Purpose: We need a single (or couple) metrics that will allow us to know if, generally, our developers are experiencing longer wait times before merges/test completion.
Where it will be used: This will hopefully inform the size of our pool of nodepool instances. We'll know when we need to increase it or if we can safely decrease it to save wmflabs capacity. This should be able to tell us something like "If we reduce the pool by 5 there will be no user-noticeable impact except for during our 3 busiest hours of the day." Or, conversely, we'll be able to tell how much developer time we saved by increasing our pool size (by comparing weekly data pre/post the change).
See also: