Examples with UniformBinaryRowGenerator org.apache.flink.table.runtime.util.UniformBinaryRowGenerator used on opensource projects

Example 11 with UniformBinaryRowGenerator

use of org.apache.flink.table.runtime.util.UniformBinaryRowGenerator in project flink by apache.

the class LongHashTableTest method testSpillingHashJoinOneRecursionValidity.

@Test
public void testSpillingHashJoinOneRecursionValidity() throws IOException {
    final int numKeys = 1000000;
    final int buildValsPerKey = 3;
    final int probeValsPerKey = 10;
    // create a build input that gives 3 million pairs with 3 values sharing the same key
    MutableObjectIterator<BinaryRowData> buildInput = new UniformBinaryRowGenerator(numKeys, buildValsPerKey, false);
    // create a probe input that gives 10 million pairs with 10 values sharing a key
    MutableObjectIterator<BinaryRowData> probeInput = new UniformBinaryRowGenerator(numKeys, probeValsPerKey, true);
    // create the map for validating the results
    HashMap<Integer, Long> map = new HashMap<>(numKeys);
    // ----------------------------------------------------------------------------------------
    final MyHashTable table = new MyHashTable(100 * PAGE_SIZE);
    BinaryRowData buildRow = buildSideSerializer.createInstance();
    while ((buildRow = buildInput.next(buildRow)) != null) {
        table.putBuildRow(buildRow);
    }
    table.endBuild();
    BinaryRowData probeRow = probeSideSerializer.createInstance();
    while ((probeRow = probeInput.next(probeRow)) != null) {
        if (table.tryProbe(probeRow)) {
            testJoin(table, map);
        }
    }
    while (table.nextMatching()) {
        testJoin(table, map);
    }
    table.close();
    Assert.assertEquals("Wrong number of keys", numKeys, 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, probeValsPerKey * buildValsPerKey, val);
    }
    // ----------------------------------------------------------------------------------------
    table.free();
}

Example 12 with UniformBinaryRowGenerator

use of org.apache.flink.table.runtime.util.UniformBinaryRowGenerator in project flink by apache.

the class LongHashTableTest method testSparseProbeSpilling.

@Test
public void testSparseProbeSpilling() throws IOException, MemoryAllocationException {
    final int numBuildKeys = 1000000;
    final int numBuildVals = 1;
    final int numProbeKeys = 20;
    final int numProbeVals = 1;
    MutableObjectIterator<BinaryRowData> buildInput = new UniformBinaryRowGenerator(numBuildKeys, numBuildVals, false);
    final MyHashTable table = new MyHashTable(100 * PAGE_SIZE);
    int expectedNumResults = (Math.min(numProbeKeys, numBuildKeys) * numBuildVals) * numProbeVals;
    int numRecordsInJoinResult = join(table, buildInput, new UniformBinaryRowGenerator(numProbeKeys, numProbeVals, true));
    Assert.assertEquals("Wrong number of records in join result.", expectedNumResults, numRecordsInJoinResult);
    table.close();
    table.free();
}

Example 13 with UniformBinaryRowGenerator

use of org.apache.flink.table.runtime.util.UniformBinaryRowGenerator in project flink by apache.

the class LongHashTableTest method testSpillingHashJoinWithTwoRecursions.

@Test
public void testSpillingHashJoinWithTwoRecursions() 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 repeatedValue1 = 40559;
    final int repeatedValue2 = 92882;
    final int repeatedValueCountBuild = 200000;
    final int repeatedValueCountProbe = 5;
    final int numKeys = 1000000;
    final int buildValsPerKey = 3;
    final int probeValsPerKey = 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<BinaryRowData> build1 = new UniformBinaryRowGenerator(numKeys, buildValsPerKey, false);
    MutableObjectIterator<BinaryRowData> build2 = new BinaryHashTableTest.ConstantsKeyValuePairsIterator(repeatedValue1, 17, repeatedValueCountBuild);
    MutableObjectIterator<BinaryRowData> build3 = new BinaryHashTableTest.ConstantsKeyValuePairsIterator(repeatedValue2, 23, repeatedValueCountBuild);
    List<MutableObjectIterator<BinaryRowData>> builds = new ArrayList<>();
    builds.add(build1);
    builds.add(build2);
    builds.add(build3);
    MutableObjectIterator<BinaryRowData> buildInput = new UnionIterator<>(builds);
    // create a probe input that gives 10 million pairs with 10 values sharing a key
    MutableObjectIterator<BinaryRowData> probe1 = new UniformBinaryRowGenerator(numKeys, probeValsPerKey, true);
    MutableObjectIterator<BinaryRowData> probe2 = new BinaryHashTableTest.ConstantsKeyValuePairsIterator(repeatedValue1, 17, 5);
    MutableObjectIterator<BinaryRowData> probe3 = new BinaryHashTableTest.ConstantsKeyValuePairsIterator(repeatedValue2, 23, 5);
    List<MutableObjectIterator<BinaryRowData>> probes = new ArrayList<>();
    probes.add(probe1);
    probes.add(probe2);
    probes.add(probe3);
    MutableObjectIterator<BinaryRowData> probeInput = new UnionIterator<>(probes);
    // create the map for validating the results
    HashMap<Integer, Long> map = new HashMap<>(numKeys);
    final MyHashTable table = new MyHashTable(896 * PAGE_SIZE);
    BinaryRowData buildRow = buildSideSerializer.createInstance();
    while ((buildRow = buildInput.next(buildRow)) != null) {
        table.putBuildRow(buildRow);
    }
    table.endBuild();
    BinaryRowData probeRow = probeSideSerializer.createInstance();
    while ((probeRow = probeInput.next(probeRow)) != null) {
        if (table.tryProbe(probeRow)) {
            testJoin(table, map);
        }
    }
    while (table.nextMatching()) {
        testJoin(table, map);
    }
    table.close();
    Assert.assertEquals("Wrong number of keys", numKeys, 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 == repeatedValue1 || key == repeatedValue2) ? (probeValsPerKey + repeatedValueCountProbe) * (buildValsPerKey + repeatedValueCountBuild) : probeValsPerKey * buildValsPerKey, val);
    }
    // ----------------------------------------------------------------------------------------
    table.free();
}

Example 14 with UniformBinaryRowGenerator

use of org.apache.flink.table.runtime.util.UniformBinaryRowGenerator in project flink by apache.

the class BinaryHashTableTest method testSpillingHashJoinWithTwoRecursions.

/*
     * 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 testSpillingHashJoinWithTwoRecursions() 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 repeatedValue1 = 40559;
    final int repeatedValue2 = 92882;
    final int repeatedValueCountBuild = 200000;
    final int repeatedValueCountProbe = 5;
    final int numKeys = 1000000;
    final int buildValsPerKey = 3;
    final int probeValsPerKey = 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<BinaryRowData> build1 = new UniformBinaryRowGenerator(numKeys, buildValsPerKey, false);
    MutableObjectIterator<BinaryRowData> build2 = new ConstantsKeyValuePairsIterator(repeatedValue1, 17, repeatedValueCountBuild);
    MutableObjectIterator<BinaryRowData> build3 = new ConstantsKeyValuePairsIterator(repeatedValue2, 23, repeatedValueCountBuild);
    List<MutableObjectIterator<BinaryRowData>> builds = new ArrayList<>();
    builds.add(build1);
    builds.add(build2);
    builds.add(build3);
    MutableObjectIterator<BinaryRowData> buildInput = new UnionIterator<>(builds);
    // create a probe input that gives 10 million pairs with 10 values sharing a key
    MutableObjectIterator<BinaryRowData> probe1 = new UniformBinaryRowGenerator(numKeys, probeValsPerKey, true);
    MutableObjectIterator<BinaryRowData> probe2 = new ConstantsKeyValuePairsIterator(repeatedValue1, 17, 5);
    MutableObjectIterator<BinaryRowData> probe3 = new ConstantsKeyValuePairsIterator(repeatedValue2, 23, 5);
    List<MutableObjectIterator<BinaryRowData>> probes = new ArrayList<>();
    probes.add(probe1);
    probes.add(probe2);
    probes.add(probe3);
    MutableObjectIterator<BinaryRowData> probeInput = new UnionIterator<>(probes);
    // create the map for validating the results
    HashMap<Integer, Long> map = new HashMap<>(numKeys);
    // ----------------------------------------------------------------------------------------
    MemoryManager memManager = MemoryManagerBuilder.newBuilder().setMemorySize(896 * PAGE_SIZE).build();
    final BinaryHashTable table = newBinaryHashTable(this.buildSideSerializer, this.probeSideSerializer, new MyProjection(), new MyProjection(), memManager, 896 * PAGE_SIZE, ioManager);
    final BinaryRowData recordReuse = new BinaryRowData(2);
    BinaryRowData buildRow = buildSideSerializer.createInstance();
    while ((buildRow = buildInput.next(buildRow)) != null) {
        table.putBuildRow(buildRow);
    }
    table.endBuild();
    BinaryRowData probeRow = probeSideSerializer.createInstance();
    while ((probeRow = probeInput.next(probeRow)) != null) {
        if (table.tryProbe(probeRow)) {
            testJoin(table, map);
        }
    }
    while (table.nextMatching()) {
        testJoin(table, map);
    }
    table.close();
    Assert.assertEquals("Wrong number of keys", numKeys, 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 == repeatedValue1 || key == repeatedValue2) ? (probeValsPerKey + repeatedValueCountProbe) * (buildValsPerKey + repeatedValueCountBuild) : probeValsPerKey * buildValsPerKey, val);
    }
    // ----------------------------------------------------------------------------------------
    table.free();
}

Example 15 with UniformBinaryRowGenerator

use of org.apache.flink.table.runtime.util.UniformBinaryRowGenerator in project flink by apache.

the class BinaryHashTableTest method testSparseProbeSpillingWithOuterJoin.

/*
     * Same test as {@link #testSparseProbeSpilling} but using a build-side outer join
     * that requires spilled build-side records to be returned and counted.
     */
@Test
public void testSparseProbeSpillingWithOuterJoin() throws IOException {
    final int numBuildKeys = 1000000;
    final int numBuildVals = 1;
    final int numProbeKeys = 20;
    final int numProbeVals = 1;
    MutableObjectIterator<BinaryRowData> buildInput = new UniformBinaryRowGenerator(numBuildKeys, numBuildVals, false);
    MemoryManager memManager = MemoryManagerBuilder.newBuilder().setMemorySize(96 * PAGE_SIZE).build();
    final BinaryHashTable table = new BinaryHashTable(conf, new Object(), this.buildSideSerializer, this.probeSideSerializer, new MyProjection(), new MyProjection(), memManager, 96 * PAGE_SIZE, ioManager, 24, 200000, true, HashJoinType.BUILD_OUTER, null, true, new boolean[] { true }, false);
    int expectedNumResults = (Math.max(numProbeKeys, numBuildKeys) * numBuildVals) * numProbeVals;
    int numRecordsInJoinResult = join(table, buildInput, new UniformBinaryRowGenerator(numProbeKeys, numProbeVals, true), true);
    Assert.assertEquals("Wrong number of records in join result.", expectedNumResults, numRecordsInJoinResult);
    table.close();
    table.free();
}