Examples with UnionIterator org.apache.flink.runtime.operators.testutils.UnionIterator used on opensource projects

Example 1 with UnionIterator

use of org.apache.flink.runtime.operators.testutils.UnionIterator in project flink by apache.

the class ReusingHashJoinIteratorITCase method testBuildSecondWithHighNumberOfCommonKeys.

@Test
public void testBuildSecondWithHighNumberOfCommonKeys() {
    // the size of the left and right inputs
    final int INPUT_1_SIZE = 200;
    final int INPUT_2_SIZE = 100;
    final int INPUT_1_DUPLICATES = 10;
    final int INPUT_2_DUPLICATES = 2000;
    final int DUPLICATE_KEY = 13;
    try {
        TestData.TupleGenerator generator1 = new TestData.TupleGenerator(SEED1, 500, 4096, KeyMode.RANDOM, ValueMode.RANDOM_LENGTH);
        TestData.TupleGenerator generator2 = new TestData.TupleGenerator(SEED2, 500, 2048, KeyMode.RANDOM, ValueMode.RANDOM_LENGTH);
        final TestData.TupleGeneratorIterator gen1Iter = new TestData.TupleGeneratorIterator(generator1, INPUT_1_SIZE);
        final TestData.TupleGeneratorIterator gen2Iter = new TestData.TupleGeneratorIterator(generator2, INPUT_2_SIZE);
        final TestData.TupleConstantValueIterator const1Iter = new TestData.TupleConstantValueIterator(DUPLICATE_KEY, "LEFT String for Duplicate Keys", INPUT_1_DUPLICATES);
        final TestData.TupleConstantValueIterator const2Iter = new TestData.TupleConstantValueIterator(DUPLICATE_KEY, "RIGHT String for Duplicate Keys", INPUT_2_DUPLICATES);
        final List<MutableObjectIterator<Tuple2<Integer, String>>> inList1 = new ArrayList<>();
        inList1.add(gen1Iter);
        inList1.add(const1Iter);
        final List<MutableObjectIterator<Tuple2<Integer, String>>> inList2 = new ArrayList<>();
        inList2.add(gen2Iter);
        inList2.add(const2Iter);
        MutableObjectIterator<Tuple2<Integer, String>> input1 = new UnionIterator<>(inList1);
        MutableObjectIterator<Tuple2<Integer, String>> input2 = new UnionIterator<>(inList2);
        // collect expected data
        final Map<Integer, Collection<TupleMatch>> expectedMatchesMap = joinTuples(collectTupleData(input1), collectTupleData(input2));
        // re-create the whole thing for actual processing
        // reset the generators and iterators
        generator1.reset();
        generator2.reset();
        const1Iter.reset();
        const2Iter.reset();
        gen1Iter.reset();
        gen2Iter.reset();
        inList1.clear();
        inList1.add(gen1Iter);
        inList1.add(const1Iter);
        inList2.clear();
        inList2.add(gen2Iter);
        inList2.add(const2Iter);
        input1 = new UnionIterator<>(inList1);
        input2 = new UnionIterator<>(inList2);
        final FlatJoinFunction matcher = new TupleMatchRemovingJoin(expectedMatchesMap);
        final Collector<Tuple2<Integer, String>> collector = new DiscardingOutputCollector<>();
        ReusingBuildSecondHashJoinIterator<Tuple2<Integer, String>, Tuple2<Integer, String>, Tuple2<Integer, String>> iterator = new ReusingBuildSecondHashJoinIterator<>(input1, input2, this.recordSerializer, this.record1Comparator, this.recordSerializer, this.record2Comparator, this.recordPairComparator, this.memoryManager, ioManager, this.parentTask, 1.0, false, false, true);
        iterator.open();
        while (iterator.callWithNextKey(matcher, collector)) ;
        iterator.close();
        // assert that each expected match was seen
        for (Entry<Integer, Collection<TupleMatch>> entry : expectedMatchesMap.entrySet()) {
            if (!entry.getValue().isEmpty()) {
                Assert.fail("Collection for key " + entry.getKey() + " is not empty");
            }
        }
    } catch (Exception e) {
        e.printStackTrace();
        Assert.fail("An exception occurred during the test: " + e.getMessage());
    }
}

Example 2 with UnionIterator

use of org.apache.flink.runtime.operators.testutils.UnionIterator in project flink by apache.

the class NonReusingHashJoinIteratorITCase method testBuildSecondWithHighNumberOfCommonKeys.

@Test
public void testBuildSecondWithHighNumberOfCommonKeys() {
    // the size of the left and right inputs
    final int INPUT_1_SIZE = 200;
    final int INPUT_2_SIZE = 100;
    final int INPUT_1_DUPLICATES = 10;
    final int INPUT_2_DUPLICATES = 2000;
    final int DUPLICATE_KEY = 13;
    try {
        TupleGenerator generator1 = new TupleGenerator(SEED1, 500, 4096, KeyMode.RANDOM, ValueMode.RANDOM_LENGTH);
        TupleGenerator generator2 = new TupleGenerator(SEED2, 500, 2048, KeyMode.RANDOM, ValueMode.RANDOM_LENGTH);
        final TestData.TupleGeneratorIterator gen1Iter = new TestData.TupleGeneratorIterator(generator1, INPUT_1_SIZE);
        final TestData.TupleGeneratorIterator gen2Iter = new TestData.TupleGeneratorIterator(generator2, INPUT_2_SIZE);
        final TestData.TupleConstantValueIterator const1Iter = new TestData.TupleConstantValueIterator(DUPLICATE_KEY, "LEFT String for Duplicate Keys", INPUT_1_DUPLICATES);
        final TestData.TupleConstantValueIterator const2Iter = new TestData.TupleConstantValueIterator(DUPLICATE_KEY, "RIGHT String for Duplicate Keys", INPUT_2_DUPLICATES);
        final List<MutableObjectIterator<Tuple2<Integer, String>>> inList1 = new ArrayList<>();
        inList1.add(gen1Iter);
        inList1.add(const1Iter);
        final List<MutableObjectIterator<Tuple2<Integer, String>>> inList2 = new ArrayList<>();
        inList2.add(gen2Iter);
        inList2.add(const2Iter);
        MutableObjectIterator<Tuple2<Integer, String>> input1 = new UnionIterator<>(inList1);
        MutableObjectIterator<Tuple2<Integer, String>> input2 = new UnionIterator<>(inList2);
        // collect expected data
        final Map<Integer, Collection<TupleMatch>> expectedMatchesMap = joinTuples(collectTupleData(input1), collectTupleData(input2));
        // re-create the whole thing for actual processing
        // reset the generators and iterators
        generator1.reset();
        generator2.reset();
        const1Iter.reset();
        const2Iter.reset();
        gen1Iter.reset();
        gen2Iter.reset();
        inList1.clear();
        inList1.add(gen1Iter);
        inList1.add(const1Iter);
        inList2.clear();
        inList2.add(gen2Iter);
        inList2.add(const2Iter);
        input1 = new UnionIterator<>(inList1);
        input2 = new UnionIterator<>(inList2);
        final TupleMatchRemovingJoin matcher = new TupleMatchRemovingJoin(expectedMatchesMap);
        final Collector<Tuple2<Integer, String>> collector = new DiscardingOutputCollector<>();
        NonReusingBuildSecondHashJoinIterator<Tuple2<Integer, String>, Tuple2<Integer, String>, Tuple2<Integer, String>> iterator = new NonReusingBuildSecondHashJoinIterator<>(input1, input2, this.recordSerializer, this.record1Comparator, this.recordSerializer, this.record2Comparator, this.recordPairComparator, this.memoryManager, ioManager, this.parentTask, 1.0, false, false, true);
        iterator.open();
        while (iterator.callWithNextKey(matcher, collector)) ;
        iterator.close();
        // assert that each expected match was seen
        for (Entry<Integer, Collection<TupleMatch>> entry : expectedMatchesMap.entrySet()) {
            if (!entry.getValue().isEmpty()) {
                Assert.fail("Collection for key " + entry.getKey() + " is not empty");
            }
        }
    } catch (Exception e) {
        e.printStackTrace();
        Assert.fail("An exception occurred during the test: " + e.getMessage());
    }
}

Example 3 with UnionIterator

use of org.apache.flink.runtime.operators.testutils.UnionIterator in project flink by apache.

the class ReusingHashJoinIteratorITCase method testBuildFirstWithHighNumberOfCommonKeys.

@Test
public void testBuildFirstWithHighNumberOfCommonKeys() {
    // the size of the left and right inputs
    final int INPUT_1_SIZE = 200;
    final int INPUT_2_SIZE = 100;
    final int INPUT_1_DUPLICATES = 10;
    final int INPUT_2_DUPLICATES = 2000;
    final int DUPLICATE_KEY = 13;
    try {
        TestData.TupleGenerator generator1 = new TestData.TupleGenerator(SEED1, 500, 4096, KeyMode.RANDOM, ValueMode.RANDOM_LENGTH);
        TestData.TupleGenerator generator2 = new TestData.TupleGenerator(SEED2, 500, 2048, KeyMode.RANDOM, ValueMode.RANDOM_LENGTH);
        final TestData.TupleGeneratorIterator gen1Iter = new TestData.TupleGeneratorIterator(generator1, INPUT_1_SIZE);
        final TestData.TupleGeneratorIterator gen2Iter = new TestData.TupleGeneratorIterator(generator2, INPUT_2_SIZE);
        final TestData.TupleConstantValueIterator const1Iter = new TestData.TupleConstantValueIterator(DUPLICATE_KEY, "LEFT String for Duplicate Keys", INPUT_1_DUPLICATES);
        final TestData.TupleConstantValueIterator const2Iter = new TestData.TupleConstantValueIterator(DUPLICATE_KEY, "RIGHT String for Duplicate Keys", INPUT_2_DUPLICATES);
        final List<MutableObjectIterator<Tuple2<Integer, String>>> inList1 = new ArrayList<>();
        inList1.add(gen1Iter);
        inList1.add(const1Iter);
        final List<MutableObjectIterator<Tuple2<Integer, String>>> inList2 = new ArrayList<>();
        inList2.add(gen2Iter);
        inList2.add(const2Iter);
        MutableObjectIterator<Tuple2<Integer, String>> input1 = new UnionIterator<>(inList1);
        MutableObjectIterator<Tuple2<Integer, String>> input2 = new UnionIterator<>(inList2);
        // collect expected data
        final Map<Integer, Collection<TupleMatch>> expectedMatchesMap = joinTuples(collectTupleData(input1), collectTupleData(input2));
        // re-create the whole thing for actual processing
        // reset the generators and iterators
        generator1.reset();
        generator2.reset();
        const1Iter.reset();
        const2Iter.reset();
        gen1Iter.reset();
        gen2Iter.reset();
        inList1.clear();
        inList1.add(gen1Iter);
        inList1.add(const1Iter);
        inList2.clear();
        inList2.add(gen2Iter);
        inList2.add(const2Iter);
        input1 = new UnionIterator<>(inList1);
        input2 = new UnionIterator<>(inList2);
        final FlatJoinFunction matcher = new TupleMatchRemovingJoin(expectedMatchesMap);
        final Collector<Tuple2<Integer, String>> collector = new DiscardingOutputCollector<>();
        ReusingBuildFirstHashJoinIterator<Tuple2<Integer, String>, Tuple2<Integer, String>, Tuple2<Integer, String>> iterator = new ReusingBuildFirstHashJoinIterator<>(input1, input2, this.recordSerializer, this.record1Comparator, this.recordSerializer, this.record2Comparator, this.recordPairComparator, this.memoryManager, ioManager, this.parentTask, 1.0, false, false, true);
        iterator.open();
        while (iterator.callWithNextKey(matcher, collector)) ;
        iterator.close();
        // assert that each expected match was seen
        for (Entry<Integer, Collection<TupleMatch>> entry : expectedMatchesMap.entrySet()) {
            if (!entry.getValue().isEmpty()) {
                Assert.fail("Collection for key " + entry.getKey() + " is not empty");
            }
        }
    } catch (Exception e) {
        e.printStackTrace();
        Assert.fail("An exception occurred during the test: " + e.getMessage());
    }
}

Example 4 with UnionIterator

use of org.apache.flink.runtime.operators.testutils.UnionIterator in project flink by apache.

the class HashTableITCase method testFailingHashJoinTooManyRecursionsIntPair.

/*
	 * This test is basically identical to the "testSpillingHashJoinWithMassiveCollisions" test, only that the number
	 * of repeated values (causing bucket collisions) are large enough to make sure that their target partition no longer
	 * fits into memory by itself and needs to be repartitioned in the recursion again.
	 */
@Test
public void testFailingHashJoinTooManyRecursionsIntPair() throws IOException {
    // the following two values are known to have a hash-code collision on the first recursion level.
    // we use them to make sure one partition grows over-proportionally large
    final int REPEATED_VALUE_1 = 40559;
    final int REPEATED_VALUE_2 = 92882;
    final int REPEATED_VALUE_COUNT = 3000000;
    final int NUM_KEYS = 1000000;
    final int BUILD_VALS_PER_KEY = 3;
    final int PROBE_VALS_PER_KEY = 10;
    // create a build input that gives 3 million pairs with 3 values sharing the same key, plus 400k pairs with two colliding keys
    MutableObjectIterator<IntPair> build1 = new UniformIntPairGenerator(NUM_KEYS, BUILD_VALS_PER_KEY, false);
    MutableObjectIterator<IntPair> build2 = new ConstantsIntPairsIterator(REPEATED_VALUE_1, 17, REPEATED_VALUE_COUNT);
    MutableObjectIterator<IntPair> build3 = new ConstantsIntPairsIterator(REPEATED_VALUE_2, 23, REPEATED_VALUE_COUNT);
    List<MutableObjectIterator<IntPair>> builds = new ArrayList<MutableObjectIterator<IntPair>>();
    builds.add(build1);
    builds.add(build2);
    builds.add(build3);
    MutableObjectIterator<IntPair> buildInput = new UnionIterator<IntPair>(builds);
    // create a probe input that gives 10 million pairs with 10 values sharing a key
    MutableObjectIterator<IntPair> probe1 = new UniformIntPairGenerator(NUM_KEYS, PROBE_VALS_PER_KEY, true);
    MutableObjectIterator<IntPair> probe2 = new ConstantsIntPairsIterator(REPEATED_VALUE_1, 17, REPEATED_VALUE_COUNT);
    MutableObjectIterator<IntPair> probe3 = new ConstantsIntPairsIterator(REPEATED_VALUE_2, 23, REPEATED_VALUE_COUNT);
    List<MutableObjectIterator<IntPair>> probes = new ArrayList<MutableObjectIterator<IntPair>>();
    probes.add(probe1);
    probes.add(probe2);
    probes.add(probe3);
    MutableObjectIterator<IntPair> probeInput = new UnionIterator<IntPair>(probes);
    // allocate the memory for the HashTable
    List<MemorySegment> memSegments;
    try {
        memSegments = this.memManager.allocatePages(MEM_OWNER, 896);
    } catch (MemoryAllocationException maex) {
        fail("Memory for the Join could not be provided.");
        return;
    }
    // ----------------------------------------------------------------------------------------
    final MutableHashTable<IntPair, IntPair> join = new MutableHashTable<IntPair, IntPair>(this.pairBuildSideAccesssor, this.pairProbeSideAccesssor, this.pairBuildSideComparator, this.pairProbeSideComparator, this.pairComparator, memSegments, ioManager);
    join.open(buildInput, probeInput);
    final IntPair recordReuse = new IntPair();
    try {
        while (join.nextRecord()) {
            MutableObjectIterator<IntPair> buildSide = join.getBuildSideIterator();
            if (buildSide.next(recordReuse) == null) {
                fail("No build side values found for a probe key.");
            }
            while (buildSide.next(recordReuse) != null) ;
        }
        fail("Hash Join must have failed due to too many recursions.");
    } catch (Exception ex) {
    // expected
    }
    join.close();
    // ----------------------------------------------------------------------------------------
    this.memManager.release(join.getFreedMemory());
}

Example 5 with UnionIterator

use of org.apache.flink.runtime.operators.testutils.UnionIterator in project flink by apache.

the class HashTableITCase method testSpillingHashJoinWithMassiveCollisionsIntPair.

@Test
public void testSpillingHashJoinWithMassiveCollisionsIntPair() throws IOException {
    // the following two values are known to have a hash-code collision on the initial level.
    // we use them to make sure one partition grows over-proportionally large
    final int REPEATED_VALUE_1 = 40559;
    final int REPEATED_VALUE_2 = 92882;
    final int REPEATED_VALUE_COUNT_BUILD = 200000;
    final int REPEATED_VALUE_COUNT_PROBE = 5;
    final int NUM_KEYS = 1000000;
    final int BUILD_VALS_PER_KEY = 3;
    final int PROBE_VALS_PER_KEY = 10;
    // create a build input that gives 3 million pairs with 3 values sharing the same key, plus 400k pairs with two colliding keys
    MutableObjectIterator<IntPair> build1 = new UniformIntPairGenerator(NUM_KEYS, BUILD_VALS_PER_KEY, false);
    MutableObjectIterator<IntPair> build2 = new ConstantsIntPairsIterator(REPEATED_VALUE_1, 17, REPEATED_VALUE_COUNT_BUILD);
    MutableObjectIterator<IntPair> build3 = new ConstantsIntPairsIterator(REPEATED_VALUE_2, 23, REPEATED_VALUE_COUNT_BUILD);
    List<MutableObjectIterator<IntPair>> builds = new ArrayList<MutableObjectIterator<IntPair>>();
    builds.add(build1);
    builds.add(build2);
    builds.add(build3);
    MutableObjectIterator<IntPair> buildInput = new UnionIterator<IntPair>(builds);
    // create a probe input that gives 10 million pairs with 10 values sharing a key
    MutableObjectIterator<IntPair> probe1 = new UniformIntPairGenerator(NUM_KEYS, PROBE_VALS_PER_KEY, true);
    MutableObjectIterator<IntPair> probe2 = new ConstantsIntPairsIterator(REPEATED_VALUE_1, 17, 5);
    MutableObjectIterator<IntPair> probe3 = new ConstantsIntPairsIterator(REPEATED_VALUE_2, 23, 5);
    List<MutableObjectIterator<IntPair>> probes = new ArrayList<MutableObjectIterator<IntPair>>();
    probes.add(probe1);
    probes.add(probe2);
    probes.add(probe3);
    MutableObjectIterator<IntPair> probeInput = new UnionIterator<IntPair>(probes);
    // allocate the memory for the HashTable
    List<MemorySegment> memSegments;
    try {
        memSegments = this.memManager.allocatePages(MEM_OWNER, 896);
    } catch (MemoryAllocationException maex) {
        fail("Memory for the Join could not be provided.");
        return;
    }
    // create the I/O access for spilling
    IOManager ioManager = new IOManagerAsync();
    // create the map for validating the results
    HashMap<Integer, Long> map = new HashMap<Integer, Long>(NUM_KEYS);
    // ----------------------------------------------------------------------------------------
    final MutableHashTable<IntPair, IntPair> join = new MutableHashTable<IntPair, IntPair>(this.pairBuildSideAccesssor, this.pairProbeSideAccesssor, this.pairBuildSideComparator, this.pairProbeSideComparator, this.pairComparator, memSegments, ioManager);
    join.open(buildInput, probeInput);
    IntPair record;
    final IntPair recordReuse = new IntPair();
    while (join.nextRecord()) {
        int numBuildValues = 0;
        final IntPair probeRec = join.getCurrentProbeRecord();
        int key = probeRec.getKey();
        MutableObjectIterator<IntPair> buildSide = join.getBuildSideIterator();
        if ((record = buildSide.next(recordReuse)) != null) {
            numBuildValues = 1;
            Assert.assertEquals("Probe-side key was different than build-side key.", key, record.getKey());
        } else {
            fail("No build side values found for a probe key.");
        }
        while ((record = buildSide.next(recordReuse)) != null) {
            numBuildValues++;
            Assert.assertEquals("Probe-side key was different than build-side key.", key, record.getKey());
        }
        Long contained = map.get(key);
        if (contained == null) {
            contained = Long.valueOf(numBuildValues);
        } else {
            contained = Long.valueOf(contained.longValue() + numBuildValues);
        }
        map.put(key, contained);
    }
    join.close();
    Assert.assertEquals("Wrong number of keys", NUM_KEYS, map.size());
    for (Map.Entry<Integer, Long> entry : map.entrySet()) {
        long val = entry.getValue();
        int key = entry.getKey();
        Assert.assertEquals("Wrong number of values in per-key cross product for key " + key, (key == REPEATED_VALUE_1 || key == REPEATED_VALUE_2) ? (PROBE_VALS_PER_KEY + REPEATED_VALUE_COUNT_PROBE) * (BUILD_VALS_PER_KEY + REPEATED_VALUE_COUNT_BUILD) : PROBE_VALS_PER_KEY * BUILD_VALS_PER_KEY, val);
    }
    // ----------------------------------------------------------------------------------------
    this.memManager.release(join.getFreedMemory());
}