Compute Pool Utilization

CPU and memory utilization for one compute pool, hour by hour.

Hourly by default because that is the grain a sizing decision is made at: a daily mean erases the short bursts that prove an application is in use. day exists for windows too long to render at hour grain.

A bucket with no telemetry is absent from data, and a metric the agent never served is null, not zero -- coverage.has_cpu_telemetry / has_mem_telemetry say which case applies. Low utilization here is also not on its own evidence a pool is unused: an SPCS container can render an application at near-zero CPU while the substantive SQL runs on a warehouse.

Ratios are a fraction of the capacity in that bucket. A pool autoscales, so capacity moves between buckets and a ratio is not comparable across them on its own; cpu_capacity_cores / mem_capacity_bytes travel with every point so a consumer can recover the absolute value.