Warm Image operator for Kubernetes

For example, you have huge image with your software and running POD on node. When POD moving to another node your image downloads to new node minute or two. This operator forces nodes to download image before rescheduling, so POD starts faster.

How it works

It runs /bin/sh with infinite loop on specified image as DaemonSet with additional options like NodeSelector, Affinity or resource limits. You can specify custom command if your image not contains /bin/sh interpreter or you want to run own script.

Deploy operator

Helm

helm repo add 21h https://charts.blindage.org
helm repo update
helm install warm-image-operator 21h/warm-image-operator -n operators

Cluster scoped installation

kubectl create ns warmimage
kubectl -f deploy/crds/blindage.org_warmimages_crd.yaml
kubectl -n warmimage -f deploy/role.yaml
kubectl -n warmimage -f deploy/role_binding.yaml
kubectl -n warmimage -f deploy/service_account.yaml
kubectl -n warmimage -f deploy/operator.yaml

Create your first warmer

kubectl -n warmimage -f deploy/example/mongo4.yaml

Settings

Required:

• Set image URL
  image: string

• Set image tag
  version: string

Additional options:

• Set custom command
  customCommand: strings list
  Default:
  - “/bin/sh”
  - “-c”
  - “while true; do sleep 86400; done”

• Add custom labels
  labels: map

• Affinity, node selection and tolerations
  nodeSelector: default empty
  affinity: default empty
  tolerations: default empty

• Resource limiting
  podRequests: default empty
  podLimits: default empty

Your first warmer

Create yaml file and place it:

apiVersion: blindage.org/v1alpha1
kind: WarmImage
metadata:
  name: mongo4
spec:
  image: mongo
  version: "4"
  nodeSelector:
    node-role.kubernetes.io/master: ""

Now you warmed mongo:4 on all master nodes.

---

Copyright by Vladimir Smagin (21h) 2020
http://blindage.org email: 21h@blindage.org
Project page: https://git.blindage.org/21h/warm-image-operator