I am testing goolge container engine and everything was fine until I found this really weird issue.
bash-3.2# kubectl get services --namespace=es
NAME CLUSTER_IP EXTERNAL_IP PORT(S) SELECTOR AGE
elasticsearch-logging 10.67.244.176 <none> 9200/TCP name=elasticsearch-logging 5m
bash-3.2# kubectl describe service elasticsearch-logging --namespace=es
Name: elasticsearch-logging
Namespace: es
Labels: k8s-app=elasticsearch-logging,kubernetes.io/cluster-service=true,kubernetes.io/name=Elasticsearch
Selector: name=elasticsearch-logging
Type: ClusterIP
IP: 10.67.248.242
Port: <unnamed> 9200/TCP
Endpoints: <none>
Session Affinity: None
No events.after exactly 5 minutes, the service was deleted automatically.
kubectl get events --namespace=es
1m 1m 1 elasticsearch-logging Service DeletingLoadBalancer {service-controller } Deleting load balancer
1m 1m 1 elasticsearch-logging Service DeletedLoadBalancer {service-controller } Deleted load balancerAnyone got a clue why? thanks.
after removing the following from metadata/labels in the yaml file, the problem went away.
**kubernetes.io/cluster-service: "true"
kubernetes.io/name: "Elasticsearch"**The label kubernetes.io/cluster-service=true is a special, reserved label that shouldn't be used by user resources. That's used by a system process that manages the cluster's addons, like the DNS and kube-ui pods that you'll see in your cluster's kube-system namespace.
The reason your service is being deleted is because the system process is checking for resources with that label, seeing one that it doesn't know about, and assuming that it's something that it started previously that isn't meant to exist anymore. This is explained a little more in this doc about cluster addons.
In general, you shouldn't have any labels that are prefixed with kubernetes.io/ on your resources, since that's a reserved namespace.