Eric Redmond is an engineer at Basho, which produces the NoSQL database Riak. We talked with Eric about design patterns for distributed systems.
Eric wanted to de-mystify the creation of highly available systems, so he wrote an example in Ruby for his OSCON presentation. He wanted to show that it's doable by an average programmer, and that you don't have to use an industrial-strength programming language like Erlang.
Eric talks about the many different tradeoffs involved in implementing some of these different distributed design patterns, including speed, capacity, uptime and data integrity
Eric describes Riak as a distributed system that persists data rather than a database that has been given distributed functionality. He explains the primary distribution mechanisms of replication and sharding, and how Riak deals with a cascading failures through primary and secondary replication.
It's that ability to deal with failure that Eric says makes Riak operations friendly: It's able to handle the loss of multiple nodes—he's seen up to half go down in a Riak deployment—while maintaining the ability to write and read all of the data.
Finally, he talks about some of the upcoming features of Riak, including commutative replicated data types (CRDT), which makes it easier to merge and resolve conflicts between replicated states.
- For more information, be sure to check out Eric's OSCON presentation "Distributed Patterns in Action"
- Check out the modules tagged "nosql" on the Puppet Forge to get started managing Riak, Cassandra, MongoDB and others with Puppet or Puppet Enterprise.
- Learn how automation can help you with all your configuration management tasks.