As part of Hypershift perf&scale exercise, we run cluster-density workloads which are control-plane intensive as well as KAS, we measure latency metrics to understand their performances. 
Surprisingly, the api latency metrics like all apiserver_request_duration_seconds_bucket are consistenlty at 990ms even during idle as well as during heavy operation. we suspect its not reporting a right state.

This is the query we used to record the linked snapshot,
histogram_quantile(0.99, sum(rate(apiserver_request_duration_seconds_bucket{namespace=~"$namespace",resource=~"$resource",subresource!="log",verb!~"WATCH|WATCHLIST|PROXY"}[2m])) by(verb,le))

4.14.0-rc.6

Always

1. Deploy a HC
2. watch apiserver_request_duration_seconds_bucket
3. They are constant at 990ms

Latency is constant, although we see spikes during workload.

During idle we don't expect 990ms latency for GET, POST, PATHC, APPLY, DELETE calls