为什么Mutable映射在Spark中的UserDefinedAggregateFunction(UDAF)中自动变为不可变

Question

我试图在Spark中定义UserDefinedAggregateFunction(UDAF),它计算组的列中每个唯一值的出现次数.

这是一个例子:假设我有一个df这样的数据帧,

+----+----+
|col1|col2|
+----+----+
|   a|  a1|
|   a|  a1|
|   a|  a2|
|   b|  b1|
|   b|  b2|
|   b|  b3|
|   b|  b1|
|   b|  b1|
+----+----+

我将有一个UDAF DistinctValues

val func = new DistinctValues

然后我将它应用于数据帧df

val agg_value = df.groupBy("col1").agg(func(col("col2")).as("DV"))

我期待有这样的事情:

+----+--------------------------+
|col1|DV                        |
+----+--------------------------+
|   a|  Map(a1->2, a2->1)       |
|   b|  Map(b1->3, b2->1, b3->1)|
+----+--------------------------+

所以我推出了像这样的UDAF,

import org.apache.spark.sql.expressions.MutableAggregationBuffer
import org.apache.spark.sql.expressions.UserDefinedAggregateFunction
import org.apache.spark.sql.Row
import org.apache.spark.sql.types.StructType
import org.apache.spark.sql.types.StructField
import org.apache.spark.sql.types.DataType
import org.apache.spark.sql.types.ArrayType
import org.apache.spark.sql.types.StringType
import org.apache.spark.sql.types.MapType
import org.apache.spark.sql.types.LongType
import Array._

class DistinctValues extends UserDefinedAggregateFunction {
  def inputSchema: org.apache.spark.sql.types.StructType = StructType(StructField("value", StringType) :: Nil)

  def bufferSchema: StructType = StructType(StructField("values", MapType(StringType, LongType))::Nil)

  def dataType: DataType =  MapType(StringType, LongType)
  def deterministic: Boolean = true

  def initialize(buffer: MutableAggregationBuffer): Unit = {
    buffer(0) = scala.collection.mutable.Map()
  }

  def update(buffer: MutableAggregationBuffer, input: Row) : Unit = {
    val str = input.getAs[String](0)
    var mp = buffer.getAs[scala.collection.mutable.Map[String, Long]](0)
    var c:Long = mp.getOrElse(str, 0)
    c = c + 1
    mp.put(str, c)
    buffer(0) = mp
  }

  def merge(buffer1: MutableAggregationBuffer, buffer2: Row) : Unit = {
    var mp1 = buffer1.getAs[scala.collection.mutable.Map[String, Long]](0)
    var mp2 = buffer2.getAs[scala.collection.mutable.Map[String, Long]](0)
    mp2 foreach {
        case (k ,v) => {
            var c:Long = mp1.getOrElse(k, 0)
            c = c + v
            mp1.put(k ,c)
        }
    }
    buffer1(0) = mp1
  }

  def evaluate(buffer: Row): Any = {
      buffer.getAs[scala.collection.mutable.Map[String, LongType]](0)
  }
}

然后我在我的数据帧上有这个功能,

val func = new DistinctValues
val agg_values = df.groupBy("col1").agg(func(col("col2")).as("DV"))

它给出了这样的错误,

func: DistinctValues = $iwC$$iwC$DistinctValues@17f48a25
org.apache.spark.SparkException: Job aborted due to stage failure: Task 1 in stage 32.0 failed 4 times, most recent failure: Lost task 1.3 in stage 32.0 (TID 884, ip-172-31-22-166.ec2.internal): java.lang.ClassCastException: scala.collection.immutable.Map$EmptyMap$ cannot be cast to scala.collection.mutable.Map
at $iwC$$iwC$DistinctValues.update(<console>:39)
at org.apache.spark.sql.execution.aggregate.ScalaUDAF.update(udaf.scala:431)
at org.apache.spark.sql.execution.aggregate.AggregationIterator$$anonfun$12.apply(AggregationIterator.scala:187)
at org.apache.spark.sql.execution.aggregate.AggregationIterator$$anonfun$12.apply(AggregationIterator.scala:180)
at org.apache.spark.sql.execution.aggregate.SortBasedAggregationIterator.processCurrentSortedGroup(SortBasedAggregationIterator.scala:116)
at org.apache.spark.sql.execution.aggregate.SortBasedAggregationIterator.next(SortBasedAggregationIterator.scala:152)
at org.apache.spark.sql.execution.aggregate.SortBasedAggregationIterator.next(SortBasedAggregationIterator.scala:29)
at scala.collection.Iterator$$anon$11.next(Iterator.scala:328)
at scala.collection.Iterator$$anon$11.next(Iterator.scala:328)
at org.apache.spark.shuffle.sort.BypassMergeSortShuffleWriter.write(BypassMergeSortShuffleWriter.java:149)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:73)
at org.apache.spark.scheduler.ShuffleMapTask.runTask(ShuffleMapTask.scala:41)
at org.apache.spark.scheduler.Task.run(Task.scala:89)
at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
at java.lang.Thread.run(Thread.java:745)

它看起来像在update(buffer: MutableAggregationBuffer, input: Row)方法中,变量buffer是a immutable.Map,程序厌倦了把它投射到mutable.Map,

但我曾经在方法中mutable.Map初始化buffer变量initialize(buffer: MutableAggregationBuffer, input:Row).它是传递给update方法的相同变量吗？也buffer就是mutableAggregationBuffer,所以它应该是可变的,对不对？

为什么我的mutable.Map变得不可变？有谁知道发生了什么？

我真的需要在这个函数中使用一个可变的Map来完成任务.我知道有一种解决方法可以从不可变映射创建可变映射,然后更新它.但我真的想知道为什么可变的自动转换为程序中的不可变的,它对我来说没有意义.

Answer 1

相信它是MapType你的StructType.buffer因此持有一个Map,这是不可改变的.

你可以转换它,但为什么不把它保持不变,并执行此操作:

mp = mp + (k -> c)

添加一个条目到immutable Map？

下面的工作示例:

class DistinctValues extends UserDefinedAggregateFunction {
  def inputSchema: org.apache.spark.sql.types.StructType = StructType(StructField("_2", IntegerType) :: Nil)

  def bufferSchema: StructType = StructType(StructField("values", MapType(StringType, LongType))::Nil)

  def dataType: DataType =  MapType(StringType, LongType)
  def deterministic: Boolean = true

  def initialize(buffer: MutableAggregationBuffer): Unit = {
    buffer(0) = Map()
  }

  def update(buffer: MutableAggregationBuffer, input: Row) : Unit = {
    val str = input.getAs[String](0)
    var mp = buffer.getAs[Map[String, Long]](0)
    var c:Long = mp.getOrElse(str, 0)
    c = c + 1
    mp = mp  + (str -> c)
    buffer(0) = mp
  }

  def merge(buffer1: MutableAggregationBuffer, buffer2: Row) : Unit = {
    var mp1 = buffer1.getAs[Map[String, Long]](0)
    var mp2 = buffer2.getAs[Map[String, Long]](0)
    mp2 foreach {
        case (k ,v) => {
            var c:Long = mp1.getOrElse(k, 0)
            c = c + v
            mp1 = mp1 + (k -> c)
        }
    }
    buffer1(0) = mp1
  }

  def evaluate(buffer: Row): Any = {
      buffer.getAs[Map[String, LongType]](0)
  }
}