[
https://issues.apache.org/jira/browse/SPARK-15415?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15292828#comment-15292828
]
Jurriaan Pruis commented on SPARK-15415:
----------------------------------------
[~rxin] I could try to work on that. The reason I ran into this issue was that
Spark was broadcasting a 8M rows table which resulted in my driver going OOM.
Reducing the autoBroadcastJoinThreshold didn't help unless I would reduce it to
very small values which would result in nothing being broadcasted at all (which
is a bug too I guess, maybe it's is calculating the sizeInBytes incorrectly?).
So I tried to fix the problem by reducing it to zero and only marking the
tables I needed to be broadcasted, but that didn't work. I guess what I really
needed was a 'do not broadcast' hint.
> Marking partitions for broadcast broken
> ---------------------------------------
>
> Key: SPARK-15415
> URL: https://issues.apache.org/jira/browse/SPARK-15415
> Project: Spark
> Issue Type: Bug
> Components: SQL
> Affects Versions: 2.0.0
> Reporter: Jurriaan Pruis
>
> I couldn't get the broadcast(DataFrame) sql function to work in Spark 2.0.
> It does work in Spark 1.6.1:
> {code}
> $ pyspark --conf spark.sql.autoBroadcastJoinThreshold=0
> >>> df = sqlCtx.range(1000);df2 =
> >>> sqlCtx.range(1000);df.join(pyspark.sql.functions.broadcast(df2),
> >>> 'id').explain()
> == Physical Plan ==
> Project [id#0L]
> +- BroadcastHashJoin [id#0L], [id#1L], BuildRight
> :- ConvertToUnsafe
> : +- Scan ExistingRDD[id#0L]
> +- ConvertToUnsafe
> +- Scan ExistingRDD[id#1L]
> {code}
> While in Spark 2.0 this results in:
> {code}
> >>> df = sqlCtx.range(1000);df2 =
> >>> sqlCtx.range(1000);df.join(pyspark.sql.functions.broadcast(df2),
> >>> 'id').explain()
> == Physical Plan ==
> WholeStageCodegen
> : +- Project [id#6L]
> : +- SortMergeJoin [id#6L], [id#9L], Inner, None
> : :- INPUT
> : +- INPUT
> :- WholeStageCodegen
> : : +- Sort [id#6L ASC], false, 0
> : : +- INPUT
> : +- Exchange hashpartitioning(id#6L, 200), None
> : +- WholeStageCodegen
> : : +- Range 0, 1, 8, 1000, [id#6L]
> +- WholeStageCodegen
> : +- Sort [id#9L ASC], false, 0
> : +- INPUT
> +- ReusedExchange [id#9L], Exchange hashpartitioning(id#6L, 200), None
> {code}
>
> While it should look like (output when you remove the
> spark.sql.autoBroadcastJoinThreshold conf):
> {code}
> == Physical Plan ==
> WholeStageCodegen
> : +- Project [id#0L]
> : +- BroadcastHashJoin [id#0L], [id#3L], Inner, BuildRight, None
> : :- Range 0, 1, 8, 1000, [id#0L]
> : +- INPUT
> +- BroadcastExchange HashedRelationBroadcastMode(List(input[0, bigint]))
> +- WholeStageCodegen
> : +- Range 0, 1, 8, 1000, [id#3L]
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
