Container App holding replicas warm with no traffic

Why it happens

Scaling to zero is the reason to choose Container Apps, and minReplicas defaults to zero. Setting it higher opts out, and Microsoft prices that opt-out explicitly: a revision configured above zero and sitting at its minimum bills every replica at the idle rate, while a revision scaled to zero incurs no resource charges at all. The idle discount applies to vCPU only, an eighth of the active rate, and memory idles at exactly the active rate, so on a small replica the memory half is the larger one. The rule fires only where the remedy would work: Flexible profiles cannot scale to zero at all, Dedicated plans bill per workload-profile instance rather than per app so zero replicas would save nothing, and an app with no ingress or with a queue-driven scale rule is not an HTTP app, which means a zero request count says the rule is measuring the wrong thing rather than that the app is idle. Azure also grants a monthly free allowance per subscription that the agent cannot apportion between apps, so the figure is reported gross with the allowance named beside it. Scaling to zero reintroduces a cold start that Microsoft publishes no SLA for, which is a real reason a team set the minimum above zero on purpose.

What the agent does about it

A Consumption-plan Container App with minReplicas above zero that served no HTTP requests in the window, so it paid idle charges to stay warm for traffic that never came.