[ 
https://issues.apache.org/jira/browse/GEODE-218?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14697393#comment-14697393
 ] 

Dan Smith commented on GEODE-218:
---------------------------------

It's certainly worth trying this in a private build to see what the impact is 
on the stability and running time of the dunit tests.

But I think if it increases the running time of the tests by a long time this 
might not be the right approach. I think we're being impacted by a small 
percentage of tests that have race conditions or bad behavior. I think that the 
way to iron those tests out is to be able to iterate faster on running the 
tests.

Currently, the single VM integration tests fork every test, I think due to a 
few bad behaving tests. But now that the build is setup that way, people are 
probably adding more tests that don't clean up after themselves, making it 
harder for us to switch that setting to make the tests run faster. 

> Change distributedTest task to fork every 1 test case
> -----------------------------------------------------
>
>                 Key: GEODE-218
>                 URL: https://issues.apache.org/jira/browse/GEODE-218
>             Project: Geode
>          Issue Type: Bug
>            Reporter: Kirk Lund
>
> Change distributedTest task in the root build.gradle from "forkEvery 30" to 
> "forkEvery 1"
> DUnit tests are continuing to have spurious failures. Two common causes are: 
> 1) stop-the-world GC pauses, 2) JVM pollution from a previous test.
> The best overall way to reduce the number of spurious failures and hopefully 
> get our automated tests running green again is to isolate each test case more 
> thoroughly by having each DUnit fork new JVMs. We can reevaluate this after 
> seeing what effect "forkEvery 1" has.
> For purposes of execution speed, we should try to prevent the test 
> (UnitTests) and integrationTest (IntegrationTests) tasks from forking. It's 
> also more important to have distributedTest (DIstributedTests) fork because 
> these are end-to-end tests using a cluster of 6 JVMs.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to