Great read comparing various #aws event source latencies.
My takeaway wasn't who is fastest so much as these are all fast enough for the variety of use cases they support (sub-second) 🚀
https://bitesizedserverless.com/bite/serverless-messaging-latency-compared/
Each of these services is for specific use cases anyways. There definitely is some overlap though so could be a useful guidepost for selecting which one you need for a given problem.
I basically default to SNS for dispatching events/commands/pub sub/fan-out, SQS for dispatching throttled tasks and EventBridge for scheduled tasks.
And that's usually more than enough for most dynamic web app use cases.
DynamoDB streams are fun but usually end up trading for SNS the times I have used them. Kinesis is badass and a thing we eyeball for refactoring to but frankly haven't had strong motivation. We do have a lot of logs tho.
On that refactoring tip, Step Functions are state machines and I can imagine upgrading large portions of our code to them as a sorta final optimization to dependency free from interpreted code.
But that's a whole other conversation.
@brianleroux a lot of our new multi step stuff is State Machine. Less boiler plate, better logs (eg state in/out) by default
@falken really feels like the final boss of cloud infra is fully declarative and deterministic