Implement FIFO Support

Implement support for FIFO queues.

Should be pretty straightforward overall. The key part should be implementing a FifoMessageSplitter with the logic to group messages from the same message group and processing each group in parallel while maintaining order within a group.

The AbstractEndpoint should be aware that it’s a fifo endpoint and instantiate the proper splitter. Maybe it’s worth having a FifoEndpoint subclass.

---

The AsyncMessageSplitter interface is responsible for splitting the batch of messages acquired from the AsyncMessagePoller and sending the messages through a processing pipeline provided by the listener container.

@FunctionalInterface
public interface AsyncMessageSplitter<T> {

	Collection<CompletableFuture<Void>> splitAndProcess(Collection<Message<T>> messages, Function<Message<T>, CompletableFuture<Void>> processingPipeline);
}

[mgusiew-guide]

One thing that can be taken under consideration is visibility timeout. In case when one gets 10 messages from same group, they all need to be completed before within the time limit. The idea is to allow to extend the time limit after every message in the group is processed.

The other interesting thing is to decide what to do in case if the message is processed with error. There are few options:

1. Process next messages in group
2. Retry the error message (possibly with extending timeout on all of messages in the group)
3. Let the messages return to queue (no ack) - the problem here is that until few retries the messages may land up in DLQ even though they were not ever processed (because of poison message in front)

It may happen that some tradeoffs will need to be made and that in order to get expected FIFO semantics, the end user will need to keep batch size set to 1 and perhaps decrease poll timeout (this is how we workaround the lack of FIFO support in Spring SQS today).

IMHO it is bit unfortunate that SQS does not allow to set poll policy to not return multiple messages from same message group as this would allow to parallelize easily (no ordering conflicts in a polled batch)

[tomazfernandes]

I’ve created a new issue that makes this scope more clear, so I’m closing this one.

Thanks!

Closed in favor of https://github.com/awspring/spring-cloud-aws/issues/444