- A: the timestamp a particular filter starts executing for a request.
- B: the timestamp the particular filter has completed and the next handler
in the chain has started executing.
We define filter latency as `B - A`. Introduce a new metric to track filter latency:
"apiserver_request_filter_duration_seconds"
This measures request filter latency distribution in seconds, for each filter type.
Kubernetes-commit: 99df0b15816cd59ed548235f33c7ab2add940eba