Container group running and talking to nobody

Why it happens

Azure Container Instances bill for the vCPU and memory a group requests, for every second it runs, and a group whose restart policy is Always runs until somebody stops it. One left behind after a demo or a migration therefore bills continuously while serving nobody. The rule reports a group that received zero bytes and transmitted zero bytes for the whole window. Traffic is the signal rather than processor use, because the processor metric is published in millicores rather than as a percentage and deciding what counts as almost none would need a threshold nobody has published: zero bytes moved is a fact and low processor use is an opinion, so the processor peak travels on the finding as evidence for a person rather than as part of the test. Whether the group was running is taken from the presence of metric samples rather than from its instance view, because a group that is not running emits nothing at all and the instance view is a field Resource Graph often does not carry. Only groups set to restart Always are considered, which is the policy Azure documents for a long running service; a task set to run once reaches a terminal state and stops billing without anyone doing anything. The saving is the requested vCPU and memory at the standard hourly meters, and requests are what Azure charges for whether or not the container uses them. One thing this cannot see is on every finding: a job that computes locally and talks to nothing looks exactly like an abandoned service over any window.

What the agent does about it

A container group with a restart policy of Always that has moved no network traffic in either direction across the window, while billing by the second for the vCPU and memory it requested.