use of org.apache.flink.streaming.util.AbstractStreamOperatorTestHarness in project flink by apache.
the class FlinkKafkaConsumerBaseMigrationTest method testRestoreFromEmptyStateWithPartitions.
/**
* Test restoring from an empty state taken using a previous Flink version, when some partitions
* could be found for topics.
*/
@Test
public void testRestoreFromEmptyStateWithPartitions() throws Exception {
final List<KafkaTopicPartition> partitions = new ArrayList<>(PARTITION_STATE.keySet());
final DummyFlinkKafkaConsumer<String> consumerFunction = new DummyFlinkKafkaConsumer<>(TOPICS, partitions, FlinkKafkaConsumerBase.PARTITION_DISCOVERY_DISABLED);
StreamSource<String, DummyFlinkKafkaConsumer<String>> consumerOperator = new StreamSource<>(consumerFunction);
final AbstractStreamOperatorTestHarness<String> testHarness = new AbstractStreamOperatorTestHarness<>(consumerOperator, 1, 1, 0);
testHarness.setTimeCharacteristic(TimeCharacteristic.ProcessingTime);
testHarness.setup();
// restore state from binary snapshot file
testHarness.initializeState(OperatorSnapshotUtil.getResourceFilename("kafka-consumer-migration-test-flink" + testMigrateVersion + "-empty-state-snapshot"));
testHarness.open();
// the expected state in "kafka-consumer-migration-test-flink1.x-snapshot-empty-state";
// all new partitions after the snapshot are considered as partitions that were created
// while the
// consumer wasn't running, and should start from the earliest offset.
final HashMap<KafkaTopicPartition, Long> expectedSubscribedPartitionsWithStartOffsets = new HashMap<>();
for (KafkaTopicPartition partition : PARTITION_STATE.keySet()) {
expectedSubscribedPartitionsWithStartOffsets.put(partition, KafkaTopicPartitionStateSentinel.EARLIEST_OFFSET);
}
// assert that there are partitions and is identical to expected list
assertTrue(consumerFunction.getSubscribedPartitionsToStartOffsets() != null);
assertTrue(!consumerFunction.getSubscribedPartitionsToStartOffsets().isEmpty());
assertEquals(expectedSubscribedPartitionsWithStartOffsets, consumerFunction.getSubscribedPartitionsToStartOffsets());
// the new partitions should have been considered as restored state
assertTrue(consumerFunction.getRestoredState() != null);
assertTrue(!consumerFunction.getSubscribedPartitionsToStartOffsets().isEmpty());
for (Map.Entry<KafkaTopicPartition, Long> expectedEntry : expectedSubscribedPartitionsWithStartOffsets.entrySet()) {
assertEquals(expectedEntry.getValue(), consumerFunction.getRestoredState().get(expectedEntry.getKey()));
}
consumerOperator.close();
consumerOperator.cancel();
}
use of org.apache.flink.streaming.util.AbstractStreamOperatorTestHarness in project flink by apache.
the class FlinkKafkaConsumerBaseMigrationTest method writeSnapshot.
private void writeSnapshot(String path, HashMap<KafkaTopicPartition, Long> state) throws Exception {
final OneShotLatch latch = new OneShotLatch();
final AbstractFetcher<String, ?> fetcher = mock(AbstractFetcher.class);
doAnswer(new Answer<Void>() {
@Override
public Void answer(InvocationOnMock invocation) throws Throwable {
latch.trigger();
return null;
}
}).when(fetcher).runFetchLoop();
when(fetcher.snapshotCurrentState()).thenReturn(state);
final List<KafkaTopicPartition> partitions = new ArrayList<>(PARTITION_STATE.keySet());
final DummyFlinkKafkaConsumer<String> consumerFunction = new DummyFlinkKafkaConsumer<>(fetcher, TOPICS, partitions, FlinkKafkaConsumerBase.PARTITION_DISCOVERY_DISABLED);
StreamSource<String, DummyFlinkKafkaConsumer<String>> consumerOperator = new StreamSource<>(consumerFunction);
final AbstractStreamOperatorTestHarness<String> testHarness = new AbstractStreamOperatorTestHarness<>(consumerOperator, 1, 1, 0);
testHarness.setTimeCharacteristic(TimeCharacteristic.ProcessingTime);
testHarness.setup();
testHarness.open();
final Throwable[] error = new Throwable[1];
// run the source asynchronously
Thread runner = new Thread() {
@Override
public void run() {
try {
consumerFunction.run(new DummySourceContext() {
@Override
public void collect(String element) {
}
});
} catch (Throwable t) {
t.printStackTrace();
error[0] = t;
}
}
};
runner.start();
if (!latch.isTriggered()) {
latch.await();
}
final OperatorSubtaskState snapshot;
synchronized (testHarness.getCheckpointLock()) {
snapshot = testHarness.snapshot(0L, 0L);
}
OperatorSnapshotUtil.writeStateHandle(snapshot, path);
consumerOperator.close();
runner.join();
}
use of org.apache.flink.streaming.util.AbstractStreamOperatorTestHarness in project flink by apache.
the class RMQSourceTest method testCheckpointing.
@Test
public void testCheckpointing() throws Exception {
source.autoAck = false;
StreamSource<String, RMQSource<String>> src = new StreamSource<>(source);
AbstractStreamOperatorTestHarness<String> testHarness = new AbstractStreamOperatorTestHarness<>(src, 1, 1, 0);
testHarness.open();
sourceThread.start();
Thread.sleep(5);
final Random random = new Random(System.currentTimeMillis());
int numSnapshots = 50;
long previousSnapshotId;
long lastSnapshotId = 0;
long totalNumberOfAcks = 0;
for (int i = 0; i < numSnapshots; i++) {
long snapshotId = random.nextLong();
OperatorSubtaskState data;
synchronized (DummySourceContext.lock) {
data = testHarness.snapshot(snapshotId, System.currentTimeMillis());
previousSnapshotId = lastSnapshotId;
lastSnapshotId = messageId;
}
// let some time pass
Thread.sleep(5);
// check if the correct number of messages have been snapshotted
final long numIds = lastSnapshotId - previousSnapshotId;
RMQTestSource sourceCopy = new RMQTestSource();
StreamSource<String, RMQTestSource> srcCopy = new StreamSource<>(sourceCopy);
AbstractStreamOperatorTestHarness<String> testHarnessCopy = new AbstractStreamOperatorTestHarness<>(srcCopy, 1, 1, 0);
testHarnessCopy.setup();
testHarnessCopy.initializeState(data);
testHarnessCopy.open();
ArrayDeque<Tuple2<Long, Set<String>>> deque = sourceCopy.getRestoredState();
Set<String> messageIds = deque.getLast().f1;
assertEquals(numIds, messageIds.size());
if (messageIds.size() > 0) {
assertTrue(messageIds.contains(Long.toString(lastSnapshotId - 1)));
}
// check if the messages are being acknowledged and the transaction committed
synchronized (DummySourceContext.lock) {
source.notifyCheckpointComplete(snapshotId);
}
totalNumberOfAcks += numIds;
}
Mockito.verify(source.channel, Mockito.times((int) totalNumberOfAcks)).basicAck(Mockito.anyLong(), Mockito.eq(false));
Mockito.verify(source.channel, Mockito.times(numSnapshots)).txCommit();
}
use of org.apache.flink.streaming.util.AbstractStreamOperatorTestHarness in project flink by apache.
the class FlinkKinesisConsumerMigrationTest method testRestoreWithReshardedStream.
@Test
public void testRestoreWithReshardedStream() throws Exception {
final List<StreamShardHandle> initialDiscoveryShards = new ArrayList<>(TEST_STATE.size());
for (StreamShardMetadata shardMetadata : TEST_STATE.keySet()) {
// setup the closed shard
Shard closedShard = new Shard();
closedShard.setShardId(shardMetadata.getShardId());
SequenceNumberRange closedSequenceNumberRange = new SequenceNumberRange();
closedSequenceNumberRange.withStartingSequenceNumber("1");
closedSequenceNumberRange.withEndingSequenceNumber(// this represents a closed shard
"1087654321");
closedShard.setSequenceNumberRange(closedSequenceNumberRange);
initialDiscoveryShards.add(new StreamShardHandle(shardMetadata.getStreamName(), closedShard));
// setup the new shards
Shard newSplitShard1 = new Shard();
newSplitShard1.setShardId(KinesisShardIdGenerator.generateFromShardOrder(1));
SequenceNumberRange newSequenceNumberRange1 = new SequenceNumberRange();
newSequenceNumberRange1.withStartingSequenceNumber("1087654322");
newSplitShard1.setSequenceNumberRange(newSequenceNumberRange1);
newSplitShard1.setParentShardId(TEST_SHARD_ID);
Shard newSplitShard2 = new Shard();
newSplitShard2.setShardId(KinesisShardIdGenerator.generateFromShardOrder(2));
SequenceNumberRange newSequenceNumberRange2 = new SequenceNumberRange();
newSequenceNumberRange2.withStartingSequenceNumber("2087654322");
newSplitShard2.setSequenceNumberRange(newSequenceNumberRange2);
newSplitShard2.setParentShardId(TEST_SHARD_ID);
initialDiscoveryShards.add(new StreamShardHandle(shardMetadata.getStreamName(), newSplitShard1));
initialDiscoveryShards.add(new StreamShardHandle(shardMetadata.getStreamName(), newSplitShard2));
}
final TestFetcher<String> fetcher = new TestFetcher<>(Collections.singletonList(TEST_STREAM_NAME), new TestSourceContext<>(), new TestRuntimeContext(true, 1, 0), TestUtils.getStandardProperties(), new KinesisDeserializationSchemaWrapper<>(new SimpleStringSchema()), null, initialDiscoveryShards);
final DummyFlinkKinesisConsumer<String> consumerFunction = new DummyFlinkKinesisConsumer<>(fetcher, new KinesisDeserializationSchemaWrapper<>(new SimpleStringSchema()));
StreamSource<String, DummyFlinkKinesisConsumer<String>> consumerOperator = new StreamSource<>(consumerFunction);
final AbstractStreamOperatorTestHarness<String> testHarness = new AbstractStreamOperatorTestHarness<>(consumerOperator, 1, 1, 0);
testHarness.setup();
testHarness.initializeState(OperatorSnapshotUtil.getResourceFilename("kinesis-consumer-migration-test-flink" + testMigrateVersion + "-snapshot"));
testHarness.open();
consumerFunction.run(new TestSourceContext<>());
// assert that state is correctly restored
assertNotEquals(null, consumerFunction.getRestoredState());
assertEquals(1, consumerFunction.getRestoredState().size());
assertEquals(TEST_STATE, removeEquivalenceWrappers(consumerFunction.getRestoredState()));
// assert that the fetcher is registered with all shards, including new shards
assertEquals(3, fetcher.getSubscribedShardsState().size());
KinesisStreamShardState restoredClosedShardState = fetcher.getSubscribedShardsState().get(0);
assertEquals(TEST_STREAM_NAME, restoredClosedShardState.getStreamShardHandle().getStreamName());
assertEquals(TEST_SHARD_ID, restoredClosedShardState.getStreamShardHandle().getShard().getShardId());
assertTrue(restoredClosedShardState.getStreamShardHandle().isClosed());
assertEquals(TEST_SEQUENCE_NUMBER, restoredClosedShardState.getLastProcessedSequenceNum());
KinesisStreamShardState restoredNewSplitShard1 = fetcher.getSubscribedShardsState().get(1);
assertEquals(TEST_STREAM_NAME, restoredNewSplitShard1.getStreamShardHandle().getStreamName());
assertEquals(KinesisShardIdGenerator.generateFromShardOrder(1), restoredNewSplitShard1.getStreamShardHandle().getShard().getShardId());
assertFalse(restoredNewSplitShard1.getStreamShardHandle().isClosed());
// new shards should be consumed from the beginning
assertEquals(SentinelSequenceNumber.SENTINEL_EARLIEST_SEQUENCE_NUM.get(), restoredNewSplitShard1.getLastProcessedSequenceNum());
KinesisStreamShardState restoredNewSplitShard2 = fetcher.getSubscribedShardsState().get(2);
assertEquals(TEST_STREAM_NAME, restoredNewSplitShard2.getStreamShardHandle().getStreamName());
assertEquals(KinesisShardIdGenerator.generateFromShardOrder(2), restoredNewSplitShard2.getStreamShardHandle().getShard().getShardId());
assertFalse(restoredNewSplitShard2.getStreamShardHandle().isClosed());
// new shards should be consumed from the beginning
assertEquals(SentinelSequenceNumber.SENTINEL_EARLIEST_SEQUENCE_NUM.get(), restoredNewSplitShard2.getLastProcessedSequenceNum());
consumerOperator.close();
consumerOperator.cancel();
}
use of org.apache.flink.streaming.util.AbstractStreamOperatorTestHarness in project flink by apache.
the class ContinuousFileProcessingTest method testFunctionRestore.
@Test
public void testFunctionRestore() throws Exception {
String testBasePath = hdfsURI + "/" + UUID.randomUUID() + "/";
org.apache.hadoop.fs.Path path = null;
long fileModTime = Long.MIN_VALUE;
for (int i = 0; i < 1; i++) {
Tuple2<org.apache.hadoop.fs.Path, String> file = createFileAndFillWithData(testBasePath, "file", i, "This is test line.");
path = file.f0;
fileModTime = hdfs.getFileStatus(file.f0).getModificationTime();
}
TextInputFormat format = new TextInputFormat(new Path(testBasePath));
final ContinuousFileMonitoringFunction<String> monitoringFunction = createTestContinuousFileMonitoringFunction(format, FileProcessingMode.PROCESS_CONTINUOUSLY);
StreamSource<TimestampedFileInputSplit, ContinuousFileMonitoringFunction<String>> src = new StreamSource<>(monitoringFunction);
final AbstractStreamOperatorTestHarness<TimestampedFileInputSplit> testHarness = new AbstractStreamOperatorTestHarness<>(src, 1, 1, 0);
testHarness.open();
final Throwable[] error = new Throwable[1];
final OneShotLatch latch = new OneShotLatch();
final DummySourceContext sourceContext = new DummySourceContext() {
@Override
public void collect(TimestampedFileInputSplit element) {
latch.trigger();
}
};
// run the source asynchronously
Thread runner = new Thread() {
@Override
public void run() {
try {
monitoringFunction.run(sourceContext);
} catch (Throwable t) {
t.printStackTrace();
error[0] = t;
}
}
};
runner.start();
// first condition for the source to have updated its state: emit at least one element
if (!latch.isTriggered()) {
latch.await();
}
// this means it has processed all the splits and updated its state.
synchronized (sourceContext.getCheckpointLock()) {
}
OperatorSubtaskState snapshot = testHarness.snapshot(0, 0);
monitoringFunction.cancel();
runner.join();
testHarness.close();
final ContinuousFileMonitoringFunction<String> monitoringFunctionCopy = createTestContinuousFileMonitoringFunction(format, FileProcessingMode.PROCESS_CONTINUOUSLY);
StreamSource<TimestampedFileInputSplit, ContinuousFileMonitoringFunction<String>> srcCopy = new StreamSource<>(monitoringFunctionCopy);
AbstractStreamOperatorTestHarness<TimestampedFileInputSplit> testHarnessCopy = new AbstractStreamOperatorTestHarness<>(srcCopy, 1, 1, 0);
testHarnessCopy.initializeState(snapshot);
testHarnessCopy.open();
Assert.assertNull(error[0]);
Assert.assertEquals(fileModTime, monitoringFunctionCopy.getGlobalModificationTime());
hdfs.delete(path, false);
}
Aggregations