[jira] [Commented] (HIVE-18525) Add explain plan to Hive on Spark Web UI

["Sahil Takiar (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-18525?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16372268#comment-16372268 ] 

Sahil Takiar commented on HIVE-18525:
-------------------------------------

Attaching a POC patch. Attached a few screenshots of what the updated UI looks like. There are some open questions I still need to work on:

* Does generating the explain plan for each query add additional overhead? If so, how much?
* Whats the best way to make the code in {{ExplainTask}} useable in {{SparkPlan}}?
* Should this be turned on by default, or disabled by default? Should it be configurable? Should it be tied to the existing configuration {{hive.log.explain.output}}?

> Add explain plan to Hive on Spark Web UI
> ----------------------------------------
>
>                 Key: HIVE-18525
>                 URL: https://issues.apache.org/jira/browse/HIVE-18525
>             Project: Hive
>          Issue Type: Sub-task
>          Components: Spark
>            Reporter: Sahil Takiar
>            Assignee: Sahil Takiar
>            Priority: Major
>         Attachments: HIVE-18525.1.patch, Map-Explain-Plan.png, Reduce-Explain-Plan.png
>
>
> More of an investigation JIRA. The Spark UI has a "long description" of each stage in the Spark DAG. Typically one stage in the Spark DAG corresponds to either a {{MapWork}} or {{ReduceWork}} object. It would be useful if the long description contained the explain plan of the corresponding work object.
> I'm not sure how much additional overhead this would introduce. If not the full explain plan, then maybe a modified one that just lists out all the operator tree along with each operator name.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)