Examples with OptimizedPlan org.apache.flink.optimizer.plan.OptimizedPlan used on opensource projects

Example 26 with OptimizedPlan

use of org.apache.flink.optimizer.plan.OptimizedPlan 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());
    }
}

Example 27 with OptimizedPlan

use of org.apache.flink.optimizer.plan.OptimizedPlan 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());
    }
}

Example 28 with OptimizedPlan

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

the class DistinctCompilationTest method testDistinctWithFieldPositionKeyCombinable.

@Test
public void testDistinctWithFieldPositionKeyCombinable() {
    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);
        DistinctOperator<Tuple2<String, Double>> reduced = data.distinct(1).name("reducer");
        reduced.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.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());
    }
}

Example 29 with OptimizedPlan

use of org.apache.flink.optimizer.plan.OptimizedPlan 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 30 with OptimizedPlan

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

the class IterationsCompilerTest method testBulkIterationWithPartialSolutionProperties.

/**
	 * Tests that interesting properties can be pushed out of the bulk iteration. This requires
	 * that a NoOp node is appended to the step function which re-establishes the properties of
	 * the initial input. If this does not work, then Flink won't find a plan, because the optimizer
	 * will not consider plans where the partitioning is done after the partial solution node in
	 * this case (because of pruning).
	 * @throws Exception
	 */
@Test
public void testBulkIterationWithPartialSolutionProperties() throws Exception {
    ExecutionEnvironment env = ExecutionEnvironment.getExecutionEnvironment();
    DataSet<Tuple1<Long>> input1 = env.generateSequence(1, 10).map(new MapFunction<Long, Tuple1<Long>>() {

        @Override
        public Tuple1<Long> map(Long value) throws Exception {
            return new Tuple1<>(value);
        }
    });
    DataSet<Tuple1<Long>> input2 = env.generateSequence(1, 10).map(new MapFunction<Long, Tuple1<Long>>() {

        @Override
        public Tuple1<Long> map(Long value) throws Exception {
            return new Tuple1<>(value);
        }
    });
    DataSet<Tuple1<Long>> distinctInput = input1.distinct();
    IterativeDataSet<Tuple1<Long>> iteration = distinctInput.iterate(10);
    DataSet<Tuple1<Long>> iterationStep = iteration.coGroup(input2).where(0).equalTo(0).with(new CoGroupFunction<Tuple1<Long>, Tuple1<Long>, Tuple1<Long>>() {

        @Override
        public void coGroup(Iterable<Tuple1<Long>> first, Iterable<Tuple1<Long>> second, Collector<Tuple1<Long>> out) throws Exception {
            Iterator<Tuple1<Long>> it = first.iterator();
            if (it.hasNext()) {
                out.collect(it.next());
            }
        }
    });
    DataSet<Tuple1<Long>> iterationResult = iteration.closeWith(iterationStep);
    iterationResult.output(new DiscardingOutputFormat<Tuple1<Long>>());
    Plan p = env.createProgramPlan();
    OptimizedPlan op = compileNoStats(p);
    new JobGraphGenerator().compileJobGraph(op);
}