Note_Tech

All technological notes.

---

Project maintained by simonangel-fong Hosted on GitHub Pages — Theme by mattgraham

Kubernetes: DaemonSets - Rollout & Rollback

Back

• Kubernetes: DaemonSets - Rollout \& Rollback
  • Imperative Command
  • Update Strategy
  • Update Strategy - RollingUpdate
    • Lab: Update Strategy
  • Update Strategy - OnDelete
    • Lab: OnDelete strategy

---

Imperative Command

Update Strategy

• When pod template gets updated, ds controller automatically deletes the old Pods and creates the new pods

• spec.updateStrategy field

  • RollingUpdate
  • OnDelete

---

Update Strategy - RollingUpdate

• RollingUpdate:
  • default
  • Pods are replaced one by one.
  • the controller waits until the new Pod is ready before updating the Pods on the other Nodes.
  • parameters:
    • maxSurge: 0
      • should be zero
      • when it > 0, it allow more than one ds running on a node, which is not supported, resulting in dead lock.
    • maxUnavailable
      • recommended 1;
      • only one Node is affected if the ds pod does not start.
    • maxSurge:0;maxUnavailable>node_number
      • act like Recreate; replace all ds
      • if the ds pod template has readiness probe but it fails, these parameters are ignored.

---

Lab: Update Strategy

# demo-ds-rollingupdate.yaml
apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: demo-ds-rollingupdate
spec:
  minReadySeconds: 30
  updateStrategy:
    type: RollingUpdate
    rollingUpdate:
      maxSurge: 0
      maxUnavailable: 1
  selector:
    matchLabels:
      app: monitor
  template:
    metadata:
      labels:
        app: monitor
    spec:
      containers:
        - name: monitor
          image: busybox
          command:
            - sleep
            - infinity

kubectl apply -f demo-ds-rollingupdate.yaml
# daemonset.apps/demo-ds-rollingupdate created

# update image
kubectl set image ds demo-ds-rollingupdate monitor=busybox:1.36
# daemonset.apps/demo-ds-rollingupdate image updated

# confirm:
#   update one by one
#   avaiable every 30s
kubectl get ds -w
# NAME                    DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
# demo-ds-rollingupdate   2         2         2       2            2           <none>          86s
# demo-ds-rollingupdate   2         2         2       2            2           <none>          102s
# demo-ds-rollingupdate   2         2         2       0            2           <none>          102s
# demo-ds-rollingupdate   2         2         1       1            1           <none>          2m12s
# demo-ds-rollingupdate   2         2         2       1            1           <none>          2m15s
# demo-ds-rollingupdate   2         2         2       1            2           <none>          2m45s
# demo-ds-rollingupdate   2         1         1       1            1           <none>          3m16s
# demo-ds-rollingupdate   2         2         1       2            1           <none>          3m16s
# demo-ds-rollingupdate   2         2         2       2            1           <none>          3m20s
# demo-ds-rollingupdate   2         2         2       2            2           <none>          3m50s

---

Update Strategy - OnDelete

• OnDelete:
  • performs the update in a semiautomatic way
  • ds controller waits for manual Pod deletion, and then replaces it with a new Pod
    • can replace Pods at your own pace.
  • used to update cluster-critical Pods with much more control

---

Lab: OnDelete strategy

# demo-ds-ondelete.yaml
apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: demo-ds-ondelete
spec:
  minReadySeconds: 30
  updateStrategy:
    type: OnDelete # strategy
  selector:
    matchLabels:
      app: monitor
  template:
    metadata:
      labels:
        app: monitor
    spec:
      containers:
        - name: monitor
          image: busybox
          command:
            - sleep
            - infinity

kubectl apply -f demo-ds-ondelete.yaml
# daemonset.apps/demo-ds-ondelete created

kubectl get ds
# NAME               DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
# demo-ds-ondelete   2         2         2       2            2           <none>          41s

kubectl set image ds demo-ds-ondelete monitor=busybox:1.36
# daemonset.apps/demo-ds-ondelete image updated

no ds get updated unless manually gets deleted

kubectl get pod
# NAME                     READY   STATUS    RESTARTS   AGE
# demo-ds-ondelete-4qvl6   1/1     Running   0          99s
# demo-ds-ondelete-wxlnm   1/1     Running   0          99s

# manually delete
kubectl delete po demo-ds-ondelete-4qvl6 --wait=false
# pod "demo-ds-ondelete-4qvl6" deleted

# confirm
kubectl get ds -w
# NAME               DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
# demo-ds-ondelete   2         2         2       2            2           <none>          96s
# demo-ds-ondelete   2         2         2       2            2           <none>          108s
# demo-ds-ondelete   2         2         2       0            2           <none>          108s
# demo-ds-ondelete   2         2         1       1            1           <none>          3m44s
# demo-ds-ondelete   2         2         2       1            1           <none>          3m48s
# demo-ds-ondelete   2         2         2       1            2           <none>          4m18s

# confirm
kubectl get pod
# NAME                     READY   STATUS    RESTARTS   AGE
# demo-ds-ondelete-bd9lg   1/1     Running   0          99s
# demo-ds-ondelete-wxlnm   1/1     Running   0          5m23s

# manually delete another one
kubectl delete po demo-ds-ondelete-wxlnm --wait=false
# pod "demo-ds-ondelete-wxlnm" deleted

# confirm
kubectl get ds -w
# NAME               DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR   AGE
# demo-ds-ondelete   2         2         1       2            1           <none>          7m17s
# demo-ds-ondelete   2         2         2       2            1           <none>          7m21s
# demo-ds-ondelete   2         2         2       2            2           <none>          7m51s

---