In this talk I show you how to set up a python and PostgreSQL based system which is easy to set up and easy to scale, provides ACID guarantees where they are needed and delays time-consistency between unrelated objects for scalability and availability where the latter are deemed more important.
The best thing is that this kind of scalability work for both OLTP and OLAP workloads, so with some planning you can have just a single large “database” which can take almost any type of load.
Also, if you hate SQL, you can do all the OLTP stuff in a pythonic way using an automagically generated ORM layer inside the database, near the data. If you are really masochistic, you can use the same ORM also for map-reduce type distributed data processing, though on this side the small effort of learning SQL usually pays off when queries get more complex. But as I said, everything runs inside the databse, near the data and thus even the ORM & map-reduce analytics works fast.