Random undefined variables (facts) since ansible 2.0 upgrade

Archives Ansible Project
1

Some ansible playbook runs will work on most hosts, but 1-3 will fail with an undefined variable that is an ansible fact. Run it a second time and it works.
Example:

fatal: [wdv-sitefinity1]: FAILED! => {“failed”: true, “msg”: “ERROR! The conditional check ‘ansible_os_family == ‘Windows’’ failed. The error was: ERROR! error while evaluating conditional (ansible_os_family == ‘Windows’): ERROR! ‘ansible_os_family’ is undefined\n\nThe error appears to have been in ‘/home/icansible/ansible/roles/newrelic/tasks/main.yml’: line 11, column 3, but may\nbe elsewhere in the file depending on the exact syntax problem.\n\nThe offending line appears to be:\n\n\n- name: create temp directory\n ^ here\n”}

2

This is still occurring during our full site.yml run, but it’s so far impossible to replicate in a small example. It only occurs on our full site.yml run with no limits, any attempt to limit to certain hosts or tags result in no errors.

3

I had something like it, not on Windows though,at the time the problem was that I set it not to gather_facts, when I actually tried to use a few facts later on. May not be the same but may be worth checking…

Alex

4

In our case, the facts are being gathered, and used other plays included in site.yml, but later on in a subsequent play the a fact that was defined previously is suddenly undefined, on some random hosts, but not all. It’s happening on windows and linux hosts.

5

Do you have fact gathering set to smart in your ansible.cfg?

6

We do, and have been looking at that. The facts for all hosts are getting gathered/refreshed during the first included playbook in site.yml, and the cache is not supposed to be expiring for 3 hours, with our entire run taking less than 2 hours. The cache files are present on disk, etcetera.

We did launch a site.yml run with fact caching commented out of ansible.cfg, waiting to see how that run works.

7

Turning off fact caching completely did not improve the situation.

8

Spoke too soon, turning off fact caching fixed or worked around the problem. Opened a bug report
https://github.com/ansible/ansible/issues/14456