/**
* Converts runtime edge data to persistent edge data (includes source/target vertex data) and
* writes it to HBase.
*
* @param epgmDatabase EPGM database instance
* @param edgeDataHandler edge data handler
* @param persistentEdgeDataFactory persistent edge data factory
* @param edgeDataTableName HBase edge data table name
* @param <PED> persistent edge data type
* @throws IOException
*/
public <PED extends PersistentEdgeData<VD>> void writeEdgeData(
final EPGMDatabase<VD, ED, GD> epgmDatabase,
final EdgeDataHandler<ED, VD> edgeDataHandler,
final PersistentEdgeDataFactory<ED, VD, PED> persistentEdgeDataFactory,
final String edgeDataTableName)
throws IOException {
Graph<Long, VD, ED> graph = epgmDatabase.getDatabaseGraph().getGellyGraph();
DataSet<PersistentEdgeData<VD>> persistentEdgeDataSet =
graph
.getVertices()
// join vertex with edges on edge source vertex id
.join(graph.getEdges())
.where(0)
.equalTo(1)
// join result with vertices on edge target vertex id
.join(graph.getVertices())
.where("1.1")
.equalTo(0)
// ((source-vertex-data, edge-data), target-vertex-data)
.with(new PersistentEdgeDataJoinFunction<>(persistentEdgeDataFactory));
// write (persistent-edge-data) to HBase table
Job job = Job.getInstance();
job.getConfiguration().set(TableOutputFormat.OUTPUT_TABLE, edgeDataTableName);
persistentEdgeDataSet
.map(new HBaseWriter.EdgeDataToHBaseMapper<>(edgeDataHandler))
.output(new HadoopOutputFormat<>(new TableOutputFormat<LongWritable>(), job));
}
@SuppressWarnings("serial")
public static void main(String[] args) throws Exception {
if (!parseParameters(args)) {
return;
}
ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
DataSet<Edge<Long, NullValue>> edges = getEdgesDataSet(env);
Graph<Long, Long, NullValue> graph =
Graph.fromDataSet(
edges,
new MapFunction<Long, Long>() {
@Override
public Long map(Long value) throws Exception {
return value;
}
},
env);
DataSet<Vertex<Long, Long>> verticesWithMinIds =
graph.run(new GSAConnectedComponents<Long, Long, NullValue>(maxIterations));
// emit result
if (fileOutput) {
verticesWithMinIds.writeAsCsv(outputPath, "\n", ",");
// since file sinks are lazy, we trigger the execution explicitly
env.execute("Connected Components Example");
} else {
verticesWithMinIds.print();
}
}
/**
* Prints a graph as two files (edges, vertices) to the file system.
*
* @param g The graph.
* @param filePathEdges File path for edges.
* @param filePathVertices File path for vertices.
*/
@SuppressWarnings("rawtypes")
public static void printGraph(Graph g, String filePathEdges, String filePathVertices) {
try {
g.getEdgesAsTuple3().writeAsCsv(filePathEdges, "\n", "\t", WriteMode.OVERWRITE);
g.getVerticesAsTuple2().writeAsCsv(filePathVertices, "\n", "\t", WriteMode.OVERWRITE);
} catch (Exception e) {
e.printStackTrace();
}
}
/**
* Prints the edges and vertices of a graph <code>g</code> to the console.
*
* @param g A graph.
* @return Result.
*/
@SuppressWarnings("rawtypes")
public static String showEdgesVertices(Graph g) {
String str = "";
try {
str = "Edges:\n";
for (Object edge : g.getEdges().collect()) str += " " + edge + "\n";
str += "\nVertices:\n";
for (Object vertex : g.getVertices().collect()) str += " " + vertex + "\n";
} catch (Exception e1) {
e1.printStackTrace();
}
return str;
}
@Override
public DataSet<Edge<K, Tuple2<EV, LongValue>>> run(Graph<K, VV, EV> input) throws Exception {
// t, d(t)
DataSet<Vertex<K, LongValue>> vertexDegrees =
input.run(
new VertexDegree<K, VV, EV>()
.setReduceOnTargetId(!reduceOnSourceId)
.setParallelism(parallelism));
// s, t, d(t)
return input
.getEdges()
.join(vertexDegrees, JoinHint.REPARTITION_HASH_SECOND)
.where(1)
.equalTo(0)
.with(new JoinEdgeWithVertexDegree<K, EV>())
.setParallelism(parallelism)
.name("Edge target degree");
}
@Test
public void testWithSimpleGraph() throws Exception {
Graph<IntValue, NullValue, NullValue> graph =
undirectedSimpleGraph.run(new MaximumDegree<IntValue, NullValue, NullValue>(3));
String expectedVerticesResult =
"(0,(null))\n" + "(1,(null))\n" + "(2,(null))\n" + "(4,(null))\n" + "(5,(null))";
TestBaseUtils.compareResultAsText(graph.getVertices().collect(), expectedVerticesResult);
String expectedEdgesResult =
"(0,1,(null))\n"
+ "(0,2,(null))\n"
+ "(1,0,(null))\n"
+ "(1,2,(null))\n"
+ "(2,0,(null))\n"
+ "(2,1,(null))";
TestBaseUtils.compareResultAsText(graph.getEdges().collect(), expectedEdgesResult);
}
@Override
public VertexMetrics<K, VV, EV> run(Graph<K, VV, EV> input) throws Exception {
super.run(input);
DataSet<Vertex<K, Degrees>> vertexDegree =
input.run(
new VertexDegrees<K, VV, EV>()
.setIncludeZeroDegreeVertices(includeZeroDegreeVertices)
.setParallelism(parallelism));
vertexDegree.output(new VertexMetricsHelper<K>(id)).name("Vertex metrics");
return this;
}
/** {@inheritDoc} */
@Override
protected LogicalGraph<VD, ED, GD> executeInternal(
LogicalGraph<VD, ED, GD> firstGraph, LogicalGraph<VD, ED, GD> secondGraph) {
final Long newGraphID = FlinkConstants.EXCLUDE_GRAPH_ID;
Graph<Long, VD, ED> graph1 = firstGraph.getGellyGraph();
Graph<Long, VD, ED> graph2 = secondGraph.getGellyGraph();
// union vertex sets, group by vertex id, filter vertices where the group
// contains exactly one vertex which belongs to the graph, the operator is
// called on
DataSet<Vertex<Long, VD>> newVertexSet =
graph1
.getVertices()
.union(graph2.getVertices())
.groupBy(new KeySelectors.VertexKeySelector<VD>())
.reduceGroup(new VertexGroupReducer<VD>(1L, firstGraph.getId(), secondGraph.getId()))
.map(new VertexToGraphUpdater<VD>(newGraphID));
JoinFunction<Edge<Long, ED>, Vertex<Long, VD>, Edge<Long, ED>> joinFunc =
new JoinFunction<Edge<Long, ED>, Vertex<Long, VD>, Edge<Long, ED>>() {
@Override
public Edge<Long, ED> join(Edge<Long, ED> leftTuple, Vertex<Long, VD> rightTuple)
throws Exception {
return leftTuple;
}
};
// In exclude(), we are only interested in edges that connect vertices
// that are in the exclusion of the vertex sets. Thus, we join the edges
// from the left graph with the new vertex set using source and target ids.
DataSet<Edge<Long, ED>> newEdgeSet =
graph1
.getEdges()
.join(newVertexSet)
.where(new KeySelectors.EdgeSourceVertexKeySelector<ED>())
.equalTo(new KeySelectors.VertexKeySelector<VD>())
.with(joinFunc)
.join(newVertexSet)
.where(new KeySelectors.EdgeTargetVertexKeySelector<ED>())
.equalTo(new KeySelectors.VertexKeySelector<VD>())
.with(joinFunc)
.map(new EdgeToGraphUpdater<ED>(newGraphID));
return LogicalGraph.fromGraph(
Graph.fromDataSet(newVertexSet, newEdgeSet, graph1.getContext()),
firstGraph.getGraphDataFactory().createGraphData(newGraphID),
firstGraph.getVertexDataFactory(),
firstGraph.getEdgeDataFactory(),
firstGraph.getGraphDataFactory());
}
@Test
public void testTranslation() {
try {
final String ITERATION_NAME = "Test Name";
final String AGGREGATOR_NAME = "AggregatorName";
final String BC_SET_GATHER_NAME = "gather messages";
final String BC_SET_SUM_NAME = "sum updates";
final String BC_SET_APLLY_NAME = "apply updates";
final int NUM_ITERATIONS = 13;
final int ITERATION_parallelism = 77;
ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
DataSet<Long> bcGather = env.fromElements(1L);
DataSet<Long> bcSum = env.fromElements(1L);
DataSet<Long> bcApply = env.fromElements(1L);
DataSet<Vertex<Long, Long>> result;
// ------------ construct the test program ------------------
{
DataSet<Edge<Long, NullValue>> edges =
env.fromElements(new Tuple3<Long, Long, NullValue>(1L, 2L, NullValue.getInstance()))
.map(new Tuple3ToEdgeMap<Long, NullValue>());
Graph<Long, Long, NullValue> graph = Graph.fromDataSet(edges, new InitVertices(), env);
GSAConfiguration parameters = new GSAConfiguration();
parameters.registerAggregator(AGGREGATOR_NAME, new LongSumAggregator());
parameters.setName(ITERATION_NAME);
parameters.setParallelism(ITERATION_parallelism);
parameters.addBroadcastSetForGatherFunction(BC_SET_GATHER_NAME, bcGather);
parameters.addBroadcastSetForSumFunction(BC_SET_SUM_NAME, bcSum);
parameters.addBroadcastSetForApplyFunction(BC_SET_APLLY_NAME, bcApply);
result =
graph
.runGatherSumApplyIteration(
new GatherNeighborIds(),
new SelectMinId(),
new UpdateComponentId(),
NUM_ITERATIONS,
parameters)
.getVertices();
result.output(new DiscardingOutputFormat<Vertex<Long, Long>>());
}
// ------------- validate the java program ----------------
assertTrue(result instanceof DeltaIterationResultSet);
DeltaIterationResultSet<?, ?> resultSet = (DeltaIterationResultSet<?, ?>) result;
DeltaIteration<?, ?> iteration = (DeltaIteration<?, ?>) resultSet.getIterationHead();
// check the basic iteration properties
assertEquals(NUM_ITERATIONS, resultSet.getMaxIterations());
assertArrayEquals(new int[] {0}, resultSet.getKeyPositions());
assertEquals(ITERATION_parallelism, iteration.getParallelism());
assertEquals(ITERATION_NAME, iteration.getName());
assertEquals(
AGGREGATOR_NAME,
iteration.getAggregators().getAllRegisteredAggregators().iterator().next().getName());
// validate that the semantic properties are set as they should
TwoInputUdfOperator<?, ?, ?, ?> solutionSetJoin =
(TwoInputUdfOperator<?, ?, ?, ?>) resultSet.getNextWorkset();
assertTrue(
solutionSetJoin.getSemanticProperties().getForwardingTargetFields(0, 0).contains(0));
assertTrue(
solutionSetJoin.getSemanticProperties().getForwardingTargetFields(1, 0).contains(0));
SingleInputUdfOperator<?, ?, ?> sumReduce =
(SingleInputUdfOperator<?, ?, ?>) solutionSetJoin.getInput1();
SingleInputUdfOperator<?, ?, ?> gatherMap =
(SingleInputUdfOperator<?, ?, ?>) sumReduce.getInput();
// validate that the broadcast sets are forwarded
assertEquals(bcGather, gatherMap.getBroadcastSets().get(BC_SET_GATHER_NAME));
assertEquals(bcSum, sumReduce.getBroadcastSets().get(BC_SET_SUM_NAME));
assertEquals(bcApply, solutionSetJoin.getBroadcastSets().get(BC_SET_APLLY_NAME));
} catch (Exception e) {
System.err.println(e.getMessage());
e.printStackTrace();
fail(e.getMessage());
}
}
/**
* Converts runtime vertex data to persistent vertex data (includes incoming and outgoing edge
* data) and writes it to HBase.
*
* @param epgmDatabase EPGM database instance
* @param vertexDataHandler vertex data handler
* @param persistentVertexDataFactory persistent vertex data factory
* @param vertexDataTableName HBase vertex data table name
* @param <PVD> persistent vertex data type
* @throws Exception
*/
public <PVD extends PersistentVertexData<ED>> void writeVertexData(
final EPGMDatabase<VD, ED, GD> epgmDatabase,
final VertexDataHandler<VD, ED> vertexDataHandler,
final PersistentVertexDataFactory<VD, ED, PVD> persistentVertexDataFactory,
final String vertexDataTableName)
throws Exception {
final Graph<Long, VD, ED> graph = epgmDatabase.getDatabaseGraph().getGellyGraph();
// group edges by source vertex id (vertex-id, [out-edge-data])
GroupReduceOperator<Edge<Long, ED>, Tuple2<Long, Set<ED>>> vertexToOutgoingEdges =
graph
.getEdges()
.groupBy(0) // group by source vertex id
.reduceGroup(
new GroupReduceFunction<Edge<Long, ED>, Tuple2<Long, Set<ED>>>() {
@Override
public void reduce(
Iterable<Edge<Long, ED>> edgeIterable,
Collector<Tuple2<Long, Set<ED>>> collector)
throws Exception {
Set<ED> outgoingEdgeData = Sets.newHashSet();
Long vertexId = null;
boolean initialized = false;
for (Edge<Long, ED> edgeData : edgeIterable) {
if (!initialized) {
vertexId = edgeData.getSource();
initialized = true;
}
outgoingEdgeData.add(edgeData.getValue());
}
collector.collect(new Tuple2<>(vertexId, outgoingEdgeData));
}
});
// group edges by target vertex id (vertex-id, [in-edge-data])
GroupReduceOperator<Edge<Long, ED>, Tuple2<Long, Set<ED>>> vertexToIncomingEdges =
graph
.getEdges()
.groupBy(1) // group by target vertex id
.reduceGroup(
new GroupReduceFunction<Edge<Long, ED>, Tuple2<Long, Set<ED>>>() {
@Override
public void reduce(
Iterable<Edge<Long, ED>> edgeIterable,
Collector<Tuple2<Long, Set<ED>>> collector)
throws Exception {
Set<ED> outgoingEdgeData = Sets.newHashSet();
Long vertexId = null;
boolean initialized = false;
for (Edge<Long, ED> edgeData : edgeIterable) {
if (!initialized) {
vertexId = edgeData.getTarget();
initialized = true;
}
outgoingEdgeData.add(edgeData.getValue());
}
collector.collect(new Tuple2<>(vertexId, outgoingEdgeData));
}
});
// co-group (vertex-data) with (vertex-id, [out-edge-data]) to simulate left
// outer join
DataSet<Tuple2<Vertex<Long, VD>, Set<ED>>> vertexDataWithOutgoingEdges =
graph
.getVertices()
.coGroup(vertexToOutgoingEdges)
.where(0)
.equalTo(0)
.with(
new CoGroupFunction<
Vertex<Long, VD>, Tuple2<Long, Set<ED>>, Tuple2<Vertex<Long, VD>, Set<ED>>>() {
@Override
public void coGroup(
Iterable<Vertex<Long, VD>> vertexIterable,
Iterable<Tuple2<Long, Set<ED>>> outEdgesIterable,
Collector<Tuple2<Vertex<Long, VD>, Set<ED>>> collector)
throws Exception {
Vertex<Long, VD> vertex = null;
Set<ED> outgoingEdgeData = null;
// read vertex data from left group
for (Vertex<Long, VD> v : vertexIterable) {
vertex = v;
}
// read outgoing edge from right group (may be empty)
for (Tuple2<Long, Set<ED>> oEdges : outEdgesIterable) {
outgoingEdgeData = oEdges.f1;
}
collector.collect(new Tuple2<>(vertex, outgoingEdgeData));
}
});
// co-group (vertex-data, (vertex-id, [out-edge-data])) with (vertex-id,
// [in-edge-data]) to simulate left outer join
DataSet<PersistentVertexData<ED>> persistentVertexDataSet =
vertexDataWithOutgoingEdges
.coGroup(vertexToIncomingEdges)
.where("0.0")
.equalTo(0)
.with(new PersistentVertexDataCoGroupFunction<>(persistentVertexDataFactory));
// write (persistent-vertex-data) to HBase table
Job job = Job.getInstance();
job.getConfiguration().set(TableOutputFormat.OUTPUT_TABLE, vertexDataTableName);
persistentVertexDataSet
.map(new HBaseWriter.VertexDataToHBaseMapper<>(vertexDataHandler))
.output(new HadoopOutputFormat<>(new TableOutputFormat<LongWritable>(), job));
}
/**
* Converts runtime graph data to persistent graph data (including vertex and edge identifiers)
* and writes it to HBase.
*
* @param epgmDatabase EPGM database instance
* @param graphDataHandler graph data handler
* @param persistentGraphDataFactory persistent graph data factory
* @param graphDataTableName HBase graph data table name
* @param <PGD> persistent graph data type
* @throws IOException
*/
public <PGD extends PersistentGraphData> void writeGraphData(
final EPGMDatabase<VD, ED, GD> epgmDatabase,
final GraphDataHandler<GD> graphDataHandler,
final PersistentGraphDataFactory<GD, PGD> persistentGraphDataFactory,
final String graphDataTableName)
throws IOException {
final Graph<Long, VD, ED> graph = epgmDatabase.getDatabaseGraph().getGellyGraph();
// build (graph-id, vertex-id) tuples from vertices
FlatMapOperator<Vertex<Long, VD>, Tuple2<Long, Long>> graphIdToVertexId =
graph
.getVertices()
.flatMap(
new FlatMapFunction<Vertex<Long, VD>, Tuple2<Long, Long>>() {
@Override
public void flatMap(
Vertex<Long, VD> vertex, Collector<Tuple2<Long, Long>> collector)
throws Exception {
if (vertex.getValue().getGraphCount() > 0) {
for (Long graphID : vertex.getValue().getGraphs()) {
collector.collect(new Tuple2<>(graphID, vertex.f0));
}
}
}
});
// build (graph-id, edge-id) tuples from vertices
FlatMapOperator<Edge<Long, ED>, Tuple2<Long, Long>> graphIdToEdgeId =
graph
.getEdges()
.flatMap(
new FlatMapFunction<Edge<Long, ED>, Tuple2<Long, Long>>() {
@Override
public void flatMap(Edge<Long, ED> edge, Collector<Tuple2<Long, Long>> collector)
throws Exception {
if (edge.getValue().getGraphCount() > 0) {
for (Long graphId : edge.getValue().getGraphs()) {
collector.collect(new Tuple2<>(graphId, edge.getValue().getId()));
}
}
}
});
// co-group (graph-id, vertex-id) and (graph-id, edge-id) tuples to
// (graph-id, {vertex-id}, {edge-id}) triples
CoGroupOperator<Tuple2<Long, Long>, Tuple2<Long, Long>, Tuple3<Long, Set<Long>, Set<Long>>>
graphToVertexIdsAndEdgeIds =
graphIdToVertexId
.coGroup(graphIdToEdgeId)
.where(0)
.equalTo(0)
.with(
new CoGroupFunction<
Tuple2<Long, Long>,
Tuple2<Long, Long>,
Tuple3<Long, Set<Long>, Set<Long>>>() {
@Override
public void coGroup(
Iterable<Tuple2<Long, Long>> graphToVertexIds,
Iterable<Tuple2<Long, Long>> graphToEdgeIds,
Collector<Tuple3<Long, Set<Long>, Set<Long>>> collector)
throws Exception {
Set<Long> vertexIds = Sets.newHashSet();
Set<Long> edgeIds = Sets.newHashSet();
Long graphId = null;
boolean initialized = false;
for (Tuple2<Long, Long> graphToVertexTuple : graphToVertexIds) {
if (!initialized) {
graphId = graphToVertexTuple.f0;
initialized = true;
}
vertexIds.add(graphToVertexTuple.f1);
}
for (Tuple2<Long, Long> graphToEdgeTuple : graphToEdgeIds) {
edgeIds.add(graphToEdgeTuple.f1);
}
collector.collect(new Tuple3<>(graphId, vertexIds, edgeIds));
}
});
// join (graph-id, {vertex-id}, {edge-id}) triples with
// (graph-id, graph-data) and build (persistent-graph-data)
JoinOperator.EquiJoin<
Tuple3<Long, Set<Long>, Set<Long>>, Subgraph<Long, GD>, PersistentGraphData>
persistentGraphDataSet =
graphToVertexIdsAndEdgeIds
.join(epgmDatabase.getCollection().getSubgraphs())
.where(0)
.equalTo(0)
.with(new PersistentGraphDataJoinFunction<>(persistentGraphDataFactory));
// write (persistent-graph-data) to HBase table
Job job = Job.getInstance();
job.getConfiguration().set(TableOutputFormat.OUTPUT_TABLE, graphDataTableName);
persistentGraphDataSet
.map(new HBaseWriter.GraphDataToHBaseMapper<>(graphDataHandler))
.output(new HadoopOutputFormat<>(new TableOutputFormat<LongWritable>(), job));
}
