use of org.apache.flink.runtime.operators.testutils.types.IntPair in project flink by apache.
the class HashTableITCase method testSpillingHashJoinOneRecursionValidityIntPair.
@Test
public void testSpillingHashJoinOneRecursionValidityIntPair() throws IOException {
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
MutableObjectIterator<IntPair> buildInput = new UniformIntPairGenerator(NUM_KEYS, BUILD_VALS_PER_KEY, false);
// create a probe input that gives 10 million pairs with 10 values sharing a key
MutableObjectIterator<IntPair> probeInput = new UniformIntPairGenerator(NUM_KEYS, PROBE_VALS_PER_KEY, true);
// 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;
int key = 0;
MutableObjectIterator<IntPair> buildSide = join.getBuildSideIterator();
if ((record = buildSide.next(recordReuse)) != null) {
numBuildValues = 1;
key = record.getKey();
} else {
fail("No build side values found for a probe key.");
}
while ((record = buildSide.next(recordReuse)) != null) {
numBuildValues++;
}
if (numBuildValues != 3) {
fail("Other than 3 build values!!!");
}
IntPair pr = join.getCurrentProbeRecord();
Assert.assertEquals("Probe-side key was different than build-side key.", key, pr.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, PROBE_VALS_PER_KEY * BUILD_VALS_PER_KEY, val);
}
// ----------------------------------------------------------------------------------------
this.memManager.release(join.getFreedMemory());
}
use of org.apache.flink.runtime.operators.testutils.types.IntPair in project flink by apache.
the class ReusingHashJoinIteratorITCase method testBuildSecondWithMixedDataTypes.
@Test
public void testBuildSecondWithMixedDataTypes() {
try {
MutableObjectIterator<IntPair> input1 = new UniformIntPairGenerator(500, 40, false);
final TestData.TupleGenerator generator2 = new TestData.TupleGenerator(SEED2, 500, 2048, KeyMode.RANDOM, ValueMode.RANDOM_LENGTH);
final TestData.TupleGeneratorIterator input2 = new TestData.TupleGeneratorIterator(generator2, INPUT_2_SIZE);
// collect expected data
final Map<Integer, Collection<TupleIntPairMatch>> expectedMatchesMap = joinIntPairs(collectIntPairData(input1), collectTupleData(input2));
final FlatJoinFunction<IntPair, Tuple2<Integer, String>, Tuple2<Integer, String>> matcher = new TupleIntPairMatchRemovingMatcher(expectedMatchesMap);
final Collector<Tuple2<Integer, String>> collector = new DiscardingOutputCollector<>();
// reset the generators
input1 = new UniformIntPairGenerator(500, 40, false);
generator2.reset();
input2.reset();
// compare with iterator values
ReusingBuildFirstHashJoinIterator<IntPair, Tuple2<Integer, String>, Tuple2<Integer, String>> iterator = new ReusingBuildFirstHashJoinIterator<>(input1, input2, this.pairSerializer, this.pairComparator, this.recordSerializer, this.record2Comparator, this.recordPairPairComparator, this.memoryManager, this.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<TupleIntPairMatch>> 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.operators.testutils.types.IntPair in project flink by apache.
the class NonReusingHashJoinIteratorITCase method testBuildSecondWithMixedDataTypes.
@Test
public void testBuildSecondWithMixedDataTypes() {
try {
MutableObjectIterator<IntPair> input1 = new UniformIntPairGenerator(500, 40, false);
final TupleGenerator generator2 = new TupleGenerator(SEED2, 500, 2048, KeyMode.RANDOM, ValueMode.RANDOM_LENGTH);
final TestData.TupleGeneratorIterator input2 = new TestData.TupleGeneratorIterator(generator2, INPUT_2_SIZE);
// collect expected data
final Map<Integer, Collection<TupleIntPairMatch>> expectedMatchesMap = joinIntPairs(collectIntPairData(input1), collectTupleData(input2));
final FlatJoinFunction<IntPair, Tuple2<Integer, String>, Tuple2<Integer, String>> matcher = new TupleIntPairMatchRemovingMatcher(expectedMatchesMap);
final Collector<Tuple2<Integer, String>> collector = new DiscardingOutputCollector<>();
// reset the generators
input1 = new UniformIntPairGenerator(500, 40, false);
generator2.reset();
input2.reset();
// compare with iterator values
NonReusingBuildFirstHashJoinIterator<IntPair, Tuple2<Integer, String>, Tuple2<Integer, String>> iterator = new NonReusingBuildFirstHashJoinIterator<>(input1, input2, this.pairSerializer, this.pairComparator, this.recordSerializer, this.record2Comparator, this.recordPairPairComparator, this.memoryManager, this.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<TupleIntPairMatch>> 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.operators.testutils.types.IntPair in project flink by apache.
the class FixedLengthRecordSorterTest method testFlushFullMemoryPage.
@Test
public void testFlushFullMemoryPage() throws Exception {
// Insert IntPair which would fill 2 memory pages.
final int NUM_RECORDS = 2 * MEMORY_PAGE_SIZE / 8;
final List<MemorySegment> memory = this.memoryManager.allocatePages(new DummyInvokable(), 3);
FixedLengthRecordSorter<IntPair> sorter = newSortBuffer(memory);
UniformIntPairGenerator generator = new UniformIntPairGenerator(Integer.MAX_VALUE, 1, false);
// write the records
IntPair record = new IntPair();
int num = -1;
do {
generator.next(record);
num++;
} while (sorter.write(record) && num < NUM_RECORDS);
FileIOChannel.ID channelID = this.ioManager.createChannelEnumerator().next();
BlockChannelWriter<MemorySegment> blockChannelWriter = this.ioManager.createBlockChannelWriter(channelID);
final List<MemorySegment> writeBuffer = this.memoryManager.allocatePages(new DummyInvokable(), 3);
ChannelWriterOutputView outputView = new ChannelWriterOutputView(blockChannelWriter, writeBuffer, writeBuffer.get(0).size());
sorter.writeToOutput(outputView, 0, NUM_RECORDS);
this.memoryManager.release(outputView.close());
BlockChannelReader<MemorySegment> blockChannelReader = this.ioManager.createBlockChannelReader(channelID);
final List<MemorySegment> readBuffer = this.memoryManager.allocatePages(new DummyInvokable(), 3);
ChannelReaderInputView readerInputView = new ChannelReaderInputView(blockChannelReader, readBuffer, false);
final List<MemorySegment> dataBuffer = this.memoryManager.allocatePages(new DummyInvokable(), 3);
ChannelReaderInputViewIterator<IntPair> iterator = new ChannelReaderInputViewIterator(readerInputView, dataBuffer, this.serializer);
record = iterator.next(record);
int i = 0;
while (record != null) {
Assert.assertEquals(i, record.getKey());
record = iterator.next(record);
i++;
}
Assert.assertEquals(NUM_RECORDS, i);
this.memoryManager.release(dataBuffer);
// release the memory occupied by the buffers
sorter.dispose();
this.memoryManager.release(memory);
}
use of org.apache.flink.runtime.operators.testutils.types.IntPair in project flink by apache.
the class FixedLengthRecordSorterTest method testSort.
@Test
public void testSort() throws Exception {
final int NUM_RECORDS = 559273;
final int numSegments = MEMORY_SIZE / MEMORY_PAGE_SIZE;
final List<MemorySegment> memory = this.memoryManager.allocatePages(new DummyInvokable(), numSegments);
FixedLengthRecordSorter<IntPair> sorter = newSortBuffer(memory);
RandomIntPairGenerator generator = new RandomIntPairGenerator(SEED);
// write the records
IntPair record = new IntPair();
int num = -1;
do {
generator.next(record);
num++;
} while (sorter.write(record) && num < NUM_RECORDS);
QuickSort qs = new QuickSort();
qs.sort(sorter);
MutableObjectIterator<IntPair> iter = sorter.getIterator();
IntPair readTarget = new IntPair();
int current;
int last;
iter.next(readTarget);
last = readTarget.getKey();
while ((readTarget = iter.next(readTarget)) != null) {
current = readTarget.getKey();
final int cmp = last - current;
if (cmp > 0) {
Assert.fail("Next key is not larger or equal to previous key.");
}
last = current;
}
// release the memory occupied by the buffers
sorter.dispose();
this.memoryManager.release(memory);
}
Aggregations