connection=local skips all remaining plays although only one host failed

[Jesse Schlüter]

Hello everyone,

I have a small problem and I am not sure whether this is a bug or "works as defined". We have a few Junos devices we are configuring and they require ansible_connection=local, but since 2.0 if one of the hosts/inventory items fails, he does not simply skip this one host, but all. Here is a simple sample to simulate this behaviour without the need for a Junos host:

Inventory (where 10.118.194.22 is down):

[junos]

10.118.194.11 ansible_connection=local

10.118.194.12 ansible_connection=local

10.118.194.21 ansible_connection=local

10.118.194.22 ansible_connection=local

Playbook (as port use something that is reachable ;) ):

---

- name: play1

  hosts: junos

  gather_facts: no

  tasks:

    - name: task1

      wait_for: host={{ inventory_hostname }} port=22 timeout=5

    - name: task2

      wait_for: host={{ inventory_hostname }} port=22 timeout=5

- name: play2

  hosts: junos

  gather_facts: no

  tasks:

    - name: task3

      wait_for: host={{ inventory_hostname }} port=22 timeout=5

What I would expect:

1. starting play1
2. running task1, failing on 10.118.194.22
3. running task2 only for the non-failed hosts
4. starting play2
5. running task3 only for the non-failed hosts

What actually happens, is that he is not even starting play2. I suspect that the connection=local makes ansible behave like all four hosts are actually only one hosts. And since this one host failed skip it in all other plays. If you remove ansible_connection=local from each host (assuming they are ssh-able) it works as expected.

Is this a bug? Should I put this in Ansible Development?

Best regards,

Jesse