Skip to content

Latest commit

 

History

History
151 lines (115 loc) · 5.63 KB

assign-pod-to-node.md

File metadata and controls

151 lines (115 loc) · 5.63 KB

Assign Pod to Node

Node Selector

You can configure pod to run on particular node by using node label along with nodeSelector

  • Deploy backend application and scale to 5 pods

    oc new-project demo
oc apply -f manifests/backend-v1.yaml -n demo
oc scale deployment backend-v1 --replicas=5 -n demo

  • Check that 5 pods is running on which worker nodes

    watch oc get pods -o 'custom-columns=Name:.metadata.name,PHASE:.status.phase,NODE:.spec.nodeName' -n demo

    Sample output

      Name                        PHASE     NODE
  backend-v1-b564ff66-6spph   Running   ip-10-0-XXX-179.us-east-2.compute.internal
  backend-v1-b564ff66-dh77c   Running   ip-10-0-XXX-101.us-east-2.compute.internal
  backend-v1-b564ff66-dtkjn   Running   ip-10-0-XXX-179.us-east-2.compute.internal
  backend-v1-b564ff66-n7kr4   Running   ip-10-0-XXX-121.us-east-2.compute.internal
  backend-v1-b564ff66-thhqv   Running   ip-10-0-XXX-121.us-east-2.compute.internal

  • Let's say we want to run backend app is belong to project app-x and project app-x need to run on node 179 only then we'll label node 179 with label app-x

    oc label node ip-10-0-XXX-179.us-east-2.compute.internal app-x=''

  • Add nodeSelector to backend-v1 deployment Remark:

    • You can use another shell/terminal to check status of backend-v1's pods with previously watch command

    • sed is use for set replicas to 5.

      cat manifests/backend-v1-with-node-selector.yaml | sed s/replicas:\ 1/replicas:\ 5/g | oc apply -n demo -f -

      Sample output from watch command

      Name                          PHASE     NODE
backend-v1-786cd59759-bcmpx   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-786cd59759-dkgxp   Pending   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-786cd59759-lpnwq   Pending   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-786cd59759-vlqp2   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-786cd59759-x4nlb   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-b564ff66-dh77c     Running   ip-10-0-XXX-101.us-east-2.compute.internal
backend-v1-b564ff66-thhqv     Running   ip-10-0-XXX-121.us-east-2.compute.internal

      When finished all pods will be run on node 179

      Name                          PHASE     NODE
backend-v1-786cd59759-bcmpx   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-786cd59759-dkgxp   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-786cd59759-lpnwq   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-786cd59759-vlqp2   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-786cd59759-x4nlb   Running   ip-10-0-XXX-179.us-east-2.compute.internal

  • Check backend-v1 deployment with nodeAffinity under template

    spec:
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: app-x
            operator: Exists

Taints and Tolerations

Taint is proporty of nodes that allow only pod with toleration can be scheduled to run on that particular node.

  • Let's say we want to allow only project app-x to run on node 179 then we will taint node 179

    oc adm taint node ip-10-0-248-179.us-east-2.compute.internal app-x='':NoSchedule

  • Taint is not effect pods those already scheduled then we will force backend-v1 to create all pods again

    oc delete pods --all -n demo

    Pods cannot schedule because there is no toleration in deployment

    Name                          PHASE     NODE
backend-v1-786cd59759-79jmh   Pending   <none>
backend-v1-786cd59759-gnwg7   Pending   <none>
backend-v1-786cd59759-lc4jr   Pending   <none>
backend-v1-786cd59759-lgk8r   Pending   <none>
backend-v1-786cd59759-vcd22   Pending   <none>

    Check Events in OpenShift Console

  • Add toleration to backend-v1 deployment

    cat manifests/backend-v1-with-node-selector-and-toleration.yaml|sed s/replicas:\ 1/replicas:\ 5/g | oc apply -n demo -f -

    When finished all pods will be run on node 179

    Name                          PHASE     NODE
backend-v1-647db46bfc-868w6   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-647db46bfc-krzvs   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-647db46bfc-nssqz   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-647db46bfc-qhvws   Running   ip-10-0-XXX-179.us-east-2.compute.internal
backend-v1-647db46bfc-vnl5l   Running   ip-10-0-XXX-179.us-east-2.compute.internal

  • Check backend-v1 deployment with nodeAffinity under template

      tolerations:
  - effect: NoSchedule
    key: app-x
    operator: Exists

Remove node label and toleration

  • Following commands show how to remove node label and toleration

    oc label node ip-10-0-XXX-179.us-east-2.compute.internal app-x-
oc adm taint nodes ip-10-0-XXX-179.us-east-2.compute.internal app-x='':NoSchedule-