Each Coordinator can have 0-3 backends running as it abstracts over {stable,beta,nightly} and each backend is lazily created and idled.
The "good" failure case was that we could be off by a factor of 3 in the upper bound. The catastrophic failure case was for the WebSocket connections where there are thousands of Coordinators idling with very few of them actively compiling at a time. Setting a low limit would prevent connections, setting a high limit was basically pointless.
Each
Coordinatorcan have 0-3 backends running as it abstracts over {stable,beta,nightly} and each backend is lazily created and idled.The "good" failure case was that we could be off by a factor of 3 in the upper bound. The catastrophic failure case was for the WebSocket connections where there are thousands of
Coordinators idling with very few of them actively compiling at a time. Setting a low limit would prevent connections, setting a high limit was basically pointless.