You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@pig.apache.org by "liyunzhang_intel (JIRA)" <ji...@apache.org> on 2016/11/04 03:14:58 UTC

[jira] [Commented] (PIG-4746) Ensure spark can be run as PIG action in Oozie

    [ https://issues.apache.org/jira/browse/PIG-4746?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15635068#comment-15635068 ] 

liyunzhang_intel commented on PIG-4746:
---------------------------------------

[~szita] and [~nkollar]:  
   i'm not familiar with oozie so not very clear about jira description. I think what [~pallavi.rao] wants is to  make pig on spark runs successfully in oozie.

{quote}
To get spork going in yarn-client mode, couple of enhancements will need to be made:
1. Right now, spark launcher reads SPARK_MASTER as env. variable. Need to make this a PIG property.
2. The spark libraries need to be in classpath of the driver in case of yarn-clientmode.
{quote}
Point1: add SPARK_MASTER to conf/pig.properties, read it by PigContext.getProperties() in SparkLauncher
Point2: After PIG-4903, i'm not sure whether we still need add spark libraries in the classpath of the driver. here the driver is pig or oozie?  if the driver is oozie, i think we need add the spark-libraries(such as spark-assembly.jar) to the classpath of oozie. If the driver is pig, i think we need not because i have add the classpath to [bin/pig|https://github.com/apache/pig/blob/spark/bin/pig#L421]


> Ensure spark can be run as PIG action in Oozie
> ----------------------------------------------
>
>                 Key: PIG-4746
>                 URL: https://issues.apache.org/jira/browse/PIG-4746
>             Project: Pig
>          Issue Type: Sub-task
>          Components: spark
>            Reporter: Pallavi Rao
>            Assignee: Prateek Vaishnav
>             Fix For: spark-branch
>
>
> I was able get PIG on SPARK going with Oozie. But, only in "local" mode. Here is what I did:
> 1. Used workflow schema version uri:oozie:workflow:0.2 and passed exectype as an argument.
> 2. Copied Spark jars + kyro jar into workflow app lib.
> To get spork going in yarn-client mode, couple of enhancements will need to be made:
> 1. Right now, spark launcher reads SPARK_MASTER as env. variable. Need to make this a PIG property.
> 2. The spark libraries need to be in classpath of the driver in case of yarn-clientmode. This will need a fix similar to PIG-4667



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