Why? Some background one your thoughts would be helpful.

It (probably) won't be ACID compliant, but it'll be extremely useful for most use-cases.

Sorry:)

Not planning to have database distributed all over the globe and even if I had to distribute the data I would do this on application level with tools like RabbitMQ.

But once you write to Rabbit, you don't know that it'll make it to _any_ db. At least with MM replication you know it's somewhere, and short of disk corruption, you should be able to retrieve it?

I have 8 years experience with Postgres and I already know that a lot of cool built-in features just do not work in production, under heavy load. Usually I have to make some logic inside application in order to protect myself from postgres-driven fuck up.

That's why I am sceptical about this new cool and ambitious built-in MM replication.

Examples please? I am aware of hard corners in places, and rough spots, but some clear discussion might be helpful.

You sound an awful lot like a troll. I have more experience with postgresql than that, and it is literally the single least problematic major software project I've ever used.