[
https://issues.apache.org/jira/browse/SPARK-16827?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15401413#comment-15401413
]
Sital Kedia commented on SPARK-16827:
-------------------------------------
That is not the case. There is no broadcast join involved, its using shuffle
join in both 1.6 and 2.0.
Plan in 1.6 -
Project [target_id#136L AS userid#115L]
+- SortMergeJoin [source_id#135L], [id#140L]
:- Sort [source_id#135L ASC], false, 0
: +- TungstenExchange hashpartitioning(source_id#135L,800), None
: +- ConvertToUnsafe
: +- HiveTableScan [target_id#136L,source_id#135L], MetastoreRelation
foo, table1, Some(a), [(ds#134 = 2016-07-15)]
+- Sort [id#140L ASC], false, 0
+- TungstenExchange hashpartitioning(id#140L,800), None
+- ConvertToUnsafe
+- HiveTableScan [id#140L], MetastoreRelation foo, table2, Some(b),
[(ds#139 = 2016-07-15)]
Plan in 2.0 -
*Project [target_id#151L AS userid#111L]
+- *SortMergeJoin [source_id#150L], [id#155L], Inner
:- *Sort [source_id#150L ASC], false, 0
: +- Exchange hashpartitioning(source_id#150L, 800)
: +- *Filter isnotnull(source_id#150L)
: +- HiveTableScan [source_id#150L, target_id#151L],
MetastoreRelation foo, table1, a, [isnotnull(ds#149), (ds#149 = 2016-07-15)]
+- *Sort [id#155L ASC], false, 0
+- Exchange hashpartitioning(id#155L, 800)
+- *Filter isnotnull(id#155L)
+- HiveTableScan [id#155L], MetastoreRelation foo, table2, b,
[isnotnull(ds#154), (ds#154 = 2016-07-15)]
> Query with Join produces excessive amount of shuffle data
> ---------------------------------------------------------
>
> Key: SPARK-16827
> URL: https://issues.apache.org/jira/browse/SPARK-16827
> Project: Spark
> Issue Type: Bug
> Components: Shuffle, Spark Core
> Affects Versions: 2.0.0
> Reporter: Sital Kedia
> Labels: performance
>
> One of our hive job which looks like this -
> {code}
> SELECT userid
> FROM table1 a
> JOIN table2 b
> ON a.ds = '2016-07-15'
> AND b.ds = '2016-07-15'
> AND a.source_id = b.id
> {code}
> After upgrade to Spark 2.0 the job is significantly slow. Digging a little
> into it, we found out that one of the stages produces excessive amount of
> shuffle data. Please note that this is a regression from Spark 1.6. Stage 2
> of the job which used to produce 32KB shuffle data with 1.6, now produces
> more than 400GB with Spark 2.0. We also tried turning off whole stage code
> generation but that did not help.
> PS - Even if the intermediate shuffle data size is huge, the job still
> produces accurate output.
--
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
