[
https://issues.apache.org/jira/browse/MAPREDUCE-5505?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Zhijie Shen updated MAPREDUCE-5505:
-----------------------------------
Attachment: MAPREDUCE-5505.4.patch
Thanks [~bikassaha] for reviewing the patch. I've updated it accordingly.
bq. Typo - already ...
Fixed
bq. Lets use a true value to be back-compatible in case it gets used.
Fixed
bq. Typo - mock
Fixed
bq. Please put a comment for this non-obvious code. We need it because noone is
calling shutdown job right. Alternatively, can MRApp.createJob() be changed to
call MRAppMaster.shutdown() or set the boolean value to true. This would be
closer to ideal that the current approach.
Agree the code is non-obvious. Instead of moving setting
safeToReportTerminationToUser to MRApp.createJob(), I moved it to the
constructor of MRApp, because safeToReportTerminationToUser is the per
MRAppMaster variable.
bq. Can we have a test that verifies the main straightline case. Job succeeds
and returns running until the boolean is set.?
Added TestMRApp#testJobSuccess
bq. How can we be sure that the previous state == RUNNING
Its a general issue. To solve it in all the transitions of JobImpl, I added the
code to remember the last non-final state. Then, whenever
safeToReportTerminationToUser is false, JobImpl returns the stored previous
state instead of the final state, i.e., SUCCEEDED, FAILED, KILLED and ERROR.
bq. Has this been tested on a single node cluster with a real job?
Tested locally. The job client saw RUNNING until AM got unregistered.
> Clients should be notified job finished only after job successfully
> unregistered
> ---------------------------------------------------------------------------------
>
> Key: MAPREDUCE-5505
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-5505
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Reporter: Jian He
> Assignee: Zhijie Shen
> Attachments: MAPREDUCE-5505.1.patch, MAPREDUCE-5505.1.patch,
> MAPREDUCE-5505.3.patch, MAPREDUCE-5505.4.patch
>
>
> This is to make sure user is notified job finished after job is really done.
> This does increase client latency but can reduce some races during unregister
> like YARN-540
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira