address comments

Author: Huaxin Gao

docs/sql-data-sources-jdbc.md

@@ -211,6 +211,13 @@ the following case-insensitive options:
      Specifies kerberos principal name for the JDBC client. If both <code>keytab</code> and <code>principal</code> are defined then Spark tries to do kerberos authentication.
     </td>
   </tr>
+
+  <tr>
+    <td><code>pushDownAggregate</code></td>
+    <td>
+     The option to enable or disable aggregate push-down into the JDBC data source. The default value is false, in which case Spark will NOT push down aggregates to the JDBC data source. Otherwise, if set to true, aggregate will be pushed down to the JDBC data source and thus aggregates will be handled by data source instead of Spark. Aggregate push-down is usually turned off when the aggregate is performed faster by Spark than by the JDBC data source. Please note that aggregates are pushed down if and only if all the aggregates and the related filters can be pushed down.
+    </td>
+  </tr>
 </table>
 
 Note that kerberos authentication with keytab is not always supported by the JDBC driver.<br>

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/Count.scala

@@ -17,11 +17,7 @@
 
 package org.apache.spark.sql.catalyst.expressions.aggregate
 
-import org.apache.spark.sql.catalyst.analysis.TypeCheckResult
-import org.apache.spark.sql.catalyst.dsl.expressions._
 import org.apache.spark.sql.catalyst.expressions._
-import org.apache.spark.sql.internal.SQLConf
-import org.apache.spark.sql.types._
 
 // scalastyle:off line.size.limit
 @ExpressionDescription(
@@ -44,62 +40,7 @@ import org.apache.spark.sql.types._
   group = "agg_funcs",
   since = "1.0.0")
 // scalastyle:on line.size.limit
-case class Count(children: Seq[Expression]) extends DeclarativeAggregate {
-
-  override def nullable: Boolean = false
-
-  // Return data type.
-  override def dataType: DataType = LongType
-
-  override def checkInputDataTypes(): TypeCheckResult = {
-    if (children.isEmpty && !SQLConf.get.getConf(SQLConf.ALLOW_PARAMETERLESS_COUNT)) {
-      TypeCheckResult.TypeCheckFailure(s"$prettyName requires at least one argument. " +
-        s"If you have to call the function $prettyName without arguments, set the legacy " +
-        s"configuration `${SQLConf.ALLOW_PARAMETERLESS_COUNT.key}` as true")
-    } else {
-      TypeCheckResult.TypeCheckSuccess
-    }
-  }
-
-  protected lazy val count = AttributeReference("count", LongType, nullable = false)()
-
-  override lazy val aggBufferAttributes = count :: Nil
-
-  override lazy val initialValues = Seq(
-    /* count = */ Literal(0L)
-  )
-
-  override lazy val mergeExpressions = Seq(
-    /* count = */ count.left + count.right
-  )
-
-  override lazy val evaluateExpression = count
-
-  override def defaultResult: Option[Literal] = Option(Literal(0L))
-
-  private[sql] var pushDown: Boolean = false
-
-  override lazy val updateExpressions = {
-    if (!pushDown) {
-      val nullableChildren = children.filter(_.nullable)
-      if (nullableChildren.isEmpty) {
-        Seq(
-          /* count = */ count + 1L
-        )
-      } else {
-        Seq(
-          /* count = */ If(nullableChildren.map(IsNull).reduce(Or), count, count + 1L)
-        )
-      }
-    } else {
-      Seq(
-        // if count is pushed down to Data Source layer, add the count result retrieved from
-        // Data Source
-        /* count = */ count + children.head
-      )
-    }
-  }
-}
+case class Count(children: Seq[Expression]) extends CountBase(children)
 
 object Count {
   def apply(child: Expression): Count = Count(child :: Nil)

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/CountBase.scala

@@ -0,0 +1,73 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.sql.catalyst.expressions.aggregate
+
+import org.apache.spark.sql.catalyst.analysis.TypeCheckResult
+import org.apache.spark.sql.catalyst.dsl.expressions._
+import org.apache.spark.sql.catalyst.expressions._
+import org.apache.spark.sql.internal.SQLConf
+import org.apache.spark.sql.types._
+
+abstract class CountBase(children: Seq[Expression]) extends DeclarativeAggregate {
+
+  override def nullable: Boolean = false
+
+  // Return data type.
+  override def dataType: DataType = LongType
+
+  override def checkInputDataTypes(): TypeCheckResult = {
+    if (children.isEmpty && !SQLConf.get.getConf(SQLConf.ALLOW_PARAMETERLESS_COUNT)) {
+      TypeCheckResult.TypeCheckFailure(s"$prettyName requires at least one argument. " +
+        s"If you have to call the function $prettyName without arguments, set the legacy " +
+        s"configuration `${SQLConf.ALLOW_PARAMETERLESS_COUNT.key}` as true")
+    } else {
+      TypeCheckResult.TypeCheckSuccess
+    }
+  }
+
+  protected lazy val count = AttributeReference("count", LongType, nullable = false)()
+
+  override lazy val aggBufferAttributes = count :: Nil
+
+  override lazy val initialValues = Seq(
+    /* count = */ Literal(0L)
+  )
+
+  override lazy val mergeExpressions = Seq(
+    /* count = */ count.left + count.right
+  )
+
+  override lazy val evaluateExpression = count
+
+  override def defaultResult: Option[Literal] = Option(Literal(0L))
+
+  override lazy val updateExpressions = {
+    val nullableChildren = children.filter(_.nullable)
+    if (nullableChildren.isEmpty) {
+      Seq(
+        /* count = */ count + 1L
+      )
+    } else {
+      Seq(
+        /* count = */ If(nullableChildren.map(IsNull).reduce(Or), count, count + 1L)
+      )
+    }
+  }
+}
+
+

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/aggregate/PushDownCount.scala

@@ -0,0 +1,41 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements.  See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License.  You may obtain a copy of the License at
+ *
+ *    http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.sql.catalyst.expressions.aggregate
+
+import org.apache.spark.sql.catalyst.dsl.expressions._
+import org.apache.spark.sql.catalyst.expressions._
+import org.apache.spark.sql.types.LongType
+
+case class PushDownCount(children: Seq[Expression]) extends CountBase(children) {
+
+  override protected lazy val count =
+    AttributeReference("PushDownCount", LongType, nullable = false)()
+
+  override lazy val updateExpressions = {
+    Seq(
+      // if count is pushed down to Data Source layer, add the count result retrieved from
+      // Data Source
+      /* count = */ count + children.head
+    )
+  }
+}
+
+object PushDownCount {
+  def apply(child: Expression): PushDownCount =
+    PushDownCount(child :: Nil)
+}

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/DataSourceStrategy.scala

@@ -721,7 +721,6 @@ object DataSourceStrategy
   }
 
   protected[sql] def translateAggregate(aggregates: AggregateExpression): Option[AggregateFunc] = {
-
     aggregates.aggregateFunction match {
       case min: aggregate.Min =>
         val colName = columnAsString(min.child)
@@ -737,8 +736,7 @@ object DataSourceStrategy
         if (colName.nonEmpty) Some(Sum(colName, sum.dataType, aggregates.isDistinct)) else None
       case count: aggregate.Count =>
         val columnName = count.children.head match {
-          case Literal(_, _) =>
-            "1"
+          case Literal(_, _) => "1"
           case _ => columnAsString(count.children.head)
         }
         if (columnName.nonEmpty) {

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/jdbc/JDBCRDD.scala

@@ -257,32 +257,30 @@ private[jdbc] class JDBCRDD(
    */
   override def getPartitions: Array[Partition] = partitions
 
-  private var updatedSchema: StructType = new StructType()
+  private val (updatedSchema, updatedCol): (StructType, Array[String]) =
+    if (aggregation.aggregateExpressions.isEmpty) {
+      (schema, columns)
+    } else {
+      getAggregateSchemaAndCol
+    }
 
   /**
    * `columns`, but as a String suitable for injection into a SQL query.
    */
   private val columnList: String = {
     val sb = new StringBuilder()
-    if(aggregation.aggregateExpressions.isEmpty) {
-      updatedSchema = schema
-      columns.foreach(x => sb.append(",").append(x))
-    } else {
-      val (compiledAgg, aggDataType) =
-        JDBCRDD.compileAggregates(aggregation.aggregateExpressions, JdbcDialects.get(url))
-        getAggregateColumnsList(sb, compiledAgg, aggDataType)
-    }
-    if (sb.length == 0) "1" else sb.substring(1)
+    updatedCol.foreach(x => sb.append(",").append(x))
+    if (sb.isEmpty) "1" else sb.substring(1)
   }
 
-  /*
+  /**
    * Build the column lists for Aggregates push down:
    * each of the Aggregates + groupBy columns
    */
-  private def getAggregateColumnsList(
-      sb: StringBuilder,
-      compiledAgg: Array[String],
-      aggDataType: Array[DataType]): Unit = {
+  private def getAggregateSchemaAndCol(): (StructType, Array[String]) = {
+    var updatedSchema: StructType = new StructType()
+    val (compiledAgg, aggDataType) =
+      JDBCRDD.compileAggregates(aggregation.aggregateExpressions, JdbcDialects.get(url))
     val colDataTypeMap: Map[String, StructField] = columns.zip(schema.fields).toMap
     val newColsBuilder = ArrayBuilder.make[String]
     for ((col, dataType) <- compiledAgg.zip(aggDataType)) {
@@ -294,7 +292,7 @@ private[jdbc] class JDBCRDD(
       newColsBuilder += quotedGroupBy
       updatedSchema = updatedSchema.add(colDataTypeMap.get(quotedGroupBy).get)
     }
-    sb.append(", ").append(newColsBuilder.result.mkString(", "))
+    (updatedSchema, newColsBuilder.result)
   }
 
   /**

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/V2ScanRelationPushDown.scala

@@ -125,9 +125,7 @@ object V2ScanRelationPushDown extends Rule[LogicalPlan] with AliasHelper {
                     } else if (agg.aggregateFunction.isInstanceOf[aggregate.Sum]) {
                       aggregate.Sum(aggOutput(i - 1))
                     } else if (agg.aggregateFunction.isInstanceOf[aggregate.Count]) {
-                      val count = aggregate.Count(aggOutput(i - 1))
-                      count.pushDown = true
-                      count
+                      aggregate.PushDownCount(aggOutput(i - 1))
                     } else {
                       agg.aggregateFunction
                     }