Filling the Gaps with CAP

Even seasoned architects have gaps in their understanding. After an interesting conversation with fellow data architects, I found a significant gap in my familiarity with CAP.

• Brewer's CAP Theorem

• Gilbert and Lynch's Formalization of CAP

• CAP Theorem Explained

• Spanner, TrueTime, and the CAP Theorem

CAP Profiles of Popular Databases

Here, I'm going through the CAP profiles of several popular databases.

Snowflake

Those who use Snowflake regularly rarely have to think about data tradeoffs. Snowflake is strongly consistent, meaning that when anyone runs a query, it will be precisely the same for any user making a query.

Also, Snowflake leverages object store systems, so it starts with a more highly scalable and robust version of data storage than other types of databases. S3, for example, offers %99.99 availability.

Finally, Snowflake distributes data and compute storage across the cloud to handle partition failures by redundancy across availability and regions.

Snowflake will, however, sacrifice availability to keep consistency and partition tolerance so it is a CP system, even though it behaves as a CAP system.