Examples with DiscardingOutputFormat org.apache.flink.api.java.io.DiscardingOutputFormat used on opensource projects

Example 16 with DiscardingOutputFormat

use of org.apache.flink.api.java.io.DiscardingOutputFormat in project flink by apache.

the class GroupOrderTest method testReduceWithGroupOrder.

@Test
public void testReduceWithGroupOrder() {
    // construct the plan
    ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
    env.setParallelism(DEFAULT_PARALLELISM);
    DataSet<Tuple4<Long, Long, Long, Long>> set1 = env.readCsvFile("/tmp/fake.csv").types(Long.class, Long.class, Long.class, Long.class);
    set1.groupBy(1).sortGroup(3, Order.DESCENDING).reduceGroup(new IdentityGroupReducer<Tuple4<Long, Long, Long, Long>>()).name("Reduce").output(new DiscardingOutputFormat<Tuple4<Long, Long, Long, Long>>()).name("Sink");
    Plan plan = env.createProgramPlan();
    OptimizedPlan oPlan;
    try {
        oPlan = compileNoStats(plan);
    } catch (CompilerException ce) {
        ce.printStackTrace();
        fail("The pact compiler is unable to compile this plan correctly.");
        // silence the compiler
        return;
    }
    OptimizerPlanNodeResolver resolver = getOptimizerPlanNodeResolver(oPlan);
    SinkPlanNode sinkNode = resolver.getNode("Sink");
    SingleInputPlanNode reducer = resolver.getNode("Reduce");
    // verify the strategies
    Assert.assertEquals(ShipStrategyType.FORWARD, sinkNode.getInput().getShipStrategy());
    Assert.assertEquals(ShipStrategyType.PARTITION_HASH, reducer.getInput().getShipStrategy());
    Channel c = reducer.getInput();
    Assert.assertEquals(LocalStrategy.SORT, c.getLocalStrategy());
    FieldList ship = new FieldList(1);
    FieldList local = new FieldList(1, 3);
    Assert.assertEquals(ship, c.getShipStrategyKeys());
    Assert.assertEquals(local, c.getLocalStrategyKeys());
    Assert.assertTrue(c.getLocalStrategySortOrder()[0] == reducer.getSortOrders(0)[0]);
    // check that we indeed sort descending
    Assert.assertEquals(false, c.getLocalStrategySortOrder()[1]);
}

Example 17 with DiscardingOutputFormat

use of org.apache.flink.api.java.io.DiscardingOutputFormat 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
        }
    });
}

Example 18 with DiscardingOutputFormat

use of org.apache.flink.api.java.io.DiscardingOutputFormat in project flink by apache.

the class UnionReplacementTest method testConsecutiveUnionsWithRebalance.

/**
	 *
	 * Checks that a plan with consecutive UNIONs followed by REBALANCE is correctly translated.
	 *
	 * The program can be illustrated as follows:
	 *
	 * Src1 -\
	 *        >-> Union12--<
	 * Src2 -/              \
	 *                       >-> Union123 -> Rebalance -> Output
	 * Src3 ----------------/
	 *
	 * In the resulting plan, the Rebalance (ShippingStrategy.PARTITION_FORCED_REBALANCE) must be
	 * pushed to the inputs of the unions (Src1, Src2, Src3).
	 *
	 */
@Test
public void testConsecutiveUnionsWithRebalance() throws Exception {
    // -----------------------------------------------------------------------------------------
    // Build test program
    // -----------------------------------------------------------------------------------------
    ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
    env.setParallelism(DEFAULT_PARALLELISM);
    DataSet<Tuple2<Long, Long>> src1 = env.fromElements(new Tuple2<>(0L, 0L));
    DataSet<Tuple2<Long, Long>> src2 = env.fromElements(new Tuple2<>(0L, 0L));
    DataSet<Tuple2<Long, Long>> src3 = env.fromElements(new Tuple2<>(0L, 0L));
    DataSet<Tuple2<Long, Long>> union12 = src1.union(src2);
    DataSet<Tuple2<Long, Long>> union123 = union12.union(src3);
    union123.rebalance().output(new DiscardingOutputFormat<Tuple2<Long, Long>>()).name("out");
    // -----------------------------------------------------------------------------------------
    // Verify optimized plan
    // -----------------------------------------------------------------------------------------
    OptimizedPlan optimizedPlan = compileNoStats(env.createProgramPlan());
    OptimizerPlanNodeResolver resolver = getOptimizerPlanNodeResolver(optimizedPlan);
    SingleInputPlanNode sink = resolver.getNode("out");
    // check partitioning is correct
    assertEquals("Sink input should be force rebalanced.", PartitioningProperty.FORCED_REBALANCED, sink.getInput().getGlobalProperties().getPartitioning());
    SingleInputPlanNode partitioner = (SingleInputPlanNode) sink.getInput().getSource();
    assertTrue(partitioner.getDriverStrategy() == DriverStrategy.UNARY_NO_OP);
    assertEquals("Partitioner input should be force rebalanced.", PartitioningProperty.FORCED_REBALANCED, partitioner.getInput().getGlobalProperties().getPartitioning());
    assertEquals("Partitioner input channel should be forwarding", ShipStrategyType.FORWARD, partitioner.getInput().getShipStrategy());
    NAryUnionPlanNode union = (NAryUnionPlanNode) partitioner.getInput().getSource();
    // all union inputs should be force rebalanced
    for (Channel c : union.getInputs()) {
        assertEquals("Union input should be force rebalanced", PartitioningProperty.FORCED_REBALANCED, c.getGlobalProperties().getPartitioning());
        assertEquals("Union input channel should be rebalancing", ShipStrategyType.PARTITION_FORCED_REBALANCE, c.getShipStrategy());
        assertTrue("Union input should be data source", c.getSource() instanceof SourcePlanNode);
    }
}

Example 19 with DiscardingOutputFormat

use of org.apache.flink.api.java.io.DiscardingOutputFormat in project flink by apache.

the class DistinctCompilationTest method testDistinctWithCombineHint.

@Test
public void testDistinctWithCombineHint() {
    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.distinct().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();
        // check wiring
        assertEquals(sourceNode, combineNode.getInput().getSource());
        assertEquals(reduceNode, sinkNode.getInput().getSource());
        // check that both reduce and combiner have the same strategy
        assertEquals(DriverStrategy.SORTED_REDUCE, reduceNode.getDriverStrategy());
        assertEquals(DriverStrategy.HASHED_PARTIAL_REDUCE, combineNode.getDriverStrategy());
        // check the keys
        assertEquals(new FieldList(0, 1), reduceNode.getKeys(0));
        assertEquals(new FieldList(0, 1), combineNode.getKeys(0));
        assertEquals(new FieldList(0, 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());
    }
}

Example 20 with DiscardingOutputFormat

use of org.apache.flink.api.java.io.DiscardingOutputFormat in project flink by apache.

the class PregelCompilerTest method testPregelCompiler.

@SuppressWarnings("serial")
@Test
public void testPregelCompiler() {
    try {
        ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
        env.setParallelism(DEFAULT_PARALLELISM);
        // compose test program
        {
            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);
            DataSet<Vertex<Long, Long>> result = graph.runVertexCentricIteration(new CCCompute(), null, 100).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());
    }
}