use of org.apache.flink.util.MutableObjectIterator in project flink by apache.
the class ExternalSortLargeRecordsITCase method testSortWithShortMediumAndLargeRecords.
@Test
public void testSortWithShortMediumAndLargeRecords() {
try {
final int NUM_RECORDS = 50000;
final int LARGE_REC_INTERVAL = 10000;
final int MEDIUM_REC_INTERVAL = 500;
final TypeInformation<?>[] types = new TypeInformation<?>[] { BasicTypeInfo.LONG_TYPE_INFO, new ValueTypeInfo<SmallOrMediumOrLargeValue>(SmallOrMediumOrLargeValue.class) };
final TupleTypeInfo<Tuple2<Long, SmallOrMediumOrLargeValue>> typeInfo = new TupleTypeInfo<Tuple2<Long, SmallOrMediumOrLargeValue>>(types);
final TypeSerializer<Tuple2<Long, SmallOrMediumOrLargeValue>> serializer = typeInfo.createSerializer(new ExecutionConfig());
final TypeComparator<Tuple2<Long, SmallOrMediumOrLargeValue>> comparator = typeInfo.createComparator(new int[] { 0 }, new boolean[] { false }, 0, new ExecutionConfig());
MutableObjectIterator<Tuple2<Long, SmallOrMediumOrLargeValue>> source = new MutableObjectIterator<Tuple2<Long, SmallOrMediumOrLargeValue>>() {
private final Random rnd = new Random(1456108743687167086L);
private int num = -1;
@Override
public Tuple2<Long, SmallOrMediumOrLargeValue> next(Tuple2<Long, SmallOrMediumOrLargeValue> reuse) {
return next();
}
@Override
public Tuple2<Long, SmallOrMediumOrLargeValue> next() {
if (++num < NUM_RECORDS) {
int size;
if (num % LARGE_REC_INTERVAL == 0) {
size = SmallOrMediumOrLargeValue.LARGE_SIZE;
} else if (num % MEDIUM_REC_INTERVAL == 0) {
size = SmallOrMediumOrLargeValue.MEDIUM_SIZE;
} else {
size = SmallOrMediumOrLargeValue.SMALL_SIZE;
}
long val = rnd.nextLong();
return new Tuple2<Long, SmallOrMediumOrLargeValue>(val, new SmallOrMediumOrLargeValue((int) val, size));
} else {
return null;
}
}
};
@SuppressWarnings("unchecked") Sorter<Tuple2<Long, SmallOrMediumOrLargeValue>> sorter = new UnilateralSortMerger<Tuple2<Long, SmallOrMediumOrLargeValue>>(this.memoryManager, this.ioManager, source, this.parentTask, new RuntimeSerializerFactory<Tuple2<Long, SmallOrMediumOrLargeValue>>(serializer, (Class<Tuple2<Long, SmallOrMediumOrLargeValue>>) (Class<?>) Tuple2.class), comparator, 1.0, 1, 128, 0.7f, true, /*use large record handler*/
false);
// check order
MutableObjectIterator<Tuple2<Long, SmallOrMediumOrLargeValue>> iterator = sorter.getIterator();
Tuple2<Long, SmallOrMediumOrLargeValue> val = serializer.createInstance();
long prevKey = Long.MAX_VALUE;
for (int i = 0; i < NUM_RECORDS; i++) {
val = iterator.next(val);
assertTrue(val.f0 <= prevKey);
assertTrue(val.f0.intValue() == val.f1.val());
}
assertNull(iterator.next(val));
sorter.close();
testSuccess = true;
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
}
}
use of org.apache.flink.util.MutableObjectIterator in project flink by apache.
the class ExternalSortLargeRecordsITCase method testSortWithMediumRecordsOnly.
@Test
public void testSortWithMediumRecordsOnly() {
try {
final int NUM_RECORDS = 70;
final TypeInformation<?>[] types = new TypeInformation<?>[] { BasicTypeInfo.LONG_TYPE_INFO, new ValueTypeInfo<SmallOrMediumOrLargeValue>(SmallOrMediumOrLargeValue.class) };
final TupleTypeInfo<Tuple2<Long, SmallOrMediumOrLargeValue>> typeInfo = new TupleTypeInfo<Tuple2<Long, SmallOrMediumOrLargeValue>>(types);
final TypeSerializer<Tuple2<Long, SmallOrMediumOrLargeValue>> serializer = typeInfo.createSerializer(new ExecutionConfig());
final TypeComparator<Tuple2<Long, SmallOrMediumOrLargeValue>> comparator = typeInfo.createComparator(new int[] { 0 }, new boolean[] { false }, 0, new ExecutionConfig());
MutableObjectIterator<Tuple2<Long, SmallOrMediumOrLargeValue>> source = new MutableObjectIterator<Tuple2<Long, SmallOrMediumOrLargeValue>>() {
private final Random rnd = new Random(62360187263087678L);
private int num = -1;
@Override
public Tuple2<Long, SmallOrMediumOrLargeValue> next(Tuple2<Long, SmallOrMediumOrLargeValue> reuse) {
return next();
}
@Override
public Tuple2<Long, SmallOrMediumOrLargeValue> next() {
if (++num < NUM_RECORDS) {
long val = rnd.nextLong();
return new Tuple2<Long, SmallOrMediumOrLargeValue>(val, new SmallOrMediumOrLargeValue((int) val, SmallOrMediumOrLargeValue.MEDIUM_SIZE));
} else {
return null;
}
}
};
@SuppressWarnings("unchecked") Sorter<Tuple2<Long, SmallOrMediumOrLargeValue>> sorter = new UnilateralSortMerger<Tuple2<Long, SmallOrMediumOrLargeValue>>(this.memoryManager, this.ioManager, source, this.parentTask, new RuntimeSerializerFactory<Tuple2<Long, SmallOrMediumOrLargeValue>>(serializer, (Class<Tuple2<Long, SmallOrMediumOrLargeValue>>) (Class<?>) Tuple2.class), comparator, 1.0, 1, 128, 0.7f, true, /*use large record handler*/
true);
// check order
MutableObjectIterator<Tuple2<Long, SmallOrMediumOrLargeValue>> iterator = sorter.getIterator();
Tuple2<Long, SmallOrMediumOrLargeValue> val = serializer.createInstance();
long prevKey = Long.MAX_VALUE;
for (int i = 0; i < NUM_RECORDS; i++) {
val = iterator.next(val);
assertTrue(val.f0 <= prevKey);
assertTrue(val.f0.intValue() == val.f1.val());
}
assertNull(iterator.next(val));
sorter.close();
testSuccess = true;
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
}
}
use of org.apache.flink.util.MutableObjectIterator in project flink by apache.
the class ExternalSortLargeRecordsITCase method testSortWithLongRecordsOnly.
// --------------------------------------------------------------------------------------------
@Test
public void testSortWithLongRecordsOnly() {
try {
final int NUM_RECORDS = 10;
final TypeInformation<?>[] types = new TypeInformation<?>[] { BasicTypeInfo.LONG_TYPE_INFO, new ValueTypeInfo<SomeMaybeLongValue>(SomeMaybeLongValue.class) };
final TupleTypeInfo<Tuple2<Long, SomeMaybeLongValue>> typeInfo = new TupleTypeInfo<Tuple2<Long, SomeMaybeLongValue>>(types);
final TypeSerializer<Tuple2<Long, SomeMaybeLongValue>> serializer = typeInfo.createSerializer(new ExecutionConfig());
final TypeComparator<Tuple2<Long, SomeMaybeLongValue>> comparator = typeInfo.createComparator(new int[] { 0 }, new boolean[] { false }, 0, new ExecutionConfig());
MutableObjectIterator<Tuple2<Long, SomeMaybeLongValue>> source = new MutableObjectIterator<Tuple2<Long, SomeMaybeLongValue>>() {
private final Random rnd = new Random(457821643089756298L);
private int num = 0;
@Override
public Tuple2<Long, SomeMaybeLongValue> next(Tuple2<Long, SomeMaybeLongValue> reuse) {
return next();
}
@Override
public Tuple2<Long, SomeMaybeLongValue> next() {
if (num++ < NUM_RECORDS) {
long val = rnd.nextLong();
return new Tuple2<Long, SomeMaybeLongValue>(val, new SomeMaybeLongValue((int) val));
} else {
return null;
}
}
};
@SuppressWarnings("unchecked") Sorter<Tuple2<Long, SomeMaybeLongValue>> sorter = new UnilateralSortMerger<Tuple2<Long, SomeMaybeLongValue>>(this.memoryManager, this.ioManager, source, this.parentTask, new RuntimeSerializerFactory<Tuple2<Long, SomeMaybeLongValue>>(serializer, (Class<Tuple2<Long, SomeMaybeLongValue>>) (Class<?>) Tuple2.class), comparator, 1.0, 1, 128, 0.7f, true, /* use large record handler */
false);
// check order
MutableObjectIterator<Tuple2<Long, SomeMaybeLongValue>> iterator = sorter.getIterator();
Tuple2<Long, SomeMaybeLongValue> val = serializer.createInstance();
long prevKey = Long.MAX_VALUE;
for (int i = 0; i < NUM_RECORDS; i++) {
val = iterator.next(val);
assertTrue(val.f0 <= prevKey);
assertTrue(val.f0.intValue() == val.f1.val());
}
assertNull(iterator.next(val));
sorter.close();
testSuccess = true;
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
}
}
use of org.apache.flink.util.MutableObjectIterator in project flink by apache.
the class AbstractSortMergeOuterJoinIteratorITCase method testOuterJoinWithHighNumberOfCommonKeys.
@SuppressWarnings("unchecked, rawtypes")
protected void testOuterJoinWithHighNumberOfCommonKeys(OuterJoinType outerJoinType, int input1Size, int input1Duplicates, int input1ValueLength, float input1KeyDensity, int input2Size, int input2Duplicates, int input2ValueLength, float input2KeyDensity) {
TypeSerializer<Tuple2<Integer, String>> serializer1 = new TupleSerializer<>((Class<Tuple2<Integer, String>>) (Class<?>) Tuple2.class, new TypeSerializer<?>[] { IntSerializer.INSTANCE, StringSerializer.INSTANCE });
TypeSerializer<Tuple2<Integer, String>> serializer2 = new TupleSerializer<>((Class<Tuple2<Integer, String>>) (Class<?>) Tuple2.class, new TypeSerializer<?>[] { IntSerializer.INSTANCE, StringSerializer.INSTANCE });
TypeComparator<Tuple2<Integer, String>> comparator1 = new TupleComparator<>(new int[] { 0 }, new TypeComparator<?>[] { new IntComparator(true) }, new TypeSerializer<?>[] { IntSerializer.INSTANCE });
TypeComparator<Tuple2<Integer, String>> comparator2 = new TupleComparator<>(new int[] { 0 }, new TypeComparator<?>[] { new IntComparator(true) }, new TypeSerializer<?>[] { IntSerializer.INSTANCE });
TypePairComparator<Tuple2<Integer, String>, Tuple2<Integer, String>> pairComparator = new GenericPairComparator<>(comparator1, comparator2);
this.memoryManager = new MemoryManager(MEMORY_SIZE, 1);
this.ioManager = new IOManagerAsync();
final int DUPLICATE_KEY = 13;
try {
final TupleGenerator generator1 = new TupleGenerator(SEED1, 500, input1KeyDensity, input1ValueLength, KeyMode.SORTED_SPARSE, ValueMode.RANDOM_LENGTH, null);
final TupleGenerator generator2 = new TupleGenerator(SEED2, 500, input2KeyDensity, input2ValueLength, KeyMode.SORTED_SPARSE, ValueMode.RANDOM_LENGTH, null);
final TupleGeneratorIterator gen1Iter = new TupleGeneratorIterator(generator1, input1Size);
final TupleGeneratorIterator gen2Iter = new TupleGeneratorIterator(generator2, input2Size);
final TupleConstantValueIterator const1Iter = new TupleConstantValueIterator(DUPLICATE_KEY, "LEFT String for Duplicate Keys", input1Duplicates);
final TupleConstantValueIterator const2Iter = new TupleConstantValueIterator(DUPLICATE_KEY, "RIGHT String for Duplicate Keys", input2Duplicates);
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 MergeIterator<>(inList1, comparator1.duplicate());
MutableObjectIterator<Tuple2<Integer, String>> input2 = new MergeIterator<>(inList2, comparator2.duplicate());
// collect expected data
final Map<Integer, Collection<Match>> expectedMatchesMap = joinValues(collectData(input1), collectData(input2), outerJoinType);
// 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 MergeIterator<>(inList1, comparator1.duplicate());
input2 = new MergeIterator<>(inList2, comparator2.duplicate());
final FlatJoinFunction<Tuple2<Integer, String>, Tuple2<Integer, String>, Tuple2<Integer, String>> joinFunction = new MatchRemovingJoiner(expectedMatchesMap);
final Collector<Tuple2<Integer, String>> collector = new DiscardingOutputCollector<>();
// we create this sort-merge iterator with little memory for the block-nested-loops fall-back to make sure it
// needs to spill for the duplicate keys
AbstractMergeOuterJoinIterator<Tuple2<Integer, String>, Tuple2<Integer, String>, Tuple2<Integer, String>> iterator = createOuterJoinIterator(outerJoinType, input1, input2, serializer1, comparator1, serializer2, comparator2, pairComparator, this.memoryManager, this.ioManager, PAGES_FOR_BNLJN, this.parentTask);
iterator.open();
while (iterator.callWithNextKey(joinFunction, collector)) ;
iterator.close();
// assert that each expected match was seen
for (Entry<Integer, Collection<Match>> 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());
}
}
use of org.apache.flink.util.MutableObjectIterator in project flink by apache.
the class NonReusingSortMergeInnerJoinIteratorITCase method testMergeWithHighNumberOfCommonKeys.
@Test
public void testMergeWithHighNumberOfCommonKeys() {
// 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 = 4000;
final int DUPLICATE_KEY = 13;
try {
final TupleGenerator generator1 = new TupleGenerator(SEED1, 500, 4096, KeyMode.SORTED, ValueMode.RANDOM_LENGTH);
final TupleGenerator generator2 = new TupleGenerator(SEED2, 500, 2048, KeyMode.SORTED, 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<MutableObjectIterator<Tuple2<Integer, String>>>();
inList1.add(gen1Iter);
inList1.add(const1Iter);
final List<MutableObjectIterator<Tuple2<Integer, String>>> inList2 = new ArrayList<MutableObjectIterator<Tuple2<Integer, String>>>();
inList2.add(gen2Iter);
inList2.add(const2Iter);
MutableObjectIterator<Tuple2<Integer, String>> input1 = new MergeIterator<Tuple2<Integer, String>>(inList1, comparator1.duplicate());
MutableObjectIterator<Tuple2<Integer, String>> input2 = new MergeIterator<Tuple2<Integer, String>>(inList2, comparator2.duplicate());
// collect expected data
final Map<Integer, Collection<Match>> expectedMatchesMap = matchValues(collectData(input1), collectData(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 MergeIterator<Tuple2<Integer, String>>(inList1, comparator1.duplicate());
input2 = new MergeIterator<Tuple2<Integer, String>>(inList2, comparator2.duplicate());
final FlatJoinFunction<Tuple2<Integer, String>, Tuple2<Integer, String>, Tuple2<Integer, String>> joinFunction = new MatchRemovingJoiner(expectedMatchesMap);
final Collector<Tuple2<Integer, String>> collector = new DiscardingOutputCollector<Tuple2<Integer, String>>();
// we create this sort-merge iterator with little memory for the block-nested-loops fall-back to make sure it
// needs to spill for the duplicate keys
NonReusingMergeInnerJoinIterator<Tuple2<Integer, String>, Tuple2<Integer, String>, Tuple2<Integer, String>> iterator = new NonReusingMergeInnerJoinIterator<Tuple2<Integer, String>, Tuple2<Integer, String>, Tuple2<Integer, String>>(input1, input2, this.serializer1, this.comparator1, this.serializer2, this.comparator2, this.pairComparator, this.memoryManager, this.ioManager, PAGES_FOR_BNLJN, this.parentTask);
iterator.open();
while (iterator.callWithNextKey(joinFunction, collector)) ;
iterator.close();
// assert that each expected match was seen
for (Entry<Integer, Collection<Match>> 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());
}
}
Aggregations