Skip to content

Latest commit

 

History

History
163 lines (127 loc) · 4.44 KB

README.adoc

File metadata and controls

163 lines (127 loc) · 4.44 KB
Raw

Automated Placement

This example illustrates how the placement of scheduled pods can be influenced. For these examples, we assume that you have Minikube installed and running as described in here.

Node Selector

The simplest way to influence the scheduling process is to use a nodeSelector.

Apply our simple example random-generator application with a node selector:

kubectl create -f https://k8spatterns.io/AutomatedPlacement/node-selector.yml

You will notice that this Pod does not get scheduled because the nodeSelector can’t find any node with the label disktype=ssd:

kubectl describe pod node-selector
....
Events:
  Type     Reason            Age              From               Message
  ----     ------            ----             ----               -------
  Warning  FailedScheduling  8s (x2 over 8s)  default-scheduler  0/1 nodes are available: 1 node(s) didnt match node selector.

Let’s change this:

kubectl label node minikube disktype=ssd
kubectl get pods
NAME                READY   STATUS    RESTARTS   AGE
random-generator    1/1     Running   0          65s

Node Affinity

Let’s now use Node affinity rules for scheduling our Pod:

kubectl create -f https://k8spatterns.io/AutomatedPlacement/node-affinity.yml

Again, our Pod will only schedule as no node fulfills the affinity rules. We can change this with

kubectl label node minikube numberCores=4

Does the Pod start up now? What if you choose two instead of 4 for the number of cores?

Pod Affinity

To test Pod affinity, we need to install a Pod to connect our Pod. We are trying to create both Pods with

kubectl create -f https://k8spatterns.io/AutomatedPlacement/pod-affinity.yml
kubectl get pods
NAME                READY   STATUS    RESTARTS   AGE
confidential-high   1/1     Running   0          22s
pod-affinity        0/1     Pending   0          22s

"confidential-high" is a placeholder pod with a label matched by our "pod-affinity" Pod. However, our node still needs to get the proper topology key. That can be changed with

kubectl label --overwrite node minikube security-zone=high
kubectl get pods
NAME                READY   STATUS    RESTARTS   AGE
confidential-high   1/1     Running   0          9m39s
pod-affinity        1/1     Running   0          9m39s

Taints and Tolerations

For testing taints and tolerations, we first have to taint our Minikube node so that, by default, no Pods are scheduled on it:

kubectl taint nodes minikube node-role.kubernetes.io/master="":NoSchedule

You can check that this taint works by reapplying the previous pod-affinity.yml example and seeing that the confidential-high Pod is not scheduled.

kubectl delete -f https://k8spatterns.io/AutomatedPlacement/pod-affinity.yml
kubectl create -f https://k8spatterns.io/AutomatedPlacement/pod-affinity.yml
kubectl get pods
NAME                READY   STATUS    RESTARTS   AGE
confidential-high   0/1     Pending   0          2s
pod-affinity        0/1     Pending   0          2s

But our Pod in tolerations.yml can be scheduled as it tolerates this new taint on Minikube:

kubectl create -f https://k8spatterns.io/AutomatedPlacement/tolerations.yml
kubectl get pods
NAME                READY   STATUS    RESTARTS   AGE
confidential-high   0/1     Pending   0          2m51s
pod-affinity        0/1     Pending   0          2m51s
tolerations         1/1     Running   0          4s