Ansible and kube-deploy

Archives Ansible Project
1

Hi,

I’m trying to integrate Ansible into kube-deploy to allow for a faster and smoother deployment. I’ve ran into a strange issue where if I’m running the script for the ansible playbook, I cant mount gluster volumes in pods. But, if I run the worker.sh/master.sh locally as root (or with sudo) everything seems to work fine.

Here is an example of the Ansible playbook for a master node. In theory, everything should run under the root user with the correct ENV variables.

- name: Download the kube-deploy files
  git: repo=https://github.com/kubernetes/kube-deploy.git dest=/opt/kube-deploy version=master

- name: Run the master deploy script
  shell: echo Y | ./master.sh
  args:
    chdir: /opt/kube-deploy/docker-multinode/
  environment:
    USE_CNI: true
    USE_CONTAINERIZED: true
    K8S_VERSION: v1.4.0-alpha.2

And here is the actual task definition

---
- hosts: k8-master
  become: yes
  become_method: sudo
  gather_facts: yes
  roles:
      #- common
    - master

As an aside, I’ve tried using the command module instead of shell, removing the chdir, removing the “echo Y” and directly removing the /var/lib/kubelet folder without success. If a master/worker is started from Ansible, it cannot mount a glusterFS volume target. It seems to be related to the mounting of the root volume from USE_CONTAINERIZED? I’m using that flag since a Glusterfs volume target would fail without it.