avoid kindnet network policies

kindnet now implements network policies that can
break the signal from the e2e tests.
Just removing the serviceaccount avoids the agents can receive the
information necessary, this is better than removing the kindnet ds
entirely because it taints the nodes as not ready.

Author: aojea

.github/workflows/e2e.yml

@@ -125,6 +125,8 @@ jobs:
         /usr/local/bin/kind load docker-image registry.k8s.io/networking/kube-network-policies:test --name ${{ env.KIND_CLUSTER_NAME}}
         sed -i s#registry.k8s.io/networking/kube-network-policies.*#registry.k8s.io/networking/kube-network-policies:test# install.yaml
         /usr/local/bin/kubectl apply -f ./install.yaml
+        # stop kindnet of checking network policies
+        kubectl -n kube-system delete clusterrolebinding kindnet
 
     - name: Get Cluster status
       run: |

.github/workflows/npa.yml

@@ -116,6 +116,8 @@ jobs:
         /usr/local/bin/kind load docker-image registry.k8s.io/networking/kube-network-policies:test --name ${{ env.KIND_CLUSTER_NAME}}
         sed -i s#registry.k8s.io/networking/kube-network-policies.*#registry.k8s.io/networking/kube-network-policies:test# install-anp.yaml
         /usr/local/bin/kubectl apply -f ./install-anp.yaml
+        # stop kindnet of checking network policies
+        kubectl -n kube-system delete clusterrolebinding kindnet
 
     - name: Get Cluster status
       run: |

tests/setup_suite.bash

@@ -34,6 +34,9 @@ EOF
   printf '%s' "${_install}" | kubectl apply -f -
   kubectl wait --for=condition=ready pods --namespace=kube-system -l k8s-app=kube-network-policies
 
+  # stop kindnet of checking network policies
+  kubectl -n kube-system delete clusterrolebinding kindnet
+  
   # Expose a webserver in the default namespace
   kubectl run web --image=httpd:2 --labels="app=web" --expose --port=80