Preparing for regional failover
It may seem like a daunting task to design a system to run in multiple regions. This is understandable from a traditional, monolithic system point of view. Before the cloud, we did not always expect our systems to run in multiple data centers. We designed them to run in one on-premises data center at a time. The presentation tier assumed that a network connection would always be available. The data tier required strong consistency. Deployments where not automated, so failover was manual and usually involved a time consuming restore of the database from the latest backup. Refitting these brittle systems to run in multiple cloud regions was painful. A primary/cold-secondary topology was usually the best we could achieve.In contract, we have been designing our serverless systems to run in multiple regions all along. In todays highly connect world, the context and expectations for our systems is very different. We have reinvented our system architecture and...