Failed to write to file .ssh/known_hosts: [Errno 2] No such file or directory: '.ssh/tmpupg6c51q'

I have task the that runs the known_hosts module with a loop and multiple items.

It mostly works, except when I run the playbook with mitogen from within a centos7 docker container (e.g., in a CI/CD pipeline). Here is the “works/fails” matrix I tested:

mitogen	Fedora WS	centos7 docker
yes	works	fails
no	works	works

System facts:

• ansible 2.8.3
• mitogen 0.2.8 and git master
• centos7 Docker container
• Target system: Python: 3.6 (system)
• Where ansible is run: Python 3.7 (conda)

The minimal playbook:

# hosts.yaml
all:
  hosts:
    myhost:
      ansible_host: host.example.com
  vars:
    ansible_python_interpreter: /usr/bin/python3.6
    ansible_user: myuser

# debug.yaml
- hosts: myhost
  gather_facts: False  # Speed-up deployments since we don't need these facts
  tasks:
    - name: Add/update public keys in kown_hosts
      known_hosts:
        name: "{{ item.host }}"
        key: "{{ item.key }}"
        path: .ssh/known_hosts
      loop:
       - {host: a, key: 'a,1.2.3.4 ecdsa-sha2-nistp256 ABCD1234'}
       - {host: b, key: 'b,5.6.7.8 ecdsa-sha2-nistp256 EFGH5678'}

The first item always works, but from the second item on, it fails like this:

...
TASK [Add/update public keys in kown_hosts] *************************************************************************
ok: [myhost] => (item={'host': 'a', 'key': 'a,1.2.3.4 ecdsa-sha2-nistp256 ABCD1234'})
failed: [myhost] (item={'host': 'b', 'key': 'b,5.6.7.8 ecdsa-sha2-nistp256 EFGH5678'}) => {"ansible_loop_var": "item", "changed": false, "item": {"host": "b", "key": "b,5.6.7.8 ecdsa-sha2-nistp256 EFGH5678"}, "msg": "Failed to write to file .ssh/known_hosts: [Errno 2] No such file or directory: '.ssh/tmpupg6c51q'"}

The the log with -vvv is attached: ansible.log

[dw]

This is a wonderfully concise bug report, thanks for for that. That it is a temp file suggests something may be going wrong with Mitogen’s “automatic leaked file descriptor cleanup” logic, but possibly it could be Yet Another Temp File issue due to the compatibility insanity inherited from Ansible 😃

I’ll try to setup a repro for this later, if it’s the auto cleanup code it’s probably a 1 liner fix. Thanks for reporting!

[dw]

That sounds fine. It’s probably something like a corner case in the extension because it’s a loop