The Cassandra community continues to impress me with their vigor and openness. 2015 is another step with September’s Cassandra Summit upsizing from a SF city hotel into the Santa Clara Convention Center.
The content came online recently on Vimeo here.
Two point stood out for me this year:
- migrating from relational database to Cassandra (NoSQL)
- multi-model persistence
The prospect of migrating a SQL RDBMS-based app to Cassandra is now a recognized practice. So much so that there was a session prosaically titled “When and how to migrate from a relational database to Cassandra” by Ben Slater.
The premise isn’t as far-fetched as it seems on first glance because the candidate application that has out-grown the architectural limits of an RDBMS. I.e., it has likely already de/re-indexed, sharded, denormalized and otherwise applied all the known tricks to wring the performance from the database. So the incremental step in Cassandra’s direction: to switch to micro-services for your database access and to rewrite the app to for CQL isn’t much of a stretch anymore.
And if you aren’t running into the limits of your RDBMS yet, you have time on your side to start making architectural shifts to prepare for that eventuality. You can take a forward-looking approach and double-write your data to familiarize and prepare for the eventuality.
And in a odd moment of serendipity, MongoDB just published a whitepaper on migrating from relational database to MongoDB which cites a number of customers who’s successfully migrated.
The second major idea is that of “multi-model persistence”.
DataStax’s senior product manager Darshan Rawal gave a series of talk on multi-model persistence that was insightful. The assertion is that polyglot persistence has run its course because of the cost and complexity of building a data access abstraction on top of multiple API and models. Instead, switch to something such as Cassandra that supports multiple models using the same API.
(Alas, Datastax seems to be in the midst of reorganizing their video content because neither talk is available on Vimeo/YouTube. Hopefully, the slides are helpful.)