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

---
# This playbook shrinks the Ceph monitors from your cluster
# It can remove any number of monitor(s) from the cluster and ALL THEIR DATA
#
# Use it like this:
# ansible-playbook shrink-mon.yml -e mon_host=ceph-mon01,ceph-mon02
#     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-cluster.yml
#     Overrides the prompt using -e option. Can be used in
#     automation scripts to avoid interactive prompt.


- name: confirm whether user really meant to remove monitor(s) from the ceph cluster

  hosts:
    - localhost

  gather_facts: false
  become: true

  vars_prompt:
    - name: ireallymeanit
      prompt: Are you sure you want to shrink the cluster?
      default: 'no'
      private: no

  tasks:
  - include_vars: roles/ceph-common/defaults/main.yml
  - include_vars: group_vars/all

  - name: exit playbook, if user did not mean to shrink cluster
    fail:
      msg: "Exiting shrink-mon playbook, no monitor(s) was/were 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: exit playbook, if no monitor(s) was/were given
    fail:
      msg: "mon_host must be declared
        Exiting shrink-cluster playbook, no monitor(s) was/were removed.
         On the command line when invoking the playbook, you can use
         -e mon_host=ceph-mon01,ceph-mon02 argument."
    when: mon_host is not defined

  - name: test if ceph command exist
    command: command -v ceph
    changed_when: false
    failed_when: false
    register: ceph_command

  - name: exit playbook, if ceph command does not exist
    debug:
      msg: "The ceph command is not available, please install it :("
    run_once: true
    when:
      - ceph_command.rc != 0

  - name: exit playbook, if cluster files do not exist
    stat:
      path: "{{ item }}"
    register: ceph_conf_key
    with_items:
      - /etc/ceph/{{ cluster }}.conf
      - /etc/ceph/{{ cluster }}.client.admin.keyring
    failed_when: false

  - fail:
      msg: "Ceph's configuration file is not present in /etc/ceph"
    with_items: "{{ceph_conf_key.results}}"
    when:
      -  item.stat.exists == false

  - name: exit playbook, if can not connect to the cluster
    command: timeout 5 ceph --cluster {{ cluster }} health
    register: ceph_health
    until: ceph_health.stdout.find("HEALTH") > -1
    retries: 5
    delay: 2

  - name: verify given monitors are reachable
    command: ping -c 1 {{ item }}
    with_items: "{{mon_host.split(',')}}"
    register: mon_reachable
    failed_when: false

  - fail:
      msg: "One or more monitors are not reachable, please check your /etc/hosts or your DNS"
    with_items: "{{mon_reachable.results}}"
    when:
      -  item.rc != 0

  - name: stop monitor service (systemd)
    service:
      name: ceph-mon@{{ item }}
      state: stopped
      enabled: no
    with_items: "{{mon_host.split(',')}}"
    delegate_to: "{{item}}"
    failed_when: false

  - name: purge monitor store
    file:
      path: /var/lib/ceph/mon/{{ cluster }}-{{ item }}
      state: absent
    with_items: "{{mon_host.split(',')}}"
    delegate_to: "{{item}}"

  - name: remove monitor from the quorum
    command: ceph --cluster {{ cluster }} mon remove {{ item }}
    failed_when: false
    with_items: "{{mon_host.split(',')}}"

  # NOTE (leseb): sorry for the 'sleep' command
  # but it will take a couple of seconds for other monitors
  # to notice that one member has left.
  # 'sleep 5' is not that bad and should be sufficient
  - name: verify the monitor is out of the cluster
    shell: "sleep 5 && ceph --cluster {{ cluster }} -s | grep monmap | sed 's/.*quorum//' | egrep -sq {{ item }}"
    with_items: "{{mon_host.split(',')}}"
    failed_when: false
    register: ceph_health_mon

  - name: please remove the monitor from your ceph configuration file
    debug:
        msg: "The monitor(s) has/have been successfully removed from the cluster.
        Please remove the monitor(s) entry(ies) from the rest of your ceph configuration files, cluster wide."
    run_once: true
    with_items: "{{ceph_health_mon.results}}"
    when:
      - item.rc != 0

  - name: please remove the monitor from your ceph configuration file
    fail:
        msg: "Monitor(s) appear(s) to still be part of the cluster, please check what happened."
    run_once: true
    with_items: "{{ceph_health_mon.results}}"
    when:
      - item.rc == 0