hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Szehon Ho (JIRA)" <>
Subject [jira] [Commented] (HIVE-10291) Hive on Spark job configuration needs to be logged [Spark Branch]
Date Fri, 10 Apr 2015 16:05:12 GMT


Szehon Ho commented on HIVE-10291:

There's no configuration for MR, because it doesn't need to appear in the log (it is sent
to MR and then appears in the jobconfiguration page of HistoryServer).  The issue is that
does not happen in Spark (it is sent to RemoteDriver), so we need a way for it to visualize
it.  The only way is the log.

Yea I'm ok with always dumping it as it would be simpler, but I was thinking to minimize the
perf impact?

> Hive on Spark job configuration needs to be logged [Spark Branch]
> -----------------------------------------------------------------
>                 Key: HIVE-10291
>                 URL:
>             Project: Hive
>          Issue Type: Sub-task
>          Components: Spark
>    Affects Versions: 1.1.0
>            Reporter: Szehon Ho
>            Assignee: Szehon Ho
>         Attachments: HIVE-10291-spark.patch, HIVE-10291.2-spark.patch
> In a Hive on MR job, all the job properties are put into the JobConf, which can then
be viewed via the MR2 HistoryServer's Job UI.
> However, in Hive on Spark we are submitting an application that is long-lived.  Hence,
we only put properties into the SparkConf relevant to application submission (spark and yarn
properties).  Only these are viewable through the Spark HistoryServer Application UI.
> It is the Hive application code (RemoteDriver, aka RemoteSparkContext) that is responsible
for serializing and deserializing the job.xml per job (ie, query) within the application.
 Thus, for supportability we also need to give an equivalent mechanism to print the job.xml
per job.

This message was sent by Atlassian JIRA

View raw message