Rolling window average on streaming data

Numbers arrive one at a time. Describe how to maintain the average of the most recent $n$ values efficiently as each new number arrives.

Keep a running sum and a circular buffer of the last $n$ values. On each arrival add the new value to the sum, subtract the value it overwrites in the buffer, and divide by $n$ (or by the count so far during warm-up). Each update is $O(1)$ time with $O(n)$ memory, versus $O(n)$ per update if you re-sum the window; periodically recompute the sum to limit floating-point drift.