启动Spark Speculative后,有时候运行任务会发现如下提示:
WARN TaskSetManager: Lost task 55.0 in stage 15.0 (TID 20815, spark047216)
org.apache.spark.executor.CommitDeniedException: attempt_201604191557_0015_m_000055_0: Not committed because the driver did not authorize commit
启动 Speculative 后,运行较慢的task会在其他executor上同时再启动一个相同的task,如果其中一个task执行完毕,相同的另一个task就会被禁止提交。因此产生了这个WARN。
这个WARN是因为task提交commit被driver拒绝引发,这个错误不会被统计在stage的failure中,这样做的目的是防止你看到一些具有欺骗性的提示。
相关源码
org.apache.spark.executor
case cDE: CommitDeniedException =>
val reason = cDE.toTaskEndReason
execBackend.statusUpdate(taskId, TaskState.FAILED, ser.serialize(reason))
org.apache.spark.executor.CommitDeniedException
private[spark] class CommitDeniedException(
msg: String,
jobID: Int,
splitID: Int,
attemptNumber: Int)
extends Exception(msg) {
def toTaskEndReason: TaskEndReason = TaskCommitDenied(jobID, splitID, attemptNumber)
}
org.apache.spark.TaskCommitDenied
case class TaskCommitDenied(
jobID: Int,
partitionID: Int,
attemptNumber: Int) extends TaskFailedReason {
override def toErrorString: String = s"TaskCommitDenied (Driver denied task commit)" +
s" for job: $jobID, partition: $partitionID, attemptNumber: $attemptNumber"
override def countTowardsTaskFailures: Boolean = false
}