release manager if
> > >>> Peter
> > >>> > Cseh
> > >>> > > does not mind.
> > >>> > >
> > >>> > > Regards,
> > >>> > > Attila
> > >>> > >
> > >>> > > On Wed, Dec 6, 2017 at 8:28 PM, Robert Kanter <
> > rkan...@cloudera.com>
> > >>> > > wrote:
> > >>> > >
> > >>> > > > Sounds good to me!
> > >>> > > >
> > >>> > > > On Wed, Dec 6, 2017 at 5:17 AM, Andras Piros <
> > >>> > andras.pi...@cloudera.com>
> > >>> > > > wrote:
> > >>> > > >
> > >>> > > > > Good idea Gezapeti!
> > >>> > > > >
> > >>> > > > > Time to wrap things up towards a stable 5.0.0 - a release
> > >>> candidate
> > >>> > on
> > >>> > > > > 5.0.0b1 is a good first step.
> > >>> > > > >
> > >>> > > > > Since other components that Oozie uses like Pig and Hive do
> not
> > >>> > (fully)
> > >>> > > > > support Hadoop 3, we have to wait with OOZIE-2973
> > >>> > > > > <https://issues.apache.org/jira/browse/OOZIE-2973>.
> > >>> > > > >
> > >>> > > > > Thanks,
> > >>> > > > >
> > >>> > > > > Andras
> > >>> > > > >
> > >>> > > > > On Wed, Dec 6, 2017 at 1:50 PM, Peter Cseh <
> > >>> gezap...@cloudera.com>
> > >>> > > > wrote:
> > >>> > > > >
> > >>> > > > > > Hi everyone!
> > >>> > > > > >
> > >>> > > > > > Now that OOZIE-2969 <https://issues.apache.org/
> > >>> > > jira/browse/OOZIE-2969>
> > >>> > > > > is
> > >>> > > > > > committed I'd like to start the process of creating the
> > branch
> > >>> for
> > >>> > > > > 5.0.0b1
> > >>> > > > > > and building a release from there.
> > >>> > > > > > It's unfortunate that we won't be able to support Hadoop 3
> in
> > >>> the
> > >>> > > beta
> > >>> > > > > for
> > >>> > > > > > reasons described in OOZIE-2973
> > >>> > > > > > <https://issues.apache.org/jira/browse/OOZIE-2973>
> > >>> > > > > > I don't see any more blockers for the beta1 and I hope we
> > won't
> > >>> > find
> > >>> > > > > > hard-to-fix major issues so we can release Oozie 5.0.0 in
> > early
> > >>> > 2018.
> > >>> > > > > > Please let me know if you have any suggestions.
> > >>> > > > > > Thanks
> > >>> > > > > > gp
> > >>> > > > > >
> > >>> > > > > > --
> > >>> > > > > > Peter Cseh
> > >>> > > > > > Software Engineer
> > >>> > > > > > <http://www.cloudera.com>
> > >>> > > > > >
> > >>> > > > >
> > >>> > > >
> > >>> > >
> > >>> > >
> > >>> > >
> > >>> > > --
> > >>> > > --
> > >>> > > Attila Sasvari
> > >>> > > Software Engineer
> > >>> > > <http://www.cloudera.com/>
> > >>> > >
> > >>> >
> > >>> >
> > >>> >
> > >>> > --
> > >>> > Peter Cseh
> > >>> > Software Engineer
> > >>> > <http://www.cloudera.com>
> > >>> >
> > >>>
> > >>
> > >>
> > >>
> > >> --
> > >> --
> > >> Attila Sasvari
> > >> Software Engineer
> > >> <http://www.cloudera.com/>
> > >>
> > >
> > >
> > >
> > > --
> > > --
> > > Attila Sasvari
> > > Software Engineer
> > > <http://www.cloudera.com/>
> > >
> >
> >
> >
> > --
> > --
> > Attila Sasvari
> > Software Engineer
> > <http://www.cloudera.com/>
> >
>



-- 
-- 
Attila Sasvari
Software Engineer
<http://www.cloudera.com/>

Reply via email to