K
Q

How to specify pod to node affinity in kubernetes

7/29/2016

How can I configure a specific pod to run on a multi-node kubernetes cluster so that it would restrict the containers of the POD to a subset of the nodes.

E.g. let's say I have A, B, C three nodes running mu kubernetes cluster.

How to limit a Pod to run its containers only on A & B, and not on C?

-- Santanu Dey
kubernetes

Similar Questions

Kubernetes deployment cannot mount volume despite equivalent gcloud/mnt works fine
Kubernetes cluster internal load balancing
Secured communication over flannel in Kubernetes
Best Practise to expose service in kubernetes using Calico

3 Answers

7/29/2016

You can add label to nodes that you want to run pod on and add nodeSelector to pod configuration. The process is described here:

http://kubernetes.io/docs/user-guide/node-selection/

So basically you want to

kubectl label nodes A node_type=foo
kubectl label nodes B node_type=foo

And you want to have this nodeSelector in your pod spec:

nodeSelector:
    node_type: foo
-- Nebril
Source: StackOverflow
4/1/2017

Firstly, you need to add label to nodes. You can refer to Nebril's answer of how to add label.

Then I recommend you to use the node affinity feature to constrain pods to nodes with particular labels. Here is an example.

apiVersion: v1
kind: Pod
metadata:
  name: with-node-affinity
spec:
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: node_type
            operator: In
            values:
            - foo
  containers:
  - name: with-node-affinity
    image: gcr.io/google_containers/pause:2.0

Compared to nodeSelector, the affinity/anti-affinity feature greatly expands the types of constraints you can express.

For more detailed information about node affinity, you can refer to the following link: https://kubernetes.io/docs/concepts/configuration/assign-pod-node/

-- Bruce Wu
Source: StackOverflow
7/28/2017

I am unable to post a comment to previous replies but I upvoted the answer that is complete. nodeSelector is not strictly enforced and relying on it may cause some grief including Master getting overwhelmed with requests as well as IOs from pods scheduled on Master. At the time the answer was provided it may still have been the only option but in the later versions that is not so.

Definitely use nodeAffinity or nodeAntiAffinity with requiredDuringSchedulingIgnoredDuringExecution or preferredDuringSchedulingIgnoredDuringExecution

For more expressive filters for scheduling peruse: Assigning pods to nodes

-- perennial_noob
Source: StackOverflow