org.apache.flink.runtime.clusterframework.standalone

Class StandaloneResourceManager

java.lang.Object

akka.actor.UntypedActor

org.apache.flink.runtime.akka.FlinkUntypedActor

org.apache.flink.runtime.clusterframework.FlinkResourceManager<ResourceID>

org.apache.flink.runtime.clusterframework.standalone.StandaloneResourceManager

All Implemented Interfaces:

akka.actor.Actor

Direct Known Subclasses:

TestingResourceManager

public class StandaloneResourceManager
extends FlinkResourceManager<ResourceID>

A standalone implementation of the resource manager. Used when the system is started in standalone mode (via scripts), rather than via a resource framework like YARN or Mesos.

Nested Class Summary

Nested classes/interfaces inherited from interface akka.actor.Actor

akka.actor.Actor.emptyBehavior$

Field Summary

Fields inherited from class org.apache.flink.runtime.clusterframework.FlinkResourceManager

config, EXIT_CODE_FATAL_ERROR, RESOURCE_MANAGER_NAME

Fields inherited from class org.apache.flink.runtime.akka.FlinkUntypedActor

LOG

Constructor Summary

Constructors

Constructor and Description
StandaloneResourceManager(org.apache.flink.configuration.Configuration flinkConfig, LeaderRetrievalService leaderRetriever)

Method Summary

Modifier and Type	Method and Description
protected void	fatalError(String message, Throwable error) Notifies the resource master of a fatal error.
protected int	getNumWorkerRequestsPending() Gets the number of requested workers that have not yet been granted.
protected int	getNumWorkersPendingRegistration() Gets the number of containers that have been started, but where the TaskManager has not yet registered at the job manager.
protected void	handleMessage(Object message) This method receives the actor messages after they have been filtered for a match with the leader session.
protected void	initialize() Initializes the framework specific components.
protected Collection<ResourceID>	reacceptRegisteredWorkers(Collection<ResourceID> toConsolidate) This method is called when the resource manager starts after a failure and reconnects to the leader JobManager, who still has some workers registered.
protected void	releasePendingWorker(ResourceID resourceID) Trigger a release of a pending worker.
protected void	releaseStartedWorker(ResourceID resourceID) Trigger a release of a started worker.
protected void	requestNewWorkers(int numWorkers) Requests to allocate a certain number of new workers.
protected void	shutdownApplication(ApplicationStatus finalStatus, String optionalDiagnostics) The framework specific code for shutting down the application.
protected void	triggerConnectingToJobManager(String leaderAddress) Causes the resource manager to announce itself at the new leader JobManager and obtains its connection information and currently known TaskManagers.
protected ResourceID	workerStarted(ResourceID resourceID) Callback when a worker was started.

Methods inherited from class org.apache.flink.runtime.clusterframework.FlinkResourceManager

allStartedWorkers, getDesignatedWorkerPoolSize, getLeaderSessionID, getNumberOfStartedTaskManagers, getStartedTaskManagers, isStarted, notifyWorkerFailed, postStop, preStart, sendInfoMessage, startResourceManagerActors, startResourceManagerActors, triggerCheckWorkers

Methods inherited from class org.apache.flink.runtime.akka.FlinkUntypedActor

decorateMessage, onReceive

Methods inherited from class akka.actor.UntypedActor

akka$actor$Actor$_setter_$context_$eq, akka$actor$Actor$_setter_$self_$eq, aroundPostRestart, aroundPostStop, aroundPreRestart, aroundPreStart, aroundReceive, context, getContext, getSelf, getSender, postRestart, preRestart, receive, self, sender, supervisorStrategy, unhandled

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

StandaloneResourceManager

public StandaloneResourceManager(org.apache.flink.configuration.Configuration flinkConfig,
                                 LeaderRetrievalService leaderRetriever)

Method Detail

triggerConnectingToJobManager

protected void triggerConnectingToJobManager(String leaderAddress)

Description copied from class: FlinkResourceManager

Causes the resource manager to announce itself at the new leader JobManager and obtains its connection information and currently known TaskManagers.

Overrides:

triggerConnectingToJobManager in class FlinkResourceManager<ResourceID>

Parameters:

leaderAddress - The akka actor URL of the new leader JobManager.

initialize

protected void initialize()
                   throws Exception

Description copied from class: FlinkResourceManager

Initializes the framework specific components.

Specified by:

initialize in class FlinkResourceManager<ResourceID>

Throws:

Exception - Exceptions during initialization cause the resource manager to fail. If the framework is able to recover this resource manager, it will be restarted.

shutdownApplication

protected void shutdownApplication(ApplicationStatus finalStatus,
                                   String optionalDiagnostics)

Description copied from class: FlinkResourceManager

The framework specific code for shutting down the application. This should report the application's final status and shut down the resource manager cleanly. This method also needs to make sure all pending containers that are not registered yet are returned.

Specified by:

shutdownApplication in class FlinkResourceManager<ResourceID>

Parameters:

finalStatus - The application status to report.

optionalDiagnostics - An optional diagnostics message.

fatalError

protected void fatalError(String message,
                          Throwable error)

Description copied from class: FlinkResourceManager

Notifies the resource master of a fatal error.

IMPORTANT: This should not cleanly shut down this master, but exit it in such a way that a high-availability setting would restart this or fail over to another master.

Specified by:

fatalError in class FlinkResourceManager<ResourceID>

reacceptRegisteredWorkers

protected Collection<ResourceID> reacceptRegisteredWorkers(Collection<ResourceID> toConsolidate)

Description copied from class: FlinkResourceManager

This method is called when the resource manager starts after a failure and reconnects to the leader JobManager, who still has some workers registered. The method is used to consolidate the view between resource manager and JobManager. The resource manager gets the list of TaskManagers that the JobManager considers available and should return a list or nodes that the resource manager considers available. After that, the JobManager is informed of loss of all TaskManagers that are not part of the returned list. It is possible that the resource manager initially confirms some TaskManagers to be alive, even through they are in an uncertain status, if it later sends necessary failure notifications via calling FlinkResourceManager.notifyWorkerFailed(ResourceID, String).

Specified by:

reacceptRegisteredWorkers in class FlinkResourceManager<ResourceID>

Parameters:

toConsolidate - The list of TaskManagers that the JobManager knows.

Returns:

The subset of TaskManagers that the resource manager can confirm to be alive.

requestNewWorkers

protected void requestNewWorkers(int numWorkers)

Description copied from class: FlinkResourceManager

Requests to allocate a certain number of new workers.

Specified by:

requestNewWorkers in class FlinkResourceManager<ResourceID>

Parameters:

numWorkers - The number of workers to allocate.

workerStarted

protected ResourceID workerStarted(ResourceID resourceID)

Description copied from class: FlinkResourceManager

Callback when a worker was started.

Specified by:

workerStarted in class FlinkResourceManager<ResourceID>

Parameters:

resourceID - The worker resource id

releaseStartedWorker

protected void releaseStartedWorker(ResourceID resourceID)

Description copied from class: FlinkResourceManager

Trigger a release of a started worker.

Specified by:

releaseStartedWorker in class FlinkResourceManager<ResourceID>

Parameters:

resourceID - The worker resource id

releasePendingWorker

protected void releasePendingWorker(ResourceID resourceID)

Description copied from class: FlinkResourceManager

Trigger a release of a pending worker.

Specified by:

releasePendingWorker in class FlinkResourceManager<ResourceID>

Parameters:

resourceID - The worker resource id

getNumWorkerRequestsPending

protected int getNumWorkerRequestsPending()

Description copied from class: FlinkResourceManager

Gets the number of requested workers that have not yet been granted.

Specified by:

getNumWorkerRequestsPending in class FlinkResourceManager<ResourceID>

Returns:

The number pending worker requests. Must never be smaller than 0.

getNumWorkersPendingRegistration

protected int getNumWorkersPendingRegistration()

Description copied from class: FlinkResourceManager

Gets the number of containers that have been started, but where the TaskManager has not yet registered at the job manager.

Specified by:

getNumWorkersPendingRegistration in class FlinkResourceManager<ResourceID>

Returns:

The number of started containers pending TaskManager registration. Must never be smaller than 0.

handleMessage

protected void handleMessage(Object message)

Description copied from class: FlinkResourceManager

This method receives the actor messages after they have been filtered for a match with the leader session.

Overrides:

handleMessage in class FlinkResourceManager<ResourceID>

Parameters:

message - The incoming actor message.