[jira] [Commented] (SPARK-4019) Repartitioning with more than 2000 partitions drops all data

["Josh Rosen (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-4019?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14177617#comment-14177617 ] 

Josh Rosen commented on SPARK-4019:
-----------------------------------

This issue is caused by a bug in HighlyCompressedMapStatus.  I think that's we're compressing a bunch of blocks whose average size is small, so this gets averaged down to zero.  As a result, we skip these blocks as empty even though they contain data.

I'm going to work on a fix ASAP, but first I'm going to use ScalaCheck to write a property-based test that would have caught this.  The invariant that we need to maintain: "if an uncompressed map output size is greater than zero, then compressing and decompressing should continue to report the map output as non-empty."

> Repartitioning with more than 2000 partitions drops all data
> ------------------------------------------------------------
>
>                 Key: SPARK-4019
>                 URL: https://issues.apache.org/jira/browse/SPARK-4019
>             Project: Spark
>          Issue Type: Bug
>          Components: Spark Core
>    Affects Versions: 1.2.0
>            Reporter: Xiangrui Meng
>            Assignee: Josh Rosen
>            Priority: Blocker
>
> {code}
> sc.makeRDD(0 until 10, 1000).repartition(2001).collect()
> {code}
> returns `Array()`.
> 1.1.0 doesn't have this issue. Tried both HASH and SORT manager.



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org