Cheng Lian created SPARK-5346:
---------------------------------
Summary: Parquet filter pushdown is not enabled when
parquet.task.side.metadata is set to true (default value)
Key: SPARK-5346
URL: https://issues.apache.org/jira/browse/SPARK-5346
Project: Spark
Issue Type: Bug
Components: SQL
Affects Versions: 1.2.0, 1.3.0
Reporter: Cheng Lian
Priority: Critical
When computing Parquet splits, reading Parquet metadata from executor side is
more memory efficient, thus Spark SQL [sets {{parquet.task.side.metadata}} to
{{true}} by
default|https://github.com/apache/spark/blob/v1.2.0/sql/core/src/main/scala/org/apache/spark/sql/parquet/ParquetTableOperations.scala#L437].
However, somehow this disables filter pushdown. Now Parquet filter pushdown is
only enabled when {{spark.sql.parquet.filterPushdown}} set to {{true}} while
{{parquet.task.side.metadata}} set to {{false}}. For large Parquet files with
thousands of part-files and/or thousands of columns, reading metadata from
driver side eats lots of memory.
--
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
