Handle single-platform container images

When using node auto-provisioning with Fleets, Cloudfleet always picks the cheapest available node to run your container. Many cloud providers offer ARM architecture nodes, which are often more cost-effective than amd64 nodes. However, before scheduling a node, Cloudfleet does not know if your container image only supports a single architecture. If your container image is only available for the amd64 architecture but Cloudfleet decides to provision a ARM node, your container will not run on ARM nodes, resulting in a crash loop.

Solution

To avoid this, use the kubernetes.io/arch label by setting it to amd64 in your pod specification to preemptively signal Cloudfleet about the architecture that your container image supports. You can do this using either the nodeSelector property or a node affinity rule. Here’s how:

Option 1: Using nodeSelector 

apiVersion: v1
kind: Pod
metadata:
  name: my-app
spec:
  containers:
    - name: my-app
      image: myregistry/my-app:latest
  nodeSelector:
    kubernetes.io/arch: amd64

Option 2: Using Node Affinity 

apiVersion: v1
kind: Pod
metadata:
  name: my-app
spec:
  containers:
    - name: my-app
      image: myregistry/my-app:latest
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
          - matchExpressions:
              - key: kubernetes.io/arch
                operator: In
                values:
                  - amd64

Both approaches ensure your container is scheduled only on nodes with the amd64 architecture, preventing crash loops due to architecture mismatches.

For the best experience, consider building multi-platform container images. This allows you to support both ARM and amd64 architectures using a single image. Cloudfleet can then schedule your containers on the cheapest compatible node, regardless of architecture.

Languages like Go, Node.js, Python, Java, and Rust are especially well-suited for multi-platform builds, as they offer strong cross-architecture support.

On this page