Pod 与 StatefulSet 中不同节点的亲和性

Question

I am creating StatefulSets and I want pods within one StatefulSet to be distributed across different nodes of the k8s cluster. In my case - one StatefulSet is one database replicaset.

sts.Spec.Template.Labels["mydb.io/replicaset-uuid"] = replicasetUUID.String()
sts.Spec.Template.Spec.Affinity.PodAntiAffinity = &corev1.PodAntiAffinity{
    RequiredDuringSchedulingIgnoredDuringExecution: []corev1.PodAffinityTerm{
        {
            LabelSelector: &metav1.LabelSelector{
                MatchExpressions: []metav1.LabelSelectorRequirement{
                    {
                        Key:      "mydb.io/replicaset-uuid",
                        Operator: metav1.LabelSelectorOpIn,
                        Values:   []string{replicasetUUID.String()},
                    },
                },
            },
            TopologyKey: "kubernetes.io/hostname",
        },
    },
}

However, with these settings, I get the opposite. storage-0-0 and storage-0-1 are on the same replicaset and on the same node...

Moreover, they have exactly the same label mydb.io/replicaset-uuid

$ kubectl -n mydb get pods -o wide
NAME                             READY   STATUS    RESTARTS   AGE   IP        NODE                           NOMINATED NODE   READINESS GATES
storage-0-0                      1/1     Running   0          40m   x.x.x.x   kubernetes-cluster-x-main-0    <none>           <none>
storage-0-1                      1/1     Running   0          39m   x.x.x.x   kubernetes-cluster-x-main-0    <none>           <none>
storage-1-0                      1/1     Running   0          40m   x.x.x.x   kubernetes-cluster-x-slave-0   <none>           <none>
storage-1-1                      1/1     Running   0          40m   x.x.x.x   kubernetes-cluster-x-slave-0   <none>           <none>
mydb-operator-58c9bfbb9b-7djml   1/1     Running   0          46m   x.x.x.x   kubernetes-cluster-x-slave-0   <none>           <none>

Answer 1

I suggest using an podAntiAffinity rule in the statefulset definition to deploy your application so that no two instances are located on the same host.

Reference: An example of a pod that uses pod affinity

Answer 2

It works correctly, as @jesmart wrote in the comment:

The description of the problem works correctly I just indicated the wrong image with the application