controllercheck.yml

- hosts: controller
  become: yes
  vars:
    verbose: "{{ verbose|default('false') }}"
  gather_facts: no
  tasks:
#    - name: show verbosity
#      shell: echo '{{ verbose }}'
#      register: verbose_value
#    - debug: var=verbose_value.stdout

    - name: galera replication status (clustercheck)
      command: clustercheck 
      register: cluschk
      failed_when: "'node is synced' not in cluschk.stdout"
      ignore_errors: yes
    - name: show galera clustercheck (verbose=true)
      debug: var=cluschk.stdout_lines
      when: verbose
    - name: rabbitmq status (rabbitmqctl node_health_check)
      command: rabbitmqctl node_health_check
      register: rabbitchk
      failed_when: rabbitchk.rc != 0
      ignore_errors: yes
    - name: show rabbitmqctl status (verbose=true)
      debug: var=rabbitchk.stdout_lines
      when: verbose
    - name: check pcs for maintenance mode
      command: pcs property show maintenance-mode
      register: pcspropchk
      failed_when: "'true' in pcspropchk.stdout"
    - name: show pcs property maintenance-mode 
      debug: var=pcspropchk.stdout
      when: verbose
    - name: get pacemaker status
      command: pcs status xml
      register: pcs_status
    - name: evaluate pacemaker status
      pacemaker: status={{ pcs_status.stdout }}
      register: result
    - name: show pacemaker status
      debug: var=result.msg
      when: verbose

- hosts: controller[0]
  become: yes
  vars:
    verbose: "{{ verbose|default('false') }}"
  gather_facts: no
  tasks:
    - name: get pacemaker resources
      shell: "pcs status xml | grep Stopped; exit 0"
      register: pcsresourcechk
    - name: find pacemaker resources Stopped
      debug: var=pcsresourcechk.stdout_lines
      when: "'Stopped' in pcsresourcechk.stdout_lines"
    - name: get pacemaker resoures (verbose=true)
      command: pcs status resources
      register: pcsresourcechk
      when: verbose
    - name: show pacemaker resources (verbose=true)
      debug: var=pcsresourcechk.stdout_lines
      when: verbose

- hosts: undercloud
  become: yes
  vars:
    verbose: "{{ verbose|default('false') }}"
  gather_facts: no
  tasks:
    - name: check openstack compute agent list
      shell: . /home/stack/overcloudrc; openstack network agent list -f value | grep -v 'True UP' 
      #shell: . /home/stack/overcloudrc; openstack network agent list -f json \
      #| jq '[.[] | select(.State!="UP"),select(.Alive!=true)'] | jq 'unique'
      register: neutronchk
      failed_when: neutronchk.rc == 99 #never fail here
    - name: show unhealthy compute agents
      debug: var=neutronchk.stdout_lines
      when: neutronchk.rc != 1
    - fail: 
        msg: "Not all compute agents are healthy" 
      when: neutronchk.rc != 1 
      ignore_errors: yes
    - name: get openstack network agent list (verbose=true)
      shell: . /home/stack/overcloudrc; openstack network agent list
      register: neutronchk
      when: verbose
    - name: show openstack network agent list (verbose=true)
      debug: var=neutronchk.stdout_lines
      when: verbose
#- import_playbook: /usr/share/openstack-tripleo-validations/validations/pacemaker-status.yaml