You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@sqoop.apache.org by "Veena Basavaraj (JIRA)" <ji...@apache.org> on 2014/09/12 18:37:33 UTC

[jira] [Commented] (SQOOP-1496) Sqoop2: Revisit/Refactor the SubmissionEngine/ExecutionEngine APIs

    [ https://issues.apache.org/jira/browse/SQOOP-1496?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14131726#comment-14131726 ] 

Veena Basavaraj commented on SQOOP-1496:
----------------------------------------


- renamed BaseCallbacks to Transferrable
- renamed SubmissionRequest to JobRequest
- fixed #2 ( execution engine now creates a job request and a submission request. submission engine submits the job request)... uuf!
- fixed #1 ( refactored to smaller methods, so that the code is readable and testable)
- fixed #3 ( added a unit test for the job manager, one of the most important class that has no tests)

> Sqoop2: Revisit/Refactor the SubmissionEngine/ExecutionEngine APIs
> ------------------------------------------------------------------
>
>                 Key: SQOOP-1496
>                 URL: https://issues.apache.org/jira/browse/SQOOP-1496
>             Project: Sqoop
>          Issue Type: Improvement
>            Reporter: Veena Basavaraj
>            Assignee: Veena Basavaraj
>         Attachments: Screen Shot 2014-09-08 at 2.35.06 PM.png
>
>
> Here are the things this ticket should address
> 1. Refactor the submit() in JobManager ( it is way too long for any good unit testing to happen)
> 2. Revisit the Execution Engine and Submission Engine relationship in the code.
> ExecutionEngine api should probably not create a submissionRequest. It should be the inverse. The Submission Engine is an abstraction of a JobTracker/ YARN/ Mesos/ OOzie. It will create a execution request for the execution engine while it submits the job into the execution engine
> a) Suggest renaming SubmissionEngine to a JobSubmissionManager. Rename MapreduceSubmissionEngine to a JobTrackerSubmisisonEngine. Since I could actually use the JobTracker with a non map reduce execution engine as well ( such as Spark )
>  
> b) SubmissionRequest is a JobRunRequest. See the java doc on the current code SubmissionRequest, it should already hint us that this object holds the job context to aid the execution engine
> c) Change the apis in Execution Engine/ SubmissionManager to reflect the same, i.e  rename prepareSubmissionRequest to prepareExecutionRequest 
> d) The Submission term is overloaded in the code base. In one case it is used to represent the JobRuns( history of the jobs) and in another place it is the submission engine such as JobTracker/YARN/ OOzie. These are unrelated and we should not be overloading this term in Sqoop
> YARN is an example of a submission engine and it can do much more than submissions. So our design and terminology should be much more generic than what it is now.
> e) define other responsibilities of the submission engine. For instance, if we are to add more monitoring/ tracking into how the job execution is progressing, we should be able to add those apis to this entity. There is more room for improvement for this api
> 3. Lastly add unit test for the  JobManager!



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