[swizzle] Nodes fail to install because pip is missing

[johnharris85]

/kind bug

What steps did you take and what happened: Install correctly

What did you expect to happen: Install fails with nodes unable to find pip

Anything else you would like to add: pip is part of the common role that is installed on all nodes. It's required by the kubernetes-cni role (which is applied to all nodes), however the add-nodes playbook is not run until after kubernetes-cni, so that fails as pip doesn't exist on the nodes at that time.

Fix is just to switch the ordering of those 2 in the install.yml playbook, will try submit a PR this afternoon.

Environment:

• Wardroom version: master
• OS (e.g. from /etc/os-release): Centos 7

[scottslowe]

@johnharris85 Do you remember the command you used to test this? The documentation for Wardroom needs some updating, and I was wondering if you tried this using ansible-playbook -i <inventory> -e wardroom_action=install main.yml or something else.

[johnharris85]

Just from the docs I believe @scottslowe - ansible-playbook install.yml --inventory inventory.ini --extra-vars @extra_vars.yml

[scottslowe]

@johnharris85 OK, thanks. I think that swapping the order of importing the add-nodes.yml playbook may affect the ability to use Wardroom to just add nodes to an existing cluster, but I'll need to test that first. (This is controlled via the wardroom_action variable I mentioned above, which isn't in the documentation yet.)

[johnharris85]

Ah OK fair enough. Needs a change of some sort though I guess as it's broken as-is.

[scottslowe]

@johnharris85 Does this affect only CentOS nodes, or have we tried on Ubuntu also?

[johnharris85]

Only tested on Centos but because it's ordering related I think it's safe to assume it affects both?