---
# This playbook removes the Ceph MDS from your cluster.
#
# Use it like this:
# ansible-playbook shrink-mds.yml -e mds_to_kill=ceph-mds01
#     Prompts for confirmation to shrink, defaults to no and
#     doesn't shrink the cluster. yes shrinks the cluster.
#
# ansible-playbook -e ireallymeanit=yes|no shrink-mds.yml
#     Overrides the prompt using -e option. Can be used in
#     automation scripts to avoid interactive prompt.
- name: gather facts and check the init system
  hosts:
    - "{{ mon_group_name | default('mons') }}"
    - "{{ mds_group_name | default('mdss') }}"
  become: true
  tasks:
    - debug:
        msg: gather facts on all Ceph hosts for following reference

- name: perform checks, remove mds and print cluster health
  hosts: localhost
  become: true
  vars_prompt:
    - name: ireallymeanit
      prompt: Are you sure you want to shrink the cluster?
      default: 'no'
      private: no
  pre_tasks:
    - import_role:
        name: ceph-defaults

    - name: exit playbook, if no mds was given
      when: mds_to_kill is not defined
      fail:
        msg: >
          mds_to_kill must be declared.
          Exiting shrink-cluster playbook, no MDS was removed. On the command
          line when invoking the playbook, you can use
          "-e mds_to_kill=ceph-mds1" argument.  You can only remove a single
          MDS each time the playbook runs."

    - name: exit playbook, if the mds is not part of the inventory
      when: mds_to_kill not in groups[mds_group_name]
      fail:
        msg: "It seems that the host given is not part of your inventory,
              please make sure it is."

    - name: exit playbook, if user did not mean to shrink cluster
      when: ireallymeanit != 'yes'
      fail:
        msg: "Exiting shrink-mds playbook, no mds was removed.
           To shrink the cluster, either say 'yes' on the prompt or
           or use `-e ireallymeanit=yes` on the command line when
           invoking the playbook"

    - import_role:
        name: ceph-facts

    - name: set_fact container_exec_cmd for mon0
      set_fact:
        container_exec_cmd: "{{ container_binary }} exec ceph-mon-{{ hostvars[groups[mon_group_name][0]]['ansible_hostname'] }}"
      when: containerized_deployment | bool

    - name: exit playbook, if can not connect to the cluster
      command: "{{ container_exec_cmd | default('') }} timeout 5 ceph --cluster {{ cluster }} health"
      register: ceph_health
      until: ceph_health is succeeded
      delegate_to: "{{ groups[mon_group_name][0] }}"
      retries: 5
      delay: 2

    - name: set_fact mds_to_kill_hostname
      set_fact:
        mds_to_kill_hostname: "{{ hostvars[mds_to_kill]['ansible_hostname'] }}"

  tasks:
    # get rid of this as soon as "systemctl stop ceph-msd@$HOSTNAME" also
    # removes the MDS from the FS map.
    - name: exit mds if it the deployment is containerized
      when: containerized_deployment | bool
      command: "{{ container_exec_cmd | default('') }} ceph tell mds.{{ mds_to_kill }} exit"
      delegate_to: "{{ groups[mon_group_name][0] }}"

    - name: stop mds service and verify it
      block:
        - name: stop mds service
          service:
            name: ceph-mds@{{ mds_to_kill_hostname }}
            state: stopped
            enabled: no
          delegate_to: "{{ mds_to_kill }}"
          failed_when: false

        - name: ensure that the mds is stopped
          command: "systemctl is-active ceph_mds@{{ mds_to_kill_hostname }}"
          register: mds_to_kill_status
          failed_when: mds_to_kill_status.rc == 0
          delegate_to: "{{ mds_to_kill }}"
          retries: 5
          delay: 2

    - name: fail if the mds is reported as active or standby
      block:
        - name: get ceph status
          command: "{{ container_exec_cmd | default('') }} ceph --cluster {{ cluster }} -s -f json"
          register: ceph_status
          delegate_to: "{{ groups[mon_group_name][0] }}"

        - name: get active mds nodes list
          set_fact:
            active_mdss: "{{ active_mdss | default([]) + [item.name] }}"
          with_items: "{{ (ceph_status.stdout | from_json)['fsmap']['by_rank'] }}"

        - name: get ceph fs dump status
          command: "{{ container_exec_cmd | default('') }} ceph --cluster {{ cluster }} fs dump -f json"
          register: ceph_fs_status
          delegate_to: "{{ groups[mon_group_name][0] }}"

        - name: create a list of standby mdss
          set_fact:
            standby_mdss: (ceph_fs_status.stdout | from_json)['standbys'] | map(attribute='name') | list

        - name: fail if mds just killed is being reported as active or standby
          fail:
            msg: "mds node {{ mds_to_kill }} still up and running."
          when:
            - (mds_to_kill in active_mdss | default([])) or
              (mds_to_kill in standby_mdss | default([]))

    - name: delete the filesystem too if deleted the last mds too
      command: "{{ container_exec_cmd | default('') }} ceph --cluster {{ cluster }} fs rm --yes-i-really-mean-it {{ cephfs }}"
      delegate_to: "{{ groups[mon_group_name][0] }}"

    - name: purge mds store
      file:
        path: /var/lib/ceph/mds/{{ cluster }}-{{ mds_to_kill_hostname }}
        state: absent
      delegate_to: "{{ mds_to_kill }}"

  post_tasks:
    - name: show ceph health
      command: "{{ container_exec_cmd | default('') }} ceph --cluster {{ cluster }} -s"
      delegate_to: "{{ groups[mon_group_name][0] }}"