This talk describes how we took Apache CouchDB and stretched it into a clustered database we call BigCouch.  BigCouch clusters behave according to concepts outlined in Amazon's Dynamo paper while supporting the better part of the CouchDB API.  The cluster nodes communicate via distributed Erlang links and use CouchDB's replication protocol to synchronize ring membership and database partition maps.

In the presentation he will share some of the finer details of Erlang/OTP gleaned from running BigCouch clusters in production, including why proper supervision trees are useful for hot code upgrades, what can go wrong with anonymous functions on remote nodes, and how to block a sender using nothing more than the ! operator.