Github user davies commented on a diff in the pull request:
https://github.com/apache/spark/pull/9016#discussion_r41802119
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/columnar/ColumnType.scala ---
@@ -449,124 +416,126 @@ private[sql] object LARGE_DECIMAL {
}
}
-private[sql] case class STRUCT(dataType: StructType)
- extends ByteArrayColumnType[InternalRow](20) {
+private[sql] case class STRUCT(dataType: StructType) extends
ColumnType[UnsafeRow] {
- private val projection: UnsafeProjection =
- UnsafeProjection.create(dataType)
private val numOfFields: Int = dataType.fields.size
- override def setField(row: MutableRow, ordinal: Int, value:
InternalRow): Unit = {
+ override def defaultSize: Int = 20
+
+ override def setField(row: MutableRow, ordinal: Int, value: UnsafeRow):
Unit = {
row.update(ordinal, value)
}
- override def getField(row: InternalRow, ordinal: Int): InternalRow = {
- row.getStruct(ordinal, numOfFields)
+ override def getField(row: InternalRow, ordinal: Int): UnsafeRow = {
+ row.getStruct(ordinal, numOfFields).asInstanceOf[UnsafeRow]
}
- override def serialize(value: InternalRow): Array[Byte] = {
- val unsafeRow = if (value.isInstanceOf[UnsafeRow]) {
- value.asInstanceOf[UnsafeRow]
- } else {
- projection(value)
- }
- unsafeRow.getBytes
+ override def actualSize(row: InternalRow, ordinal: Int): Int = {
+ 4 + getField(row, ordinal).getSizeInBytes
+ }
+
+ override def append(value: UnsafeRow, buffer: ByteBuffer): Unit = {
+ buffer.putInt(value.getSizeInBytes)
+ value.writeTo(buffer)
}
- override def deserialize(bytes: Array[Byte]): InternalRow = {
+ override def extract(buffer: ByteBuffer): UnsafeRow = {
+ val sizeInBytes = buffer.getInt()
+ assert(buffer.hasArray)
+ val base = buffer.array()
+ val offset = buffer.arrayOffset()
+ val cursor = buffer.position()
+ buffer.position(cursor + sizeInBytes)
val unsafeRow = new UnsafeRow
- unsafeRow.pointTo(bytes, numOfFields, bytes.length)
+ unsafeRow.pointTo(base, Platform.BYTE_ARRAY_OFFSET + offset + cursor,
numOfFields, sizeInBytes)
unsafeRow
}
- override def clone(v: InternalRow): InternalRow = v.copy()
+ override def clone(v: UnsafeRow): UnsafeRow = v.copy()
}
-private[sql] case class ARRAY(dataType: ArrayType)
- extends ByteArrayColumnType[ArrayData](16) {
+private[sql] case class ARRAY(dataType: ArrayType) extends
ColumnType[UnsafeArrayData] {
- private lazy val projection =
UnsafeProjection.create(Array[DataType](dataType))
- private val mutableRow = new GenericMutableRow(new Array[Any](1))
+ override def defaultSize: Int = 16
- override def setField(row: MutableRow, ordinal: Int, value: ArrayData):
Unit = {
+ override def setField(row: MutableRow, ordinal: Int, value:
UnsafeArrayData): Unit = {
row.update(ordinal, value)
}
- override def getField(row: InternalRow, ordinal: Int): ArrayData = {
- row.getArray(ordinal)
+ override def getField(row: InternalRow, ordinal: Int): UnsafeArrayData =
{
+ row.getArray(ordinal).asInstanceOf[UnsafeArrayData]
}
- override def serialize(value: ArrayData): Array[Byte] = {
- val unsafeArray = if (value.isInstanceOf[UnsafeArrayData]) {
- value.asInstanceOf[UnsafeArrayData]
- } else {
- mutableRow(0) = value
- projection(mutableRow).getArray(0)
- }
- val outputBuffer =
- ByteBuffer.allocate(4 +
unsafeArray.getSizeInBytes).order(ByteOrder.nativeOrder())
- outputBuffer.putInt(unsafeArray.numElements())
- val underlying = outputBuffer.array()
- unsafeArray.writeToMemory(underlying, Platform.BYTE_ARRAY_OFFSET + 4)
- underlying
+ override def actualSize(row: InternalRow, ordinal: Int): Int = {
+ val unsafeArray = getField(row, ordinal)
+ 4 + 4 + unsafeArray.getSizeInBytes
}
- override def deserialize(bytes: Array[Byte]): ArrayData = {
- val buffer = ByteBuffer.wrap(bytes).order(ByteOrder.nativeOrder())
+ override def append(value: UnsafeArrayData, buffer: ByteBuffer): Unit = {
+ buffer.putInt(value.numElements())
+ buffer.putInt(value.getSizeInBytes)
+ value.writeTo(buffer)
--- End diff --
We could consolidate the format for array/map later.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
