v0.10 to v1.0

Follow the Regular Upgrading Process.

Upgrading Notable Changes

Introduced karmada-aggregated-apiserver component

In the releases before v1.0.0, we are using CRD to extend the Cluster API, and starts v1.0.0 we use API Aggregation(AA) to extend it.

Based on the above change, perform the following operations during the upgrade:

Step 1: Stop karmada-apiserver

You can stop karmada-apiserver by updating its replica to 0.

Step 2: Remove Cluster CRD from ETCD

Remove the Cluster CRD from ETCD directly by running the following command.

etcdctl --cert="/etc/kubernetes/pki/etcd/karmada.crt" \
--key="/etc/kubernetes/pki/etcd/karmada.key" \
--cacert="/etc/kubernetes/pki/etcd/server-ca.crt" \
del /registry/apiextensions.k8s.io/customresourcedefinitions/clusters.cluster.karmada.io

Note: This command only removed the CRD resource, all the CR (Cluster objects) not changed. That's the reason why we don't remove CRD by karmada-apiserver.

Step 3: Prepare the certificate for the karmada-aggregated-apiserver

To avoid CA Reusage and Conflicts, create CA signer and sign a certificate to enable the aggregation layer.

Update karmada-cert-secret secret in karmada-system namespace:

apiVersion: v1
kind: Secret
metadata:
  name: karmada-cert-secret
  namespace: karmada-system
type: Opaque
data:
  ...
+  front-proxy-ca.crt: |
+    {{front_proxy_ca_crt}}
+  front-proxy-client.crt: |
+    {{front_proxy_client_crt}}
+  front-proxy-client.key: |
+    {{front_proxy_client_key}}

Then update karmada-apiserver deployment's container command:

-            - --proxy-client-cert-file=/etc/kubernetes/pki/karmada.crt
-            - --proxy-client-key-file=/etc/kubernetes/pki/karmada.key
+            - --proxy-client-cert-file=/etc/kubernetes/pki/front-proxy-client.crt
+            - --proxy-client-key-file=/etc/kubernetes/pki/front-proxy-client.key
-            - --requestheader-client-ca-file=/etc/kubernetes/pki/server-ca.crt
+            - --requestheader-client-ca-file=/etc/kubernetes/pki/front-proxy-ca.crt

After the update, restore the replicas of karmada-apiserver instances.

Step 4: Deploy karmada-aggregated-apiserver:

Deploy karmada-aggregated-apiserver instance to your host cluster by following manifests:

unfold me to see the yaml

---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: karmada-aggregated-apiserver
  namespace: karmada-system
  labels:
    app: karmada-aggregated-apiserver
    apiserver: "true"
spec:
  selector:
    matchLabels:
      app: karmada-aggregated-apiserver
      apiserver: "true"
  replicas: 1
  template:
    metadata:
      labels:
        app: karmada-aggregated-apiserver
        apiserver: "true"
    spec:
      automountServiceAccountToken: false
      containers:
        - name: karmada-aggregated-apiserver
          image: swr.ap-southeast-1.myhuaweicloud.com/karmada/karmada-aggregated-apiserver:v1.0.0
          imagePullPolicy: IfNotPresent
          volumeMounts:
            - name: k8s-certs
              mountPath: /etc/kubernetes/pki
              readOnly: true
            - name: kubeconfig
              subPath: kubeconfig
              mountPath: /etc/kubeconfig
          command:
            - /bin/karmada-aggregated-apiserver
            - --kubeconfig=/etc/kubeconfig
            - --authentication-kubeconfig=/etc/kubeconfig
            - --authorization-kubeconfig=/etc/kubeconfig
            - --karmada-config=/etc/kubeconfig
            - --etcd-servers=https://etcd-client.karmada-system.svc.cluster.local:2379
            - --etcd-cafile=/etc/kubernetes/pki/server-ca.crt
            - --etcd-certfile=/etc/kubernetes/pki/karmada.crt
            - --etcd-keyfile=/etc/kubernetes/pki/karmada.key
            - --tls-cert-file=/etc/kubernetes/pki/karmada.crt
            - --tls-private-key-file=/etc/kubernetes/pki/karmada.key
            - --audit-log-path=-
            - --feature-gates=APIPriorityAndFairness=false
            - --audit-log-maxage=0
            - --audit-log-maxbackup=0
          resources:
            requests:
              cpu: 100m
      volumes:
        - name: k8s-certs
          secret:
            secretName: karmada-cert-secret
        - name: kubeconfig
          secret:
            secretName: kubeconfig
---
apiVersion: v1
kind: Service
metadata:
  name: karmada-aggregated-apiserver
  namespace: karmada-system
  labels:
    app: karmada-aggregated-apiserver
    apiserver: "true"
spec:
  ports:
    - port: 443
      protocol: TCP
      targetPort: 443
  selector:
    app: karmada-aggregated-apiserver

Then, deploy APIService to karmada-apiserver by following manifests.

unfold me to see the yaml

apiVersion: apiregistration.k8s.io/v1
kind: APIService
metadata:
  name: v1alpha1.cluster.karmada.io
  labels:
    app: karmada-aggregated-apiserver
    apiserver: "true"
spec:
  insecureSkipTLSVerify: true
  group: cluster.karmada.io
  groupPriorityMinimum: 2000
  service:
    name: karmada-aggregated-apiserver
    namespace: karmada-system
  version: v1alpha1
  versionPriority: 10
---
apiVersion: v1
kind: Service
metadata:
  name: karmada-aggregated-apiserver
  namespace: karmada-system
spec:
  type: ExternalName
  externalName: karmada-aggregated-apiserver.karmada-system.svc.cluster.local

Step 5: check clusters status

If everything goes well, you can see all your clusters just as before the upgrading.

kubectl get clusters

karmada-agent requires an extra impersonate verb

In order to proxy user's request, the karmada-agent now request an extra impersonate verb. Please check the ClusterRole configuration or apply the following manifest.

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: karmada-agent
rules:
  - apiGroups: ['*']
    resources: ['*']
    verbs: ['*']
  - nonResourceURLs: ['*']
    verbs: ["get"]

MCS feature now supports Kubernetes v1.21+

Since the discovery.k8s.io/v1beta1 of EndpointSlices has been deprecated in favor of discovery.k8s.io/v1, in Kubernetes v1.21, Karmada adopt this change at release v1.0.0. Now the MCS feature requires member cluster version no less than v1.21.