skip wait_for in reboot handler ( for ansible-pull )

[martbhell]

The handler to reboot has a "wait_for" - this needelessly makes this role pause for 300s before rebooting - also when ansible-pull is running. Can we skip that one somehow on ansible-pull runs?

For push mode then it's nice to keep this one as the play shouldn't fail - just reboot and then wait for the node to come back online.

https://github.com/CSC-IT-Center-for-Science/ansible-role-cuda/blob/master/handlers/main.yml

[martbhell]

As far as I can see there's no special variable in ansible_facts one could use to figure out if we are running ansible-pull. So instead we set a variable when running ansible-pull.

Conditional updated for the wait_for handler in e1af63b

[martbhell]

Testing the change on our GPU node

[martbhell]

Not perfect. It didn't restart at all now after the cuda role on a new install.
Maybe one needs to remove the async/poll settings in the handler.

[martbhell]

The wait_for does not apply anymore. Also added a handler 1 and task 2 which runs cuda_init.sh in ansible too (not only from rc.local). But after the file has changed or if /dev/nvidia0 does not exist. Idea here is to have a /dev/nvidia0 available on the GPU nodes before we start slurm (which fails if those files in gres.conf does not exist).