use of org.apache.flink.runtime.io.disk.iomanager.IOManagerAsync 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.runtime.io.disk.iomanager.IOManagerAsync in project flink by apache.
the class AbstractSortMergeOuterJoinIteratorITCase method beforeTest.
@Before
public void beforeTest() {
ExecutionConfig config = new ExecutionConfig();
config.disableObjectReuse();
TupleTypeInfo<Tuple2<String, String>> typeInfo1 = TupleTypeInfo.getBasicTupleTypeInfo(String.class, String.class);
TupleTypeInfo<Tuple2<String, Integer>> typeInfo2 = TupleTypeInfo.getBasicTupleTypeInfo(String.class, Integer.class);
serializer1 = typeInfo1.createSerializer(config);
serializer2 = typeInfo2.createSerializer(config);
comparator1 = typeInfo1.createComparator(new int[] { 0 }, new boolean[] { true }, 0, config);
comparator2 = typeInfo2.createComparator(new int[] { 0 }, new boolean[] { true }, 0, config);
pairComp = new GenericPairComparator<>(comparator1, comparator2);
this.memoryManager = new MemoryManager(MEMORY_SIZE, 1);
this.ioManager = new IOManagerAsync();
}
use of org.apache.flink.runtime.io.disk.iomanager.IOManagerAsync in project flink by apache.
the class ExternalSortITCase method beforeTest.
// --------------------------------------------------------------------------------------------
@SuppressWarnings("unchecked")
@Before
public void beforeTest() {
this.memoryManager = new MemoryManager(MEMORY_SIZE, 1);
this.ioManager = new IOManagerAsync();
this.pactRecordSerializer = TestData.getIntStringTupleSerializerFactory();
this.pactRecordComparator = TestData.getIntStringTupleComparator();
}
use of org.apache.flink.runtime.io.disk.iomanager.IOManagerAsync in project flink by apache.
the class LargeRecordHandlerTest method testRecordHandlerCompositeKey.
@Test
public void testRecordHandlerCompositeKey() {
final IOManager ioMan = new IOManagerAsync();
final int PAGE_SIZE = 4 * 1024;
final int NUM_PAGES = 24;
final int NUM_RECORDS = 25000;
try {
final MemoryManager memMan = new MemoryManager(NUM_PAGES * PAGE_SIZE, 1, PAGE_SIZE, MemoryType.HEAP, true);
final AbstractInvokable owner = new DummyInvokable();
final List<MemorySegment> initialMemory = memMan.allocatePages(owner, 6);
final List<MemorySegment> sortMemory = memMan.allocatePages(owner, NUM_PAGES - 6);
final TupleTypeInfo<Tuple3<Long, String, Byte>> typeInfo = (TupleTypeInfo<Tuple3<Long, String, Byte>>) TypeInfoParser.<Tuple3<Long, String, Byte>>parse("Tuple3<Long, String, Byte>");
final TypeSerializer<Tuple3<Long, String, Byte>> serializer = typeInfo.createSerializer(new ExecutionConfig());
final TypeComparator<Tuple3<Long, String, Byte>> comparator = typeInfo.createComparator(new int[] { 2, 0 }, new boolean[] { true, true }, 0, new ExecutionConfig());
LargeRecordHandler<Tuple3<Long, String, Byte>> handler = new LargeRecordHandler<Tuple3<Long, String, Byte>>(serializer, comparator, ioMan, memMan, initialMemory, owner, 128);
assertFalse(handler.hasData());
// add the test data
Random rnd = new Random();
for (int i = 0; i < NUM_RECORDS; i++) {
long val = rnd.nextLong();
handler.addRecord(new Tuple3<Long, String, Byte>(val, String.valueOf(val), (byte) val));
assertTrue(handler.hasData());
}
MutableObjectIterator<Tuple3<Long, String, Byte>> sorted = handler.finishWriteAndSortKeys(sortMemory);
try {
handler.addRecord(new Tuple3<Long, String, Byte>(92L, "peter pepper", (byte) 1));
fail("should throw an exception");
} catch (IllegalStateException e) {
// expected
}
Tuple3<Long, String, Byte> previous = null;
Tuple3<Long, String, Byte> next;
while ((next = sorted.next(null)) != null) {
// key and value must be equal
assertTrue(next.f0.equals(Long.parseLong(next.f1)));
assertTrue(next.f0.byteValue() == next.f2);
// order must be correct
if (previous != null) {
assertTrue(previous.f2 <= next.f2);
assertTrue(previous.f2.byteValue() != next.f2.byteValue() || previous.f0 <= next.f0);
}
previous = next;
}
handler.close();
assertFalse(handler.hasData());
handler.close();
try {
handler.addRecord(new Tuple3<Long, String, Byte>(92L, "peter pepper", (byte) 1));
fail("should throw an exception");
} catch (IllegalStateException e) {
// expected
}
assertTrue(memMan.verifyEmpty());
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
} finally {
ioMan.shutdown();
}
}
use of org.apache.flink.runtime.io.disk.iomanager.IOManagerAsync in project flink by apache.
the class NonReusingSortMergeInnerJoinIteratorITCase method beforeTest.
@SuppressWarnings("unchecked")
@Before
public void beforeTest() {
serializer1 = new TupleSerializer<Tuple2<Integer, String>>((Class<Tuple2<Integer, String>>) (Class<?>) Tuple2.class, new TypeSerializer<?>[] { IntSerializer.INSTANCE, StringSerializer.INSTANCE });
serializer2 = new TupleSerializer<Tuple2<Integer, String>>((Class<Tuple2<Integer, String>>) (Class<?>) Tuple2.class, new TypeSerializer<?>[] { IntSerializer.INSTANCE, StringSerializer.INSTANCE });
comparator1 = new TupleComparator<Tuple2<Integer, String>>(new int[] { 0 }, new TypeComparator<?>[] { new IntComparator(true) }, new TypeSerializer<?>[] { IntSerializer.INSTANCE });
comparator2 = new TupleComparator<Tuple2<Integer, String>>(new int[] { 0 }, new TypeComparator<?>[] { new IntComparator(true) }, new TypeSerializer<?>[] { IntSerializer.INSTANCE });
pairComparator = new GenericPairComparator<Tuple2<Integer, String>, Tuple2<Integer, String>>(comparator1, comparator2);
this.memoryManager = new MemoryManager(MEMORY_SIZE, 1);
this.ioManager = new IOManagerAsync();
}
Aggregations