org.apache.spark.scheduler

Interface TaskScheduler

public interface TaskScheduler

Low-level task scheduler interface, currently implemented exclusively by TaskSchedulerImpl. This interface allows plugging in different task schedulers. Each TaskScheduler schedules tasks for a single SparkContext. These schedulers get sets of tasks submitted to them from the DAGScheduler for each stage, and are responsible for sending the tasks to the cluster, running them, retrying if there are failures, and mitigating stragglers. They return events to the DAGScheduler.

Method Summary

Modifier and Type	Method and Description
scala.Option<String>	applicationAttemptId() Get an application's attempt ID associated with the job.
String	applicationId() Get an application ID associated with the job.
void	cancelTasks(int stageId, boolean interruptThread)
int	defaultParallelism()
void	executorDecommission(String executorId, org.apache.spark.scheduler.ExecutorDecommissionInfo decommissionInfo) Process a decommissioning executor.
boolean	executorHeartbeatReceived(String execId, scala.Tuple2<Object,scala.collection.Seq<AccumulatorV2<?,?>>>[] accumUpdates, BlockManagerId blockManagerId, scala.collection.mutable.Map<scala.Tuple2<Object,Object>,org.apache.spark.executor.ExecutorMetrics> executorUpdates) Update metrics for in-progress tasks and executor metrics, and let the master know that the BlockManager is still alive.
void	executorLost(String executorId, org.apache.spark.scheduler.ExecutorLossReason reason) Process a lost executor
scala.Option<org.apache.spark.scheduler.ExecutorDecommissionState>	getExecutorDecommissionState(String executorId) If an executor is decommissioned, return its corresponding decommission info
void	killAllTaskAttempts(int stageId, boolean interruptThread, String reason)
boolean	killTaskAttempt(long taskId, boolean interruptThread, String reason) Kills a task attempt.
void	notifyPartitionCompletion(int stageId, int partitionId)
void	postStartHook()
org.apache.spark.scheduler.Pool	rootPool()
scala.Enumeration.Value	schedulingMode()
void	setDAGScheduler(org.apache.spark.scheduler.DAGScheduler dagScheduler)
void	start()
void	stop(int exitCode)
void	submitTasks(org.apache.spark.scheduler.TaskSet taskSet)
void	workerRemoved(String workerId, String host, String message) Process a removed worker

Method Detail

applicationAttemptId

scala.Option<String> applicationAttemptId()

Get an application's attempt ID associated with the job.

Returns:

An application's Attempt ID

applicationId

String applicationId()

Get an application ID associated with the job.

Returns:

An application ID

cancelTasks

void cancelTasks(int stageId,
                 boolean interruptThread)

defaultParallelism

int defaultParallelism()

executorDecommission

void executorDecommission(String executorId,
                          org.apache.spark.scheduler.ExecutorDecommissionInfo decommissionInfo)

Process a decommissioning executor.

Parameters:

executorId - (undocumented)

decommissionInfo - (undocumented)

executorHeartbeatReceived

boolean executorHeartbeatReceived(String execId,
                                  scala.Tuple2<Object,scala.collection.Seq<AccumulatorV2<?,?>>>[] accumUpdates,
                                  BlockManagerId blockManagerId,
                                  scala.collection.mutable.Map<scala.Tuple2<Object,Object>,org.apache.spark.executor.ExecutorMetrics> executorUpdates)

Update metrics for in-progress tasks and executor metrics, and let the master know that the BlockManager is still alive. Return true if the driver knows about the given block manager. Otherwise, return false, indicating that the block manager should re-register.

Parameters:

execId - (undocumented)

accumUpdates - (undocumented)

blockManagerId - (undocumented)

executorUpdates - (undocumented)

Returns:

(undocumented)

executorLost

void executorLost(String executorId,
                  org.apache.spark.scheduler.ExecutorLossReason reason)

Process a lost executor

Parameters:

executorId - (undocumented)

reason - (undocumented)

getExecutorDecommissionState

scala.Option<org.apache.spark.scheduler.ExecutorDecommissionState> getExecutorDecommissionState(String executorId)

If an executor is decommissioned, return its corresponding decommission info

Parameters:

executorId - (undocumented)

Returns:

(undocumented)

killAllTaskAttempts

void killAllTaskAttempts(int stageId,
                         boolean interruptThread,
                         String reason)

killTaskAttempt

boolean killTaskAttempt(long taskId,
                        boolean interruptThread,
                        String reason)

Kills a task attempt. Throw UnsupportedOperationException if the backend doesn't support kill a task.

Parameters:

taskId - (undocumented)

interruptThread - (undocumented)

reason - (undocumented)

Returns:

Whether the task was successfully killed.

notifyPartitionCompletion

void notifyPartitionCompletion(int stageId,
                               int partitionId)

postStartHook

void postStartHook()

rootPool

org.apache.spark.scheduler.Pool rootPool()

schedulingMode

scala.Enumeration.Value schedulingMode()

setDAGScheduler

void setDAGScheduler(org.apache.spark.scheduler.DAGScheduler dagScheduler)

start

void start()

stop

void stop(int exitCode)

submitTasks

void submitTasks(org.apache.spark.scheduler.TaskSet taskSet)

workerRemoved

void workerRemoved(String workerId,
                   String host,
                   String message)

Process a removed worker

Parameters:

workerId - (undocumented)

host - (undocumented)

message - (undocumented)