org.apache.flink.optimizer.dag

Class SingleInputNode

java.lang.Object

org.apache.flink.optimizer.dag.OptimizerNode

org.apache.flink.optimizer.dag.SingleInputNode

All Implemented Interfaces:

EstimateProvider, DumpableNode<OptimizerNode>, Visitable<OptimizerNode>

Direct Known Subclasses:

BulkIterationNode, FilterNode, FlatMapNode, GroupCombineNode, GroupReduceNode, MapNode, MapPartitionNode, PartitionNode, ReduceNode, SortPartitionNode, UnaryOperatorNode

public abstract class SingleInputNode
extends OptimizerNode

A node in the optimizer's program representation for an operation with a single input.

This class contains all the generic logic for handling branching flows, as well as to enumerate candidate execution plans. The subclasses for specific operators simply add logic for cost estimates and specify possible strategies for their execution.

Nested Class Summary

Nested classes/interfaces inherited from class org.apache.flink.optimizer.dag.OptimizerNode

OptimizerNode.UnclosedBranchDescriptor

Field Summary

Fields

Modifier and Type	Field and Description
protected DagConnection	inConn
protected FieldSet	keys

Fields inherited from class org.apache.flink.optimizer.dag.OptimizerNode

cachedPlans, closedBranchingNodes, costWeight, estimatedNumRecords, estimatedOutputSize, hereJoinedBranches, id, MAX_DYNAMIC_PATH_COST_WEIGHT, onDynamicPath, openBranches, uniqueFields

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	SingleInputNode()
protected	SingleInputNode(FieldSet keys)
protected	SingleInputNode(SingleInputNode toCopy)
protected	SingleInputNode(SingleInputOperator<?,?,?> programOperator) Creates a new node with a single input for the optimizer plan.

Method Summary

Modifier and Type	Method and Description
void	accept(Visitor<OptimizerNode> visitor) This method implements the visit of a depth-first graph traversing visitor.
protected void	addLocalCandidates(Channel template, List<Set<? extends NamedChannel>> broadcastPlanChannels, RequestedGlobalProperties rgps, List<PlanNode> target, CostEstimator estimator)
void	computeInterestingPropertiesForInputs(CostEstimator estimator) Tells the node to compute the interesting properties for its inputs.
void	computeUnclosedBranchStack() This method causes the node to compute the description of open branches in its sub-plan.
List<PlanNode>	getAlternativePlans(CostEstimator estimator) Computes the plan alternatives for this node, an implicitly for all nodes that are children of this node.
DagConnection	getIncomingConnection() Gets the input of this operator.
List<DagConnection>	getIncomingConnections() Gets all incoming connections of this node.
SingleInputOperator<?,?,?>	getOperator() Gets the operator represented by this optimizer node.
protected abstract List<OperatorDescriptorSingle>	getPossibleProperties()
OptimizerNode	getPredecessorNode() Gets the predecessor of this node.
SemanticProperties	getSemanticProperties()
protected SemanticProperties	getSemanticPropertiesForGlobalPropertyFiltering()
protected SemanticProperties	getSemanticPropertiesForLocalPropertyFiltering()
protected void	instantiateCandidate(OperatorDescriptorSingle dps, Channel in, List<Set<? extends NamedChannel>> broadcastPlanChannels, List<PlanNode> target, CostEstimator estimator, RequestedGlobalProperties globPropsReq, RequestedLocalProperties locPropsReq)
void	setIncomingConnection(DagConnection inConn) Sets the connection through which this node receives its input.
void	setInput(Map<Operator<?>,OptimizerNode> contractToNode, ExecutionMode defaultExchangeMode) This function connects the predecessors to this operator.

Methods inherited from class org.apache.flink.optimizer.dag.OptimizerNode

addBroadcastConnection, addClosedBranch, addClosedBranches, addOutgoingConnection, areBranchCompatible, clearInterestingProperties, computeOperatorSpecificDefaultEstimates, computeOutputEstimates, computeUnclosedBranchStackForBroadcastInputs, computeUnionOfInterestingPropertiesFromSuccessors, getBranchesForParent, getBroadcastConnectionNames, getBroadcastConnections, getClosedBranchingNodes, getCostWeight, getDumpableInputs, getEstimatedAvgWidthPerOutputRecord, getEstimatedNumRecords, getEstimatedOutputSize, getId, getInterestingProperties, getMaxDepth, getMinimalMemoryAcrossAllSubTasks, getOpenBranches, getOperatorName, getOptimizerNode, getOutgoingConnections, getParallelism, getPlanNode, getPredecessors, getUniqueFields, hasUnclosedBranches, haveAllOutputConnectionInterestingProperties, identifyDynamicPath, initId, isBranching, isOnDynamicPath, markAllOutgoingConnectionsAsPipelineBreaking, mergeLists, prunePlanAlternatives, prunePlanAlternativesWithCommonBranching, readStubAnnotations, readUniqueFieldsAnnotation, removeClosedBranches, setBroadcastInputs, setEstimatedNumRecords, setEstimatedOutputSize, setParallelism, toString

Methods inherited from class java.lang.Object

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

Field Detail

keys

protected final FieldSet keys

inConn

protected DagConnection inConn

Constructor Detail

SingleInputNode

protected SingleInputNode(SingleInputOperator<?,?,?> programOperator)

Creates a new node with a single input for the optimizer plan.

Parameters:

programOperator - The PACT that the node represents.

SingleInputNode

protected SingleInputNode(FieldSet keys)

SingleInputNode

protected SingleInputNode()

SingleInputNode

protected SingleInputNode(SingleInputNode toCopy)

Method Detail

getOperator

public SingleInputOperator<?,?,?> getOperator()

Description copied from class: OptimizerNode

Gets the operator represented by this optimizer node.

Overrides:

getOperator in class OptimizerNode

Returns:

This node's operator.

getIncomingConnection

public DagConnection getIncomingConnection()

Gets the input of this operator.

Returns:

The input.

setIncomingConnection

public void setIncomingConnection(DagConnection inConn)

Sets the connection through which this node receives its input.

Parameters:

inConn - The input connection to set.

getPredecessorNode

public OptimizerNode getPredecessorNode()

Gets the predecessor of this node.

Returns:

The predecessor of this node.

getIncomingConnections

public List<DagConnection> getIncomingConnections()

Description copied from class: OptimizerNode

Gets all incoming connections of this node. This method needs to be overridden by subclasses to return the children.

Specified by:

getIncomingConnections in class OptimizerNode

Returns:

The list of incoming connections.

getSemanticProperties

public SemanticProperties getSemanticProperties()

Specified by:

getSemanticProperties in class OptimizerNode

getSemanticPropertiesForLocalPropertyFiltering

protected SemanticProperties getSemanticPropertiesForLocalPropertyFiltering()

getSemanticPropertiesForGlobalPropertyFiltering

protected SemanticProperties getSemanticPropertiesForGlobalPropertyFiltering()

setInput

public void setInput(Map<Operator<?>,OptimizerNode> contractToNode,
                     ExecutionMode defaultExchangeMode)
              throws CompilerException

Description copied from class: OptimizerNode

This function connects the predecessors to this operator.

Specified by:

setInput in class OptimizerNode

Parameters:

contractToNode - The map from program operators to optimizer nodes.

defaultExchangeMode - The data exchange mode to use, if the operator does not specify one.

Throws:

CompilerException

getPossibleProperties

protected abstract List<OperatorDescriptorSingle> getPossibleProperties()

computeInterestingPropertiesForInputs

public void computeInterestingPropertiesForInputs(CostEstimator estimator)

Description copied from class: OptimizerNode

Tells the node to compute the interesting properties for its inputs. The interesting properties for the node itself must have been computed before. The node must then see how many of interesting properties it preserves and add its own.

Specified by:

computeInterestingPropertiesForInputs in class OptimizerNode

Parameters:

estimator - The CostEstimator instance to use for plan cost estimation.

getAlternativePlans

public List<PlanNode> getAlternativePlans(CostEstimator estimator)

Description copied from class: OptimizerNode

Computes the plan alternatives for this node, an implicitly for all nodes that are children of this node. This method must determine for each alternative the global and local properties and the costs. This method may recursively call getAlternatives() on its children to get their plan alternatives, and build its own alternatives on top of those.

Specified by:

getAlternativePlans in class OptimizerNode

Parameters:

estimator - The cost estimator used to estimate the costs of each plan alternative.

Returns:

A list containing all plan alternatives.

addLocalCandidates

protected void addLocalCandidates(Channel template,
                                  List<Set<? extends NamedChannel>> broadcastPlanChannels,
                                  RequestedGlobalProperties rgps,
                                  List<PlanNode> target,
                                  CostEstimator estimator)

instantiateCandidate

protected void instantiateCandidate(OperatorDescriptorSingle dps,
                                    Channel in,
                                    List<Set<? extends NamedChannel>> broadcastPlanChannels,
                                    List<PlanNode> target,
                                    CostEstimator estimator,
                                    RequestedGlobalProperties globPropsReq,
                                    RequestedLocalProperties locPropsReq)

computeUnclosedBranchStack

public void computeUnclosedBranchStack()

Description copied from class: OptimizerNode

This method causes the node to compute the description of open branches in its sub-plan. An open branch describes, that a (transitive) child node had multiple outputs, which have not all been re-joined in the sub-plan. This method needs to set the openBranches field to a stack of unclosed branches, the latest one top. A branch is considered closed, if some later node sees all of the branching node's outputs, no matter if there have been more branches to different paths in the meantime.

Specified by:

computeUnclosedBranchStack in class OptimizerNode

accept

public void accept(Visitor<OptimizerNode> visitor)

Description copied from class: OptimizerNode

This method implements the visit of a depth-first graph traversing visitor. Implementers must first call the preVisit() method, then hand the visitor to their children, and finally call the postVisit() method.

Specified by:

accept in interface Visitable<OptimizerNode>

Specified by:

accept in class OptimizerNode

Parameters:

visitor - The graph traversing visitor.

See Also:

Visitable.accept(org.apache.flink.util.Visitor)