Ansible doesn't see an installed Python module

Question

I am trying to create an Ansible playbook which would setup a local Postgres database to be used for local/dev testing of a certain app.

postgresql_db seems to be the Ansible module I need, and psycopg2 is the Python module listed as a dependency.

So in the same virtualenv where I have installed Ansible, I also installed psycopg2 (I'm running on Mac with pipenv).

But when I run my playbook with this command:

ansible-playbook pg_local.yml --connection=local

I get the error:

"msg": "the python psycopg2 module is required"

The playbook is tiny:

---
- hosts: my_ws

  tasks:
    - name: create a db
      postgresql_db:
        name: mynewdb

and so is /etc/ansible/hosts:

[my_ws]
localhost

I suspect that somehow the "remote" machine, which is really local, is trying to import psycopg2, which running in a python environment which doesn't have the module. Is the --connection=local to blame?

I have added it to solve the ssh: connect to host localhost port 22: Connection refused error, and as I do intend to run this only locally. I don't think it's wrong - but I do wonder if it messes up the environment for Ansible.

---

I have added a 'test and report' task to the playbook and no problems were detected:

changed: [localhost] => {
    "changed": true,
    "cmd": [
        "python",
        "-c",
        "import psycopg2; print(psycopg2.__version__)"
    ],
    "delta": "0:00:00.087664",
    "end": "2018-08-22 13:36:17.046624",
    "invocation": {
        "module_args": {
            "_raw_params": "python -c 'import psycopg2; print(psycopg2.__version__)'",
            "_uses_shell": false,
            "argv": null,
            "chdir": null,
            "creates": null,
            "executable": null,
            "removes": null,
            "stdin": null,
            "warn": true
        }
    },
    "rc": 0,
    "start": "2018-08-22 13:36:16.958960",
    "stderr": "/Users/lsh783/.local/share/virtualenvs/docker-ansible-setup--HsJmUMv/lib/python3.6/site-packages/psycopg2/__init__.py:144: UserWarning: The psycopg2 wheel package will be renamed from release 2.8; in order to keep installing from binary please use \"pip install psycopg2-binary\" instead. For details see: <http://initd.org/psycopg/docs/install.html#binary-install-from-pypi>.\n  \"\"\")",
    "stderr_lines": [
        "/Users/lsh783/.local/share/virtualenvs/docker-ansible-setup--HsJmUMv/lib/python3.6/site-packages/psycopg2/__init__.py:144: UserWarning: The psycopg2 wheel package will be renamed from release 2.8; in order to keep installing from binary please use \"pip install psycopg2-binary\" instead. For details see: <http://initd.org/psycopg/docs/install.html#binary-install-from-pypi>.",
        "  \"\"\")"
    ],
    "stdout": "2.7.5 (dt dec pq3 ext lo64)",
    "stdout_lines": [
        "2.7.5 (dt dec pq3 ext lo64)"
    ]
}

---

I do see this line in the output with -vvv:

<localhost> EXEC /bin/sh -c '/usr/bin/python /Users/lsh783/.ansible/tmp/ansible-tmp-1534957231.272713-121756549613883/postgresql_db.py > && sleep 0'

and it bothers me that it's not the Python inside the virtual environment I'm under.

Answer 1

This is caused by a well-known and described behaviour of Ansible.

---

In short, if you specify localhost wherever in your inventory, Ansible will default to using /usr/bin/python for running the modules regardless of the connection: local setting.

This in turn will cause problems if additional libraries were installed in a Python environment used to execute a playbook, but not for the /usr/bin/python.

The solution is to specify ansible_python_interpreter for the localhost. In your case:

[my_ws]
localhost ansible_python_interpreter={{ansible_playbook_python}}

---

Because of the above, the test to verify module presence should be:

- command: "{{ ansible_python_interpreter | default('/usr/bin/python') }} -c 'import {{ module }}; print({{ module }}.__version__)'"
  vars:
    module: psycopg2
  register: test
- debug:
    var: test.stdout