Hi,
I am writing third party modules in Python using the tag
“#<<INCLUDE_ANSIBLE_MODULE_COMMON>>”
so that I get parameters for my modules in a nice ANSIBLE_MODULE_ARG= {… } json list, read by a call to AnsibleModule().
However, I am using a virtual env for python : /srv/ansible/venv/bin/python
Actually, I need to set in “vars” of all my playbooks, the following variable :
vars:
ansible_python_interpreter: /srv/ansible/venv/bin/python
Or (even better) in the [all:vars] section of my inventory
[all:vars]
If I don’t do it, lib/ansible/executor/module_common.py is handling my module in a way that leads to using /usr/bin/python by default
(which is obviously not what I want)
Would a PR useful to tell module_common.py to use “sys.executable” (the current python executable path), instead of “/usr/bin/python” as the “default” python executable path ?
Ansible will already utilize sys.executable for localhost, as long as you do not explicitly define localhost in your inventory, and instead allow ansible to create it’s implicit localhost. See:
Telling ansible to use sys.executable as the default would likely not be what anyone wants. Since the local sys.executable, is likely not the same path as the remote python binary. /usr/bin/python is predictable for remote target systems, allowing to be overridden by ansible_python_interpreter.
The example I mentioned was about launching modules on the localhost system.
I don’t think “module_common.py” was meant to launch modules on remote target systems, is it ?
I might be mistaking. Can anyone help us on this subject ?
Thank you for your concern. I understand (I hope better now) that “module_common” is run to build modules that will be launched either on local or on remote system.
When building for a local execution of the module, it should take care to search for the ‘implicit’ localhost “ansible_python_interpreter”, or look for sys.executor.
When building for a remote execution of the module, it’s ok to call “_get_shebang()” and get the ansible_python_interpreter of the remote system from task_vars.
My playbook is run with a list of remote hosts each possibly having different “ansible_python_interpreter”.
I am stuck with that problem, since my remote systems don’t have the same python interpreter path as the local one running ansible and … my modules are run on my local system when “playing” the remote systems.
Any help wanted
We already have functionality to support what you want.
I think you need to move this over to ansible-project, share your playbook, and ask for help on how to get what you want.
Again, use implicit localhost, don’t define localhost in your inventory. Use delegate_to: localhost, or local_action, or target localhost in your hosts specification to use implicit localhost.
What I intended to do is very simple : I need to write a module similar “ios_command”, launched in a very simple playbook :
The module (ios_command or my own module) runs on the localhost (thus needs to use my dedicated environment for python /srv/ansible/venv/bin/python), then connects to the target host to do things there.
Normal play
the ios_command runs well, my own module is called locally with /usr/bin/python, runs out of the dedicated environment (thus do not find its entry point and abort)
Play with ansible_python_interpreter=/srv/ansible/venv/bin/python in inventory for each target host (which is not what I should do, but then)
the ios_command runs well, and my own module also
Normal play - my own module includes the “WANT_JSON” tag
the ios_command runs well, and my own module also
The tag “# WANT_JSON” mainly breaks the way ansible-play will send the parameters to my own module (but I can handle that), and as a side effect, tells module_common.py to treat my own module as it is without trying to ANSIBALLZ it or rewrite the shebang.
For the time being, I will use either 3. or 2. … but I guess there is a better way to handle the problem (since “native” ansible module are working well).
This is a “how to write a module, so that it is launched with the correct shebang” issue
In the meantime, I studied your proposals and the thing about “modules are run on the remote targets” :
I thought “modules” were run on the local controller by default :
Modules are run on the remote targets by default using “Normal action plugin”.
What I need is “Another action plugin” to run my modules on the local ansible controller.
I will check in detail how ios_command do the job and see how to do the same
