Are there any rules of thumb where such an architecture should be considered? >X TPS? >Y milliseconds per txn? >Z milliseconds between write and subsequent read? Eventual consistency OK?
It depends! FWIW the spots I've used event patterns most often are high write loads that benefit either from fan-in micro-batching before hitting the data stores or polyglot systems where a client/system event need to fan-out to multiple systems and we want to insulate the producer from downstream latency/outages.
