org.apache.flink.runtime.checkpoint

Class SavepointCoordinator

java.lang.Object

org.apache.flink.runtime.checkpoint.CheckpointCoordinator

org.apache.flink.runtime.checkpoint.SavepointCoordinator

public class SavepointCoordinator
extends CheckpointCoordinator

The savepoint coordinator is a slightly modified variant of the regular checkpoint coordinator. Checkpoints are not triggered periodically, but manually. The actual checkpointing mechanism is the same as for periodic checkpoints, only the control flow is modified.

The savepoint coordinator is meant to be used as a separate coordinator instance. Otherwise, there can be unwanted queueing effects like discarding savepoints, because of in-progress periodic checkpoints.

The savepoint coordinator registers callbacks on the regular checkpoint life-cycle and manages a map of promises, which are completed/failed as soon as the trigged checkpoint is done.

Important: it's necessary that both the periodic checkpoint coordinator and the savepoint coordinator share the same instance of the CheckpointIDCounter to ensure that all task managers see ascending checkpoints IDs.

Field Summary

Fields inherited from class org.apache.flink.runtime.checkpoint.CheckpointCoordinator

checkpointIdCounter, lock

Constructor Summary

Constructors

Constructor and Description
SavepointCoordinator(org.apache.flink.api.common.JobID jobId, long baseInterval, long checkpointTimeout, ExecutionVertex[] tasksToTrigger, ExecutionVertex[] tasksToWaitFor, ExecutionVertex[] tasksToCommitTo, ClassLoader userClassLoader, CheckpointIDCounter checkpointIDCounter, StateStore<CompletedCheckpoint> savepointStore, CheckpointStatsTracker statsTracker)

Method Summary

Modifier and Type	Method and Description
ActorGateway	createActivatorDeactivator(akka.actor.ActorSystem actorSystem, UUID leaderSessionID)
String	getSavepointRestorePath()
protected void	onCancelCheckpoint(long canceledCheckpointId) Callback on cancellation of a checkpoint.
protected void	onFullyAcknowledgedCheckpoint(CompletedCheckpoint checkpoint) Callback on full acknowledgement of a checkpoint.
protected void	onShutdown() Callback on shutdown of the coordinator.
void	restoreSavepoint(Map<JobVertexID,ExecutionJobVertex> tasks, String savepointPath) Resets the state of Execution instances back to the state of a savepoint.
scala.concurrent.Future<String>	triggerSavepoint(long timestamp) Triggers a new savepoint using the current system time as the checkpoint timestamp.

Methods inherited from class org.apache.flink.runtime.checkpoint.CheckpointCoordinator

getAndIncrementCheckpointId, getJobStatusListener, getNumberOfPendingCheckpoints, getNumberOfRetainedSuccessfulCheckpoints, getPendingCheckpoints, getSuccessfulCheckpoints, isShutdown, receiveAcknowledgeMessage, receiveDeclineMessage, restoreLatestCheckpointedState, setJobStatusListener, shutdown, startCheckpointScheduler, stopCheckpointScheduler, triggerCheckpoint, triggerCheckpoint

Methods inherited from class java.lang.Object

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

Constructor Detail

SavepointCoordinator

public SavepointCoordinator(org.apache.flink.api.common.JobID jobId,
                            long baseInterval,
                            long checkpointTimeout,
                            ExecutionVertex[] tasksToTrigger,
                            ExecutionVertex[] tasksToWaitFor,
                            ExecutionVertex[] tasksToCommitTo,
                            ClassLoader userClassLoader,
                            CheckpointIDCounter checkpointIDCounter,
                            StateStore<CompletedCheckpoint> savepointStore,
                            CheckpointStatsTracker statsTracker)
                     throws Exception

Throws:

Exception

Method Detail

getSavepointRestorePath

public String getSavepointRestorePath()

triggerSavepoint

public scala.concurrent.Future<String> triggerSavepoint(long timestamp)
                                                 throws Exception

Triggers a new savepoint using the current system time as the checkpoint timestamp.

Throws:

Exception

restoreSavepoint

public void restoreSavepoint(Map<JobVertexID,ExecutionJobVertex> tasks,
                             String savepointPath)
                      throws Exception

Resets the state of Execution instances back to the state of a savepoint.

The execution vertices need to be in state ExecutionState.CREATED when calling this method. The operation might block. Make sure that calls don't block the job manager actor.

Parameters:

tasks - Tasks that will possibly be reset

savepointPath - The path of the savepoint to rollback to

Throws:

IllegalStateException - If coordinator is shut down

IllegalStateException - If mismatch between program and savepoint state

Exception - If savepoint store failure

onShutdown

protected void onShutdown()

Description copied from class: CheckpointCoordinator

Callback on shutdown of the coordinator. Called in lock scope.

Overrides:

onShutdown in class CheckpointCoordinator

onCancelCheckpoint

protected void onCancelCheckpoint(long canceledCheckpointId)

Description copied from class: CheckpointCoordinator

Callback on cancellation of a checkpoint. Called in lock scope.

Overrides:

onCancelCheckpoint in class CheckpointCoordinator

onFullyAcknowledgedCheckpoint

protected void onFullyAcknowledgedCheckpoint(CompletedCheckpoint checkpoint)

Description copied from class: CheckpointCoordinator

Callback on full acknowledgement of a checkpoint. Called in lock scope.

Overrides:

onFullyAcknowledgedCheckpoint in class CheckpointCoordinator

createActivatorDeactivator

public ActorGateway createActivatorDeactivator(akka.actor.ActorSystem actorSystem,
                                               UUID leaderSessionID)

Overrides:

createActivatorDeactivator in class CheckpointCoordinator