0

Having problems correctly deploying a trivial config. Not suggesting this is an ideal configuration, but it ought to be a valid one:

[k8s-cluster:children]
kube-node
[all]
node1   ansible_host=X.Y.Z.A
node2   ansible_host=X.Y.Z.B
node3   ansible_host=X.Y.Z.C

[kube-master]
node1
node2
node3

[kube-node]
node1
node2
node3

[etcd]
node1
node2
node3

[calico-rr]

[vault]
node1

The ansible_hosts all refer to machines w/ identical resources. The first playbook run stalled on starting the API server on host2 and host3. After re-running, the playbook completed successfully, but hosts 2 & 3 are running only quay.io/coreos/etcd:v3.2.24 - host1 is running a ton of k8s images. Why is the API server not running on all masters?

Deploying via ansible-playbook -i inventory/cluster/hosts.ini --user=ubuntu --become cluster.yml where cluster.yml is from master@HEAD.

moe
  • 452
  • 2
  • 10
  • do you think this information is enough to resolve your issue? `Kubespray` has ton of options, variables, versions and no one can help you with the only one inventory file attached. – Konstantin Vustin Oct 09 '18 at 11:07
  • Added some details. – moe Oct 09 '18 at 11:11
  • it is _highly questionable_ to have a Node be both a master and a worker role. You will almost certainly want to segregate your cluster workloads from the master nodes, because if the masters fall under duress, things will only get worse from there. Separately, you're living life on the edge trying to deploy kubespray from `master` – mdaniel Oct 09 '18 at 21:02

0 Answers0