use of org.apache.gobblin.util.LoggingUncaughtExceptionHandler in project incubator-gobblin by apache.
the class StreamModelTaskRunner method run.
protected void run() throws Exception {
long maxWaitInMinute = taskState.getPropAsLong(ConfigurationKeys.FORK_MAX_WAIT_MININUTES, ConfigurationKeys.DEFAULT_FORK_MAX_WAIT_MININUTES);
long initialDelay = taskState.getPropAsLong(ConfigurationKeys.FORK_FINISHED_CHECK_INTERVAL, ConfigurationKeys.DEFAULT_FORK_FINISHED_CHECK_INTERVAL);
// Get the fork operator. By default IdentityForkOperator is used with a single branch.
ForkOperator forkOperator = closer.register(this.taskContext.getForkOperator());
RecordStreamWithMetadata<?, ?> stream = this.extractor.recordStream(this.shutdownRequested);
// This prevents emitting records until a connect() call is made on the connectable stream
ConnectableFlowable connectableStream = stream.getRecordStream().publish();
// The cancel is not propagated to the extractor's record generator when it has been turned into a hot Flowable
// by publish, and in the case that extractor stuck in reading record when cancel get called,
// we directly call shutdown to force it instead of setting the shutdownRequested flag on cancel to stop the extractor
Flowable streamWithShutdownOnCancel = connectableStream.doOnCancel(this.extractor::shutdown);
stream = stream.withRecordStream(streamWithShutdownOnCancel);
stream = stream.mapRecords(r -> {
this.task.onRecordExtract();
return r;
});
if (this.task.isStreamingTask()) {
// Start watermark manager and tracker
if (this.watermarkTracker.isPresent()) {
this.watermarkTracker.get().start();
}
this.watermarkManager.get().start();
((StreamingExtractor) this.taskContext.getRawSourceExtractor()).start(this.watermarkStorage.get());
stream = stream.mapRecords(r -> {
AcknowledgableWatermark ackableWatermark = new AcknowledgableWatermark(r.getWatermark());
if (watermarkTracker.isPresent()) {
watermarkTracker.get().track(ackableWatermark);
}
r.addCallBack(ackableWatermark);
return r;
});
}
// Use the recordStreamProcessor list if it is configured. This list can contain both all RecordStreamProcessor types
if (!this.recordStreamProcessors.isEmpty()) {
for (RecordStreamProcessor streamProcessor : this.recordStreamProcessors) {
stream = streamProcessor.processStream(stream, this.taskState);
}
} else {
if (this.converter instanceof MultiConverter) {
// if multiconverter, unpack it
for (Converter cverter : ((MultiConverter) this.converter).getConverters()) {
stream = cverter.processStream(stream, this.taskState);
}
} else {
stream = this.converter.processStream(stream, this.taskState);
}
}
stream = this.rowChecker.processStream(stream, this.taskState);
Forker.ForkedStream<?, ?> forkedStreams = new Forker().forkStream(stream, forkOperator, this.taskState);
boolean isForkAsync = !this.task.areSingleBranchTasksSynchronous(this.taskContext) || forkedStreams.getForkedStreams().size() > 1;
int bufferSize = this.taskState.getPropAsInt(ConfigurationKeys.FORK_RECORD_QUEUE_CAPACITY_KEY, ConfigurationKeys.DEFAULT_FORK_RECORD_QUEUE_CAPACITY);
for (int fidx = 0; fidx < forkedStreams.getForkedStreams().size(); fidx++) {
RecordStreamWithMetadata<?, ?> forkedStream = forkedStreams.getForkedStreams().get(fidx);
if (forkedStream != null) {
if (isForkAsync) {
forkedStream = forkedStream.mapStream(f -> f.observeOn(Schedulers.from(this.taskExecutor.getForkExecutor()), false, bufferSize));
}
Fork fork = new Fork(this.taskContext, forkedStream.getGlobalMetadata().getSchema(), forkedStreams.getForkedStreams().size(), fidx, this.taskMode);
fork.consumeRecordStream(forkedStream);
this.forks.put(Optional.of(fork), Optional.of(Futures.immediateFuture(null)));
this.task.configureStreamingFork(fork);
}
}
Thread thread = new Thread(() -> connectableStream.connect());
thread.setName(this.getClass().getSimpleName());
// Log uncaught exceptions (e.g.OOMEs) to prevent threads from dying silently
thread.setUncaughtExceptionHandler(new LoggingUncaughtExceptionHandler(Optional.absent()));
thread.start();
if (!ExponentialBackoff.awaitCondition().callable(() -> this.forks.keySet().stream().map(Optional::get).allMatch(Fork::isDone)).initialDelay(initialDelay).maxDelay(initialDelay).maxWait(TimeUnit.MINUTES.toMillis(maxWaitInMinute)).await()) {
throw new TimeoutException("Forks did not finish withing specified timeout.");
}
}
Aggregations