When a problem occurs in a running system, the first question we always ask is the same. "What's wrong right now, and how should we have known?" The entire field of monitoring is actually an attempt to automate these two sentences. AWS CloudWatch is a service that places this automation on top of a single abstraction: "everything ultimately becomes a time-series number (metric)". CPU utilization, Lambda invocation count, error counts extracted from logs, and even custom business metrics — everything inside CloudWatch reduces to the same form of data point: (timestamp, value, dimensions)