CODEWITHSUNDEEP
kubernetesk3d
jasonlg3d
jasonlg3d

Reputation: 504

Kubernetes Weighted Node Affinity always prioritized over higher weight pod affinity

I have the following affinity defined on a pod. When this pod gets scheduled, it is always respecting the NodeAffinity first. Even if both pod affinities are satisfied, they are ignored. Are weights between node and pod affinity always ignored and node affinity always wins? or am I doing something else wrong?

K3d v5.7.5

Kubernetes Version: v1.30.6+k3s1

   spec:
      affinity:
        nodeAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - preference:
              matchExpressions:
              - key: example.com/capacity-remaining
                operator: Gt
                values:
                - "2457600"
            weight: 1
        podAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - podAffinityTerm:
              labelSelector:
                matchExpressions:
                - key: example.com/lead-pod
                  operator: In
                  values:
                  - "receiver-0"
              topologyKey: kubernetes.io/hostname
            weight: 100
          - podAffinityTerm:
              labelSelector:
                matchExpressions:
                - key: example.com/imagery-destination
                  operator: In
                  values:
                  - live-stitcher
              topologyKey: kubernetes.io/hostname
            weight: 80
        podAntiAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
          - labelSelector:
              matchExpressions:
              - key: example.com/regime
                operator: In
                values:
                - ingest
            topologyKey: kubernetes.io/hostname

Upvotes: 0

Views: 28

Answers (1)

ralphyjade
ralphyjade

Reputation: 197

Pod affinity are not ignored, it's just that weights for soft affinity are important and respected first because it influences the scheduler to find a node that meets the rule as concept of node affinity:

preferredDuringSchedulingIgnoredDuringExecution

Configs are good. I suggest removing your weights with your pod affinity as it is only applicable with preferred affinity config and experiment with your affinity. Setting pod affinity as hard affinity for stricter conditions wherein if pod conditions aren’t met it will not be scheduled at all:

requiredDuringSchedulingIgnoredDuringExecution

Upvotes: 0

Related Questions