Selecting nodes based on architecture

General Discussions
1

I’m running a test cluster on Raspberry Pis and one bigger machine as master and agent.

Cluster information:

Kubernetes version: v1.17.2+k3s1
Cloud being used: bare-metal (raspberry pis and x86 machine)
Installation method: k3s
Host OS: Ubuntu 18.04.4 LTS or 18.04.3 LTS
CNI and version: flannel v0.3.1 I believe
CRI and version:

Version:  0.1.0
RuntimeName:  containerd
RuntimeVersion:  v1.3.3-k3s1
RuntimeApiVersion:  v1alpha2

Some helm charts however don’t have arm or arm64 images and I would like to limit scheduling of those charts to machine with specific architectures.
What would be the most appropriate way to do so?
I see there are several labels and some other info which could be used, so I’m unsure which one to select and how.

$ kubectl get node rpi3p -o json
{
    "apiVersion": "v1",
    "kind": "Node",
    "metadata": {
        "..." // some other info
        "labels": {
            "beta.kubernetes.io/arch": "arm64",
            "beta.kubernetes.io/instance-type": "k3s",
            "beta.kubernetes.io/os": "linux",
            "k3s.io/hostname": "rpi3p",
            "k3s.io/internal-ip": "10.0.0.105",
            "kubernetes.io/arch": "arm64",
            "kubernetes.io/hostname": "rpi3p",
            "kubernetes.io/os": "linux",
            "node-role.kubernetes.io/master": "true",
            "node.kubernetes.io/instance-type": "k3s"
        },
        "name": "rpi3p",
        "..." // other info
    },
    "spec": {
        "podCIDR": "10.42.0.0/24",
        "podCIDRs": [
            "10.42.0.0/24"
        ],
        "providerID": "k3s://rpi3p"
    },
    "status": {
        "addresses": [
            {
                "address": "10.0.0.105",
                "type": "InternalIP"
            },
            {
                "address": "rpi3p",
                "type": "Hostname"
            }
        ],
        "allocatable": {
            "cpu": "4",
            "ephemeral-storage": "29642188777",
            "memory": "933992Ki",
            "pods": "110"
        },
        "capacity": {
            "cpu": "4",
            "ephemeral-storage": "30471000Ki",
            "memory": "933992Ki",
            "pods": "110"
        },
        "conditions": [ ... ],
        "daemonEndpoints": { ... },
        "images": [ ... ]
        "nodeInfo": {
            "architecture": "arm64",
            "bootID": "37de8048-af37-419e-89cb-c02187d29280",
            "containerRuntimeVersion": "containerd://1.3.3-k3s1",
            "kernelVersion": "4.15.0-1054-raspi2",
            "kubeProxyVersion": "v1.17.2+k3s1",
            "kubeletVersion": "v1.17.2+k3s1",
            "machineID": "a3c14b5eccab4034a96aa9de3ae0b371",
            "operatingSystem": "linux",
            "osImage": "Ubuntu 18.04.4 LTS",
            "systemUUID": "a3c14b5eccab4034a96aa9de3ae0b371"
        }
    }
}

Can I select by status.nodeInfo.architecture? Are labels better?
Thank you.

2

You can use nodeSelector on your pods to specify labels of the nodes you want to deploy to. The simplest solution is probably to add a specific label on the nodes you want to deploy your pods. See Assigning Pods to Nodes - Kubernetes for more information

3

Thank you.
So nodeSelector works on labels only and can’t select based on status.nodeInfo.architecture.
In that case kubernetes.io/arch would seem the most appropriate.

apiVersion: v1
kind: Pod
metadata:
  name: nginx
  labels:
    env: test
spec:
  containers:
  - name: nginx
    image: nginx
    imagePullPolicy: IfNotPresent
  nodeSelector:
    kubernetes.io/arch: arm64
1 Like
4

JSON PATH query to retrieve the osImages of all the nodes

kubectl get nodes -o jsonpath='{.items[*].status.nodeInfo.osImage}'