org.apache.flink.runtime.resourcemanager

Class ResourceManager<WorkerType extends ResourceIDRetrievable>

java.lang.Object

org.apache.flink.runtime.rpc.RpcEndpoint

org.apache.flink.runtime.rpc.FencedRpcEndpoint<ResourceManagerId>

org.apache.flink.runtime.resourcemanager.ResourceManager<WorkerType>

All Implemented Interfaces:

AutoCloseable, LeaderContender, ResourceManagerGateway, FencedRpcGateway<ResourceManagerId>, RpcGateway, org.apache.flink.util.AutoCloseableAsync

Direct Known Subclasses:

ActiveResourceManager, StandaloneResourceManager

public abstract class ResourceManager<WorkerType extends ResourceIDRetrievable>
extends FencedRpcEndpoint<ResourceManagerId>
implements ResourceManagerGateway, LeaderContender

ResourceManager implementation. The resource manager is responsible for resource de-/allocation and bookkeeping.

It offers the following methods as part of its rpc interface to interact with him remotely:

• registerJobManager(JobMasterId, ResourceID, String, JobID, Time) registers a JobMaster at the resource manager
• requestSlot(JobMasterId, SlotRequest, Time) requests a slot from the resource manager

Nested Class Summary

Nested classes/interfaces inherited from class org.apache.flink.runtime.rpc.RpcEndpoint

RpcEndpoint.MainThreadExecutor

Field Summary

Fields

Modifier and Type	Field and Description
static String	RESOURCE_MANAGER_NAME

Fields inherited from class org.apache.flink.runtime.rpc.RpcEndpoint

log, rpcServer

Constructor Summary

Constructors

Constructor and Description
ResourceManager(RpcService rpcService, String resourceManagerEndpointId, ResourceID resourceId, HighAvailabilityServices highAvailabilityServices, HeartbeatServices heartbeatServices, SlotManager slotManager, JobLeaderIdService jobLeaderIdService, ClusterInformation clusterInformation, FatalErrorHandler fatalErrorHandler, ResourceManagerMetricGroup resourceManagerMetricGroup)

Method Summary

Modifier and Type	Method and Description
void	cancelSlotRequest(AllocationID allocationID) Cancel the slot allocation requests from the resource manager.
protected CompletableFuture<Void>	clearStateAsync() This method can be overridden to add a (non-blocking) state clearing routine to the ResourceManager that will be called when leadership is revoked.
protected void	closeJobManagerConnection(org.apache.flink.api.common.JobID jobId, Exception cause) This method should be called by the framework once it detects that a currently registered job manager has failed.
protected void	closeTaskManagerConnection(ResourceID resourceID, Exception cause) This method should be called by the framework once it detects that a currently registered task executor has failed.
CompletableFuture<Acknowledge>	deregisterApplication(ApplicationStatus finalStatus, String diagnostics) Cleanup application and shut down cluster.
void	disconnectJobManager(org.apache.flink.api.common.JobID jobId, Exception cause) Disconnects a JobManager specified by the given resourceID from the ResourceManager.
void	disconnectTaskManager(ResourceID resourceId, Exception cause) Disconnects a TaskManager specified by the given resourceID from the ResourceManager.
CompletableFuture<Integer>	getNumberOfRegisteredTaskManagers() Gets the currently registered number of TaskManagers.
protected int	getNumberRequiredTaskManagerSlots()
void	grantLeadership(UUID newLeaderSessionID) Callback method when current resourceManager is granted leadership.
void	handleError(Exception exception) Handles error occurring in the leader election service.
void	heartbeatFromJobManager(ResourceID resourceID) Sends the heartbeat to resource manager from job manager
void	heartbeatFromTaskManager(ResourceID resourceID, TaskExecutorHeartbeatPayload heartbeatPayload) Sends the heartbeat to resource manager from task manager
protected abstract void	initialize() Initializes the framework specific components.
protected abstract void	internalDeregisterApplication(ApplicationStatus finalStatus, String optionalDiagnostics) The framework specific code to deregister the application.
protected void	jobLeaderLostLeadership(org.apache.flink.api.common.JobID jobId, JobMasterId oldJobMasterId)
void	notifySlotAvailable(InstanceID instanceID, SlotID slotId, AllocationID allocationId) Sent by the TaskExecutor to notify the ResourceManager that a slot has become available.
protected void	onFatalError(Throwable t) Notifies the ResourceManager that a fatal error has occurred and it cannot proceed.
void	onStart() User overridable callback which is called from RpcEndpoint.internalCallOnStart().
CompletableFuture<Void>	onStop() User overridable callback which is called from RpcEndpoint.internalCallOnStop().
protected CompletableFuture<Void>	prepareLeadershipAsync() This method can be overridden to add a (non-blocking) initialization routine to the ResourceManager that will be called when leadership is granted but before leadership is confirmed.
CompletableFuture<RegistrationResponse>	registerJobManager(JobMasterId jobMasterId, ResourceID jobManagerResourceId, String jobManagerAddress, org.apache.flink.api.common.JobID jobId, org.apache.flink.api.common.time.Time timeout) Register a JobMaster at the resource manager.
CompletableFuture<RegistrationResponse>	registerTaskExecutor(TaskExecutorRegistration taskExecutorRegistration, org.apache.flink.api.common.time.Time timeout) Register a TaskExecutor at the resource manager.
protected void	releaseResource(InstanceID instanceId, Exception cause)
protected void	removeJob(org.apache.flink.api.common.JobID jobId)
CompletableFuture<ResourceOverview>	requestResourceOverview(org.apache.flink.api.common.time.Time timeout) Requests the resource overview.
CompletableFuture<Acknowledge>	requestSlot(JobMasterId jobMasterId, SlotRequest slotRequest, org.apache.flink.api.common.time.Time timeout) Requests a slot from the resource manager.
CompletableFuture<TransientBlobKey>	requestTaskManagerFileUpload(ResourceID taskManagerId, FileType fileType, org.apache.flink.api.common.time.Time timeout) Request the file upload from the given TaskExecutor to the cluster's BlobServer.
CompletableFuture<TaskManagerInfo>	requestTaskManagerInfo(ResourceID resourceId, org.apache.flink.api.common.time.Time timeout) Requests information about the given TaskExecutor.
CompletableFuture<Collection<TaskManagerInfo>>	requestTaskManagerInfo(org.apache.flink.api.common.time.Time timeout) Requests information about the registered TaskExecutor.
CompletableFuture<Collection<org.apache.flink.api.java.tuple.Tuple2<ResourceID,String>>>	requestTaskManagerMetricQueryServiceAddresses(org.apache.flink.api.common.time.Time timeout) Requests the paths for the TaskManager's MetricQueryService to query.
void	revokeLeadership() Callback method when current resourceManager loses leadership.
CompletableFuture<Acknowledge>	sendSlotReport(ResourceID taskManagerResourceId, InstanceID taskManagerRegistrationId, SlotReport slotReport, org.apache.flink.api.common.time.Time timeout) Sends the given SlotReport to the ResourceManager.
protected void	setFailUnfulfillableRequest(boolean failUnfulfillableRequest) Set SlotManager whether to fail unfulfillable slot requests.
abstract Collection<ResourceProfile>	startNewWorker(ResourceProfile resourceProfile) Allocates a resource using the resource profile.
protected void	startServicesOnLeadership()
abstract boolean	stopWorker(WorkerType worker) Stops the given worker.
protected abstract WorkerType	workerStarted(ResourceID resourceID) Callback when a worker was started.

Methods inherited from class org.apache.flink.runtime.rpc.FencedRpcEndpoint

callAsyncWithoutFencing, getFencingToken, getMainThreadExecutor, getUnfencedMainThreadExecutor, runAsyncWithoutFencing, setFencingToken

Methods inherited from class org.apache.flink.runtime.rpc.RpcEndpoint

callAsync, closeAsync, getAddress, getEndpointId, getHostname, getRpcService, getSelfGateway, getTerminationFuture, internalCallOnStart, internalCallOnStop, isRunning, runAsync, scheduleRunAsync, scheduleRunAsync, start, stop, validateRunsInMainThread

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.apache.flink.runtime.rpc.FencedRpcGateway

getFencingToken

Methods inherited from interface org.apache.flink.runtime.rpc.RpcGateway

getAddress, getHostname

Methods inherited from interface org.apache.flink.runtime.leaderelection.LeaderContender

getDescription

Methods inherited from interface org.apache.flink.util.AutoCloseableAsync

close

Field Detail

RESOURCE_MANAGER_NAME

public static final String RESOURCE_MANAGER_NAME

See Also:

Constant Field Values

Constructor Detail

ResourceManager

public ResourceManager(RpcService rpcService,
                       String resourceManagerEndpointId,
                       ResourceID resourceId,
                       HighAvailabilityServices highAvailabilityServices,
                       HeartbeatServices heartbeatServices,
                       SlotManager slotManager,
                       JobLeaderIdService jobLeaderIdService,
                       ClusterInformation clusterInformation,
                       FatalErrorHandler fatalErrorHandler,
                       ResourceManagerMetricGroup resourceManagerMetricGroup)

Method Detail

onStart

public void onStart()
             throws Exception

Description copied from class: RpcEndpoint

User overridable callback which is called from RpcEndpoint.internalCallOnStart().

This method is called when the RpcEndpoint is being started. The method is guaranteed to be executed in the main thread context and can be used to start the rpc endpoint in the context of the rpc endpoint's main thread.

IMPORTANT: This method should never be called directly by the user.

Overrides:

onStart in class RpcEndpoint

Throws:

Exception - indicating that the rpc endpoint could not be started. If an exception occurs, then the rpc endpoint will automatically terminate.

onStop

public CompletableFuture<Void> onStop()

Description copied from class: RpcEndpoint

User overridable callback which is called from RpcEndpoint.internalCallOnStop().

This method is called when the RpcEndpoint is being shut down. The method is guaranteed to be executed in the main thread context and can be used to clean up internal state.

IMPORTANT: This method should never be called directly by the user.

Overrides:

onStop in class RpcEndpoint

Returns:

Future which is completed once all post stop actions are completed. If an error occurs this future is completed exceptionally

registerJobManager

public CompletableFuture<RegistrationResponse> registerJobManager(JobMasterId jobMasterId,
                                                                  ResourceID jobManagerResourceId,
                                                                  String jobManagerAddress,
                                                                  org.apache.flink.api.common.JobID jobId,
                                                                  org.apache.flink.api.common.time.Time timeout)

Description copied from interface: ResourceManagerGateway

Register a JobMaster at the resource manager.

Specified by:

registerJobManager in interface ResourceManagerGateway

Parameters:

jobMasterId - The fencing token for the JobMaster leader

jobManagerResourceId - The resource ID of the JobMaster that registers

jobManagerAddress - The address of the JobMaster that registers

jobId - The Job ID of the JobMaster that registers

timeout - Timeout for the future to complete

Returns:

Future registration response

registerTaskExecutor

public CompletableFuture<RegistrationResponse> registerTaskExecutor(TaskExecutorRegistration taskExecutorRegistration,
                                                                    org.apache.flink.api.common.time.Time timeout)

Description copied from interface: ResourceManagerGateway

Register a TaskExecutor at the resource manager.

Specified by:

registerTaskExecutor in interface ResourceManagerGateway

Parameters:

taskExecutorRegistration - the task executor registration.

timeout - The timeout for the response.

Returns:

The future to the response by the ResourceManager.

sendSlotReport

public CompletableFuture<Acknowledge> sendSlotReport(ResourceID taskManagerResourceId,
                                                     InstanceID taskManagerRegistrationId,
                                                     SlotReport slotReport,
                                                     org.apache.flink.api.common.time.Time timeout)

Description copied from interface: ResourceManagerGateway

Sends the given SlotReport to the ResourceManager.

Specified by:

sendSlotReport in interface ResourceManagerGateway

taskManagerRegistrationId - id identifying the sending TaskManager

slotReport - which is sent to the ResourceManager

timeout - for the operation

Returns:

Future which is completed with Acknowledge once the slot report has been received.

heartbeatFromTaskManager

public void heartbeatFromTaskManager(ResourceID resourceID,
                                     TaskExecutorHeartbeatPayload heartbeatPayload)

Description copied from interface: ResourceManagerGateway

Sends the heartbeat to resource manager from task manager

Specified by:

heartbeatFromTaskManager in interface ResourceManagerGateway

Parameters:

resourceID - unique id of the task manager

heartbeatPayload - payload from the originating TaskManager

heartbeatFromJobManager

public void heartbeatFromJobManager(ResourceID resourceID)

Description copied from interface: ResourceManagerGateway

Sends the heartbeat to resource manager from job manager

Specified by:

heartbeatFromJobManager in interface ResourceManagerGateway

Parameters:

resourceID - unique id of the job manager

disconnectTaskManager

public void disconnectTaskManager(ResourceID resourceId,
                                  Exception cause)

Description copied from interface: ResourceManagerGateway

Disconnects a TaskManager specified by the given resourceID from the ResourceManager.

Specified by:

disconnectTaskManager in interface ResourceManagerGateway

Parameters:

resourceId - identifying the TaskManager to disconnect

cause - for the disconnection of the TaskManager

disconnectJobManager

public void disconnectJobManager(org.apache.flink.api.common.JobID jobId,
                                 Exception cause)

Description copied from interface: ResourceManagerGateway

Disconnects a JobManager specified by the given resourceID from the ResourceManager.

Specified by:

disconnectJobManager in interface ResourceManagerGateway

Parameters:

jobId - JobID for which the JobManager was the leader

cause - for the disconnection of the JobManager

requestSlot

public CompletableFuture<Acknowledge> requestSlot(JobMasterId jobMasterId,
                                                  SlotRequest slotRequest,
                                                  org.apache.flink.api.common.time.Time timeout)

Description copied from interface: ResourceManagerGateway

Requests a slot from the resource manager.

Specified by:

requestSlot in interface ResourceManagerGateway

Parameters:

jobMasterId - id of the JobMaster

slotRequest - The slot to request

Returns:

The confirmation that the slot gets allocated

cancelSlotRequest

public void cancelSlotRequest(AllocationID allocationID)

Description copied from interface: ResourceManagerGateway

Cancel the slot allocation requests from the resource manager.

Specified by:

cancelSlotRequest in interface ResourceManagerGateway

Parameters:

allocationID - The slot to request

notifySlotAvailable

public void notifySlotAvailable(InstanceID instanceID,
                                SlotID slotId,
                                AllocationID allocationId)

Description copied from interface: ResourceManagerGateway

Sent by the TaskExecutor to notify the ResourceManager that a slot has become available.

Specified by:

notifySlotAvailable in interface ResourceManagerGateway

Parameters:

instanceID - TaskExecutor's instance id

slotId - The SlotID of the freed slot

allocationId - to which the slot has been allocated

deregisterApplication

public CompletableFuture<Acknowledge> deregisterApplication(ApplicationStatus finalStatus,
                                                            @Nullable
                                                            String diagnostics)

Cleanup application and shut down cluster.

Specified by:

deregisterApplication in interface ResourceManagerGateway

Parameters:

finalStatus - of the Flink application

diagnostics - diagnostics message for the Flink application or null

getNumberOfRegisteredTaskManagers

public CompletableFuture<Integer> getNumberOfRegisteredTaskManagers()

Description copied from interface: ResourceManagerGateway

Gets the currently registered number of TaskManagers.

Specified by:

getNumberOfRegisteredTaskManagers in interface ResourceManagerGateway

Returns:

The future to the number of registered TaskManagers.

requestTaskManagerInfo

public CompletableFuture<Collection<TaskManagerInfo>> requestTaskManagerInfo(org.apache.flink.api.common.time.Time timeout)

Description copied from interface: ResourceManagerGateway

Requests information about the registered TaskExecutor.

Specified by:

requestTaskManagerInfo in interface ResourceManagerGateway

Parameters:

timeout - of the request

Returns:

Future collection of TaskManager information

requestTaskManagerInfo

public CompletableFuture<TaskManagerInfo> requestTaskManagerInfo(ResourceID resourceId,
                                                                 org.apache.flink.api.common.time.Time timeout)

Description copied from interface: ResourceManagerGateway

Requests information about the given TaskExecutor.

Specified by:

requestTaskManagerInfo in interface ResourceManagerGateway

Parameters:

resourceId - identifying the TaskExecutor for which to return information

timeout - of the request

Returns:

Future TaskManager information

requestResourceOverview

public CompletableFuture<ResourceOverview> requestResourceOverview(org.apache.flink.api.common.time.Time timeout)

Description copied from interface: ResourceManagerGateway

Requests the resource overview. The resource overview provides information about the connected TaskManagers, the total number of slots and the number of available slots.

Specified by:

requestResourceOverview in interface ResourceManagerGateway

Parameters:

timeout - of the request

Returns:

Future containing the resource overview

requestTaskManagerMetricQueryServiceAddresses

public CompletableFuture<Collection<org.apache.flink.api.java.tuple.Tuple2<ResourceID,String>>> requestTaskManagerMetricQueryServiceAddresses(org.apache.flink.api.common.time.Time timeout)

Description copied from interface: ResourceManagerGateway

Requests the paths for the TaskManager's MetricQueryService to query.

Specified by:

requestTaskManagerMetricQueryServiceAddresses in interface ResourceManagerGateway

Parameters:

timeout - for the asynchronous operation

Returns:

Future containing the collection of resource ids and the corresponding metric query service path

requestTaskManagerFileUpload

public CompletableFuture<TransientBlobKey> requestTaskManagerFileUpload(ResourceID taskManagerId,
                                                                        FileType fileType,
                                                                        org.apache.flink.api.common.time.Time timeout)

Description copied from interface: ResourceManagerGateway

Request the file upload from the given TaskExecutor to the cluster's BlobServer. The corresponding TransientBlobKey is returned.

Specified by:

requestTaskManagerFileUpload in interface ResourceManagerGateway

Parameters:

taskManagerId - identifying the TaskExecutor to upload the specified file

fileType - type of the file to upload

timeout - for the asynchronous operation

Returns:

Future which is completed with the TransientBlobKey after uploading the file to the BlobServer.

closeJobManagerConnection

protected void closeJobManagerConnection(org.apache.flink.api.common.JobID jobId,
                                         Exception cause)

This method should be called by the framework once it detects that a currently registered job manager has failed.

Parameters:

jobId - identifying the job whose leader shall be disconnected.

cause - The exception which cause the JobManager failed.

closeTaskManagerConnection

protected void closeTaskManagerConnection(ResourceID resourceID,
                                          Exception cause)

This method should be called by the framework once it detects that a currently registered task executor has failed.

Parameters:

resourceID - Id of the TaskManager that has failed.

cause - The exception which cause the TaskManager failed.

removeJob

protected void removeJob(org.apache.flink.api.common.JobID jobId)

jobLeaderLostLeadership

protected void jobLeaderLostLeadership(org.apache.flink.api.common.JobID jobId,
                                       JobMasterId oldJobMasterId)

releaseResource

protected void releaseResource(InstanceID instanceId,
                               Exception cause)

onFatalError

protected void onFatalError(Throwable t)

Notifies the ResourceManager that a fatal error has occurred and it cannot proceed.

Parameters:

t - The exception describing the fatal error

grantLeadership

public void grantLeadership(UUID newLeaderSessionID)

Callback method when current resourceManager is granted leadership.

Specified by:

grantLeadership in interface LeaderContender

Parameters:

newLeaderSessionID - unique leadershipID

startServicesOnLeadership

protected void startServicesOnLeadership()

revokeLeadership

public void revokeLeadership()

Callback method when current resourceManager loses leadership.

Specified by:

revokeLeadership in interface LeaderContender

handleError

public void handleError(Exception exception)

Handles error occurring in the leader election service.

Specified by:

handleError in interface LeaderContender

Parameters:

exception - Exception being thrown in the leader election service

initialize

protected abstract void initialize()
                            throws ResourceManagerException

Initializes the framework specific components.

Throws:

ResourceManagerException - which occurs during initialization and causes the resource manager to fail.

prepareLeadershipAsync

protected CompletableFuture<Void> prepareLeadershipAsync()

This method can be overridden to add a (non-blocking) initialization routine to the ResourceManager that will be called when leadership is granted but before leadership is confirmed.

Returns:

Returns a CompletableFuture that completes when the computation is finished.

clearStateAsync

protected CompletableFuture<Void> clearStateAsync()

This method can be overridden to add a (non-blocking) state clearing routine to the ResourceManager that will be called when leadership is revoked.

Returns:

Returns a CompletableFuture that completes when the state clearing routine is finished.

internalDeregisterApplication

protected abstract void internalDeregisterApplication(ApplicationStatus finalStatus,
                                                      @Nullable
                                                      String optionalDiagnostics)
                                               throws ResourceManagerException

The framework specific code to deregister 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.

Parameters:

finalStatus - The application status to report.

optionalDiagnostics - A diagnostics message or null.

Throws:

ResourceManagerException - if the application could not be shut down.

startNewWorker

@VisibleForTesting
public abstract Collection<ResourceProfile> startNewWorker(ResourceProfile resourceProfile)

Allocates a resource using the resource profile.

Parameters:

resourceProfile - The resource description

Returns:

Collection of ResourceProfile describing the launched slots

workerStarted

protected abstract WorkerType workerStarted(ResourceID resourceID)

Callback when a worker was started.

Parameters:

resourceID - The worker resource id

stopWorker

public abstract boolean stopWorker(WorkerType worker)

Stops the given worker.

Parameters:

worker - The worker.

Returns:

True if the worker was stopped, otherwise false

setFailUnfulfillableRequest

protected void setFailUnfulfillableRequest(boolean failUnfulfillableRequest)

Set SlotManager whether to fail unfulfillable slot requests.

Parameters:

failUnfulfillableRequest - whether to fail unfulfillable requests

getNumberRequiredTaskManagerSlots

protected int getNumberRequiredTaskManagerSlots()