Search in sources :

Example 76 with SingleInputPlanNode

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

the class ReduceCompilationTest method testGroupedReduceWithFieldPositionKey.

@Test
public void testGroupedReduceWithFieldPositionKey() {
    try {
        ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
        env.setParallelism(8);
        DataSet<Tuple2<String, Double>> data = env.readCsvFile("file:///will/never/be/read").types(String.class, Double.class).name("source").setParallelism(6);
        data.groupBy(1).reduce(new RichReduceFunction<Tuple2<String, Double>>() {

            @Override
            public Tuple2<String, Double> reduce(Tuple2<String, Double> value1, Tuple2<String, Double> value2) {
                return null;
            }
        }).name("reducer").output(new DiscardingOutputFormat<Tuple2<String, Double>>()).name("sink");
        Plan p = env.createProgramPlan();
        OptimizedPlan op = compileNoStats(p);
        OptimizerPlanNodeResolver resolver = getOptimizerPlanNodeResolver(op);
        // get the original nodes
        SourcePlanNode sourceNode = resolver.getNode("source");
        SingleInputPlanNode reduceNode = resolver.getNode("reducer");
        SinkPlanNode sinkNode = resolver.getNode("sink");
        // get the combiner
        SingleInputPlanNode combineNode = (SingleInputPlanNode) reduceNode.getInput().getSource();
        // check wiring
        assertEquals(sourceNode, combineNode.getInput().getSource());
        assertEquals(reduceNode, sinkNode.getInput().getSource());
        // check the strategies
        assertEquals(DriverStrategy.SORTED_REDUCE, reduceNode.getDriverStrategy());
        assertEquals(DriverStrategy.SORTED_PARTIAL_REDUCE, combineNode.getDriverStrategy());
        // check the keys
        assertEquals(new FieldList(1), reduceNode.getKeys(0));
        assertEquals(new FieldList(1), combineNode.getKeys(0));
        assertEquals(new FieldList(1), reduceNode.getInput().getLocalStrategyKeys());
        // check parallelism
        assertEquals(6, sourceNode.getParallelism());
        assertEquals(6, combineNode.getParallelism());
        assertEquals(8, reduceNode.getParallelism());
        assertEquals(8, sinkNode.getParallelism());
    } catch (Exception e) {
        System.err.println(e.getMessage());
        e.printStackTrace();
        fail(e.getClass().getSimpleName() + " in test: " + e.getMessage());
    }
}
Also used : ExecutionEnvironment(org.apache.flink.api.java.ExecutionEnvironment) 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) RichReduceFunction(org.apache.flink.api.common.functions.RichReduceFunction) Tuple2(org.apache.flink.api.java.tuple.Tuple2) SourcePlanNode(org.apache.flink.optimizer.plan.SourcePlanNode) SinkPlanNode(org.apache.flink.optimizer.plan.SinkPlanNode) Test(org.junit.Test)

Example 77 with SingleInputPlanNode

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

the class ReduceCompilationTest method testAllReduceNoCombiner.

@Test
public void testAllReduceNoCombiner() {
    try {
        ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
        env.setParallelism(8);
        DataSet<Double> data = env.fromElements(0.2, 0.3, 0.4, 0.5).name("source");
        data.reduce(new RichReduceFunction<Double>() {

            @Override
            public Double reduce(Double value1, Double value2) {
                return value1 + value2;
            }
        }).name("reducer").output(new DiscardingOutputFormat<Double>()).name("sink");
        Plan p = env.createProgramPlan();
        OptimizedPlan op = compileNoStats(p);
        OptimizerPlanNodeResolver resolver = getOptimizerPlanNodeResolver(op);
        // the all-reduce has no combiner, when the parallelism of the input is one
        SourcePlanNode sourceNode = resolver.getNode("source");
        SingleInputPlanNode reduceNode = resolver.getNode("reducer");
        SinkPlanNode sinkNode = resolver.getNode("sink");
        // check wiring
        assertEquals(sourceNode, reduceNode.getInput().getSource());
        assertEquals(reduceNode, sinkNode.getInput().getSource());
        // check parallelism
        assertEquals(1, sourceNode.getParallelism());
        assertEquals(1, reduceNode.getParallelism());
        assertEquals(1, sinkNode.getParallelism());
    } catch (Exception e) {
        System.err.println(e.getMessage());
        e.printStackTrace();
        fail(e.getClass().getSimpleName() + " in test: " + e.getMessage());
    }
}
Also used : SingleInputPlanNode(org.apache.flink.optimizer.plan.SingleInputPlanNode) ExecutionEnvironment(org.apache.flink.api.java.ExecutionEnvironment) RichReduceFunction(org.apache.flink.api.common.functions.RichReduceFunction) SourcePlanNode(org.apache.flink.optimizer.plan.SourcePlanNode) SinkPlanNode(org.apache.flink.optimizer.plan.SinkPlanNode) 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) Test(org.junit.Test)

Example 78 with SingleInputPlanNode

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

the class ReduceCompilationTest method testGroupedReduceWithHint.

@Test
public void testGroupedReduceWithHint() {
    try {
        ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
        env.setParallelism(8);
        DataSet<Tuple2<String, Double>> data = env.readCsvFile("file:///will/never/be/read").types(String.class, Double.class).name("source").setParallelism(6);
        data.groupBy(new KeySelector<Tuple2<String, Double>, String>() {

            public String getKey(Tuple2<String, Double> value) {
                return value.f0;
            }
        }).reduce(new RichReduceFunction<Tuple2<String, Double>>() {

            @Override
            public Tuple2<String, Double> reduce(Tuple2<String, Double> value1, Tuple2<String, Double> value2) {
                return null;
            }
        }).setCombineHint(CombineHint.HASH).name("reducer").output(new DiscardingOutputFormat<Tuple2<String, Double>>()).name("sink");
        Plan p = env.createProgramPlan();
        OptimizedPlan op = compileNoStats(p);
        OptimizerPlanNodeResolver resolver = getOptimizerPlanNodeResolver(op);
        // get the original nodes
        SourcePlanNode sourceNode = resolver.getNode("source");
        SingleInputPlanNode reduceNode = resolver.getNode("reducer");
        SinkPlanNode sinkNode = resolver.getNode("sink");
        // get the combiner
        SingleInputPlanNode combineNode = (SingleInputPlanNode) reduceNode.getInput().getSource();
        // get the key extractors and projectors
        SingleInputPlanNode keyExtractor = (SingleInputPlanNode) combineNode.getInput().getSource();
        SingleInputPlanNode keyProjector = (SingleInputPlanNode) sinkNode.getInput().getSource();
        // check wiring
        assertEquals(sourceNode, keyExtractor.getInput().getSource());
        assertEquals(keyProjector, sinkNode.getInput().getSource());
        // check the strategies
        assertEquals(DriverStrategy.SORTED_REDUCE, reduceNode.getDriverStrategy());
        assertEquals(DriverStrategy.HASHED_PARTIAL_REDUCE, combineNode.getDriverStrategy());
        // check the keys
        assertEquals(new FieldList(0), reduceNode.getKeys(0));
        assertEquals(new FieldList(0), combineNode.getKeys(0));
        assertEquals(new FieldList(0), reduceNode.getInput().getLocalStrategyKeys());
        // check parallelism
        assertEquals(6, sourceNode.getParallelism());
        assertEquals(6, keyExtractor.getParallelism());
        assertEquals(6, combineNode.getParallelism());
        assertEquals(8, reduceNode.getParallelism());
        assertEquals(8, keyProjector.getParallelism());
        assertEquals(8, sinkNode.getParallelism());
    } catch (Exception e) {
        System.err.println(e.getMessage());
        e.printStackTrace();
        fail(e.getClass().getSimpleName() + " in test: " + e.getMessage());
    }
}
Also used : ExecutionEnvironment(org.apache.flink.api.java.ExecutionEnvironment) KeySelector(org.apache.flink.api.java.functions.KeySelector) 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) Tuple2(org.apache.flink.api.java.tuple.Tuple2) SourcePlanNode(org.apache.flink.optimizer.plan.SourcePlanNode) SinkPlanNode(org.apache.flink.optimizer.plan.SinkPlanNode) Test(org.junit.Test)

Example 79 with SingleInputPlanNode

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

the class WorksetIterationsJavaApiCompilerTest method testJavaApiWithDeferredSoltionSetUpdateWithMapper.

@Test
public void testJavaApiWithDeferredSoltionSetUpdateWithMapper() {
    try {
        Plan plan = getJavaTestPlan(false, true);
        OptimizedPlan oPlan = compileNoStats(plan);
        OptimizerPlanNodeResolver resolver = getOptimizerPlanNodeResolver(oPlan);
        DualInputPlanNode joinWithInvariantNode = resolver.getNode(JOIN_WITH_INVARIANT_NAME);
        DualInputPlanNode joinWithSolutionSetNode = resolver.getNode(JOIN_WITH_SOLUTION_SET);
        SingleInputPlanNode worksetReducer = resolver.getNode(NEXT_WORKSET_REDUCER_NAME);
        SingleInputPlanNode deltaMapper = resolver.getNode(SOLUTION_DELTA_MAPPER_NAME);
        // iteration preserves partitioning in reducer, so the first partitioning is out of the loop, 
        // the in-loop partitioning is before the final reducer
        // verify joinWithInvariant
        assertEquals(ShipStrategyType.FORWARD, joinWithInvariantNode.getInput1().getShipStrategy());
        assertEquals(ShipStrategyType.PARTITION_HASH, joinWithInvariantNode.getInput2().getShipStrategy());
        assertEquals(new FieldList(1, 2), joinWithInvariantNode.getKeysForInput1());
        assertEquals(new FieldList(1, 2), joinWithInvariantNode.getKeysForInput2());
        // verify joinWithSolutionSet
        assertEquals(ShipStrategyType.PARTITION_HASH, joinWithSolutionSetNode.getInput1().getShipStrategy());
        assertEquals(ShipStrategyType.FORWARD, joinWithSolutionSetNode.getInput2().getShipStrategy());
        assertEquals(new FieldList(1, 0), joinWithSolutionSetNode.getKeysForInput1());
        // verify reducer
        assertEquals(ShipStrategyType.PARTITION_HASH, worksetReducer.getInput().getShipStrategy());
        assertEquals(new FieldList(1, 2), worksetReducer.getKeys(0));
        // currently, the system may partition before or after the mapper
        ShipStrategyType ss1 = deltaMapper.getInput().getShipStrategy();
        ShipStrategyType ss2 = deltaMapper.getOutgoingChannels().get(0).getShipStrategy();
        assertTrue((ss1 == ShipStrategyType.FORWARD && ss2 == ShipStrategyType.PARTITION_HASH) || (ss2 == ShipStrategyType.FORWARD && ss1 == ShipStrategyType.PARTITION_HASH));
        new JobGraphGenerator().compileJobGraph(oPlan);
    } catch (Exception e) {
        System.err.println(e.getMessage());
        e.printStackTrace();
        fail("Test errored: " + e.getMessage());
    }
}
Also used : DualInputPlanNode(org.apache.flink.optimizer.plan.DualInputPlanNode) SingleInputPlanNode(org.apache.flink.optimizer.plan.SingleInputPlanNode) JobGraphGenerator(org.apache.flink.optimizer.plantranslate.JobGraphGenerator) Plan(org.apache.flink.api.common.Plan) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) InvalidProgramException(org.apache.flink.api.common.InvalidProgramException) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) FieldList(org.apache.flink.api.common.operators.util.FieldList) ShipStrategyType(org.apache.flink.runtime.operators.shipping.ShipStrategyType) Test(org.junit.Test)

Example 80 with SingleInputPlanNode

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

the class WorksetIterationsJavaApiCompilerTest method testJavaApiWithDirectSoltionSetUpdate.

@Test
public void testJavaApiWithDirectSoltionSetUpdate() {
    try {
        Plan plan = getJavaTestPlan(true, false);
        OptimizedPlan oPlan = compileNoStats(plan);
        OptimizerPlanNodeResolver resolver = getOptimizerPlanNodeResolver(oPlan);
        DualInputPlanNode joinWithInvariantNode = resolver.getNode(JOIN_WITH_INVARIANT_NAME);
        DualInputPlanNode joinWithSolutionSetNode = resolver.getNode(JOIN_WITH_SOLUTION_SET);
        SingleInputPlanNode worksetReducer = resolver.getNode(NEXT_WORKSET_REDUCER_NAME);
        // iteration preserves partitioning in reducer, so the first partitioning is out of the loop, 
        // the in-loop partitioning is before the final reducer
        // verify joinWithInvariant
        assertEquals(ShipStrategyType.FORWARD, joinWithInvariantNode.getInput1().getShipStrategy());
        assertEquals(ShipStrategyType.PARTITION_HASH, joinWithInvariantNode.getInput2().getShipStrategy());
        assertEquals(new FieldList(1, 2), joinWithInvariantNode.getKeysForInput1());
        assertEquals(new FieldList(1, 2), joinWithInvariantNode.getKeysForInput2());
        // verify joinWithSolutionSet
        assertEquals(ShipStrategyType.PARTITION_HASH, joinWithSolutionSetNode.getInput1().getShipStrategy());
        assertEquals(ShipStrategyType.FORWARD, joinWithSolutionSetNode.getInput2().getShipStrategy());
        assertEquals(new FieldList(1, 0), joinWithSolutionSetNode.getKeysForInput1());
        // verify reducer
        assertEquals(ShipStrategyType.FORWARD, worksetReducer.getInput().getShipStrategy());
        assertEquals(new FieldList(1, 2), worksetReducer.getKeys(0));
        // verify solution delta
        assertEquals(1, joinWithSolutionSetNode.getOutgoingChannels().size());
        assertEquals(ShipStrategyType.FORWARD, joinWithSolutionSetNode.getOutgoingChannels().get(0).getShipStrategy());
        new JobGraphGenerator().compileJobGraph(oPlan);
    } catch (Exception e) {
        System.err.println(e.getMessage());
        e.printStackTrace();
        fail("Test errored: " + e.getMessage());
    }
}
Also used : DualInputPlanNode(org.apache.flink.optimizer.plan.DualInputPlanNode) SingleInputPlanNode(org.apache.flink.optimizer.plan.SingleInputPlanNode) JobGraphGenerator(org.apache.flink.optimizer.plantranslate.JobGraphGenerator) Plan(org.apache.flink.api.common.Plan) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) InvalidProgramException(org.apache.flink.api.common.InvalidProgramException) OptimizedPlan(org.apache.flink.optimizer.plan.OptimizedPlan) FieldList(org.apache.flink.api.common.operators.util.FieldList) Test(org.junit.Test)

Aggregations

SingleInputPlanNode (org.apache.flink.optimizer.plan.SingleInputPlanNode)104 Test (org.junit.Test)83 OptimizedPlan (org.apache.flink.optimizer.plan.OptimizedPlan)81 Plan (org.apache.flink.api.common.Plan)73 SinkPlanNode (org.apache.flink.optimizer.plan.SinkPlanNode)72 ExecutionEnvironment (org.apache.flink.api.java.ExecutionEnvironment)71 Tuple2 (org.apache.flink.api.java.tuple.Tuple2)38 Channel (org.apache.flink.optimizer.plan.Channel)32 SourcePlanNode (org.apache.flink.optimizer.plan.SourcePlanNode)32 FieldList (org.apache.flink.api.common.operators.util.FieldList)31 DualInputPlanNode (org.apache.flink.optimizer.plan.DualInputPlanNode)28 DiscardingOutputFormat (org.apache.flink.api.java.io.DiscardingOutputFormat)26 InvalidProgramException (org.apache.flink.api.common.InvalidProgramException)18 JobGraphGenerator (org.apache.flink.optimizer.plantranslate.JobGraphGenerator)16 NAryUnionPlanNode (org.apache.flink.optimizer.plan.NAryUnionPlanNode)14 PlanNode (org.apache.flink.optimizer.plan.PlanNode)14 IdentityGroupReducerCombinable (org.apache.flink.optimizer.testfunctions.IdentityGroupReducerCombinable)14 IdentityMapper (org.apache.flink.optimizer.testfunctions.IdentityMapper)14 GlobalProperties (org.apache.flink.optimizer.dataproperties.GlobalProperties)13 LocalProperties (org.apache.flink.optimizer.dataproperties.LocalProperties)13