org.apache.flink.runtime.checkpoint

Class MappingBasedRepartitioner<T>

java.lang.Object

org.apache.flink.runtime.checkpoint.MappingBasedRepartitioner<T>

All Implemented Interfaces:

OperatorStateRepartitioner<T>

@NotThreadSafe
public class MappingBasedRepartitioner<T>
extends Object
implements OperatorStateRepartitioner<T>

A repartitioner that assigns the same channel state to multiple subtasks according to some mapping.

The replicated data will then be filtered before processing the record.

Note that channel mappings are cached for the same parallelism changes.

Constructor Summary

Constructors

Constructor and Description
MappingBasedRepartitioner(RescaleMappings newToOldSubtasksMapping)

Method Summary

Modifier and Type	Method and Description
List<List<T>>	repartitionState(List<List<T>> previousParallelSubtaskStates, int oldParallelism, int newParallelism)

Methods inherited from class java.lang.Object

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

Constructor Detail

MappingBasedRepartitioner

public MappingBasedRepartitioner(RescaleMappings newToOldSubtasksMapping)

Method Detail

repartitionState

public List<List<T>> repartitionState(List<List<T>> previousParallelSubtaskStates,
                                      int oldParallelism,
                                      int newParallelism)

Specified by:

repartitionState in interface OperatorStateRepartitioner<T>

Parameters:

previousParallelSubtaskStates - List with one entry of state handles per parallel subtask of an operator, as they have been checkpointed.

oldParallelism - The parallelism before we start redistribution.

newParallelism - The parallelism that we consider for the state redistribution. Determines the size of the returned list.

Returns:

List with one entry per parallel subtask. Each subtask receives now one collection of states that build of the new total state for this subtask.