Disable kube-controller-manager client-side rate limiting

Author: timebertt

docs/evaluation.md

@@ -125,6 +125,7 @@ In the cluster itself, kyverno policies are used for scheduling the sharder and
 This makes sure that these components run on machines isolated from other system components and don't content for compute resources during load tests.
 
 Furthermore, kyverno policies are added to the control plane to ensure a static size of etcd, kube-apiserver, and kube-controller-manager (requests=limits for guaranteed resources, disable vertical autoscaling, 4 replicas of kube-apiserver and disable horizontal autoscaling).
+Also, kube-controller-manager's client-side rate limiting is disabled (ref https://github.com/timebertt/kubernetes-controller-sharding/pull/610, [SIG api-machinery recommendation](https://kubernetes.slack.com/archives/C0EG7JC6T/p1680889646346859?thread_ts=1680791299.631439&cid=C0EG7JC6T)) and HTTP/2 is disabled so that API requests are distributed across API server instances (ref https://github.com/gardener/gardener/issues/8810).
 This is done to make load test experiments more stable and their results more reproducible.
 
 ## Measurements

hack/config/policy/controlplane/kube-controller-manager.yaml

@@ -46,8 +46,8 @@ spec:
         spec:
           updatePolicy:
             updateMode: Off
-  # increases kube-controller-manager's client-side rate limits to speed up garbage collection after executing load tests
-  - name: increase-rate-limits
+  # disable kube-controller-manager's client-side rate limits similar to webhosting-operator
+  - name: disable-rate-limits
     match:
       any:
       - resources:
@@ -59,10 +59,7 @@ spec:
       patchesJson6902: |-
         - op: add
           path: /spec/template/spec/containers/0/command/-
-          value: "--kube-api-qps=2000"
-        - op: add
-          path: /spec/template/spec/containers/0/command/-
-          value: "--kube-api-burst=2200"
+          value: "--kube-api-qps=-1"
   # disable HTTP2 in kube-controller-manager's so that API requests are distributed across API server instances
   - name: disable-http2
     match: