spark-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From cloud-fan <...@git.apache.org>
Subject [GitHub] spark pull request #15821: [SPARK-13534][PySpark] Using Apache Arrow to incr...
Date Tue, 20 Jun 2017 09:49:33 GMT
Github user cloud-fan commented on a diff in the pull request:

    https://github.com/apache/spark/pull/15821#discussion_r122929684
  
    --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/arrow/ArrowConverters.scala
---
    @@ -0,0 +1,429 @@
    +/*
    +* 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.execution.arrow
    +
    +import java.io.ByteArrayOutputStream
    +import java.nio.channels.Channels
    +
    +import scala.collection.JavaConverters._
    +
    +import io.netty.buffer.ArrowBuf
    +import org.apache.arrow.memory.{BufferAllocator, RootAllocator}
    +import org.apache.arrow.vector._
    +import org.apache.arrow.vector.BaseValueVector.BaseMutator
    +import org.apache.arrow.vector.file._
    +import org.apache.arrow.vector.schema.{ArrowFieldNode, ArrowRecordBatch}
    +import org.apache.arrow.vector.types.FloatingPointPrecision
    +import org.apache.arrow.vector.types.pojo.{ArrowType, Field, FieldType, Schema}
    +import org.apache.arrow.vector.util.ByteArrayReadableSeekableByteChannel
    +
    +import org.apache.spark.sql.catalyst.InternalRow
    +import org.apache.spark.sql.types._
    +import org.apache.spark.util.Utils
    +
    +
    +/**
    + * Store Arrow data in a form that can be serialized by Spark and served to a Python
process.
    + */
    +private[sql] class ArrowPayload private[arrow] (payload: Array[Byte]) extends Serializable
{
    +
    +  /**
    +   * Convert the ArrowPayload to an ArrowRecordBatch.
    +   */
    +  def loadBatch(allocator: BufferAllocator): ArrowRecordBatch = {
    +    ArrowConverters.byteArrayToBatch(payload, allocator)
    +  }
    +
    +  /**
    +   * Get the ArrowPayload as a type that can be served to Python.
    +   */
    +  def asPythonSerializable: Array[Byte] = payload
    +}
    +
    +private[sql] object ArrowPayload {
    +
    +  /**
    +   * Create an ArrowPayload from an ArrowRecordBatch and Spark schema.
    +   */
    +  def apply(
    +      batch: ArrowRecordBatch,
    +      schema: StructType,
    +      allocator: BufferAllocator): ArrowPayload = {
    +    new ArrowPayload(ArrowConverters.batchToByteArray(batch, schema, allocator))
    +  }
    +}
    +
    +private[sql] object ArrowConverters {
    +
    +  /**
    +   * Map a Spark DataType to ArrowType.
    +   */
    +  private[arrow] def sparkTypeToArrowType(dataType: DataType): ArrowType = {
    +    dataType match {
    +      case BooleanType => ArrowType.Bool.INSTANCE
    +      case ShortType => new ArrowType.Int(8 * ShortType.defaultSize, true)
    +      case IntegerType => new ArrowType.Int(8 * IntegerType.defaultSize, true)
    +      case LongType => new ArrowType.Int(8 * LongType.defaultSize, true)
    +      case FloatType => new ArrowType.FloatingPoint(FloatingPointPrecision.SINGLE)
    +      case DoubleType => new ArrowType.FloatingPoint(FloatingPointPrecision.DOUBLE)
    +      case ByteType => new ArrowType.Int(8, true)
    +      case StringType => ArrowType.Utf8.INSTANCE
    +      case BinaryType => ArrowType.Binary.INSTANCE
    +      case _ => throw new UnsupportedOperationException(s"Unsupported data type: $dataType")
    +    }
    +  }
    +
    +  /**
    +   * Convert a Spark Dataset schema to Arrow schema.
    +   */
    +  private[arrow] def schemaToArrowSchema(schema: StructType): Schema = {
    +    val arrowFields = schema.fields.map { f =>
    +      new Field(f.name, f.nullable, sparkTypeToArrowType(f.dataType), List.empty[Field].asJava)
    +    }
    +    new Schema(arrowFields.toList.asJava)
    +  }
    +
    +  /**
    +   * Maps Iterator from InternalRow to ArrowPayload. Limit ArrowRecordBatch size in ArrowPayload
    +   * by setting maxRecordsPerBatch or use 0 to fully consume rowIter.
    +   */
    +  private[sql] def toPayloadIterator(
    +      rowIter: Iterator[InternalRow],
    +      schema: StructType,
    +      maxRecordsPerBatch: Int): Iterator[ArrowPayload] = {
    +    new Iterator[ArrowPayload] {
    +      private val _allocator = new RootAllocator(Long.MaxValue)
    +      private var _nextPayload = if (rowIter.nonEmpty) convert() else null
    +
    +      override def hasNext: Boolean = _nextPayload != null
    +
    +      override def next(): ArrowPayload = {
    +        val obj = _nextPayload
    +        if (hasNext) {
    +          if (rowIter.hasNext) {
    +            _nextPayload = convert()
    +          } else {
    +            _allocator.close()
    +            _nextPayload = null
    +          }
    +        }
    +        obj
    +      }
    +
    +      private def convert(): ArrowPayload = {
    +        val batch = internalRowIterToArrowBatch(rowIter, schema, _allocator, maxRecordsPerBatch)
    +        ArrowPayload(batch, schema, _allocator)
    +      }
    +    }
    +  }
    +
    +  /**
    +   * Iterate over InternalRows and write to an ArrowRecordBatch, stopping when rowIter
is consumed
    +   * or the number of records in the batch equals maxRecordsInBatch.  If maxRecordsPerBatch
is 0,
    +   * then rowIter will be fully consumed.
    +   */
    +  private def internalRowIterToArrowBatch(
    +      rowIter: Iterator[InternalRow],
    +      schema: StructType,
    +      allocator: BufferAllocator,
    +      maxRecordsPerBatch: Int = 0): ArrowRecordBatch = {
    +
    +    val columnWriters = schema.fields.zipWithIndex.map { case (field, ordinal) =>
    +      ColumnWriter(field.dataType, ordinal, allocator).init()
    --- End diff --
    
    I'm not quite sure about the best practice of arrow writing, but wouldn't it be better
to write the [root batch](https://github.com/apache/spark/pull/15821/files#diff-52cca47e7a940849b28d476ddf99d65eR182)
directly? 
    
    We can also address this in followup


---
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 infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org


Mime
View raw message