"If you go in the direction of microservices, err on the side of too big then
rather too small."

This looks familiar:

Read / Write: use HTTP GET / PUT. "It just™️ works in your stack."

"For events use polling." Ask the microservice you want events and GET an
endpoint that returns the events, optionally with a range key.

"This approach gives you a simple, and robust way for collaboration, without a
single point of failure (which you would get with centralized queuing systems).
But it's slow, and needs both services to be up at the same time."

This talk reminded me a lot about the principles described in "REST in
Practice": http://shop.oreilly.com/product/9780596805838.do #ndcoslo