The proxy exposes a process_cpu_seconds_total metric. It also exposes a process_start_time_seconds metric. But given a random metrics snapshot, we have no way of knowing the proxy's uptime unless we know the current time at the time of the snapshot.

I propose adding a process_uptime_seconds_total metric that reflects the number of (fractional) seconds since the proxy started. This can help signal, for instance, if a proxy's CPU usage is low (i.e., if the cpu seconds is much smaller than the uptime value).