autoscaler always tries to manage the Master ASG

[magnusboman]

I deployed the auto scaler without --include-master-nodes, but it still tries to manage the master ASG

2017-06-05 03:24:17,137 WARNING: Desired capacity for ASG ASGMaster is 2, but exceeds max 1

Deployment used:

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  labels:
    application: kube-aws-autoscaler
    version: v0.9
  name: kube-aws-autoscaler
spec:
  replicas: 1
  selector:
    matchLabels:
      application: kube-aws-autoscaler
  template:
    metadata:
      labels:
        application: kube-aws-autoscaler
        version: v0.9
      annotations:
        # FIXME: using hardcoded IAM role
        iam.amazonaws.com/role: auto-scaler-role
    spec:
      containers:
      - name: autoscaler
        image: hjacobs/kube-aws-autoscaler:0.9
        resources:
          limits:
            cpu: 200m
            memory: 100Mi
          requests:
            cpu: 50m
            memory: 50Mi

[hjacobs]

Hi @magnusboman, the autoscaler currently relies on the master label having the value true. See https://github.com/zalando-incubator/kubernetes-on-aws/blob/dev/cluster/userdata-master.yaml#L97 on how we configure this for Zalando (I think kube-aws does the same IIRC).

We should document this :smirk:

Anyway, even without the master node label, the autoscaler won't do anything for your master ASG as it respects the min/max settings.

[kaazoo]

You could additionally check the EC2 tag 'k8s.io/role/master' (set to 1 for masters) for clusters created by kops.

[hjacobs]

@kaazoo yes, I also realized that the way of marking/labeling nodes is not consistent right now across kube-aws, kops and kubeadm. Supporting the "standard" role label makes sense.