Take for example a load balancer that is configured to send traffic to the fastest server at the time. Makes sense, if a server is fast it has more available capacity to handle requests. But when the front end code fails the server short circuits, skipping all the expensive code as it unwinds the stack and returns a 5XX error, the load balancer that just sends to the fastest now sends ALL traffic to the failing node.
The antipattern is a feedback loop that is not measuring the right thing. In this case, they should be timing _successful_ calls to the front end.
Take for example a load balancer that is configured to send traffic to the fastest server at the time. Makes sense, if a server is fast it has more available capacity to handle requests. But when the front end code fails the server short circuits, skipping all the expensive code as it unwinds the stack and returns a 5XX error, the load balancer that just sends to the fastest now sends ALL traffic to the failing node.
The antipattern is a feedback loop that is not measuring the right thing. In this case, they should be timing _successful_ calls to the front end.