❤

Topology Spread Constraints.

Plan your pod placement across the cluster with ease.

Control how pods are spread across your cluster.

Specify the spread and how the pods should be placed across the cluster.

Let us see how the template looks like..

example-template.yaml

---
apiVersion: v1
kind: Pod
metadata:
  name: example-pod
spec:
  # Configure a topology spread constraint
  topologySpreadConstraints:
    - maxSkew: <integer>
      minDomains: <integer> # optional; beta since v1.25
      topologyKey: <string>
      whenUnsatisfiable: <string>
      labelSelector: <object>
      matchLabelKeys: <list> # optional; beta since v1.27
      nodeAffinityPolicy: [Honor|Ignore] # optional; beta since v1.26
      nodeTaintsPolicy: [Honor|Ignore] # optional; beta since v1.26
  ### other Pod fields go here

• maxSkew: degree of uneveness of the pod placement.
• topologyKey: labels that all pods should have (region=regionA) kind of like a lable key.
• whenUnsatisfiable: what to do when the constraints are not satisfied, either: schedule or donotschedule.
• labelSelector: helps in keeping track of the other pods on node.
• matchLabelKeys: match additional labels in the other pods.
• nodeAffinityPolicy: honor the policy or ignore it.
• nodeTaintsPolicy: honor the policy or ignore it.

Single pod on all the nodes

So we have a 4 node cluster with 1 control plane node.

Let us remove the default taint and all the taints from all the nodes.

Essentially we have now 4 nodes where pods can be scheduled normally.

Let’s divide these 4 nodes in to two groups :-

zone1 and zone2

So we will use labels to do this.

kubectl label node db1 zone=zone1
kubectl label node db2 zone=zone1

kubectl label node db3 zone=zone2
kubectl label node db4 zone=zone2

a little fuckup.

The pods are not being scheduled on the control plane node.

though the pods are scheduled on the worker nodes evenly.

Let’s divide these 3 nodes in to two groups :-

zone1 and zone2

So we will use labels to do this.

kubectl label node db2 zone=zone1
kubectl label node db3 zone=zone2
kubectl label node db4 zone=zone2

topology.yaml

kind: Pod
apiVersion: v1
metadata:
  name: mypod
  labels:
    app: mypod 
spec:
  topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: zone
    whenUnsatisfiable: DoNotSchedule
    labelSelector:
      matchLabels:
        app: mypod
  containers:
  - name: pause
    image: registry.k8s.io/pause:3.1

So I created 6 similar pod files with diffent name as mypod1 and label for creating 6 similar pods.

so we can see that 3 nodes are there in the db2 node while 3 are spread in db3 and db4 nodes.

query: topology

ref: https://kubernetes.io/docs/concepts/scheduling-eviction/topology-spread-constraints/