[SPARK-24762][SQL] Enable Option of Product encoders

## What changes were proposed in this pull request?

SparkSQL doesn't support to encode `Option[Product]` as a top-level row now, because in SparkSQL entire top-level row can't be null.

However for use cases like Aggregator, it is reasonable to use `Option[Product]` as buffer and output column types. Due to above limitation, we don't do it for now.

This patch proposes to encode `Option[Product]` at top-level as single struct column. So we can work around the issue that entire top-level row can't be null.

To summarize encoding of `Product` and `Option[Product]`.

For `Product`, 1. at root level, the schema is all fields are flatten it into multiple columns. The `Product ` can't be null, otherwise it throws an exception.

```scala
val df = Seq((1 -> "a"), (2 -> "b")).toDF()
df.printSchema()

root
 |-- _1: integer (nullable = false)
 |-- _2: string (nullable = true)
```

2. At non-root level, `Product` is a struct type column.

```scala
val df = Seq((1, (1 -> "a")), (2, (2 -> "b")), (3, null)).toDF()
df.printSchema()

root
 |-- _1: integer (nullable = false)
 |-- _2: struct (nullable = true)
 |    |-- _1: integer (nullable = false)
 |    |-- _2: string (nullable = true)
```

For `Option[Product]`, 1. it was not supported at root level. After this change, it is a struct type column.

```scala
val df = Seq(Some(1 -> "a"), Some(2 -> "b"), None).toDF()
df.printSchema

root
 |-- value: struct (nullable = true)
 |    |-- _1: integer (nullable = false)
 |    |-- _2: string (nullable = true)
```

2. At non-root level, it is also a struct type column.

```scala
val df = Seq((1, Some(1 -> "a")), (2, Some(2 -> "b")), (3, None)).toDF()
df.printSchema

root
 |-- _1: integer (nullable = false)
 |-- _2: struct (nullable = true)
 |    |-- _1: integer (nullable = false)
 |    |-- _2: string (nullable = true)
```

3. For use case like Aggregator, it was not supported too. After this change, we support to use `Option[Product]` as buffer/output column type.

```scala
val df = Seq(
    OptionBooleanIntData("bob", Some((true, 1))),
    OptionBooleanIntData("bob", Some((false, 2))),
    OptionBooleanIntData("bob", None)).toDF()

val group = df
    .groupBy("name")
    .agg(OptionBooleanIntAggregator("isGood").toColumn.alias("isGood"))
group.printSchema

root
 |-- name: string (nullable = true)
 |-- isGood: struct (nullable = true)
 |    |-- _1: boolean (nullable = false)
 |    |-- _2: integer (nullable = false)
```

The buffer and output type of `OptionBooleanIntAggregator` is both `Option[(Boolean, Int)`.

## How was this patch tested?

Added test.

Closes apache#21732 from viirya/SPARK-24762.

Authored-by: Liang-Chi Hsieh <[email protected]>
Signed-off-by: Wenchen Fan <[email protected]>

Author: viirya

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/encoders/ExpressionEncoder.scala

@@ -49,15 +49,6 @@ object ExpressionEncoder {
     val mirror = ScalaReflection.mirror
     val tpe = typeTag[T].in(mirror).tpe
 
-    if (ScalaReflection.optionOfProductType(tpe)) {
-      throw new UnsupportedOperationException(
-        "Cannot create encoder for Option of Product type, because Product type is represented " +
-          "as a row, and the entire row can not be null in Spark SQL like normal databases. " +
-          "You can wrap your type with Tuple1 if you do want top level null Product objects, " +
-          "e.g. instead of creating `Dataset[Option[MyClass]]`, you can do something like " +
-          "`val ds: Dataset[Tuple1[MyClass]] = Seq(Tuple1(MyClass(...)), Tuple1(null)).toDS`")
-    }
-
     val cls = mirror.runtimeClass(tpe)
     val serializer = ScalaReflection.serializerForType(tpe)
     val deserializer = ScalaReflection.deserializerForType(tpe)
@@ -198,7 +189,7 @@ case class ExpressionEncoder[T](
   val serializer: Seq[NamedExpression] = {
     val clsName = Utils.getSimpleName(clsTag.runtimeClass)
 
-    if (isSerializedAsStruct) {
+    if (isSerializedAsStructForTopLevel) {
       val nullSafeSerializer = objSerializer.transformUp {
         case r: BoundReference =>
           // For input object of Product type, we can't encode it to row if it's null, as Spark SQL
@@ -213,6 +204,9 @@ case class ExpressionEncoder[T](
     } else {
       // For other input objects like primitive, array, map, etc., we construct a struct to wrap
       // the serializer which is a column of an row.
+      //
+      // Note: Because Spark SQL doesn't allow top-level row to be null, to encode
+      // top-level Option[Product] type, we make it as a top-level struct column.
       CreateNamedStruct(Literal("value") :: objSerializer :: Nil)
     }
   }.flatten
@@ -226,7 +220,7 @@ case class ExpressionEncoder[T](
    * `GetColumnByOrdinal` with corresponding ordinal.
    */
   val deserializer: Expression = {
-    if (isSerializedAsStruct) {
+    if (isSerializedAsStructForTopLevel) {
       // We serialized this kind of objects to root-level row. The input of general deserializer
       // is a `GetColumnByOrdinal(0)` expression to extract first column of a row. We need to
       // transform attributes accessors.
@@ -253,10 +247,24 @@ case class ExpressionEncoder[T](
   })
 
   /**
-   * Returns true if the type `T` is serialized as a struct.
+   * Returns true if the type `T` is serialized as a struct by `objSerializer`.
    */
   def isSerializedAsStruct: Boolean = objSerializer.dataType.isInstanceOf[StructType]
 
+  /**
+   * Returns true if the type `T` is an `Option` type.
+   */
+  def isOptionType: Boolean = classOf[Option[_]].isAssignableFrom(clsTag.runtimeClass)
+
+  /**
+   * If the type `T` is serialized as a struct, when it is encoded to a Spark SQL row, fields in
+   * the struct are naturally mapped to top-level columns in a row. In other words, the serialized
+   * struct is flattened to row. But in case of the `T` is also an `Option` type, it can't be
+   * flattened to top-level row, because in Spark SQL top-level row can't be null. This method
+   * returns true if `T` is serialized as struct and is not `Option` type.
+   */
+  def isSerializedAsStructForTopLevel: Boolean = isSerializedAsStruct && !isOptionType
+
   // serializer expressions are used to encode an object to a row, while the object is usually an
   // intermediate value produced inside an operator, not from the output of the child operator. This
   // is quite different from normal expressions, and `AttributeReference` doesn't work here

sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala

@@ -1093,7 +1093,7 @@ class Dataset[T] private[sql](
     // Note that we do this before joining them, to enable the join operator to return null for one
     // side, in cases like outer-join.
     val left = {
-      val combined = if (!this.exprEnc.isSerializedAsStruct) {
+      val combined = if (!this.exprEnc.isSerializedAsStructForTopLevel) {
         assert(joined.left.output.length == 1)
         Alias(joined.left.output.head, "_1")()
       } else {
@@ -1103,7 +1103,7 @@ class Dataset[T] private[sql](
     }
 
     val right = {
-      val combined = if (!other.exprEnc.isSerializedAsStruct) {
+      val combined = if (!other.exprEnc.isSerializedAsStructForTopLevel) {
         assert(joined.right.output.length == 1)
         Alias(joined.right.output.head, "_2")()
       } else {
@@ -1116,14 +1116,14 @@ class Dataset[T] private[sql](
     // combine the outputs of each join side.
     val conditionExpr = joined.condition.get transformUp {
       case a: Attribute if joined.left.outputSet.contains(a) =>
-        if (!this.exprEnc.isSerializedAsStruct) {
+        if (!this.exprEnc.isSerializedAsStructForTopLevel) {
           left.output.head
         } else {
           val index = joined.left.output.indexWhere(_.exprId == a.exprId)
           GetStructField(left.output.head, index)
         }
       case a: Attribute if joined.right.outputSet.contains(a) =>
-        if (!other.exprEnc.isSerializedAsStruct) {
+        if (!other.exprEnc.isSerializedAsStructForTopLevel) {
           right.output.head
         } else {
           val index = joined.right.output.indexWhere(_.exprId == a.exprId)
@@ -1396,7 +1396,7 @@ class Dataset[T] private[sql](
     implicit val encoder = c1.encoder
     val project = Project(c1.withInputType(exprEnc, logicalPlan.output).named :: Nil, logicalPlan)
 
-    if (!encoder.isSerializedAsStruct) {
+    if (!encoder.isSerializedAsStructForTopLevel) {
       new Dataset[U1](sparkSession, project, encoder)
     } else {
       // Flattens inner fields of U1

sql/core/src/main/scala/org/apache/spark/sql/KeyValueGroupedDataset.scala

@@ -458,7 +458,7 @@ class KeyValueGroupedDataset[K, V] private[sql](
     val encoders = columns.map(_.encoder)
     val namedColumns =
       columns.map(_.withInputType(vExprEnc, dataAttributes).named)
-    val keyColumn = if (!kExprEnc.isSerializedAsStruct) {
+    val keyColumn = if (!kExprEnc.isSerializedAsStructForTopLevel) {
       assert(groupingAttributes.length == 1)
       if (SQLConf.get.nameNonStructGroupingKeyAsValue) {
         groupingAttributes.head

sql/core/src/main/scala/org/apache/spark/sql/execution/aggregate/TypedAggregateExpression.scala

@@ -40,9 +40,9 @@ object TypedAggregateExpression {
     val outputEncoder = encoderFor[OUT]
     val outputType = outputEncoder.objSerializer.dataType
 
-    // Checks if the buffer object is simple, i.e. the buffer encoder is flat and the serializer
-    // expression is an alias of `BoundReference`, which means the buffer object doesn't need
-    // serialization.
+    // Checks if the buffer object is simple, i.e. the `BUF` type is not serialized as struct
+    // and the serializer expression is an alias of `BoundReference`, which means the buffer
+    // object doesn't need serialization.
     val isSimpleBuffer = {
       bufferSerializer.head match {
         case Alias(_: BoundReference, _) if !bufferEncoder.isSerializedAsStruct => true
@@ -76,7 +76,7 @@ object TypedAggregateExpression {
         None,
         bufferSerializer,
         bufferEncoder.resolveAndBind().deserializer,
-        outputEncoder.serializer,
+        outputEncoder.objSerializer,
         outputType,
         outputEncoder.objSerializer.nullable)
     }
@@ -213,7 +213,7 @@ case class ComplexTypedAggregateExpression(
     inputSchema: Option[StructType],
     bufferSerializer: Seq[NamedExpression],
     bufferDeserializer: Expression,
-    outputSerializer: Seq[Expression],
+    outputSerializer: Expression,
     dataType: DataType,
     nullable: Boolean,
     mutableAggBufferOffset: Int = 0,
@@ -245,13 +245,7 @@ case class ComplexTypedAggregateExpression(
     aggregator.merge(buffer, input)
   }
 
-  private lazy val resultObjToRow = dataType match {
-    case _: StructType =>
-      UnsafeProjection.create(CreateStruct(outputSerializer))
-    case _ =>
-      assert(outputSerializer.length == 1)
-      UnsafeProjection.create(outputSerializer.head)
-  }
+  private lazy val resultObjToRow = UnsafeProjection.create(outputSerializer)
 
   override def eval(buffer: Any): Any = {
     val resultObj = aggregator.finish(buffer)

sql/core/src/test/scala/org/apache/spark/sql/DatasetAggregatorSuite.scala

@@ -22,7 +22,7 @@ import org.apache.spark.sql.expressions.Aggregator
 import org.apache.spark.sql.expressions.scalalang.typed
 import org.apache.spark.sql.functions._
 import org.apache.spark.sql.test.SharedSQLContext
-import org.apache.spark.sql.types.StringType
+import org.apache.spark.sql.types.{BooleanType, IntegerType, StringType, StructType}
 
 
 object ComplexResultAgg extends Aggregator[(String, Int), (Long, Long), (Long, Long)] {
@@ -149,6 +149,7 @@ object VeryComplexResultAgg extends Aggregator[Row, String, ComplexAggData] {
 
 
 case class OptionBooleanData(name: String, isGood: Option[Boolean])
+case class OptionBooleanIntData(name: String, isGood: Option[(Boolean, Int)])
 
 case class OptionBooleanAggregator(colName: String)
     extends Aggregator[Row, Option[Boolean], Option[Boolean]] {
@@ -183,6 +184,43 @@ case class OptionBooleanAggregator(colName: String)
   def OptionalBoolEncoder: Encoder[Option[Boolean]] = ExpressionEncoder()
 }
 
+case class OptionBooleanIntAggregator(colName: String)
+    extends Aggregator[Row, Option[(Boolean, Int)], Option[(Boolean, Int)]] {
+
+  override def zero: Option[(Boolean, Int)] = None
+
+  override def reduce(buffer: Option[(Boolean, Int)], row: Row): Option[(Boolean, Int)] = {
+    val index = row.fieldIndex(colName)
+    val value = if (row.isNullAt(index)) {
+      Option.empty[(Boolean, Int)]
+    } else {
+      val nestedRow = row.getStruct(index)
+      Some((nestedRow.getBoolean(0), nestedRow.getInt(1)))
+    }
+    merge(buffer, value)
+  }
+
+  override def merge(
+      b1: Option[(Boolean, Int)],
+      b2: Option[(Boolean, Int)]): Option[(Boolean, Int)] = {
+    if ((b1.isDefined && b1.get._1) || (b2.isDefined && b2.get._1)) {
+      val newInt = b1.map(_._2).getOrElse(0) + b2.map(_._2).getOrElse(0)
+      Some((true, newInt))
+    } else if (b1.isDefined) {
+      b1
+    } else {
+      b2
+    }
+  }
+
+  override def finish(reduction: Option[(Boolean, Int)]): Option[(Boolean, Int)] = reduction
+
+  override def bufferEncoder: Encoder[Option[(Boolean, Int)]] = OptionalBoolIntEncoder
+  override def outputEncoder: Encoder[Option[(Boolean, Int)]] = OptionalBoolIntEncoder
+
+  def OptionalBoolIntEncoder: Encoder[Option[(Boolean, Int)]] = ExpressionEncoder()
+}
+
 class DatasetAggregatorSuite extends QueryTest with SharedSQLContext {
   import testImplicits._
 
@@ -393,4 +431,28 @@ class DatasetAggregatorSuite extends QueryTest with SharedSQLContext {
     assert(grouped.schema == df.schema)
     checkDataset(grouped.as[OptionBooleanData], OptionBooleanData("bob", Some(true)))
   }
+
+  test("SPARK-24762: Aggregator should be able to use Option of Product encoder") {
+    val df = Seq(
+      OptionBooleanIntData("bob", Some((true, 1))),
+      OptionBooleanIntData("bob", Some((false, 2))),
+      OptionBooleanIntData("bob", None)).toDF()
+
+    val group = df
+      .groupBy("name")
+      .agg(OptionBooleanIntAggregator("isGood").toColumn.alias("isGood"))
+
+    val expectedSchema = new StructType()
+      .add("name", StringType, nullable = true)
+      .add("isGood",
+        new StructType()
+          .add("_1", BooleanType, nullable = false)
+          .add("_2", IntegerType, nullable = false),
+        nullable = true)
+
+    assert(df.schema == expectedSchema)
+    assert(group.schema == expectedSchema)
+    checkAnswer(group, Row("bob", Row(true, 3)) :: Nil)
+    checkDataset(group.as[OptionBooleanIntData], OptionBooleanIntData("bob", Some((true, 3))))
+  }
 }

sql/core/src/test/scala/org/apache/spark/sql/DatasetSuite.scala

@@ -1312,15 +1312,6 @@ class DatasetSuite extends QueryTest with SharedSQLContext {
     checkDataset(dsString, arrayString)
   }
 
-  test("SPARK-18251: the type of Dataset can't be Option of Product type") {
-    checkDataset(Seq(Some(1), None).toDS(), Some(1), None)
-
-    val e = intercept[UnsupportedOperationException] {
-      Seq(Some(1 -> "a"), None).toDS()
-    }
-    assert(e.getMessage.contains("Cannot create encoder for Option of Product type"))
-  }
-
   test ("SPARK-17460: the sizeInBytes in Statistics shouldn't overflow to a negative number") {
     // Since the sizeInBytes in Statistics could exceed the limit of an Int, we should use BigInt
     // instead of Int for avoiding possible overflow.
@@ -1558,6 +1549,74 @@ class DatasetSuite extends QueryTest with SharedSQLContext {
       Seq(Row("Amsterdam")))
   }
 
+  test("SPARK-24762: Enable top-level Option of Product encoders") {
+    val data = Seq(Some((1, "a")), Some((2, "b")), None)
+    val ds = data.toDS()
+
+    checkDataset(
+      ds,
+      data: _*)
+
+    val schema = new StructType().add(
+      "value",
+      new StructType()
+        .add("_1", IntegerType, nullable = false)
+        .add("_2", StringType, nullable = true),
+      nullable = true)
+
+    assert(ds.schema == schema)
+
+    val nestedOptData = Seq(Some((Some((1, "a")), 2.0)), Some((Some((2, "b")), 3.0)))
+    val nestedDs = nestedOptData.toDS()
+
+    checkDataset(
+      nestedDs,
+      nestedOptData: _*)
+
+    val nestedSchema = StructType(Seq(
+      StructField("value", StructType(Seq(
+        StructField("_1", StructType(Seq(
+          StructField("_1", IntegerType, nullable = false),
+          StructField("_2", StringType, nullable = true)))),
+        StructField("_2", DoubleType, nullable = false)
+      )), nullable = true)
+    ))
+    assert(nestedDs.schema == nestedSchema)
+  }
+
+  test("SPARK-24762: Resolving Option[Product] field") {
+    val ds = Seq((1, ("a", 1.0)), (2, ("b", 2.0)), (3, null)).toDS()
+      .as[(Int, Option[(String, Double)])]
+    checkDataset(ds,
+      (1, Some(("a", 1.0))), (2, Some(("b", 2.0))), (3, None))
+  }
+
+  test("SPARK-24762: select Option[Product] field") {
+    val ds = Seq(("a", 1), ("b", 2), ("c", 3)).toDS()
+    val ds1 = ds.select(expr("struct(_2, _2 + 1)").as[Option[(Int, Int)]])
+    checkDataset(ds1,
+      Some((1, 2)), Some((2, 3)), Some((3, 4)))
+
+    val ds2 = ds.select(expr("if(_2 > 2, struct(_2, _2 + 1), null)").as[Option[(Int, Int)]])
+    checkDataset(ds2,
+      None, None, Some((3, 4)))
+  }
+
+  test("SPARK-24762: joinWith on Option[Product]") {
+    val ds1 = Seq(Some((1, 2)), Some((2, 3)), None).toDS().as("a")
+    val ds2 = Seq(Some((1, 2)), Some((2, 3)), None).toDS().as("b")
+    val joined = ds1.joinWith(ds2, $"a.value._1" === $"b.value._2", "inner")
+    checkDataset(joined, (Some((2, 3)), Some((1, 2))))
+  }
+
+  test("SPARK-24762: typed agg on Option[Product] type") {
+    val ds = Seq(Some((1, 2)), Some((2, 3)), Some((1, 3))).toDS()
+    assert(ds.groupByKey(_.get._1).count().collect() === Seq((1, 2), (2, 1)))
+
+    assert(ds.groupByKey(x => x).count().collect() ===
+      Seq((Some((1, 2)), 1), (Some((2, 3)), 1), (Some((1, 3)), 1)))
+  }
+
   test("SPARK-25942: typed aggregation on primitive type") {
     val ds = Seq(1, 2, 3).toDS()