I'm thinking about the following use-case and would like to validate if this approach is conceptually valid.
The goal is to expose a long-running Server-Sent-Event (SSE) endpoint in Spring replaying the same Kafka topic for each incoming connection (with some user-specific filtering).
The SSE is exposed in this way:
#GetMapping("/sse")
public SseEmitter sse() {
SseEmitter sseEmitter = new SseEmitter();
Executors
.newSingleThreadExecutor()
.execute(() -> dummyDataProducer.generate() // kafka ultimately
.forEach(payload -> {
try {
sseEmitter.send(payload);
} catch (IOException ex) {
sseEmitter.completeWithError(ex);
}
}));
return sseEmitter;
}
From the other side, there is a KafkaListener method (ConcurrentKafkaListenerContainerFactory is used) :
#KafkaListener(topics = "${app.kafka.topic1}")
public void receive(
#Header(KafkaHeaders.RECEIVED_MESSAGE_KEY) Integer id,
#Payload Object payload) {
// do something ...
}
As far as I know, the Kafka consumer application uses one thread for reading data from a single topic. This somehow violates the idea of using SSE, where for each incoming connection a dedicated long-running thread is created.
Is it a valid approach for this use-case? If so, how to accomplish this properly?
Related
I was reading about Spring Integration's FluxMessageChannel here and here, but I still don't understand exactly what are the differences between using a DirectChannel and FluxMessageChannel when using Project Reactor. Since the DirectChannel is stateless and controlled by its pollers, I'd expect the FluxMessageChannel to not be needed. I'm trying to understand when exactly should I use each and why, when speaking on Reactive Streams applications that are implemented with Spring Integration.
I currently have a reactive project that uses DirectChannel, and it seems to work fine, even the documentation says:
the flow behavior is changed from an imperative push model to a reactive pull model
I'd like to understand when to use each of the channels and what is the exact difference when working with Reactive Streams.
The DirectChannel does not have any poller and its implementation is very simple: as long as a message is sent to it, the handler is called. In the same caller's thread:
public class DirectChannel extends AbstractSubscribableChannel {
private final UnicastingDispatcher dispatcher = new UnicastingDispatcher();
private volatile Integer maxSubscribers;
/**
* Create a channel with default {#link RoundRobinLoadBalancingStrategy}.
*/
public DirectChannel() {
this(new RoundRobinLoadBalancingStrategy());
}
Where that UnicastingDispatcher is:
public final boolean dispatch(final Message<?> message) {
if (this.executor != null) {
Runnable task = createMessageHandlingTask(message);
this.executor.execute(task);
return true;
}
return this.doDispatch(message);
}
(There is no executor option for the DirectChannel)
private boolean doDispatch(Message<?> message) {
if (tryOptimizedDispatch(message)) {
return true;
}
...
protected boolean tryOptimizedDispatch(Message<?> message) {
MessageHandler handler = this.theOneHandler;
if (handler != null) {
try {
handler.handleMessage(message);
return true;
}
catch (Exception e) {
throw IntegrationUtils.wrapInDeliveryExceptionIfNecessary(message,
() -> "Dispatcher failed to deliver Message", e);
}
}
return false;
}
That's why I call it " imperative push model". The caller is this case is going to wait until the handler finishes its job. And if you have a big flow, everything is going to be stopped in the sender thread until a sent message has reached the end of the flow of direct channels. In two simple words: the publisher is in charge for the whole execution and it is blocked in this case. You haven't faced any problems with your solution based on the DirectChannel just because you didn't use reactive non-blocking threads yet like Netty in WebFlux or MongoDB reactive driver.
The FluxMessageChannel was really designed for Reactive Streams purposes where the subscriber is in charge for handling a message which it pulls from the Flux on demand. This way just after sending the publisher is free to do anything else. Just because it is already a subscriber responsibility to handle the message.
I would say it is definitely OK to use DirectChannel as long as your handlers are not blocking. As long as they are blocking you should go with FluxMessageChannel. Although don't forget that there are other channel types for different tasks: https://docs.spring.io/spring-integration/docs/current/reference/html/core.html#channel-implementations
I'm creating a spring reactor application to consume messages from websockets server, transform them and later save them to redis and some sql database, saving to redis and sql database is also reactive. Also, before writing to redis and sql database, messages will be windowed (with different timespans) and aggregated.
I'm not sure if the way I've accomplished what I want to achieve is a proper reactive wise, it means, I'm not losing reactive benefits (performance).
First, let me show you what I got:
#Service
class WebSocketsConsumer {
public ConnectableFlux<String> webSocketFlux() {
return Flux.<String>create(emitter -> {
createWebSocketClient()
.execute(URI.create("wss://some-url-goes-here.com"), session -> {
WebSocketMessage initialMessage = session.textMessage("SOME_MSG_HERE");
Flux<String> flux = session.send(Mono.just(initialMessage))
.thenMany(session.receive())
.map(WebSocketMessage::getPayloadAsText)
.doOnNext(emitter::next);
Flux<String> sessionStatus = session.closeStatus()
.switchIfEmpty(Mono.just(CloseStatus.GOING_AWAY))
.map(CloseStatus::toString)
.doOnNext(emitter::next)
.flatMapMany(Flux::just);
return flux
.mergeWith(sessionStatus)
.then();
})
.subscribe(); //1: highlighted by Intellij Idea: `Calling subsribe in not blocking context`
})
.publish();
}
private ReactorNettyWebSocketClient createWebSocketClient() {
return new ReactorNettyWebSocketClient(
HttpClient.create(),
() -> WebsocketClientSpec.builder().maxFramePayloadLength(131072 * 100)
);
}
}
And
#Service
class WebSocketMessageDispatcher {
private final WebSocketsConsumer webSocketsConsumer;
private final Consumer<String> reactiveRedisConsumer;
private final Consumer<String> reactiveJdbcConsumer;
private Disposable webSocketsDisposable;
WebSocketMessageDispatcher(WebSocketsConsumer webSocketsConsumer, Consumer<String> redisConsumer, Consumer<String> dbConsumer) {
this.webSocketsConsumer = webSocketsConsumer;
this.reactiveRedisConsumer = redisConsumer;
this.reactiveJdbcConsumer = dbConsumer;
}
#EventListener(ApplicationReadyEvent.class)
public void onReady() {
ConnectableFlux<String> messages = webSocketsConsumer.webSocketFlux();
messages.subscribe(reactiveRedisConsumer);
messages.subscribe(reactiveJdbcConsumer);
webSocketsDisposable = messages.connect();
}
#PreDestroy
public void onDestroy() {
if (webSocketsDisposable != null) webSocketsDisposable.dispose();
}
}
Questions:
Is it a proper use of reactive streams? Maybe redis and database writes should be done in flatMap, however IMO they can't as I want them to happen in the background and they will also aggregate messages with different time windows. Also note comment 1 from the code above where idea lints my code, code works however I wonder what this lint may result in? Maybe I should use doOnNext not to call emitter::next but to invoke some dispatcher of messages there with some funcion like doOnNext(dispatcher::dispatchMessage) ?
I want websockets client to start immediately after application is ready and stop consuming messages when application shuts down, are #EventListener(ApplicationReadyEvent.class) and #PreDestroy annotations and code shown above a proper way to handle this scenario in reactive world?
As I said saving to redis and sql database is also reactive, i.e. those saves are also producing Mono<T> is subscribing to those Monos inside subscribe of websockets flux ok or it should be accomplished some other way (comments 2 and 3 in code above)
I currently have a spring cloud stream application that has a listener function that mainly listens to a certain topic and executes the following in sequence:
Consume messages from a topic
Store consumed message in the DB
Call an external service for some information
Process the data
Record the results in DB
Send the message to another topic
Acknowledge the message (I have the acknowledge mode set to manual)
We have decided to move to Spring cloud function, and I have been already able to already do almost all the steps above using the Function interface, with the source topic as input and the sink topic as an output.
#Bean
public Function<Message<NotificationMessage>, Message<ValidatedEvent>> validatedProducts() {
return message -> {
Acknowledgment acknowledgment = message.getHeaders().get(KafkaHeaders.ACKNOWLEDGMENT, Acknowledgment.class);
notificationMessageService.saveOrUpdate(notificationMessage, 0, false);
String status = restEndpoint.getStatusFor(message.getPayload());
ValidatedEvent event = getProcessingResult(message.getPayload(), status);
notificationMessageService.saveOrUpdate(notificationMessage, 1, true);
Optional.ofNullable(acknowledgment).ifPresent(Acknowledgment::acknowledge);
return MessageBuilder
.withPayload(event)
.setHeader(KafkaHeaders.MESSAGE_KEY, event.getKey().getBytes())
.build();
}
}
My problem goes with exception handling in step 7 (Acknowledge the message). We only acknowledge the message if we are sure that it was sent successfully to the sink queue, otherwise we do no acknowledge the message.
My question is, how can such a thing be implemented within Spring cloud function, specially that the send method is fully dependant on the Spring Framework (as the result of the function interface implementation evaluation).
earlier, we could do this through try/catch
#StreamListener(value = NotificationMesage.INPUT)
public void onMessage(Message<NotificationMessage> message) {
try {
Acknowledgment acknowledgment = message.getHeaders().get(KafkaHeaders.ACKNOWLEDGMENT, Acknowledgment.class);
notificationMessageService.saveOrUpdate(notificationMessage, 0, false);
String status = restEndpoint.getStatusFor(message.getPayload());
ValidatedEvent event = getProcessingResult(message.getPayload(), status);
Message message = MessageBuilder
.withPayload(event)
.setHeader(KafkaHeaders.MESSAGE_KEY, event.getKey().getBytes())
.build();
kafkaTemplate.send(message);
notificationMessageService.saveOrUpdate(notificationMessage, 1, true);
Optional.ofNullable(acknowledgment).ifPresent(Acknowledgment::acknowledge);
}catch (Exception exception){
notificationMessageService.saveOrUpdate(notificationMessage, 1, false);
}
}
Is there a listener that triggers after the Function interface have returned successfully, something like KafkaSendCallback but without specifying a template
Building upon what Oleg mentioned above, if you want to strictly restore the behavior in your StreamListener code, here is something you can try. Instead of using a function, you can switch to a consumer and then use KafkaTemplate to send on the outbound as you had previously.
#Bean
public Consumer<Message<NotificationMessage>> validatedProducts() {
return message -> {
try{
Acknowledgment acknowledgment = message.getHeaders().get(KafkaHeaders.ACKNOWLEDGMENT, Acknowledgment.class);
notificationMessageService.saveOrUpdate(notificationMessage, 0, false);
String status = restEndpoint.getStatusFor(message.getPayload());
ValidatedEvent event = getProcessingResult(message.getPayload(), status);
Message message = MessageBuilder
.withPayload(event)
.setHeader(KafkaHeaders.MESSAGE_KEY, event.getKey().getBytes())
.build();
kafkaTemplate.send(message); //here, you make sure that the data was sent successfully by using some callback.
//only ack if the data was sent successfully.
Optional.ofNullable(acknowledgment).ifPresent(Acknowledgment::acknowledge);
}
catch (Exception exception){
notificationMessageService.saveOrUpdate(notificationMessage, 1, false);
}
};
}
Another thing that is worth looking into is using Kafka transactions, in which case if it doesn't work end-to-end, no acknowledgment will happen. Spring Cloud Stream binder has support for this based on the foundations in Spring for Apache Kafka. More details here. Here is the Spring Cloud Stream doc on this.
Spring cloud stream has no knowledge of function. It is just the same message handler as it was before, so the same approach with callback as you used before would work with functions. So perhaps you can share some code that could clarify what you mean? I also don't understand what do you mean by ..send method is fully dependant on the Spring Framework..
Alright, So what I opted in was actually not to use KafkaTemplate (Or streamBridge)for that matter. While it is a feasible solution it would mean that my Function is going to be split into Consumer and some sort of an improvised supplied (the KafkaTemplate in this case).
As I wanted to adhere to the design goals of the functional interface, I have isolated the behaviour for Database update in a ProducerListener interface implementation
#Configuration
public class ProducerListenerConfiguration {
private final MongoTemplate mongoTemplate;
public ProducerListenerConfiguration(MongoTemplate mongoTemplate) {
this.mongoTemplate = mongoTemplate;
}
#Bean
public ProducerListener myProducerListener() {
return new ProducerListener() {
#SneakyThrows
#Override
public void onSuccess(ProducerRecord producerRecord, RecordMetadata recordMetadata) {
final ValidatedEvent event = new ObjectMapper().readerFor(ValidatedEvent.class).readValue((byte[]) producerRecord.value());
final var updateResult = updateDocumentProcessedState(event.getKey(), event.getPayload().getVersion(), true);
}
#SneakyThrows
#Override
public void onError(ProducerRecord producerRecord, #Nullable RecordMetadata recordMetadata, Exception exception) {
ProducerListener.super.onError(producerRecord, recordMetadata, exception);
}
};
}
public UpdateResult updateDocumentProcessedState(String id, long version, boolean isProcessed) {
Query query = new Query();
query.addCriteria(Criteria.where("_id").is(id));
Update update = new Update();
update.set("processed", isProcessed);
update.set("version", version);
return mongoTemplate.updateFirst(query, update, ProductChangedEntity.class);
}
}
Then with each successful attempt, the DB is updated with the processing result and the updated version number.
I am trying to perform the spring kafka batch process error handling. First of all i have few questions.
what is difference between listener and container error handlers and what errors comes into these two categories ?
Could you please help some samples on this to understand better ?
Here is our design:
Poll every certain interval
consume messages in a batch mode
push to local cache (application cache) based on key (to avoid duplicate events)
push all values one by one to another topic once batch process done.
clear the the cache once the operation 3 done and acknowledge the offsets manually.
Here is my plan to have error handling:
public ConcurrentKafkaListenerContainerFactory<String, String> myListenerPartitionContainerFactory(String groupId) {
ConcurrentKafkaListenerContainerFactory<String, String> factory = new ConcurrentKafkaListenerContainerFactory<>();
factory.setConsumerFactory(consumerFactory(groupId));
factory.setConcurrency(partionCount);
factory.getContainerProperties().setAckMode(ContainerProperties.AckMode.MANUAL);
factory.getContainerProperties().setIdleBetweenPolls(pollInterval);
factory.setBatchListener(true);
return factory;
}
#Bean
public ConcurrentKafkaListenerContainerFactory<String, String> myPartitionsListenerContainerFactory()
{
return myListenerPartitionContainerFactory(groupIdPO);
}
#Bean
public RecoveringBatchErrorHandler(KafkaTemplate<String, String> errorKafkaTemplate) {
DeadLetterPublishingRecoverer recoverer =
new DeadLetterPublishingRecoverer(errorKakfaTemplate);
RecoveringBatchErrorHandler errorHandler =
new RecoveringBatchErrorHandler(recoverer, new FixedBackOff(2L, 5000)); // push error event to the error topic
}
#KafkaListener(id = "mylistener", topics = "someTopic", containerFactory = "myPartitionsListenerContainerFactory"))
public void listen(List<ConsumerRecord<String, String>> records, #Header(KafkaHeaders.MESSAGE_KEY) String key, Acknowledgement ack) {
Map hashmap = new Hashmap<>();
records.forEach(record -> {
try {
//key will be formed based on the input record - it will be id.
hashmap.put(key, record);
}
catch (Exception e) {
throw new BatchListenerFailedException("Failed to process", record);
}
});
// Once success each messages to another topic.
try {
hashmap.forEach( (key,value) -> { push to another topic })
hashmap.clear();
ack.acknowledge();
} catch(Exception ex) {
//handle producer exceptions
}
}
is the direction good or any improvements needs to be done? And also what type of container and listener handlers need to be implemented?
#Gary Russell.. could you please help on this ?
The listener error handler is intended for request/reply situations where the error handler can return a meaningful reply to the sender.
You need to throw an exception to trigger the container error handler and you need to know in the index in the original batch to tell it which record failed.
If you are using manual acks like that, you can use the nack() method to indicate which record failed (and don't throw an exception in that case).
We are designing a microservices architecture, we would like to use RabbitMQ as message broker.
We wanted each service to have one specific queue, lets say applicationQueue.
We also defined that our messages would be of two kinds:
Events: Messages that are routed to every service. If a service is interested in some specific event, it will intercept it and create a task from it.
Tasks: Messages representing jobs created from the service to himself, they should be publish only to the queue of the service itself
We are struggling to implement that so far using Spring AMQP.
We designed a message producer, so after a given http request, it would create a task for the service itself:
RestController:
#PostMapping
public void saveProduct(#RequestBody Product product) {
messageProducer.message("subscriptions.product.create", product)
.fromHttpRequest(requestContext)
.send();
}
our send method of the message producer:
public void send() {
template.convertAndSend(exchange, routingKey, payload, message -> {
if (requestContext != null) {
extractHttpRequestInfo(message);
message.getMessageProperties().getHeaders()
.put(MessageDictionary.TRANSACTION_ID, generateTransactionId());
} else if (originalMessage != null) {
extractMessageInfo(message);
}
return message;
});
}
RabbitMQ Configuration:
#Bean
List<Binding> binding(Queue queue, TopicExchange exchange) {
return Arrays.asList(
BindingBuilder.bind(queue).to(exchange).with("*.*"),
BindingBuilder.bind(queue).to(exchange).with("${condohub.rabbitmq.queue.name}.#")
);
}
and then subscribe elsewhere (The #Digest annotation is a custom annotation):
#Digest("${condohub.rabbitmq.queue.name}.product.create")
public void createProduct(Product product) {
service.save(product);
}
Any help is welcome.
Your bindings don't make sense; the first one will match all keys with the form foo.bar, baz.qux etc, so the second one is irrelevant.
You should probably just use a fanout exchange for the events and each service has 2 queues, one on the fanout for events and one on the topic exchange for jobs (with a narrow binding for just its own jobs).