@SuppressWarnings({"unchecked", "rawtypes"})
static void addStormConfigToTopology(FlinkTopology topology, Map conf)
throws ClassNotFoundException {
if (conf != null) {
ExecutionConfig flinkConfig = topology.getExecutionEnvironment().getConfig();
flinkConfig.setGlobalJobParameters(new StormConfig(conf));
// add all registered types to ExecutionConfig
List<?> registeredClasses = (List<?>) conf.get(Config.TOPOLOGY_KRYO_REGISTER);
if (registeredClasses != null) {
for (Object klass : registeredClasses) {
if (klass instanceof String) {
flinkConfig.registerKryoType(Class.forName((String) klass));
} else {
for (Entry<String, String> register : ((Map<String, String>) klass).entrySet()) {
flinkConfig.registerTypeWithKryoSerializer(
Class.forName(register.getKey()),
(Class<? extends Serializer<?>>) Class.forName(register.getValue()));
}
}
}
}
}
}
@Override
public void prepare() {
ExecutionConfig executionConfig = taskContext.getExecutionConfig();
this.objectReuseEnabled = executionConfig.isObjectReuseEnabled();
if (LOG.isDebugEnabled()) {
LOG.debug(
"NoOpDriver object reuse: " + (this.objectReuseEnabled ? "ENABLED" : "DISABLED") + ".");
}
}
@Parameterized.Parameters
public static Collection<Object[]> getConfigurations() throws FileNotFoundException, IOException {
LinkedList<Object[]> configs = new LinkedList<Object[]>();
ExecutionConfig withReuse = new ExecutionConfig();
withReuse.enableObjectReuse();
ExecutionConfig withoutReuse = new ExecutionConfig();
withoutReuse.disableObjectReuse();
Object[] a = {withoutReuse};
configs.add(a);
Object[] b = {withReuse};
configs.add(b);
return configs;
}
@Test
public void testJoinPlain() {
final FlatJoinFunction<String, String, Integer> joiner =
new FlatJoinFunction<String, String, Integer>() {
@Override
public void join(String first, String second, Collector<Integer> out) throws Exception {
out.collect(first.length());
out.collect(second.length());
}
};
@SuppressWarnings({"rawtypes", "unchecked"})
JoinOperatorBase<String, String, Integer, FlatJoinFunction<String, String, Integer>> base =
new JoinOperatorBase(
joiner,
new BinaryOperatorInformation(
BasicTypeInfo.STRING_TYPE_INFO,
BasicTypeInfo.STRING_TYPE_INFO,
BasicTypeInfo.INT_TYPE_INFO),
new int[0],
new int[0],
"TestJoiner");
List<String> inputData1 = new ArrayList<String>(Arrays.asList("foo", "bar", "foobar"));
List<String> inputData2 = new ArrayList<String>(Arrays.asList("foobar", "foo"));
List<Integer> expected = new ArrayList<Integer>(Arrays.asList(3, 3, 6, 6));
try {
ExecutionConfig executionConfig = new ExecutionConfig();
executionConfig.disableObjectReuse();
List<Integer> resultSafe =
base.executeOnCollections(inputData1, inputData2, null, executionConfig);
executionConfig.enableObjectReuse();
List<Integer> resultRegular =
base.executeOnCollections(inputData1, inputData2, null, executionConfig);
assertEquals(expected, resultSafe);
assertEquals(expected, resultRegular);
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
}
}
public void run(final Object lockingObject, final Output<StreamRecord<T>> collector)
throws Exception {
final ExecutionConfig executionConfig = getExecutionConfig();
if (userFunction instanceof EventTimeSourceFunction) {
ctx = new ManualWatermarkContext<T>(lockingObject, collector);
} else if (executionConfig.getAutoWatermarkInterval() > 0) {
ctx = new AutomaticWatermarkContext<T>(lockingObject, collector, executionConfig);
} else if (executionConfig.areTimestampsEnabled()) {
ctx = new NonWatermarkContext<T>(lockingObject, collector);
} else {
ctx = new NonTimestampContext<T>(lockingObject, collector);
}
userFunction.run(ctx);
// This will mostly emit a final +Inf Watermark to make the Watermark logic work
// when some sources finish before others do
ctx.close();
}
public AutomaticWatermarkContext(
Object lockingObjectParam,
Output<StreamRecord<T>> outputParam,
ExecutionConfig executionConfig) {
this.lockingObject = lockingObjectParam;
this.output = outputParam;
this.reuse = new StreamRecord<T>(null);
watermarkInterval = executionConfig.getAutoWatermarkInterval();
scheduleExecutor = Executors.newScheduledThreadPool(1);
watermarkTimer =
scheduleExecutor.scheduleAtFixedRate(
new Runnable() {
@Override
public void run() {
long currentTime = System.currentTimeMillis();
// align the watermarks across all machines. this will ensure that we
// don't have watermarks that creep along at different intervals because
// the machine clocks are out of sync
long watermarkTime = currentTime - (currentTime % watermarkInterval);
if (currentTime > watermarkTime
&& watermarkTime - lastWatermarkTime >= watermarkInterval) {
synchronized (lockingObject) {
if (currentTime > watermarkTime
&& watermarkTime - lastWatermarkTime >= watermarkInterval) {
output.emitWatermark(new Watermark(watermarkTime));
lastWatermarkTime = watermarkTime;
}
}
}
}
},
0,
watermarkInterval,
TimeUnit.MILLISECONDS);
}
@Test
public void testJoinRich() {
final AtomicBoolean opened = new AtomicBoolean(false);
final AtomicBoolean closed = new AtomicBoolean(false);
final String taskName = "Test rich join function";
final RichFlatJoinFunction<String, String, Integer> joiner =
new RichFlatJoinFunction<String, String, Integer>() {
@Override
public void open(Configuration parameters) throws Exception {
opened.compareAndSet(false, true);
assertEquals(0, getRuntimeContext().getIndexOfThisSubtask());
assertEquals(1, getRuntimeContext().getNumberOfParallelSubtasks());
}
@Override
public void close() throws Exception {
closed.compareAndSet(false, true);
}
@Override
public void join(String first, String second, Collector<Integer> out) throws Exception {
out.collect(first.length());
out.collect(second.length());
}
};
JoinOperatorBase<String, String, Integer, RichFlatJoinFunction<String, String, Integer>> base =
new JoinOperatorBase<
String, String, Integer, RichFlatJoinFunction<String, String, Integer>>(
joiner,
new BinaryOperatorInformation<String, String, Integer>(
BasicTypeInfo.STRING_TYPE_INFO,
BasicTypeInfo.STRING_TYPE_INFO,
BasicTypeInfo.INT_TYPE_INFO),
new int[0],
new int[0],
taskName);
final List<String> inputData1 = new ArrayList<String>(Arrays.asList("foo", "bar", "foobar"));
final List<String> inputData2 = new ArrayList<String>(Arrays.asList("foobar", "foo"));
final List<Integer> expected = new ArrayList<Integer>(Arrays.asList(3, 3, 6, 6));
try {
final HashMap<String, Accumulator<?, ?>> accumulatorMap =
new HashMap<String, Accumulator<?, ?>>();
final HashMap<String, Future<Path>> cpTasks = new HashMap<>();
ExecutionConfig executionConfig = new ExecutionConfig();
executionConfig.disableObjectReuse();
List<Integer> resultSafe =
base.executeOnCollections(
inputData1,
inputData2,
new RuntimeUDFContext(taskName, 1, 0, null, executionConfig, cpTasks, accumulatorMap),
executionConfig);
executionConfig.enableObjectReuse();
List<Integer> resultRegular =
base.executeOnCollections(
inputData1,
inputData2,
new RuntimeUDFContext(taskName, 1, 0, null, executionConfig, cpTasks, accumulatorMap),
executionConfig);
assertEquals(expected, resultSafe);
assertEquals(expected, resultRegular);
} catch (Exception e) {
e.printStackTrace();
fail(e.getMessage());
}
assertTrue(opened.get());
assertTrue(closed.get());
}
@Override
public void invoke() throws Exception {
if (LOG.isDebugEnabled()) {
LOG.debug(getLogString("Starting data sink operator"));
}
ExecutionConfig executionConfig;
try {
ExecutionConfig c =
(ExecutionConfig)
InstantiationUtil.readObjectFromConfig(
getJobConfiguration(), ExecutionConfig.CONFIG_KEY, getUserCodeClassLoader());
if (c != null) {
executionConfig = c;
} else {
LOG.warn("The execution config returned by the configuration was null");
executionConfig = new ExecutionConfig();
}
} catch (IOException e) {
throw new RuntimeException("Could not load ExecutionConfig from Job Configuration: " + e);
} catch (ClassNotFoundException e) {
throw new RuntimeException("Could not load ExecutionConfig from Job Configuration: " + e);
}
boolean objectReuseEnabled = executionConfig.isObjectReuseEnabled();
try {
// initialize local strategies
MutableObjectIterator<IT> input1;
switch (this.config.getInputLocalStrategy(0)) {
case NONE:
// nothing to do
localStrategy = null;
input1 = reader;
break;
case SORT:
// initialize sort local strategy
try {
// get type comparator
TypeComparatorFactory<IT> compFact =
this.config.getInputComparator(0, getUserCodeClassLoader());
if (compFact == null) {
throw new Exception("Missing comparator factory for local strategy on input " + 0);
}
// initialize sorter
UnilateralSortMerger<IT> sorter =
new UnilateralSortMerger<IT>(
getEnvironment().getMemoryManager(),
getEnvironment().getIOManager(),
this.reader,
this,
this.inputTypeSerializerFactory,
compFact.createComparator(),
this.config.getRelativeMemoryInput(0),
this.config.getFilehandlesInput(0),
this.config.getSpillingThresholdInput(0));
this.localStrategy = sorter;
input1 = sorter.getIterator();
} catch (Exception e) {
throw new RuntimeException(
"Initializing the input processing failed"
+ (e.getMessage() == null ? "." : ": " + e.getMessage()),
e);
}
break;
default:
throw new RuntimeException("Invalid local strategy for DataSinkTask");
}
// read the reader and write it to the output
final TypeSerializer<IT> serializer = this.inputTypeSerializerFactory.getSerializer();
final MutableObjectIterator<IT> input = input1;
final OutputFormat<IT> format = this.format;
// check if task has been canceled
if (this.taskCanceled) {
return;
}
if (LOG.isDebugEnabled()) {
LOG.debug(getLogString("Starting to produce output"));
}
// open
format.open(
this.getEnvironment().getIndexInSubtaskGroup(),
this.getEnvironment().getNumberOfSubtasks());
if (objectReuseEnabled) {
IT record = serializer.createInstance();
// work!
while (!this.taskCanceled && ((record = input.next(record)) != null)) {
format.writeRecord(record);
}
} else {
IT record;
// work!
while (!this.taskCanceled && ((record = input.next()) != null)) {
format.writeRecord(record);
}
}
// close. We close here such that a regular close throwing an exception marks a task as
// failed.
if (!this.taskCanceled) {
this.format.close();
this.format = null;
}
} catch (Exception ex) {
// make a best effort to clean up
try {
if (!cleanupCalled && format instanceof CleanupWhenUnsuccessful) {
cleanupCalled = true;
((CleanupWhenUnsuccessful) format).tryCleanupOnError();
}
} catch (Throwable t) {
LOG.error("Cleanup on error failed.", t);
}
ex = ExceptionInChainedStubException.exceptionUnwrap(ex);
if (ex instanceof CancelTaskException) {
// forward canceling exception
throw ex;
}
// drop, if the task was canceled
else if (!this.taskCanceled) {
if (LOG.isErrorEnabled()) {
LOG.error(getLogString("Error in user code: " + ex.getMessage()), ex);
}
throw ex;
}
} finally {
if (this.format != null) {
// close format, if it has not been closed, yet.
// This should only be the case if we had a previous error, or were canceled.
try {
this.format.close();
} catch (Throwable t) {
if (LOG.isWarnEnabled()) {
LOG.warn(getLogString("Error closing the output format"), t);
}
}
}
// close local strategy if necessary
if (localStrategy != null) {
try {
this.localStrategy.close();
} catch (Throwable t) {
LOG.error("Error closing local strategy", t);
}
}
RegularPactTask.clearReaders(new MutableReader[] {inputReader});
}
if (!this.taskCanceled) {
if (LOG.isDebugEnabled()) {
LOG.debug(getLogString("Finished data sink operator"));
}
} else {
if (LOG.isDebugEnabled()) {
LOG.debug(getLogString("Data sink operator cancelled"));
}
}
}
@Override
public void prepare() throws Exception {
final TaskConfig config = this.taskContext.getTaskConfig();
if (config.getDriverStrategy() != DriverStrategy.CO_GROUP) {
throw new Exception(
"Unrecognized driver strategy for CoGoup driver: " + config.getDriverStrategy().name());
}
final MutableObjectIterator<IT1> in1 = this.taskContext.getInput(0);
final MutableObjectIterator<IT2> in2 = this.taskContext.getInput(1);
// get the key positions and types
final TypeSerializer<IT1> serializer1 =
this.taskContext.<IT1>getInputSerializer(0).getSerializer();
final TypeSerializer<IT2> serializer2 =
this.taskContext.<IT2>getInputSerializer(1).getSerializer();
final TypeComparator<IT1> groupComparator1 = this.taskContext.getDriverComparator(0);
final TypeComparator<IT2> groupComparator2 = this.taskContext.getDriverComparator(1);
final TypePairComparatorFactory<IT1, IT2> pairComparatorFactory =
config.getPairComparatorFactory(this.taskContext.getUserCodeClassLoader());
if (pairComparatorFactory == null) {
throw new Exception("Missing pair comparator factory for CoGroup driver");
}
ExecutionConfig executionConfig = taskContext.getExecutionConfig();
this.objectReuseEnabled = executionConfig.isObjectReuseEnabled();
if (LOG.isDebugEnabled()) {
LOG.debug(
"CoGroupDriver object reuse: "
+ (this.objectReuseEnabled ? "ENABLED" : "DISABLED")
+ ".");
}
if (objectReuseEnabled) {
// create CoGropuTaskIterator according to provided local strategy.
this.coGroupIterator =
new ReusingSortMergeCoGroupIterator<IT1, IT2>(
in1,
in2,
serializer1,
groupComparator1,
serializer2,
groupComparator2,
pairComparatorFactory.createComparator12(groupComparator1, groupComparator2));
} else {
// create CoGropuTaskIterator according to provided local strategy.
this.coGroupIterator =
new NonReusingSortMergeCoGroupIterator<IT1, IT2>(
in1,
in2,
serializer1,
groupComparator1,
serializer2,
groupComparator2,
pairComparatorFactory.createComparator12(groupComparator1, groupComparator2));
}
// open CoGroupTaskIterator - this triggers the sorting and blocks until the iterator is ready
this.coGroupIterator.open();
if (LOG.isDebugEnabled()) {
LOG.debug(this.taskContext.formatLogString("CoGroup task iterator ready."));
}
}
