[
https://issues.apache.org/jira/browse/MAHOUT-1345?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13831026#comment-13831026
]
Dawid Weiss commented on MAHOUT-1345:
-------------------------------------
There is no way to reliably "kill" threads in Java. You can try to interrupt
them, but in general threads can catch exceptions and respin in a loop. It is
the test's duty to clean up any resources it starts -- including shutting down
any threads or thread pools it starts up.
If you have suite-level initialized stuff (thread pools or resources), like
BeforeClass or class rules, then detect thread leaks at the suite level (after
all tests of a class are over), not in every test. This is done by declaring
{code}
@ThreadLeakScope(Scope.SUITE)
{code}
Thread lingering is meant only for those corner cases when one cannot join all
threads that are to be terminated before the test is over (and in effect the
thread may be still in an alive state). Thread pools are one example of when
this is the case (no direct control over thread pool threads, although one can
shutdown the threadpool itself).
In general all this is a bit complex, but getting it to work is well worth
overcoming the initial learning curve. You get a (soft, although valuable)
guarantee that no test stomps over each other by leaving threads that do
something in the background.
> Enable randomised testing for all Mahout modules
> ------------------------------------------------
>
> Key: MAHOUT-1345
> URL: https://issues.apache.org/jira/browse/MAHOUT-1345
> Project: Mahout
> Issue Type: Improvement
> Affects Versions: 0.8
> Reporter: Isabel Drost-Fromm
> Priority: Minor
> Fix For: 0.9
>
> Attachments: MAHOUT-1345.diff, MAHOUT-1345.patch
>
>
> When enabling randomised testing for all modules I found a few tests became
> unstable or even fail deterministically due to lingering threads. The
> attached patch:
> * defines the randomised testing dependency in our parent pom
> * re-uses said dependencies in all depending modules (makes upgrading easier
> as the version number needs to be changed in just one place)
> * adds several code changes that fixed the failures due to lingering threads
> for me on my machine. I'd greatly appreciate input a) from those who wrote
> the respective code and b) others who ran the tests with these changes to
> make sure there are no other tests that suffer from the same issues.
> Warning: I touched quite a few bits and pieces I'm not intimately familiar
> with over the last few weeks (whenever I had a few spare minutes) - second
> pair of eyes needed.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
