SignalR service nobody connects to

Why it happens

Azure SignalR bills per unit per day whether or not a client ever connects, so an instance left behind after a prototype keeps charging for capacity nobody reaches. The rule reads ConnectionCount, whose documented default aggregation is Maximum, and reports an instance whose PEAK concurrent connections across the whole window did not exceed the threshold. The peak is the right test rather than the average: an average of zero can hide a burst, while a maximum of zero means no client connected at any minute of the window. The units are what Azure charges for, not the connections, so the saving is the instance sku capacity at the per-unit meter for its tier. Those meters are published per day rather than per hour, which is a twenty-four fold difference if read as hourly, and the free tier is excluded because it bills nothing. What this cannot see travels on every finding: an instance provisioned ahead of a launch, or one serving a seasonal workload, looks exactly like an abandoned one over any window, which is why it reports and a person decides.

What the agent does about it

A Standard or Premium SignalR instance whose highest concurrent connection count across the window was at or below the threshold, while its units bill every day.