了解Flink - 任务不可序列化

我正在研究一个Flink项目，并遇到了一个问题，我在Stackoverflow的帮助下设法解决了这个问题。但是，我不清楚为什么提出的解决方案实际可行，我发现有关该主题的信息很少。请考虑以下代码：

object DeCP {
def main(args: Array[String]): Unit = {

val params: ParameterTool = ParameterTool.fromArgs(args)

// Get the execution environment and read the data
val env: ExecutionEnvironment = ExecutionEnvironment.getExecutionEnvironment
val queryPoints: DataSet[Point] = readQueryPoints(env, params)
val points: DataSet[Point] = readFeatureVector(env, params)

// Process the query points
queryPoints
  .map(new KNNRich)
  .withBroadcastSet(points, "pointsIn")
  .print

}

final class KNNRich extends RichMapFunction[Point, (Point, Vector[Point])]{

private var pointsIn: Traversable[Point] = _

override def open(parameters: Configuration): Unit =
  pointsIn = getRuntimeContext.getBroadcastVariable[Point]("pointsIn").asScala

def map(queryPoint: Point): (Point, Vector[Point]) = {
  val dataSetIn = ExecutionEnvironment.getExecutionEnvironment
                                      .fromCollection(pointsIn.toVector)
  val cluster = new Cluster(dataSetIn, queryPoint)
  val knn = cluster.kNearestNeighbor(queryPoint, 3) // This call causes problems
  (queryPoint, knn.collect.toVector)
}

}
}
Cluster类和伴随对象定义为：

class Cluster(var points: DataSet[Point],

          var clusterLeader: Point) extends Serializable {

private var queryPoint: Point = _

def distance(p: Point): Point = {

p.eucDist(queryPoint)

}

def kNearestNeighbor(queryPoint: Point, k: Int): DataSet[Point] = {

this.queryPoint = queryPoint

this.points.map{p => distance(p)} // Task not serializable
this.points.map{p => p.eucDist(queryPoint)} // Works
this.points.map{p => Cluster.staticDistance(queryPoint, p)} // Works

}
}

object Cluster {
def staticDistance(queryPoint: Point, p: Point): Point = {

p.eucDist(queryPoint)

}
}
对distance方法的调用导致任务不可序列化异常，但用定义替换方法调用可以解决问题。类似地，定义与伴随对象的成员完全相同的方法允许代码正常运行。

为什么第一个call不起作用，但另外两个call有效？如果你在类上有一个更复杂的执行流程会发生什么，这不容易被替换为伴随对象上的方法？