Example architectures

The rest of the architecture section describes patterns one at a time. This section works through designs where several of them combine to meet a stated requirement, and — more useful — where the requirement rules some of them out.

Global Accelerator for latency-sensitive multi-Region workloads covers applications deployed in many Regions where latency differences are unacceptable and the transport may not be HTTP. It is the case where DNS-based failover is not good enough and anycast entry points are.

SNS fan-out with per-subscriber message filtering delivers one event to several independent consumers simultaneously, each buffered by its own queue and filtered to the event types it handles. It also documents the filter-scope default that makes this pattern fail silently.

Analytics and observability for an event-driven system is a larger design: capturing everything an event-driven workload produces, separating operational signals from business records, and serving both engineers and business users without building a warehouse first.

A further worked example — a highly available TCP service with static IP addresses, compared five ways — sits with the availability material in high availability.