Search in sources :

Example 6 with PlanNode

use of org.apache.flink.optimizer.plan.PlanNode in project flink by apache.

the class DataSourceNode method getAlternativePlans.

@Override
public List<PlanNode> getAlternativePlans(CostEstimator estimator) {
    if (this.cachedPlans != null) {
        return this.cachedPlans;
    }
    SourcePlanNode candidate = new SourcePlanNode(this, "DataSource (" + this.getOperator().getName() + ")", this.gprops, this.lprops);
    if (!replicatedInput) {
        candidate.updatePropertiesWithUniqueSets(getUniqueFields());
        final Costs costs = new Costs();
        if (FileInputFormat.class.isAssignableFrom(getOperator().getFormatWrapper().getUserCodeClass()) && this.estimatedOutputSize >= 0) {
            estimator.addFileInputCost(this.estimatedOutputSize, costs);
        }
        candidate.setCosts(costs);
    } else {
        // replicated input
        final Costs costs = new Costs();
        InputFormat<?, ?> inputFormat = ((ReplicatingInputFormat<?, ?>) getOperator().getFormatWrapper().getUserCodeObject()).getReplicatedInputFormat();
        if (FileInputFormat.class.isAssignableFrom(inputFormat.getClass()) && this.estimatedOutputSize >= 0) {
            estimator.addFileInputCost(this.estimatedOutputSize * this.getParallelism(), costs);
        }
        candidate.setCosts(costs);
    }
    // since there is only a single plan for the data-source, return a list with that element only
    List<PlanNode> plans = new ArrayList<PlanNode>(1);
    plans.add(candidate);
    this.cachedPlans = plans;
    return plans;
}
Also used : ReplicatingInputFormat(org.apache.flink.api.common.io.ReplicatingInputFormat) Costs(org.apache.flink.optimizer.costs.Costs) SourcePlanNode(org.apache.flink.optimizer.plan.SourcePlanNode) PlanNode(org.apache.flink.optimizer.plan.PlanNode) ArrayList(java.util.ArrayList) SourcePlanNode(org.apache.flink.optimizer.plan.SourcePlanNode) FileInputFormat(org.apache.flink.api.common.io.FileInputFormat)

Example 7 with PlanNode

use of org.apache.flink.optimizer.plan.PlanNode 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 8 with PlanNode

use of org.apache.flink.optimizer.plan.PlanNode 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 9 with PlanNode

use of org.apache.flink.optimizer.plan.PlanNode 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 10 with PlanNode

use of org.apache.flink.optimizer.plan.PlanNode in project flink by apache.

the class ParallelismChangeTest method checkPropertyHandlingWithTwoInputs.

/**
	 * Checks that re-partitioning happens when the inputs of a two-input contract have different parallelisms.
	 * 
	 * Test Plan:
	 * <pre>
	 * 
	 * (source) -> reduce -\
	 *                      Match -> (sink)
	 * (source) -> reduce -/
	 * 
	 * </pre>
	 * 
	 */
@Test
public void checkPropertyHandlingWithTwoInputs() {
    // construct the plan
    ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
    env.setParallelism(DEFAULT_PARALLELISM);
    DataSet<Long> set1 = env.generateSequence(0, 1).setParallelism(5);
    DataSet<Long> set2 = env.generateSequence(0, 1).setParallelism(7);
    DataSet<Long> reduce1 = set1.groupBy("*").reduceGroup(new IdentityGroupReducer<Long>()).withForwardedFields("*").setParallelism(5);
    DataSet<Long> reduce2 = set2.groupBy("*").reduceGroup(new IdentityGroupReducer<Long>()).withForwardedFields("*").setParallelism(7);
    reduce1.join(reduce2).where("*").equalTo("*").with(new IdentityJoiner<Long>()).setParallelism(5).output(new DiscardingOutputFormat<Long>()).setParallelism(5);
    Plan plan = env.createProgramPlan();
    // submit the plan to the compiler
    OptimizedPlan oPlan = compileNoStats(plan);
    JobGraphGenerator jobGen = new JobGraphGenerator();
    //Compile plan to verify that no error is thrown
    jobGen.compileJobGraph(oPlan);
    oPlan.accept(new Visitor<PlanNode>() {

        @Override
        public boolean preVisit(PlanNode visitable) {
            if (visitable instanceof DualInputPlanNode) {
                DualInputPlanNode node = (DualInputPlanNode) visitable;
                Channel c1 = node.getInput1();
                Channel c2 = node.getInput2();
                Assert.assertEquals("Incompatible shipping strategy chosen for match", ShipStrategyType.FORWARD, c1.getShipStrategy());
                Assert.assertEquals("Incompatible shipping strategy chosen for match", ShipStrategyType.PARTITION_HASH, c2.getShipStrategy());
                return false;
            }
            return true;
        }

        @Override
        public void postVisit(PlanNode visitable) {
        // DO NOTHING
        }
    });
}
Also used : ExecutionEnvironment(org.apache.flink.api.java.ExecutionEnvironment) Channel(org.apache.flink.optimizer.plan.Channel) 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) DualInputPlanNode(org.apache.flink.optimizer.plan.DualInputPlanNode) 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) JobGraphGenerator(org.apache.flink.optimizer.plantranslate.JobGraphGenerator) IdentityJoiner(org.apache.flink.optimizer.testfunctions.IdentityJoiner) Test(org.junit.Test)

Aggregations

PlanNode (org.apache.flink.optimizer.plan.PlanNode)43 SingleInputPlanNode (org.apache.flink.optimizer.plan.SingleInputPlanNode)27 DualInputPlanNode (org.apache.flink.optimizer.plan.DualInputPlanNode)25 Channel (org.apache.flink.optimizer.plan.Channel)24 SinkPlanNode (org.apache.flink.optimizer.plan.SinkPlanNode)22 WorksetIterationPlanNode (org.apache.flink.optimizer.plan.WorksetIterationPlanNode)21 CompilerException (org.apache.flink.optimizer.CompilerException)16 NamedChannel (org.apache.flink.optimizer.plan.NamedChannel)16 OptimizedPlan (org.apache.flink.optimizer.plan.OptimizedPlan)15 BulkIterationPlanNode (org.apache.flink.optimizer.plan.BulkIterationPlanNode)14 BulkPartialSolutionPlanNode (org.apache.flink.optimizer.plan.BulkPartialSolutionPlanNode)13 IterationPlanNode (org.apache.flink.optimizer.plan.IterationPlanNode)13 NAryUnionPlanNode (org.apache.flink.optimizer.plan.NAryUnionPlanNode)13 SolutionSetPlanNode (org.apache.flink.optimizer.plan.SolutionSetPlanNode)13 SourcePlanNode (org.apache.flink.optimizer.plan.SourcePlanNode)13 WorksetPlanNode (org.apache.flink.optimizer.plan.WorksetPlanNode)13 Plan (org.apache.flink.api.common.Plan)12 ExecutionEnvironment (org.apache.flink.api.java.ExecutionEnvironment)12 Test (org.junit.Test)12 ArrayList (java.util.ArrayList)11