[SPARK-6554] [SQL] Don't push down predicates which reference partition column(s)

There are two cases for the new Parquet data source:

1. Partition columns exist in the Parquet data files

   We don't need to push-down these predicates since partition pruning already handles them.

1. Partition columns don't exist in the Parquet data files

   We can't push-down these predicates since they are considered as invalid columns by Parquet.

<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/5210)
<!-- Reviewable:end -->

Author: Cheng Lian <lian@databricks.com>

Closes #5210 from liancheng/spark-6554 and squashes the following commits:

4f7ec03 [Cheng Lian] Adds comments
e134ced [Cheng Lian] Don't push down predicates which reference partition column(s)

2 files changed, 29 insertions, 5 deletions

diff --git a/sql/core/src/main/scala/org/apache/spark/sql/parquet/newParquet.scala b/sql/core/src/main/scala/org/apache/spark/sql/parquet/newParquet.scala
index 3516cfe680..0d68810ec6 100644
--- a/sql/core/src/main/scala/org/apache/spark/sql/parquet/newParquet.scala
+++ b/sql/core/src/main/scala/org/apache/spark/sql/parquet/newParquet.scala
@@ -435,11 +435,18 @@ private[sql] case class ParquetRelation2(
     // Push down filters when possible. Notice that not all filters can be converted to Parquet
     // filter predicate. Here we try to convert each individual predicate and only collect those
     // convertible ones.
-    predicates
-      .flatMap(ParquetFilters.createFilter)
-      .reduceOption(FilterApi.and)
-      .filter(_ => sqlContext.conf.parquetFilterPushDown)
-      .foreach(ParquetInputFormat.setFilterPredicate(jobConf, _))
+    if (sqlContext.conf.parquetFilterPushDown) {
+      predicates
+        // Don't push down predicates which reference partition columns
+        .filter { pred =>
+          val partitionColNames = partitionColumns.map(_.name).toSet
+          val referencedColNames = pred.references.map(_.name).toSet
+          referencedColNames.intersect(partitionColNames).isEmpty
+        }
+        .flatMap(ParquetFilters.createFilter)
+        .reduceOption(FilterApi.and)
+        .foreach(ParquetInputFormat.setFilterPredicate(jobConf, _))
+    }
 
     if (isPartitioned) {
       logInfo {
diff --git a/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetFilterSuite.scala b/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetFilterSuite.scala
index 4d32e84fc1..6a2c2a7c40 100644
--- a/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetFilterSuite.scala
+++ b/sql/core/src/test/scala/org/apache/spark/sql/parquet/ParquetFilterSuite.scala
@@ -321,6 +321,23 @@ class ParquetDataSourceOnFilterSuite extends ParquetFilterSuiteBase with BeforeA
   override protected def afterAll(): Unit = {
     sqlContext.setConf(SQLConf.PARQUET_USE_DATA_SOURCE_API, originalConf.toString)
   }
+
+  test("SPARK-6554: don't push down predicates which reference partition columns") {
+    import sqlContext.implicits._
+
+    withSQLConf(SQLConf.PARQUET_FILTER_PUSHDOWN_ENABLED -> "true") {
+      withTempPath { dir =>
+        val path = s"${dir.getCanonicalPath}/part=1"
+        (1 to 3).map(i => (i, i.toString)).toDF("a", "b").saveAsParquetFile(path)
+
+        // If the "part = 1" filter gets pushed down, this query will throw an exception since
+        // "part" is not a valid column in the actual Parquet file
+        checkAnswer(
+          sqlContext.parquetFile(path).filter("part = 1"),
+          (1 to 3).map(i => Row(i, i.toString, 1)))
+      }
+    }
+  }
 }
 
 class ParquetDataSourceOffFilterSuite extends ParquetFilterSuiteBase with BeforeAndAfterAll {