Use case is that we want to publish events to redis and requirement is that events published from single publisher should be in order.
So to achieve this, we have a code like this for single publisher:
source .mapAsync(1)(publish) .to(Sink.ignore) .run()
2. But when i ran a benchmark with this code on single machine (8GB RAA, 8 vCPUS, MacOs) with 40 Publishers and 40 Subscribers, I got only 25k/sec throughput.
After little investigation, I thought of creating single threaded execution context per Publisher to maintain ordering and increased mapAsync to 8 and ran the same benchmark, in this case, I am getting around 45k/sec throughput and code looks like this:
implicit val synchronousEc = ExecutionContext.fromExecutor((task: Runnable) => task.run()) source .mapAsync(8)(publish) .to(Sink.ignore) .run()
So the question is which one is the recommended way? What are pros and cons for both the approaches or is there any other way I can guarantee ordering?