I work with someone who is an ex-Datastax employee. He swears by Cassandra. We end up debating about it a lot, and I usually bring up Riak as a better alternative - in theory. Cassandra has so many peculiarities and gotchas, and almost requires a dedicated team to run at any non-trivial size, and good knowledge of its behaviour among the development team.

His point, which I can't refute, is that Cassandra is probably the most widely used and battle tested open source distributed database out there - there are massive clusters (10s of thousands of nodes) in production use by many of the most well known engineering firms.

Does Riak have similar success stories? Are there any mega clusters in use at big firms? How would you argue for using Riak?