Rearchitecting for cloud native; Or, All we changed was everything

Description

Twilio SendGrid, a business unit of Twilio, processes more than 50 billion marketing and transactional emails per month for companies like Airbnb, Pandora, Uber, and Spotify. Since its founding in 2009, the company has run the majority of its operations on-premises in a private cloud, dependent on public clouds only for ancillary workloads. In 2016, the company began a multiyear journey to rearchitect its most critical systems as a precursor to an anticipated migration to the public cloud (now underway).

The protocols, software, and systems that collectively drive email predate modern notions of the cloud by several decades. Classic email software architecture typically follows the assumptions of legacy infrastructure—its monolithic, stateful, scale-up posture is ill-suited to run in the cloud. Like pulling a thread on a knitted sweater, it quickly became apparent that sweeping changes would be required. Some systems were completely rebuilt from the ground up. Others were significantly redesigned. Systems, software, and data architecture; networking infrastructure and routing methodologies; messaging and eventing topologies; statefulness and persistence—virtually nothing remains as it was.

J.R. Jasperson covers some of the challenges SendGrid faced and their technical outcomes, including obviating legacy scale-up networking to eliminate choke points and route traffic through the most efficient path; eradicating state from CPU-bound components to take advantage of ephemeral compute; decoupling discrete components in the system to allow for independent scaling events; solving for nonlinear messaging and eventing complexity to accommodate hypergrowth on a higher-latency, jittery network; and developing solutions for distributed complex queuing and deduplication capabilities by building atop public cloud primitives and capabilities.