Mixed order of operations when using completionStage

[einarjohnson]

Hi,
I have a reactive Quarkus 2.16.3.Final application(lambda) that is processing a batch of events and inserting them to AWS DynamoDB and then forwarding them to an AWS SQS queue. One of the requirements is that the events must be persisted in DynamoDB before we forward them to SQS for further processing because downstream handlers rely on this event store to be up to date.

The DynamoDB persist code is like this:

public Uni<Void> storeEvent(MyRecord record) {
	return Uni.createFrom().completionStage(() -> myTable.putItem(record));
}

And the SQS message queue code is like this:

public Uni<String> queueMessage(String message, String destinationQueue) {
	return Uni.createFrom()
			.completionStage(sqs.sendMessage(m -> m.queueUrl(destinationQueue).messageBody(message)))
			.onItem().transform(SendMessageResponse::messageId);
}

The main event processor/lambda handler is simple and the code is pretty much doing something like this.

UniJoin.Builder<String> uniBuilder = Uni.join().builder();
for (MyRecord record : listOfRecords) {
	uniBuilder.add(dynamoDBService.storeEvent(record).onItem().transformToUni(
			storeResult -> sqsService.queueMessage(record.toJson(), myQueue)));
}
var result = uniBuilder.joinAll().andCollectFailures().await().atMost(Duration.ofSeconds(30));

The problem I am facing is that it seems to me (according to our production logs) that the sqsService.queueMessage is sometimes executed before the dynamoDBService.storeLoggerEvent call has completed.
Do I need to structure the Mutiny call chain in a different way if I want to guarantee that these two nested calls are called one after another?

[jponge]

Hi, the pipeline looks correct: storeEvent is to be called first then the result is passed to queueMessage so you do have something like lambda -> dynamodb -> sqs.

Now what you may see in your logs (depending on what level of detail you have here) is concurrent work because you are joining several Uni instances, and there's a level of concurrency involved.

[einarjohnson]

you are absolutely right, i've had my head buried in this bug for two days now and I found the root cause and it was totally unrelated to this uni pipeline. so all is working as expected, I was just barking up the wrong tree here. thanks for chiming in anyway.