For a service built to handle millions of requests/hour, it's insufficient to rely on latest trendy components or datastores to save you from system failures, instead it's necessary to deeply understand the properties and the mechanics of your system, and to partition its different dimensions to avoid a domino style failure cascade.

Partitioning time is about uncoupling subsystems that don't absolutely need to be updated in sync, whereas partitioning space is achieved by separating subsystems that don't need to share their data.

In this talk we'll look at these and other design principles and how they were applied to build prismic.io, an API-based CMS platform.