[
https://issues.apache.org/jira/browse/SPARK-16321?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15383785#comment-15383785
]
Maciej Bryński edited comment on SPARK-16321 at 7/19/16 8:27 AM:
-----------------------------------------------------------------
Bingo. I think I found the answer.
I did following query
{code}
o.where("id > 9000000").where('nested_column.id % 100000 =
0').select("id").collect()
{code}
Second where is to force Spark to read whole nested_column structure.
SQL Query
*Spark 1.6*
Time: *32 sec*
{code}
== Physical Plan ==
Project [id#0]
+- Filter (((nested_column#1.id % 100000) = 0) && (id#0 > 9000000))
+- Scan ParquetRelation[id#0,nested_column#1] InputPaths: file:some_file,
PushedFilters: [GreaterThan(id,9000000)]
{code}
*Spark 2.0*
Time: *65 sec*
{code}
== Physical Plan ==
Project [id#0]
+- Filter ((isnotnull(id#0) && (id#0 > 9000000)) && ((nested_column#1.id %
100000) = 0))
+- Scan parquet [id#0,nested_column#1] Format: ParquetFormat, InputPaths:
file:some_file, PushedFilters: [IsNotNull(id), GreaterThan(id,9000000)],
ReadSchema: struct<id:int,nested_column:struct<id:int...
{code}
As you can see there is a difference in predicates.
Spark 1.6: GreaterThan(id,9000000)
Spark 2.0: IsNotNull(id), GreaterThan(id,9000000)
was (Author: maver1ck):
Bingo. I think I found the answer.
I did following query
{code}
o.where("id > 9000000").where('nested_column.id % 100000 =
0').select("id").collect()
{code}
Second where is to force Spark to read whole nested_column structure.
SQL Query
*Spark 1.6*
Time: *32 sec*
{code}
== Physical Plan ==
Project [id#0]
+- Filter (((nested_column#1.id % 100000) = 0) && (id#0 > 9000000))
+- Scan ParquetRelation[id#0,nested_column#1] InputPaths: file:some_file,
PushedFilters: [GreaterThan(id,9000000)]
{code}
*Spark 2.0
Time: *65 sec*
{code}
== Physical Plan ==
Project [id#0]
+- Filter ((isnotnull(id#0) && (id#0 > 9000000)) && ((nested_column#1.id %
100000) = 0))
+- Scan parquet [id#0,nested_column#1] Format: ParquetFormat, InputPaths:
file:some_file, PushedFilters: [IsNotNull(id), GreaterThan(id,9000000)],
ReadSchema: struct<id:int,nested_column:struct<id:int...
{code}
As you can see there is a difference in predicates.
Spark 1.6: GreaterThan(id,9000000)
Spark 2.0: IsNotNull(id), GreaterThan(id,9000000)
> Pyspark 2.0 performance drop vs pyspark 1.6
> -------------------------------------------
>
> Key: SPARK-16321
> URL: https://issues.apache.org/jira/browse/SPARK-16321
> Project: Spark
> Issue Type: Bug
> Components: PySpark
> Affects Versions: 2.0.0
> Reporter: Maciej Bryński
> Attachments: Spark16.nps, Spark2.nps, spark16._trace.png,
> spark2_nofilterpushdown.nps, spark2_trace.png, visualvm_spark16.png,
> visualvm_spark2.png, visualvm_spark2_G1GC.png
>
>
> I did some test on parquet file with many nested columns (about 30G in
> 400 partitions) and Spark 2.0 is 2x slower.
> {code}
> df = sqlctx.read.parquet(path)
> df.where('id > some_id').rdd.flatMap(lambda r: [r.id] if not r.id %100000
> else []).collect()
> {code}
> Spark 1.6 -> 2.3 min
> Spark 2.0 -> 4.6 min (2x slower)
> I used BasicProfiler for this task and cumulative time was:
> Spark 1.6 - 4300 sec
> Spark 2.0 - 5800 sec
> Should I expect such a drop in performance ?
> I don't know how to prepare sample data to show the problem.
> Any ideas ? Or public data with many nested columns ?
--
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
