We need to be able to measure when the server is reaching capacity and will soon start to run into error scenarios.

This needs to be measurable and specific to the environment the user is running under.

Eg if we know in a lab environment with powerful hardware that we can support X many agents and Y many requests/minute, we cannot just this number. Users will be running our applications on less powerful hardware, on virtual machines with slow io.

We might not be able to pin point what the exact bottleneck is (eg you need to run with a faster CPU, or your disk speed is too slow) but we should at least be able to measure if we are going to soon run into problems.