[
https://issues.apache.org/jira/browse/HADOOP-4059?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12627959#action_12627959
]
Steve Loughran commented on HADOOP-4059:
----------------------------------------
This may actually be easiest to test using AntUnit, as ant has much of the
process management code needed: http://ant.apache.org/antlibs/antunit/
It could be added as a contribs/ section that has nothing but tests that get
run in a series of test build files. The various entry points do have to run in
their own process (we can't call JobTracker.main() etc) as the tests really do
have to check the scripts end-to-end. Go via main() in a sandbox to catch
System.exit() and you aren't testing the scripts any more, which always create
more trouble than 50 lines of code really should
> Add tests that try starting the various hadoop command line scripts
> -------------------------------------------------------------------
>
> Key: HADOOP-4059
> URL: https://issues.apache.org/jira/browse/HADOOP-4059
> Project: Hadoop Core
> Issue Type: Test
> Components: test
> Affects Versions: 0.19.0
> Reporter: Steve Loughran
> Assignee: Steve Loughran
> Original Estimate: 72h
> Remaining Estimate: 72h
>
> Hadoop has lots of tests that start clusters, but I don't see any that test
> the command line scripts working. Which means that changes to the scripts and
> the code behind them may not get picked up rapidly, and regressions won't get
> picked up by Hudson and apportioned to the specific patches.
> Propose:
> * an abstract test case that can exec scripts on startup; wait for them to
> finish, kill them if they take too long
> * test cases for every service (namenode, tasktracker, datanode, etc)
> * tests that try invalid commands, -help options
> * tests that start the services, assert that they are live, and shut them down
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
