我对Flink还不熟悉，我想知道Flink是怎么安排电话的 processElement() 在它的 KeyedProcessFunction 并行下的抽象。考虑以下生成部分和流的示例：

package sample

import org.apache.flink.api.common.state.{ValueState, ValueStateDescriptor}
import org.apache.flink.streaming.api.functions.KeyedProcessFunction
import org.apache.flink.streaming.api.scala.{DataStream, StreamExecutionEnvironment, createTypeInformation}
import org.apache.flink.util.Collector

object Playground {
  case class Record(groupId: String, score: Int) {}

  def main(args: Array[String]): Unit = {
    // 1. Create the environment
    val env: StreamExecutionEnvironment = StreamExecutionEnvironment.createLocalEnvironment()
    env.setParallelism(10)

    // 2. Source
    val record1 = Record("groupX", 1)
    val record2 = Record("groupX", 2)
    val record3 = Record("groupX", 3)
    val records: DataStream[Record] = env.fromElements(record1, record2, record3, record1, record2, record3)

    // 3. Application Logic
    val partialSums: DataStream[Int] = records
      .keyBy(record => record.groupId)
      .process(new KeyedProcessFunction[String, Record, Int] {
        // Store partial sum of score for Records seen
        lazy val partialSum: ValueState[Int] = getRuntimeContext.getState(
          new ValueStateDescriptor[Int]("partialSum", classOf[Int]))

        // Ingest new record
        override
        def processElement(value: Record,
                           ctx: KeyedProcessFunction[String, Record, Int]#Context,
                           out: Collector[Int]): Unit =
        {
          val currentSum: Int = partialSum.value()
          partialSum.update(currentSum + value.score)
          out.collect(partialSum.value())
        }
      })

    // 4. Sink
    partialSums.print()

    // 5. Build JobGraph and execute
    env.execute("sample-job")
  }
}

我希望它的输出是流： 1, 3, 6, 7, 9, 12 . 事实上，就在这里。
我们是否可以安全地假设情况总是这样，特别是在读取具有大量并行性的源代码时？