use of org.apache.flink.core.io.InputStatus in project flink by apache.
the class KafkaSourceReaderTest method testCommitOffsetsWithoutAliveFetchers.
// -----------------------------------------
@Test
void testCommitOffsetsWithoutAliveFetchers() throws Exception {
final String groupId = "testCommitOffsetsWithoutAliveFetchers";
try (KafkaSourceReader<Integer> reader = (KafkaSourceReader<Integer>) createReader(Boundedness.CONTINUOUS_UNBOUNDED, groupId)) {
KafkaPartitionSplit split = new KafkaPartitionSplit(new TopicPartition(TOPIC, 0), 0, NUM_RECORDS_PER_SPLIT);
reader.addSplits(Collections.singletonList(split));
reader.notifyNoMoreSplits();
ReaderOutput<Integer> output = new TestingReaderOutput<>();
InputStatus status;
do {
status = reader.pollNext(output);
} while (status != InputStatus.NOTHING_AVAILABLE);
pollUntil(reader, output, () -> reader.getNumAliveFetchers() == 0, "The split fetcher did not exit before timeout.");
reader.snapshotState(100L);
reader.notifyCheckpointComplete(100L);
// Due to a bug in KafkaConsumer, when the consumer closes, the offset commit callback
// won't be fired, so the offsetsToCommit map won't be cleaned. To make the test
// stable, we add a split whose starting offset is the log end offset, so the
// split fetcher won't become idle and exit after commitOffsetAsync is invoked from
// notifyCheckpointComplete().
reader.addSplits(Collections.singletonList(new KafkaPartitionSplit(new TopicPartition(TOPIC, 0), NUM_RECORDS_PER_SPLIT)));
pollUntil(reader, output, () -> reader.getOffsetsToCommit().isEmpty(), "The offset commit did not finish before timeout.");
}
// Verify the committed offsets.
try (AdminClient adminClient = KafkaSourceTestEnv.getAdminClient()) {
Map<TopicPartition, OffsetAndMetadata> committedOffsets = adminClient.listConsumerGroupOffsets(groupId).partitionsToOffsetAndMetadata().get();
assertThat(committedOffsets).hasSize(1);
assertThat(committedOffsets.values()).extracting(OffsetAndMetadata::offset).allMatch(offset -> offset == NUM_RECORDS_PER_SPLIT);
}
}
use of org.apache.flink.core.io.InputStatus in project flink by apache.
the class FileSourceHeavyThroughputTest method testHeavyThroughput.
// ------------------------------------------------------------------------
@Test
public void testHeavyThroughput() throws Exception {
final Path path = new Path("testfs:///testpath");
// 20 GB
final long fileSize = 20L << 30;
final FileSourceSplit split = new FileSourceSplit("testsplitId", path, 0, fileSize, 0, fileSize);
testFs = TestingFileSystem.createForFileStatus(path.toUri().getScheme(), TestingFileSystem.TestFileStatus.forFileWithStream(path, fileSize, new GeneratingInputStream(fileSize)));
testFs.register();
final FileSource<byte[]> source = FileSource.forRecordStreamFormat(new ArrayReaderFormat(), path).build();
final SourceReader<byte[], FileSourceSplit> reader = source.createReader(new NoOpReaderContext());
reader.addSplits(Collections.singletonList(split));
reader.notifyNoMoreSplits();
final ReaderOutput<byte[]> out = new NoOpReaderOutput<>();
InputStatus status;
while ((status = reader.pollNext(out)) != InputStatus.END_OF_INPUT) {
// if nothing is available currently, wait for more
if (status == InputStatus.NOTHING_AVAILABLE) {
reader.isAvailable().get();
}
}
}
use of org.apache.flink.core.io.InputStatus in project flink by apache.
the class PulsarSourceReaderTestBase method assigningEmptySplits.
@TestTemplate
void assigningEmptySplits(PulsarSourceReaderBase<Integer> reader, Boundedness boundedness, String topicName) throws Exception {
final PulsarPartitionSplit emptySplit = createPartitionSplit(topicName, 0, Boundedness.CONTINUOUS_UNBOUNDED, MessageId.latest);
reader.addSplits(Collections.singletonList(emptySplit));
TestingReaderOutput<Integer> output = new TestingReaderOutput<>();
InputStatus status = reader.pollNext(output);
assertThat(status).isEqualTo(InputStatus.NOTHING_AVAILABLE);
reader.close();
}
use of org.apache.flink.core.io.InputStatus in project flink by apache.
the class PulsarOrderedSourceReaderTest method offsetCommitOnCheckpointComplete.
@TestTemplate
void offsetCommitOnCheckpointComplete(PulsarSourceReaderBase<Integer> baseReader, Boundedness boundedness, String topicName) throws Exception {
PulsarOrderedSourceReader<Integer> reader = (PulsarOrderedSourceReader<Integer>) baseReader;
// consume more than 1 partition
reader.addSplits(createPartitionSplits(topicName, DEFAULT_PARTITIONS, Boundedness.CONTINUOUS_UNBOUNDED));
reader.notifyNoMoreSplits();
TestingReaderOutput<Integer> output = new TestingReaderOutput<>();
long checkpointId = 0;
int emptyResultTime = 0;
InputStatus status;
do {
checkpointId++;
status = reader.pollNext(output);
// Create a checkpoint for each message consumption, but not complete them.
reader.snapshotState(checkpointId);
// the first couple of pollNext() might return NOTHING_AVAILABLE before data appears
if (InputStatus.NOTHING_AVAILABLE == status) {
emptyResultTime++;
sleepUninterruptibly(1, TimeUnit.SECONDS);
}
} while (emptyResultTime < MAX_EMPTY_POLLING_TIMES && status != InputStatus.END_OF_INPUT && output.getEmittedRecords().size() < NUM_RECORDS_PER_PARTITION * DEFAULT_PARTITIONS);
// The completion of the last checkpoint should subsume all previous checkpoints.
assertThat(reader.cursorsToCommit).hasSize((int) checkpointId);
long lastCheckpointId = checkpointId;
// notify checkpoint complete and expect all cursors committed
assertThatCode(() -> reader.notifyCheckpointComplete(lastCheckpointId)).doesNotThrowAnyException();
assertThat(reader.cursorsToCommit).isEmpty();
// Verify the committed offsets.
reader.close();
for (int i = 0; i < DEFAULT_PARTITIONS; i++) {
verifyAllMessageAcknowledged(NUM_RECORDS_PER_PARTITION, TopicNameUtils.topicNameWithPartition(topicName, i));
}
}
use of org.apache.flink.core.io.InputStatus in project flink by apache.
the class HybridSourceReaderTest method testReader.
@Test
public void testReader() throws Exception {
TestingReaderContext readerContext = new TestingReaderContext();
TestingReaderOutput<Integer> readerOutput = new TestingReaderOutput<>();
MockBaseSource source = new MockBaseSource(1, 1, Boundedness.BOUNDED);
// 2 underlying readers to exercise switch
SourceReader<Integer, MockSourceSplit> mockSplitReader1 = source.createReader(readerContext);
SourceReader<Integer, MockSourceSplit> mockSplitReader2 = source.createReader(readerContext);
HybridSourceReader<Integer> reader = new HybridSourceReader<>(readerContext);
Assert.assertThat(readerContext.getSentEvents(), Matchers.emptyIterable());
reader.start();
assertAndClearSourceReaderFinishedEvent(readerContext, -1);
Assert.assertNull(currentReader(reader));
Assert.assertEquals(InputStatus.NOTHING_AVAILABLE, reader.pollNext(readerOutput));
Source source1 = new MockSource(null, 0) {
@Override
public SourceReader<Integer, MockSourceSplit> createReader(SourceReaderContext readerContext) {
return mockSplitReader1;
}
};
reader.handleSourceEvents(new SwitchSourceEvent(0, source1, false));
MockSourceSplit mockSplit = new MockSourceSplit(0, 0, 1);
mockSplit.addRecord(0);
SwitchedSources switchedSources = new SwitchedSources();
switchedSources.put(0, source);
HybridSourceSplit hybridSplit = HybridSourceSplit.wrapSplit(mockSplit, 0, switchedSources);
reader.addSplits(Collections.singletonList(hybridSplit));
// drain splits
InputStatus status = reader.pollNext(readerOutput);
while (readerOutput.getEmittedRecords().isEmpty() || status == InputStatus.MORE_AVAILABLE) {
status = reader.pollNext(readerOutput);
Thread.sleep(10);
}
Assert.assertThat(readerOutput.getEmittedRecords(), Matchers.contains(0));
reader.pollNext(readerOutput);
Assert.assertEquals("before notifyNoMoreSplits", InputStatus.NOTHING_AVAILABLE, reader.pollNext(readerOutput));
reader.notifyNoMoreSplits();
reader.pollNext(readerOutput);
assertAndClearSourceReaderFinishedEvent(readerContext, 0);
Assert.assertEquals("reader before switch source event", mockSplitReader1, currentReader(reader));
Source source2 = new MockSource(null, 0) {
@Override
public SourceReader<Integer, MockSourceSplit> createReader(SourceReaderContext readerContext) {
return mockSplitReader2;
}
};
reader.handleSourceEvents(new SwitchSourceEvent(1, source2, true));
Assert.assertEquals("reader after switch source event", mockSplitReader2, currentReader(reader));
reader.notifyNoMoreSplits();
Assert.assertEquals("reader 1 after notifyNoMoreSplits", InputStatus.END_OF_INPUT, reader.pollNext(readerOutput));
reader.close();
}
Aggregations