use of io.mantisrx.connector.kafka.source.checkpoint.strategy.CheckpointStrategy in project mantis by Netflix.
the class KafkaSource method createBackPressuredConsumerObs.
/**
* Create an observable with back pressure semantics from the consumer records fetched using consumer.
*
* @param mantisKafkaConsumer non thread-safe KafkaConsumer
* @param kafkaSourceConfig configuration for the Mantis Kafka Source
*/
private Observable<KafkaAckable> createBackPressuredConsumerObs(final MantisKafkaConsumer<?> mantisKafkaConsumer, final MantisKafkaSourceConfig kafkaSourceConfig) {
CheckpointStrategy checkpointStrategy = mantisKafkaConsumer.getStrategy();
final CheckpointTrigger trigger = mantisKafkaConsumer.getTrigger();
final ConsumerMetrics consumerMetrics = mantisKafkaConsumer.getConsumerMetrics();
final TopicPartitionStateManager partitionStateManager = mantisKafkaConsumer.getPartitionStateManager();
int mantisKafkaConsumerId = mantisKafkaConsumer.getConsumerId();
SyncOnSubscribe<Iterator<ConsumerRecord<String, byte[]>>, KafkaAckable> syncOnSubscribe = SyncOnSubscribe.createStateful(() -> {
final ConsumerRecords<String, byte[]> records = mantisKafkaConsumer.poll(kafkaSourceConfig.getConsumerPollTimeoutMs());
if (LOGGER.isDebugEnabled()) {
LOGGER.debug("topic listing for consumer {}", mantisKafkaConsumer.listTopics());
}
LOGGER.info("consumer subscribed to topic-partitions {}", mantisKafkaConsumer.assignment());
return records.iterator();
}, (consumerRecordIterator, observer) -> {
Iterator<ConsumerRecord<String, byte[]>> it = consumerRecordIterator;
final Set<TopicPartition> partitions = mantisKafkaConsumer.assignment();
if (trigger.shouldCheckpoint()) {
long startTime = System.currentTimeMillis();
final Map<TopicPartition, OffsetAndMetadata> checkpoint = partitionStateManager.createCheckpoint(partitions);
checkpointStrategy.persistCheckpoint(checkpoint);
long now = System.currentTimeMillis();
consumerMetrics.recordCheckpointDelay(now - startTime);
consumerMetrics.incrementCommitCount();
trigger.reset();
}
if (!done.get()) {
try {
if (!consumerRecordIterator.hasNext()) {
final ConsumerRecords<String, byte[]> consumerRecords = mantisKafkaConsumer.poll(kafkaSourceConfig.getConsumerPollTimeoutMs());
if (LOGGER.isDebugEnabled()) {
LOGGER.debug("poll returned {} records", consumerRecords.count());
}
it = consumerRecords.iterator();
}
if (it.hasNext()) {
final ConsumerRecord<String, byte[]> m = it.next();
final TopicPartition topicPartition = new TopicPartition(m.topic(), m.partition());
consumerMetrics.incrementInCount();
if (LOGGER.isDebugEnabled()) {
LOGGER.debug("updating read offset to " + m.offset() + " read " + m.value());
}
if (m.value() != null) {
try {
trigger.update(getPayloadSize(m));
if (kafkaSourceConfig.getParseMessageInSource()) {
final Parser parser = ParserType.parser(kafkaSourceConfig.getMessageParserType()).getParser();
if (parser.canParse(m.value())) {
final Map<String, Object> parsedKafkaValue = parser.parseMessage(m.value());
final KafkaData kafkaData = new KafkaData(m, Optional.ofNullable(parsedKafkaValue), Optional.ofNullable(m.key()), mantisKafkaConsumerId);
final KafkaAckable ackable = new KafkaAckable(kafkaData, ackSubject);
// record offset consumed in TopicPartitionStateManager before onNext to avoid race condition with Ack being processed before the consume is recorded
partitionStateManager.recordMessageRead(topicPartition, m.offset());
consumerMetrics.recordReadOffset(topicPartition, m.offset());
observer.onNext(ackable);
} else {
consumerMetrics.incrementParseFailureCount();
}
} else {
final KafkaData kafkaData = new KafkaData(m, Optional.empty(), Optional.ofNullable(m.key()), mantisKafkaConsumerId);
final KafkaAckable ackable = new KafkaAckable(kafkaData, ackSubject);
// record offset consumed in TopicPartitionStateManager before onNext to avoid race condition with Ack being processed before the consume is recorded
partitionStateManager.recordMessageRead(topicPartition, m.offset());
consumerMetrics.recordReadOffset(topicPartition, m.offset());
observer.onNext(ackable);
}
} catch (ParseException pe) {
consumerMetrics.incrementErrorCount();
LOGGER.warn("failed to parse {}:{} message {}", m.topic(), m.partition(), m.value(), pe);
}
} else {
consumerMetrics.incrementKafkaMessageValueNullCount();
}
} else {
consumerMetrics.incrementWaitForDataCount();
if (LOGGER.isDebugEnabled()) {
LOGGER.debug("Reached head of partition, waiting for more data");
}
TimeUnit.MILLISECONDS.sleep(200);
}
} catch (TimeoutException toe) {
consumerMetrics.incrementWaitForDataCount();
if (LOGGER.isDebugEnabled()) {
LOGGER.debug("Reached head of partition waiting for more data");
}
} catch (OffsetOutOfRangeException oore) {
LOGGER.warn("offsets out of range " + oore.partitions() + " will seek to beginning", oore);
final Set<TopicPartition> topicPartitionSet = oore.partitions();
for (TopicPartition tp : topicPartitionSet) {
LOGGER.info("partition {} consumer position {}", tp, mantisKafkaConsumer.position(tp));
}
mantisKafkaConsumer.seekToBeginning(oore.partitions().toArray(new TopicPartition[oore.partitions().size()]));
} catch (InvalidRecordException ire) {
consumerMetrics.incrementErrorCount();
LOGGER.warn("iterator error with invalid message. message will be dropped " + ire.getMessage());
} catch (KafkaException e) {
consumerMetrics.incrementErrorCount();
LOGGER.warn("Other Kafka exception, message will be dropped. " + e.getMessage());
} catch (InterruptedException ie) {
LOGGER.error("consumer interrupted", ie);
Thread.currentThread().interrupt();
} catch (Exception e) {
consumerMetrics.incrementErrorCount();
LOGGER.warn("caught exception", e);
}
} else {
mantisKafkaConsumer.close();
}
return it;
}, consumerRecordIterator -> {
LOGGER.info("closing Kafka consumer on unsubscribe" + mantisKafkaConsumer.toString());
mantisKafkaConsumer.close();
});
return Observable.create(syncOnSubscribe).subscribeOn(Schedulers.newThread()).doOnUnsubscribe(() -> LOGGER.info("consumer {} stopped due to unsubscribe", mantisKafkaConsumerId)).doOnError((t) -> {
LOGGER.error("consumer {} stopped due to error", mantisKafkaConsumerId, t);
consumerMetrics.incrementErrorCount();
}).doOnTerminate(() -> LOGGER.info("consumer {} terminated", mantisKafkaConsumerId));
}
Aggregations