add-osd: fix fact gathering in add-osd

This commit makes this playbook gathering facts from all other nodes but
clients.
When collocating OSDs on other nodes it can fail like following:

```
fatal: [vm252-11]: FAILED! => {
    "msg": "'ansible.vars.hostvars.HostVarsVars object' has no attribute 'ansible_hostname'"
}
```

In that case, a fact from a RGW node is called when rendering the
`ceph.conf.j2` but it fails because facts are gathered only from mon and
osd nodes.

Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1806765

Signed-off-by: Guillaume Abrioux <gabrioux@redhat.com>

infrastructure-playbooks/add-osd.yml

@@ -35,9 +35,7 @@
       setup:
       delegate_to: "{{ item }}"
       delegate_facts: True
-      with_items:
+      with_items: "{{ groups['all'] | difference(groups.get('clients', [])) }}"
-        - "{{ groups['mons'] }}"
-        - "{{ groups['osds'] }}"
       run_once: True
       when:
         - delegate_facts_host | bool
@@ -65,9 +63,7 @@
       setup:
       delegate_to: "{{ item }}"
       delegate_facts: True
-      with_items:
+      with_items: "{{ groups['all'] | difference(groups.get('clients', [])) }}"
-        - "{{ groups['mons'] }}"
-        - "{{ groups['osds'] }}"
       run_once: True
       when:
         - delegate_facts_host | bool
@@ -77,12 +73,6 @@
       set_fact:
         add_osd: True
 
-    - name: set noup flag
-      command: "{{ docker_exec_cmd | default('') }} ceph --cluster {{ cluster }} osd set noup"
-      delegate_to: "{{ groups['mons'][0] }}"
-      run_once: True
-      changed_when: False
-
   roles:
     - role: ceph-defaults
     - role: ceph-handler
@@ -94,10 +84,3 @@
       when: not containerized_deployment | bool
     - role: ceph-config
     - role: ceph-osd
-
-  post_tasks:
-    - name: unset noup flag
-      command: "{{ docker_exec_cmd | default('') }} ceph --cluster {{ cluster }} osd unset noup"
-      delegate_to: "{{ groups['mons'][0] }}"
-      run_once: True
-      changed_when: False