Search in sources :

Example 11 with NullValue

use of org.apache.flink.types.NullValue in project flink by apache.

the class PregelCompilerTest method testPregelCompilerWithBroadcastVariable.

@SuppressWarnings("serial")
@Test
public void testPregelCompilerWithBroadcastVariable() {
    try {
        final String BC_VAR_NAME = "borat variable";
        ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
        env.setParallelism(DEFAULT_PARALLELISM);
        // compose test program
        {
            DataSet<Long> bcVar = env.fromElements(1L);
            DataSet<Vertex<Long, Long>> initialVertices = env.fromElements(new Tuple2<>(1L, 1L), new Tuple2<>(2L, 2L)).map(new Tuple2ToVertexMap<Long, Long>());
            DataSet<Edge<Long, NullValue>> edges = env.fromElements(new Tuple2<>(1L, 2L)).map(new MapFunction<Tuple2<Long, Long>, Edge<Long, NullValue>>() {

                public Edge<Long, NullValue> map(Tuple2<Long, Long> edge) {
                    return new Edge<>(edge.f0, edge.f1, NullValue.getInstance());
                }
            });
            Graph<Long, Long, NullValue> graph = Graph.fromDataSet(initialVertices, edges, env);
            VertexCentricConfiguration parameters = new VertexCentricConfiguration();
            parameters.addBroadcastSet(BC_VAR_NAME, bcVar);
            DataSet<Vertex<Long, Long>> result = graph.runVertexCentricIteration(new CCCompute(), null, 100, parameters).getVertices();
            result.output(new DiscardingOutputFormat<Vertex<Long, Long>>());
        }
        Plan p = env.createProgramPlan("Pregel Connected Components");
        OptimizedPlan op = compileNoStats(p);
        // check the sink
        SinkPlanNode sink = op.getDataSinks().iterator().next();
        assertEquals(ShipStrategyType.FORWARD, sink.getInput().getShipStrategy());
        assertEquals(DEFAULT_PARALLELISM, sink.getParallelism());
        // check the iteration
        WorksetIterationPlanNode iteration = (WorksetIterationPlanNode) sink.getInput().getSource();
        assertEquals(DEFAULT_PARALLELISM, iteration.getParallelism());
        // check the solution set delta
        PlanNode ssDelta = iteration.getSolutionSetDeltaPlanNode();
        assertTrue(ssDelta instanceof SingleInputPlanNode);
        SingleInputPlanNode ssFlatMap = (SingleInputPlanNode) ((SingleInputPlanNode) (ssDelta)).getInput().getSource();
        assertEquals(DEFAULT_PARALLELISM, ssFlatMap.getParallelism());
        assertEquals(ShipStrategyType.FORWARD, ssFlatMap.getInput().getShipStrategy());
        // check the computation coGroup
        DualInputPlanNode computationCoGroup = (DualInputPlanNode) (ssFlatMap.getInput().getSource());
        assertEquals(DEFAULT_PARALLELISM, computationCoGroup.getParallelism());
        assertEquals(ShipStrategyType.FORWARD, computationCoGroup.getInput1().getShipStrategy());
        assertEquals(ShipStrategyType.PARTITION_HASH, computationCoGroup.getInput2().getShipStrategy());
        assertTrue(computationCoGroup.getInput2().getTempMode().isCached());
        assertEquals(new FieldList(0), computationCoGroup.getInput2().getShipStrategyKeys());
        // check that the initial partitioning is pushed out of the loop
        assertEquals(ShipStrategyType.PARTITION_HASH, iteration.getInput1().getShipStrategy());
        assertEquals(new FieldList(0), iteration.getInput1().getShipStrategyKeys());
    } catch (Exception e) {
        System.err.println(e.getMessage());
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : ExecutionEnvironment(org.apache.flink.api.java.ExecutionEnvironment) Tuple2ToVertexMap(org.apache.flink.graph.utils.Tuple2ToVertexMap) DataSet(org.apache.flink.api.java.DataSet) WorksetIterationPlanNode(org.apache.flink.optimizer.plan.WorksetIterationPlanNode) MapFunction(org.apache.flink.api.common.functions.MapFunction) Plan(org.apache.flink.api.common.Plan) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) DiscardingOutputFormat(org.apache.flink.api.java.io.DiscardingOutputFormat) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) FieldList(org.apache.flink.api.common.operators.util.FieldList) SingleInputPlanNode(org.apache.flink.optimizer.plan.SingleInputPlanNode) DualInputPlanNode(org.apache.flink.optimizer.plan.DualInputPlanNode) NullValue(org.apache.flink.types.NullValue) Graph(org.apache.flink.graph.Graph) WorksetIterationPlanNode(org.apache.flink.optimizer.plan.WorksetIterationPlanNode) DualInputPlanNode(org.apache.flink.optimizer.plan.DualInputPlanNode) PlanNode(org.apache.flink.optimizer.plan.PlanNode) SinkPlanNode(org.apache.flink.optimizer.plan.SinkPlanNode) SingleInputPlanNode(org.apache.flink.optimizer.plan.SingleInputPlanNode) Tuple2(org.apache.flink.api.java.tuple.Tuple2) SinkPlanNode(org.apache.flink.optimizer.plan.SinkPlanNode) Edge(org.apache.flink.graph.Edge) Test(org.junit.Test)

Example 12 with NullValue

use of org.apache.flink.types.NullValue in project flink by apache.

the class SpargelCompilerTest method testSpargelCompilerWithBroadcastVariable.

@SuppressWarnings("serial")
@Test
public void testSpargelCompilerWithBroadcastVariable() {
    try {
        final String BC_VAR_NAME = "borat variable";
        ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
        env.setParallelism(DEFAULT_PARALLELISM);
        // compose test program
        {
            DataSet<Long> bcVar = env.fromElements(1L);
            DataSet<Vertex<Long, Long>> initialVertices = env.fromElements(new Tuple2<>(1L, 1L), new Tuple2<>(2L, 2L)).map(new Tuple2ToVertexMap<Long, Long>());
            DataSet<Edge<Long, NullValue>> edges = env.fromElements(new Tuple2<>(1L, 2L)).map(new MapFunction<Tuple2<Long, Long>, Edge<Long, NullValue>>() {

                public Edge<Long, NullValue> map(Tuple2<Long, Long> edge) {
                    return new Edge<>(edge.f0, edge.f1, NullValue.getInstance());
                }
            });
            Graph<Long, Long, NullValue> graph = Graph.fromDataSet(initialVertices, edges, env);
            ScatterGatherConfiguration parameters = new ScatterGatherConfiguration();
            parameters.addBroadcastSetForScatterFunction(BC_VAR_NAME, bcVar);
            parameters.addBroadcastSetForGatherFunction(BC_VAR_NAME, bcVar);
            DataSet<Vertex<Long, Long>> result = graph.runScatterGatherIteration(new ConnectedComponents.CCMessenger<Long, Long>(BasicTypeInfo.LONG_TYPE_INFO), new ConnectedComponents.CCUpdater<Long, Long>(), 100).getVertices();
            result.output(new DiscardingOutputFormat<Vertex<Long, Long>>());
        }
        Plan p = env.createProgramPlan("Spargel Connected Components");
        OptimizedPlan op = compileNoStats(p);
        // check the sink
        SinkPlanNode sink = op.getDataSinks().iterator().next();
        assertEquals(ShipStrategyType.FORWARD, sink.getInput().getShipStrategy());
        assertEquals(DEFAULT_PARALLELISM, sink.getParallelism());
        // check the iteration
        WorksetIterationPlanNode iteration = (WorksetIterationPlanNode) sink.getInput().getSource();
        assertEquals(DEFAULT_PARALLELISM, iteration.getParallelism());
        // check the solution set join and the delta
        PlanNode ssDelta = iteration.getSolutionSetDeltaPlanNode();
        // this is only true if the update functions preserves the partitioning
        assertTrue(ssDelta instanceof DualInputPlanNode);
        DualInputPlanNode ssJoin = (DualInputPlanNode) ssDelta;
        assertEquals(DEFAULT_PARALLELISM, ssJoin.getParallelism());
        assertEquals(ShipStrategyType.PARTITION_HASH, ssJoin.getInput1().getShipStrategy());
        assertEquals(new FieldList(0), ssJoin.getInput1().getShipStrategyKeys());
        // check the workset set join
        DualInputPlanNode edgeJoin = (DualInputPlanNode) ssJoin.getInput1().getSource();
        assertEquals(DEFAULT_PARALLELISM, edgeJoin.getParallelism());
        assertEquals(ShipStrategyType.PARTITION_HASH, edgeJoin.getInput1().getShipStrategy());
        assertEquals(ShipStrategyType.FORWARD, edgeJoin.getInput2().getShipStrategy());
        assertTrue(edgeJoin.getInput1().getTempMode().isCached());
        assertEquals(new FieldList(0), edgeJoin.getInput1().getShipStrategyKeys());
        // check that the initial partitioning is pushed out of the loop
        assertEquals(ShipStrategyType.PARTITION_HASH, iteration.getInput1().getShipStrategy());
        assertEquals(ShipStrategyType.PARTITION_HASH, iteration.getInput2().getShipStrategy());
        assertEquals(new FieldList(0), iteration.getInput1().getShipStrategyKeys());
        assertEquals(new FieldList(0), iteration.getInput2().getShipStrategyKeys());
    } catch (Exception e) {
        System.err.println(e.getMessage());
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : ExecutionEnvironment(org.apache.flink.api.java.ExecutionEnvironment) Tuple2ToVertexMap(org.apache.flink.graph.utils.Tuple2ToVertexMap) DataSet(org.apache.flink.api.java.DataSet) WorksetIterationPlanNode(org.apache.flink.optimizer.plan.WorksetIterationPlanNode) MapFunction(org.apache.flink.api.common.functions.MapFunction) Plan(org.apache.flink.api.common.Plan) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) DiscardingOutputFormat(org.apache.flink.api.java.io.DiscardingOutputFormat) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) FieldList(org.apache.flink.api.common.operators.util.FieldList) DualInputPlanNode(org.apache.flink.optimizer.plan.DualInputPlanNode) NullValue(org.apache.flink.types.NullValue) Graph(org.apache.flink.graph.Graph) WorksetIterationPlanNode(org.apache.flink.optimizer.plan.WorksetIterationPlanNode) DualInputPlanNode(org.apache.flink.optimizer.plan.DualInputPlanNode) PlanNode(org.apache.flink.optimizer.plan.PlanNode) SinkPlanNode(org.apache.flink.optimizer.plan.SinkPlanNode) ConnectedComponents(org.apache.flink.graph.library.ConnectedComponents) Tuple2(org.apache.flink.api.java.tuple.Tuple2) SinkPlanNode(org.apache.flink.optimizer.plan.SinkPlanNode) Edge(org.apache.flink.graph.Edge) Test(org.junit.Test)

Example 13 with NullValue

use of org.apache.flink.types.NullValue in project flink by apache.

the class GSACompilerTest method testGSACompiler.

@Test
public void testGSACompiler() {
    try {
        ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
        env.setParallelism(DEFAULT_PARALLELISM);
        // compose test program
        {
            DataSet<Edge<Long, NullValue>> edges = env.fromElements(new Tuple3<>(1L, 2L, NullValue.getInstance())).map(new Tuple3ToEdgeMap<Long, NullValue>());
            Graph<Long, Long, NullValue> graph = Graph.fromDataSet(edges, new InitVertices(), env);
            DataSet<Vertex<Long, Long>> result = graph.runGatherSumApplyIteration(new GatherNeighborIds(), new SelectMinId(), new UpdateComponentId(), 100).getVertices();
            result.output(new DiscardingOutputFormat<Vertex<Long, Long>>());
        }
        Plan p = env.createProgramPlan("GSA Connected Components");
        OptimizedPlan op = compileNoStats(p);
        // check the sink
        SinkPlanNode sink = op.getDataSinks().iterator().next();
        assertEquals(ShipStrategyType.FORWARD, sink.getInput().getShipStrategy());
        assertEquals(DEFAULT_PARALLELISM, sink.getParallelism());
        assertEquals(PartitioningProperty.HASH_PARTITIONED, sink.getGlobalProperties().getPartitioning());
        // check the iteration
        WorksetIterationPlanNode iteration = (WorksetIterationPlanNode) sink.getInput().getSource();
        assertEquals(DEFAULT_PARALLELISM, iteration.getParallelism());
        // check the solution set join and the delta
        PlanNode ssDelta = iteration.getSolutionSetDeltaPlanNode();
        // this is only true if the update function preserves the partitioning
        assertTrue(ssDelta instanceof DualInputPlanNode);
        DualInputPlanNode ssJoin = (DualInputPlanNode) ssDelta;
        assertEquals(DEFAULT_PARALLELISM, ssJoin.getParallelism());
        assertEquals(ShipStrategyType.PARTITION_HASH, ssJoin.getInput1().getShipStrategy());
        assertEquals(new FieldList(0), ssJoin.getInput1().getShipStrategyKeys());
        // check the workset set join
        SingleInputPlanNode sumReducer = (SingleInputPlanNode) ssJoin.getInput1().getSource();
        SingleInputPlanNode gatherMapper = (SingleInputPlanNode) sumReducer.getInput().getSource();
        DualInputPlanNode edgeJoin = (DualInputPlanNode) gatherMapper.getInput().getSource();
        assertEquals(DEFAULT_PARALLELISM, edgeJoin.getParallelism());
        // input1 is the workset
        assertEquals(ShipStrategyType.FORWARD, edgeJoin.getInput1().getShipStrategy());
        // input2 is the edges
        assertEquals(ShipStrategyType.PARTITION_HASH, edgeJoin.getInput2().getShipStrategy());
        assertTrue(edgeJoin.getInput2().getTempMode().isCached());
        assertEquals(new FieldList(0), edgeJoin.getInput2().getShipStrategyKeys());
    } catch (Exception e) {
        System.err.println(e.getMessage());
        e.printStackTrace();
        fail(e.getMessage());
    }
}
Also used : ExecutionEnvironment(org.apache.flink.api.java.ExecutionEnvironment) DataSet(org.apache.flink.api.java.DataSet) WorksetIterationPlanNode(org.apache.flink.optimizer.plan.WorksetIterationPlanNode) Plan(org.apache.flink.api.common.Plan) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) DiscardingOutputFormat(org.apache.flink.api.java.io.DiscardingOutputFormat) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) FieldList(org.apache.flink.api.common.operators.util.FieldList) DualInputPlanNode(org.apache.flink.optimizer.plan.DualInputPlanNode) SingleInputPlanNode(org.apache.flink.optimizer.plan.SingleInputPlanNode) NullValue(org.apache.flink.types.NullValue) Graph(org.apache.flink.graph.Graph) WorksetIterationPlanNode(org.apache.flink.optimizer.plan.WorksetIterationPlanNode) DualInputPlanNode(org.apache.flink.optimizer.plan.DualInputPlanNode) PlanNode(org.apache.flink.optimizer.plan.PlanNode) SinkPlanNode(org.apache.flink.optimizer.plan.SinkPlanNode) SingleInputPlanNode(org.apache.flink.optimizer.plan.SingleInputPlanNode) Tuple3ToEdgeMap(org.apache.flink.graph.utils.Tuple3ToEdgeMap) SinkPlanNode(org.apache.flink.optimizer.plan.SinkPlanNode) Test(org.junit.Test)

Example 14 with NullValue

use of org.apache.flink.types.NullValue in project flink by apache.

the class GSATranslationTest method testTranslation.

@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<>(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 = 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());
    }
}
Also used : Vertex(org.apache.flink.graph.Vertex) ExecutionEnvironment(org.apache.flink.api.java.ExecutionEnvironment) DataSet(org.apache.flink.api.java.DataSet) LongSumAggregator(org.apache.flink.api.common.aggregators.LongSumAggregator) DeltaIterationResultSet(org.apache.flink.api.java.operators.DeltaIterationResultSet) TwoInputUdfOperator(org.apache.flink.api.java.operators.TwoInputUdfOperator) DiscardingOutputFormat(org.apache.flink.api.java.io.DiscardingOutputFormat) NullValue(org.apache.flink.types.NullValue) SingleInputUdfOperator(org.apache.flink.api.java.operators.SingleInputUdfOperator) Graph(org.apache.flink.graph.Graph) Tuple3ToEdgeMap(org.apache.flink.graph.utils.Tuple3ToEdgeMap) Test(org.junit.Test)

Example 15 with NullValue

use of org.apache.flink.types.NullValue in project flink by apache.

the class EdgeSourceDegreeTest method testWithRMatGraph.

@Test
public void testWithRMatGraph() throws Exception {
    DataSet<Edge<LongValue, Tuple2<NullValue, LongValue>>> sourceDegreeOnSourceId = undirectedRMatGraph.run(new EdgeSourceDegree<LongValue, NullValue, NullValue>());
    Checksum checksumOnSourceId = new ChecksumHashCode<Edge<LongValue, Tuple2<NullValue, LongValue>>>().run(sourceDegreeOnSourceId).execute();
    assertEquals(20884, checksumOnSourceId.getCount());
    assertEquals(0x000000019d8f0070L, checksumOnSourceId.getChecksum());
    DataSet<Edge<LongValue, Tuple2<NullValue, LongValue>>> sourceDegreeOnTargetId = undirectedRMatGraph.run(new EdgeSourceDegree<LongValue, NullValue, NullValue>().setReduceOnTargetId(true));
    Checksum checksumOnTargetId = new ChecksumHashCode<Edge<LongValue, Tuple2<NullValue, LongValue>>>().run(sourceDegreeOnTargetId).execute();
    assertEquals(checksumOnTargetId, checksumOnTargetId);
}
Also used : NullValue(org.apache.flink.types.NullValue) Checksum(org.apache.flink.graph.asm.dataset.ChecksumHashCode.Checksum) Tuple2(org.apache.flink.api.java.tuple.Tuple2) LongValue(org.apache.flink.types.LongValue) ChecksumHashCode(org.apache.flink.graph.asm.dataset.ChecksumHashCode) Edge(org.apache.flink.graph.Edge) Test(org.junit.Test)

Aggregations

NullValue (org.apache.flink.types.NullValue)49 Test (org.junit.Test)39 ExecutionEnvironment (org.apache.flink.api.java.ExecutionEnvironment)33 LongValue (org.apache.flink.types.LongValue)23 Edge (org.apache.flink.graph.Edge)18 Vertex (org.apache.flink.graph.Vertex)18 Tuple2 (org.apache.flink.api.java.tuple.Tuple2)15 Checksum (org.apache.flink.graph.asm.dataset.ChecksumHashCode.Checksum)13 Graph (org.apache.flink.graph.Graph)12 DataSet (org.apache.flink.api.java.DataSet)11 ChecksumHashCode (org.apache.flink.graph.asm.dataset.ChecksumHashCode)11 DiscardingOutputFormat (org.apache.flink.api.java.io.DiscardingOutputFormat)7 JDKRandomGeneratorFactory (org.apache.flink.graph.generator.random.JDKRandomGeneratorFactory)7 NumberFormat (java.text.NumberFormat)6 JobExecutionResult (org.apache.flink.api.common.JobExecutionResult)6 Plan (org.apache.flink.api.common.Plan)6 MapFunction (org.apache.flink.api.common.functions.MapFunction)6 FieldList (org.apache.flink.api.common.operators.util.FieldList)6 ParameterTool (org.apache.flink.api.java.utils.ParameterTool)6 ProgramParametrizationException (org.apache.flink.client.program.ProgramParametrizationException)6