ceph-ansible/infrastructure-playbooks/shrink-mgr.yml

---
# This playbook shrinks the Ceph manager from your cluster
#
# Use it like this:
# ansible-playbook shrink-mgr.yml -e mgr_to_kill=ceph-mgr1
#     Prompts for confirmation to shrink, defaults to no and
#     doesn't shrink the cluster and yes shrinks the cluster.
#
# ansible-playbook -e ireallymeanit=yes|no shrink-mgr.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') }}"
    - "{{ mgr_group_name | default('mgrs') }}"
  become: true
  tasks:
    - name: Gather facts on all Ceph hosts for following reference
      ansible.builtin.debug:
        msg: gather facts on all Ceph hosts for following reference

- name: Confirm if user really meant to remove manager from the ceph cluster
  hosts: mons[0]
  become: true
  vars_prompt:
    - name: ireallymeanit  # noqa: name[casing]
      prompt: Are you sure you want to shrink the cluster?
      default: 'no'
      private: false
  pre_tasks:
    - name: Import ceph-defaults role
      ansible.builtin.import_role:
        name: ceph-defaults

    - name: Import ceph-facts role
      ansible.builtin.import_role:
        name: ceph-facts
        tasks_from: container_binary

    - name: Set_fact container_exec_cmd
      when: containerized_deployment | bool
      ansible.builtin.set_fact:
        container_exec_cmd: "{{ container_binary }} exec ceph-mon-{{ ansible_facts['hostname'] }}"

    - name: Exit playbook, if can not connect to the cluster
      ansible.builtin.command: "{{ container_exec_cmd | default('') }} timeout 5 ceph --cluster {{ cluster }} health"
      register: ceph_health
      changed_when: false
      until: ceph_health is succeeded
      retries: 5
      delay: 2

    - name: Get total number of mgrs in cluster
      block:
        - name: Save mgr dump output
          ansible.builtin.command: "{{ container_exec_cmd | default('') }} ceph --cluster {{ cluster }} mgr dump -f json"
          register: mgr_dump
          changed_when: false

        - name: Get active and standbys mgr list
          ansible.builtin.set_fact:
            active_mgr: "{{ [mgr_dump.stdout | from_json] | map(attribute='active_name') | list }}"
            standbys_mgr: "{{ (mgr_dump.stdout | from_json)['standbys'] | map(attribute='name') | list }}"

    - name: Exit playbook, if there's no standby manager
      ansible.builtin.fail:
        msg: "You are about to shrink the only manager present in the cluster."
      when: standbys_mgr | length | int < 1

    - name: Exit playbook, if no manager was given
      ansible.builtin.fail:
        msg: "mgr_to_kill must be declared
              Exiting shrink-cluster playbook, no manager was removed.
              On the command line when invoking the playbook, you can use
              -e mgr_to_kill=ceph-mgr01 argument. You can only remove a single
              manager each time the playbook runs."
      when: mgr_to_kill is not defined

    - name: Exit playbook, if user did not mean to shrink cluster
      ansible.builtin.fail:
        msg: "Exiting shrink-mgr playbook, no manager 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"
      when: ireallymeanit != 'yes'

    - name: Set_fact mgr_to_kill_hostname
      ansible.builtin.set_fact:
        mgr_to_kill_hostname: "{{ hostvars[mgr_to_kill]['ansible_facts']['hostname'] }}"

    - name: Exit playbook, if the selected manager is not present in the cluster
      ansible.builtin.fail:
        msg: "It seems that the host given is not present in the cluster."
      when:
        - mgr_to_kill_hostname not in active_mgr
        - mgr_to_kill_hostname not in standbys_mgr

  tasks:
    - name: Stop manager services and verify it
      block:
        - name: Stop manager service
          ansible.builtin.service:
            name: ceph-mgr@{{ mgr_to_kill_hostname }}
            state: stopped
            enabled: false
          delegate_to: "{{ mgr_to_kill }}"
          failed_when: false

        - name: Ensure that the mgr is stopped
          ansible.builtin.command: "systemctl is-active ceph-mgr@{{ mgr_to_kill_hostname }}"  # noqa command-instead-of-module
          register: mgr_to_kill_status
          failed_when: mgr_to_kill_status.rc == 0
          delegate_to: "{{ mgr_to_kill }}"
          changed_when: false
          retries: 5
          delay: 2

    - name: Fail if the mgr is reported in ceph mgr dump
      ansible.builtin.command: "{{ container_exec_cmd | default('') }} ceph --cluster {{ cluster }} mgr dump -f json"
      register: mgr_dump
      changed_when: false
      failed_when: mgr_to_kill_hostname in (([mgr_dump.stdout | from_json] | map(attribute='active_name') | list) + (mgr_dump.stdout | from_json)['standbys'] | map(attribute='name') | list)
      until: mgr_to_kill_hostname not in (([mgr_dump.stdout | from_json] | map(attribute='active_name') | list) + (mgr_dump.stdout | from_json)['standbys'] | map(attribute='name') | list)
      retries: 12
      delay: 10

    - name: Purge manager store
      ansible.builtin.file:
        path: /var/lib/ceph/mgr/{{ cluster }}-{{ mgr_to_kill_hostname }}
        state: absent
      delegate_to: "{{ mgr_to_kill }}"

  post_tasks:
    - name: Show ceph health
      ansible.builtin.command: "{{ container_exec_cmd | default('') }} ceph --cluster {{ cluster }} -s"
      changed_when: false