Skip to content
This repository has been archived by the owner on Sep 18, 2023. It is now read-only.

[NSE-523] Support ArrayType in ArrowColumnarToRow optimization #524

Merged
merged 2 commits into from
Oct 13, 2021
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
Expand Up @@ -102,15 +102,16 @@ class ArrowColumnarToRowExec(child: SparkPlan) extends ColumnarToRowExec(child =
} else {
val bufAddrs = new ListBuffer[Long]()
val bufSizes = new ListBuffer[Long]()
val recordBatch = ConverterUtils.createArrowRecordBatch(batch)
recordBatch.getBuffers().asScala.foreach { buffer => bufAddrs += buffer.memoryAddress() }
recordBatch.getBuffersLayout().asScala.foreach { bufLayout =>
bufSizes += bufLayout.getSize()
}

val fields = new ListBuffer[Field]()
(0 until batch.numCols).foreach { idx =>
val column = batch.column(idx).asInstanceOf[ArrowWritableColumnVector]
fields += column.getValueVector.getField
column.getValueVector.getBuffers(false)
.foreach { buffer =>
bufAddrs += buffer.memoryAddress()
bufSizes += buffer.readableBytes()
}
}

if (arrowSchema == null) {
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,357 @@
/*
* 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 com.intel.oap.execution

import java.time.ZoneId

import com.intel.oap.expression.ConverterUtils
import com.intel.oap.sql.execution.RowToColumnConverter
import com.intel.oap.vectorized.{ArrowColumnarToRowInfo, ArrowColumnarToRowJniWrapper, ArrowWritableColumnVector}
import org.apache.arrow.vector.types.pojo.{Field, Schema}
import org.apache.spark.sql.catalyst.InternalRow
import org.apache.spark.sql.catalyst.expressions.UnsafeRow
import org.apache.spark.sql.catalyst.util.{DateTimeUtils, GenericArrayData}
import org.apache.spark.sql.execution.datasources.v2.arrow.SparkMemoryUtils
import org.apache.spark.sql.execution.vectorized.WritableColumnVector
import org.apache.spark.sql.test.SharedSparkSession
import org.apache.spark.sql.types._
import org.apache.spark.sql.vectorized.ColumnarBatch
import org.apache.spark.unsafe.types.UTF8String

import scala.collection.JavaConverters._
import scala.collection.mutable.ListBuffer

class ArrowColumnarToRowExecSuite extends SharedSparkSession {

test("ArrowColumnarToRowExec: Boolean type with array list") {
val schema = StructType(Seq(StructField("boolean type with array", ArrayType(BooleanType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(Seq(true, false)))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(true == array.getBoolean(0))
assert(false == array.getBoolean(1))
}
}

test("ArrowColumnarToRowExec: Byte type with array list") {
val schema = StructType(Seq(StructField("boolean type with array", ArrayType(ByteType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(Seq(1.toByte, 2.toByte)))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(1.toByte == array.getByte(0))
assert(2.toByte == array.getByte(1))
}
}

test("ArrowColumnarToRowExec: Short type with array list") {
val schema = StructType(Seq(StructField("short type with array", ArrayType(ShortType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(Seq(1.toShort, 2.toShort)))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(1.toShort == array.getShort(0))
assert(2.toShort == array.getShort(1))
}
}

test("ArrowColumnarToRowExec: Int type with array list") {
val schema = StructType(Seq(StructField("Int type with array", ArrayType(IntegerType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(Seq(1, 2)))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(1 == array.getInt(0))
assert(2 == array.getInt(1))
}
}

test("ArrowColumnarToRowExec: Long type with array list") {
val schema = StructType(Seq(StructField("Long type with array", ArrayType(LongType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(Seq(1.toLong, 2.toLong)))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(1.toLong == array.getLong(0))
assert(2.toLong == array.getLong(1))
}
}

test("ArrowColumnarToRowExec: Float type with array list") {
val schema = StructType(Seq(StructField("Float type with array", ArrayType(FloatType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(Seq(1.toFloat, 2.toFloat)))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(1.toFloat == array.getFloat(0))
assert(2.toFloat == array.getFloat(1))
}
}

test("ArrowColumnarToRowExec: Double type with array list") {
val schema = StructType(Seq(StructField("Double type with array", ArrayType(DoubleType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(Seq(1.toDouble, 2.toDouble)))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(1.toDouble == array.getDouble(0))
assert(2.toDouble == array.getDouble(1))
}
}

test("ArrowColumnarToRowExec: String type with array list") {
val schema = StructType(Seq(StructField("String type with array", ArrayType(StringType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(
Seq(UTF8String.fromString("abc"), UTF8String.fromString("def"))))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(UTF8String.fromString("abc") == array.getUTF8String(0))
assert(UTF8String.fromString("def") == array.getUTF8String(1))
}
}

test("ArrowColumnarToRowExec: Decimal type with array list precision <= 18") {
val schema = StructType(
Seq(StructField("Decimal type with array", ArrayType(DecimalType(10, 2)))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(
Seq(new Decimal().set(BigDecimal("1.00")), new Decimal().set(BigDecimal("1.00")))))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(new Decimal().set(BigDecimal("1.00")) == array.getDecimal(0, 10, 2))
assert(new Decimal().set(BigDecimal("1.00")) == array.getDecimal(0, 10, 2))
}
}

test("ArrowColumnarToRowExec: Decimal type with array list precision > 18") {
val schema = StructType(
Seq(StructField("Decimal type with array", ArrayType(DecimalType(19, 2)))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(
Seq(new Decimal().set(BigDecimal("1.00")), new Decimal().set(BigDecimal("1.00")))))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(new Decimal().set(BigDecimal("1.00")) == array.getDecimal(0, 19, 2))
assert(new Decimal().set(BigDecimal("1.00")) == array.getDecimal(0, 19, 2))
}
}

test("ArrowColumnarToRowExec: Timestamp type with array list ") {
val defaultZoneId = ZoneId.systemDefault()
val schema = StructType(
Seq(StructField("Timestamp type with array", ArrayType(TimestampType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(
Seq( DateTimeUtils.stringToTimestamp(
UTF8String.fromString("1970-1-1 00:00:00"), defaultZoneId).get, DateTimeUtils.stringToTimestamp(
UTF8String.fromString("1970-1-1 00:00:00"), defaultZoneId).get)))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert( DateTimeUtils.stringToTimestamp(
UTF8String.fromString("1970-1-1 00:00:00"), defaultZoneId).get == array.get(0, TimestampType).asInstanceOf[Long])
assert( DateTimeUtils.stringToTimestamp(
UTF8String.fromString("1970-1-1 00:00:00"), defaultZoneId).get == array.get(1, TimestampType).asInstanceOf[Long])
}
}

test("ArrowColumnarToRowExec: Date32 type with array list ") {
val defaultZoneId = ZoneId.systemDefault()
val schema = StructType(
Seq(StructField("Date type with array", ArrayType(DateType))))
val rowIterator = (0 until 2).map { i =>
InternalRow(new GenericArrayData(
Seq(DateTimeUtils.stringToDate(UTF8String.fromString("1970-1-1"), defaultZoneId).get,
DateTimeUtils.stringToDate(UTF8String.fromString("1970-1-1"), defaultZoneId).get)))
}.toIterator

val cb = ArrowColumnarToRowExecSuite.createColumnarBatch(schema, rowIterator)
val info = ArrowColumnarToRowExecSuite.nativeOp(cb)

var rowId = 0
val row = new UnsafeRow(cb.numCols())
while (rowId < cb.numRows()) {
val (offset, length) = (info.offsets(rowId), info.lengths(rowId))
row.pointTo(null, info.memoryAddress + offset, length.toInt)
rowId += 1
val array = row.getArray(0)
assert(DateTimeUtils.stringToDate(UTF8String.fromString("1970-1-1"), defaultZoneId).get ==
array.get(0, DateType).asInstanceOf[Int])
assert(DateTimeUtils.stringToDate(UTF8String.fromString("1970-1-1"), defaultZoneId).get ==
array.get(1, DateType).asInstanceOf[Int])
}
}
}

object ArrowColumnarToRowExecSuite {

def serializeSchema(fields: Seq[Field]): Array[Byte] = {
val schema = new Schema(fields.asJava)
ConverterUtils.getSchemaBytesBuf(schema)
}

def nativeOp(batch: ColumnarBatch): ArrowColumnarToRowInfo = {
val bufAddrs = new ListBuffer[Long]()
val bufSizes = new ListBuffer[Long]()
val fields = new ListBuffer[Field]()

val recordBatch = ConverterUtils.createArrowRecordBatch(batch)
recordBatch.getBuffers().asScala.foreach { buffer => bufAddrs += buffer.memoryAddress() }
recordBatch.getBuffersLayout().asScala.foreach { bufLayout =>
bufSizes += bufLayout.getSize()
}

(0 until batch.numCols).foreach { idx =>
val column = batch.column(idx).asInstanceOf[ArrowWritableColumnVector]
fields += column.getValueVector.getField
}
val arrowSchema = serializeSchema(fields)
val jniWrapper = new ArrowColumnarToRowJniWrapper()
jniWrapper.nativeConvertColumnarToRow(
arrowSchema, batch.numRows(), bufAddrs.toArray, bufSizes.toArray,
SparkMemoryUtils.contextMemoryPool().getNativeInstanceId)
}

def createColumnarBatch(schema: StructType, rowIterator: Iterator[InternalRow]): ColumnarBatch = {
val converters = new RowToColumnConverter(schema)
val vectors: Seq[WritableColumnVector] =
ArrowWritableColumnVector.allocateColumns(1, schema)

var rowCount = 0
while (rowIterator.hasNext) {
val row = rowIterator.next()
converters.convert(row, vectors.toArray)
rowCount += 1
}

vectors.foreach(v => v.asInstanceOf[ArrowWritableColumnVector].setValueCount(rowCount))
new ColumnarBatch(vectors.toArray, rowCount)
}
}

Loading