org.apache.tinkerpop.gremlin.process.computer

Interface VertexProgram<M>

All Superinterfaces:

Cloneable

All Known Implementing Classes:

BulkDumperVertexProgram, BulkLoaderVertexProgram, GraphComputerTest.VertexProgramA, GraphComputerTest.VertexProgramB, GraphComputerTest.VertexProgramC, GraphComputerTest.VertexProgramD, GraphComputerTest.VertexProgramE, GraphComputerTest.VertexProgramF, GraphComputerTest.VertexProgramG, GraphComputerTest.VertexProgramH, GraphComputerTest.VertexProgramI, GraphComputerTest.VertexProgramJ, GraphComputerTest.VertexProgramK, GraphComputerTest.VertexProgramL, GraphComputerTest.VertexProgramM, PageRankVertexProgram, PeerPressureVertexProgram, ProgramTest.TestProgram, StaticVertexProgram, TraversalVertexProgram

public interface VertexProgram<M>
extends Cloneable

A VertexProgram represents one component of a distributed graph computation. Each vertex in the graph (logically) executes the VertexProgram instance in parallel. The collective behavior yields the computational result. In practice, a "worker" (i.e. task, thread, etc.) is responsible for executing the VertexProgram against each vertex that it has in its vertex set (a subset of the full graph vertex set). At minimum there is one "worker" for each vertex, though this is impractical in practice and GraphComputer implementations that leverage such a design are not expected to perform well due to the excess object creation. Any local state/fields in a VertexProgram is static to the vertices within the same worker set. It is not safe to assume that the VertexProgram's "worker" state will remain stable between iterations. Hence, the existence of workerIterationStart(org.apache.tinkerpop.gremlin.process.computer.Memory) and workerIterationEnd(org.apache.tinkerpop.gremlin.process.computer.Memory).

Author:

Marko A. Rodriguez (http://markorodriguez.com), Matthias Broecheler (me@matthiasb.com)

Nested Class Summary

Nested Classes

Modifier and Type	Interface and Description
static interface	VertexProgram.Builder
static interface	VertexProgram.Features

Field Summary

Fields

Modifier and Type	Field and Description
static String	VERTEX_PROGRAM

Method Summary

Modifier and Type	Method and Description
VertexProgram<M>	clone() When multiple workers on a single machine need VertexProgram instances, it is possible to use clone.
static <V extends VertexProgram> V	createVertexProgram(Graph graph, Configuration configuration) A helper method to construct a VertexProgram given the content of the supplied configuration.
void	execute(Vertex vertex, Messenger<M> messenger, Memory memory) This method denotes the main body of the computation and is executed on each vertex in the graph.
default VertexProgram.Features	getFeatures()
default Set<MapReduce>	getMapReducers() The set of MapReduce jobs that are associated with the VertexProgram.
default Set<MemoryComputeKey>	getMemoryComputeKeys() The Memory keys that will be used during the computation.
default Optional<MessageCombiner<M>>	getMessageCombiner() Combine the messages in route to a particular vertex.
Set<MessageScope>	getMessageScopes(Memory memory) This method returns all the MessageScope possibilities for a particular iteration of the vertex program.
GraphComputer.Persist	getPreferredPersist()
GraphComputer.ResultGraph	getPreferredResultGraph()
default Set<VertexComputeKey>	getVertexComputeKeys() The Element properties that will be mutated during the computation.
default void	loadState(Graph graph, Configuration configuration) When it is necessary to load the state of the VertexProgram, this method is called.
void	setup(Memory memory) The method is called at the beginning of the computation.
default void	storeState(Configuration configuration) When it is necessary to store the state of the VertexProgram, this method is called.
boolean	terminate(Memory memory) The method is called at the end of each iteration to determine if the computation is complete.
default void	workerIterationEnd(Memory memory) This method is called at the end of each iteration of each "computational chunk." The set of vertices in the graph are typically not processed with full parallelism.
default void	workerIterationStart(Memory memory) This method is called at the start of each iteration of each "computational chunk." The set of vertices in the graph are typically not processed with full parallelism.

Field Detail

VERTEX_PROGRAM

static final String VERTEX_PROGRAM

See Also:

Constant Field Values

Method Detail

storeState

default void storeState(Configuration configuration)

When it is necessary to store the state of the VertexProgram, this method is called. This is typically required when the VertexProgram needs to be serialized to another machine. Note that what is stored is simply the instance/configuration state, not any processed data. The default implementation provided simply stores the VertexProgram class name for reflective reconstruction. It is typically a good idea to VertexProgram.super.storeState().

Parameters:

configuration - the configuration to store the state of the VertexProgram in.

loadState

default void loadState(Graph graph,
                       Configuration configuration)

When it is necessary to load the state of the VertexProgram, this method is called. This is typically required when the VertexProgram needs to be serialized to another machine. Note that what is loaded is simply the instance state, not any processed data.

Parameters:

graph - the graph that the VertexProgram will run against

configuration - the configuration to load the state of the VertexProgram from.

setup

void setup(Memory memory)

The method is called at the beginning of the computation. The method is global to the GraphComputer and as such, is not called for each vertex. During this stage, the Memory should be initialized to to its "start state."

Parameters:

memory - The global memory of the GraphComputer

execute

void execute(Vertex vertex,
             Messenger<M> messenger,
             Memory memory)

This method denotes the main body of the computation and is executed on each vertex in the graph. This method is logically executed in parallel on all vertices in the graph. When the Memory is read, it is according to the aggregated state yielded in the previous iteration. When the Memory is written, the data will be aggregated at the end of the iteration for reading in the next iteration.

Parameters:

vertex - the Vertex to execute the VertexProgram on

messenger - the messenger that moves data between vertices

memory - the shared state between all vertices in the computation

terminate

boolean terminate(Memory memory)

The method is called at the end of each iteration to determine if the computation is complete. The method is global to the GraphComputer and as such, is not called for each Vertex. The Memory maintains the aggregated data from the last execute() iteration.

Parameters:

memory - The global memory of the GraphComputer

Returns:

whether or not to halt the computation

workerIterationStart

default void workerIterationStart(Memory memory)

This method is called at the start of each iteration of each "computational chunk." The set of vertices in the graph are typically not processed with full parallelism. The vertex set is split into subsets and a worker is assigned to call the execute(org.apache.tinkerpop.gremlin.structure.Vertex, org.apache.tinkerpop.gremlin.process.computer.Messenger<M>, org.apache.tinkerpop.gremlin.process.computer.Memory) method. The default implementation is a no-op.

Parameters:

memory - The memory at the start of the iteration.

workerIterationEnd

default void workerIterationEnd(Memory memory)

This method is called at the end of each iteration of each "computational chunk." The set of vertices in the graph are typically not processed with full parallelism. The vertex set is split into subsets and a worker is assigned to call the execute(org.apache.tinkerpop.gremlin.structure.Vertex, org.apache.tinkerpop.gremlin.process.computer.Messenger<M>, org.apache.tinkerpop.gremlin.process.computer.Memory) method. The default implementation is a no-op.

Parameters:

memory - The memory at the end of the iteration.

getVertexComputeKeys

default Set<VertexComputeKey> getVertexComputeKeys()

The Element properties that will be mutated during the computation. All properties in the graph are readable, but only the keys specified here are writable. The default is an empty set.

Returns:

the set of element keys that will be mutated during the vertex program's execution

getMemoryComputeKeys

default Set<MemoryComputeKey> getMemoryComputeKeys()

The Memory keys that will be used during the computation. These are the only keys that can be read or written throughout the life of the GraphComputer. The default is an empty set.

Returns:

the set of memory keys that will be read/written

getMessageCombiner

default Optional<MessageCombiner<M>> getMessageCombiner()

Combine the messages in route to a particular vertex. Useful to reduce the amount of data transmitted over the wire. For example, instead of sending two objects that will ultimately be merged at the vertex destination, merge/combine into one and send that object. If no message combiner is provider, then no messages will be combined. Furthermore, it is not guaranteed the all messages in route to the vertex will be combined and thus, combiner-state should not be used. The result of the vertex program algorithm should be the same regardless of whether message combining is executed or not.

Returns:

A optional denoting whether or not their is a message combine associated with the vertex program.

getMessageScopes

Set<MessageScope> getMessageScopes(Memory memory)

This method returns all the MessageScope possibilities for a particular iteration of the vertex program. The returned messages scopes are the scopes that will be used to send messages during the stated iteration. It is not a requirement that all stated messages scopes be used, just that it is possible that they be used during the iteration.

Parameters:

memory - an immutable form of the Memory

Returns:

all possible message scopes during said vertex program iteration

getMapReducers

default Set<MapReduce> getMapReducers()

The set of MapReduce jobs that are associated with the VertexProgram. This is not necessarily the exhaustive list over the life of the GraphComputer. If MapReduce jobs are declared by GraphComputer.mapReduce(), they are not contained in this set. The default is an empty set.

Returns:

the set of MapReduce jobs associated with this VertexProgram

clone

VertexProgram<M> clone()

When multiple workers on a single machine need VertexProgram instances, it is possible to use clone. This will provide a speedier way of generating instances, over the storeState(Configuration) and loadState(org.apache.tinkerpop.gremlin.structure.Graph, Configuration) model. The default implementation simply returns the object as it assumes that the VertexProgram instance is a stateless singleton.

Returns:

A clone of the VertexProgram object

getPreferredResultGraph

GraphComputer.ResultGraph getPreferredResultGraph()

getPreferredPersist

GraphComputer.Persist getPreferredPersist()

createVertexProgram

static <V extends VertexProgram> V createVertexProgram(Graph graph,
                                                       Configuration configuration)

A helper method to construct a VertexProgram given the content of the supplied configuration. The class of the VertexProgram is read from the VERTEX_PROGRAM static configuration key. Once the VertexProgram is constructed, loadState(org.apache.tinkerpop.gremlin.structure.Graph, Configuration) method is called with the provided graph and configuration.

Type Parameters:

V - The vertex program type

Parameters:

graph - The graph that the vertex program will execute against

configuration - A configuration with requisite information to build a vertex program

Returns:

the newly constructed vertex program

getFeatures

default VertexProgram.Features getFeatures()