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

---
# This playbook shrinks Ceph OSDs.
# It can remove any number of OSD(s) from the cluster and ALL THEIR DATA
#
# Use it like this:
# ansible-playbook shrink-osd.yml -e osd_to_kill=0,2,6
#     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-osd.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') }}"
    - "{{ osd_group_name|default('osds') }}"

  become: True
  tasks:
    - debug: msg="gather facts on all Ceph hosts for following reference"

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

  hosts:
    - localhost

  become: true

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

  vars:
    mon_group_name: mons
    osd_group_name: osds

  pre_tasks:
    - name: exit playbook, if user did not mean to shrink cluster
      fail:
        msg: "Exiting shrink-osd playbook, no osd(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 osd(s) was/were given
      fail:
        msg: "osd_to_kill must be declared
          Exiting shrink-osd playbook, no OSD(s) was/were removed.
           On the command line when invoking the playbook, you can use
           -e osd_to_kill=0,1,2,3 argument."
      when: osd_to_kill is not defined

  roles:
    - ceph-defaults

  post_tasks:

    - name: set_fact docker_exec_cmd build docker exec command (containerized)
      set_fact:
        docker_exec_cmd: "docker exec ceph-mon-{{ hostvars[groups[mon_group_name][0]]['ansible_hostname'] }}"
      when: containerized_deployment

    - name: exit playbook, if can not connect to the cluster
      command: "{{ docker_exec_cmd }} timeout 5 ceph --cluster {{ cluster }} health"
      register: ceph_health
      until: ceph_health.stdout.find("HEALTH") > -1
      delegate_to: "{{ groups[mon_group_name][0] }}"
      retries: 5
      delay: 2

    - name: find the host(s) where the osd(s) is/are running on
      command: "{{ docker_exec_cmd }} ceph --cluster {{ cluster }} osd find {{ item }}"
      with_items: "{{ osd_to_kill.split(',') }}"
      delegate_to: "{{ groups[mon_group_name][0] }}"
      register: find_osd_hosts

    - name: set_fact osd_hosts
      set_fact:
        osd_hosts: "{{ osd_hosts | default([]) + [ (item.stdout | from_json).crush_location.host ] }}"
      with_items: "{{ find_osd_hosts.results }}"

    - name: check if ceph admin key exists on the osd nodes
      stat:
        path: "/etc/ceph/{{ cluster }}.client.admin.keyring"
      register: ceph_admin_key
      with_items: "{{ osd_hosts }}"
      delegate_to: "{{ item }}"
      failed_when: false
      when:
        - not containerized_deployment

    - name: fail when admin key is not present
      fail:
        msg: "The Ceph admin key is not present on the OSD node, please add it and remove it after the playbook is done."
      with_items: "{{ ceph_admin_key.results }}"
      when:
        - not containerized_deployment
        - item.stat.exists == false

    # NOTE(leseb): using '>' is the only way I could have the command working
    - name: find osd device based on the id
      shell: >
        docker run --privileged=true -v /dev:/dev --entrypoint /usr/sbin/ceph-disk
        {{ ceph_docker_registry}}/{{ ceph_docker_image }}:{{ ceph_docker_image_tag }}
        list | awk  -v pattern=osd.{{ item.0 }} '$0 ~ pattern {print $1}'
      with_together:
        - "{{ osd_to_kill.split(',') }}"
        - "{{ osd_hosts }}"
      register: osd_to_kill_disks
      delegate_to: "{{ item.1 }}"
      when:
        - containerized_deployment

    # if nvme then osd_to_kill_disks is nvme0n1, we need nvme0
    # if ssd or hdd then osd_to_kill_disks is sda1, we need sda
    - name: stop osd services (container)
      service:
        name: "ceph-osd@{{ item.0.stdout[:-2] | regex_replace('/dev/', '') if 'nvme' in item.0.stdout else item.0.stdout[:-1] | regex_replace('/dev/', '') }}"
        state: stopped
        enabled: no
      with_together:
        - "{{ osd_to_kill_disks.results }}"
        - "{{ osd_hosts }}"
      delegate_to: "{{ item.1 }}"
      when:
        - containerized_deployment

    - name: resolve parent device
      command: lsblk --nodeps -no pkname "{{ item.stdout }}"
      register: resolved_parent_device
      with_items:
        - "{{ osd_to_kill_disks.results }}"
      when:
        - containerized_deployment

    - name: zap ceph osd disks
      shell: |
        docker run --rm \
        --privileged=true \
        --name ceph-osd-zap-{{ ansible_hostname }}-{{ item.stdout }} \
        -v /dev/:/dev/ \
        -e OSD_DEVICE=/dev/{{ item.stdout }} \
        {{ ceph_docker_registry }}/{{ ceph_docker_image }}:{{ ceph_docker_image_tag }} \
        zap_device
      with_items:
        - "{{ resolved_parent_device.results }}"
      when:
        - containerized_deployment

    - name: deactivating osd(s)
      command: ceph-disk deactivate --cluster {{ cluster }} --deactivate-by-id {{ item.0 }} --mark-out
      run_once: true
      with_together:
        - "{{ osd_to_kill.split(',') }}"
        - "{{ osd_hosts }}"
      delegate_to: "{{ item.1 }}"
      when:
        - not containerized_deployment

    - name: destroying osd(s)
      command: ceph-disk destroy --cluster {{ cluster }} --destroy-by-id {{ item.0 }} --zap
      run_once: true
      with_together:
        - "{{ osd_to_kill.split(',') }}"
        - "{{ osd_hosts }}"
      delegate_to: "{{ item.1 }}"
      when:
        - not containerized_deployment

    - name: remove osd(s) from crush_map when ceph-disk destroy fail
      command: "{{ docker_exec_cmd }} ceph --cluster {{ cluster }} osd crush remove osd.{{ item }}"
      run_once: true
      delegate_to: "{{ groups[mon_group_name][0] }}"
      with_items: "{{ osd_to_kill.split(',') }}"

    - name: delete osd(s) auth key when ceph-disk destroy fail
      command: "{{ docker_exec_cmd }} ceph --cluster {{ cluster }} auth del osd.{{ item }}"
      delegate_to: "{{ groups[mon_group_name][0] }}"
      with_items: "{{ osd_to_kill.split(',') }}"

    - name: deallocate osd(s) id when ceph-disk destroy fail
      command: "{{ docker_exec_cmd }} ceph --cluster {{ cluster }} osd rm {{ item }}"
      delegate_to: "{{ groups[mon_group_name][0] }}"
      with_items: "{{ osd_to_kill.split(',') }}"

    - name: show ceph health
      command: "{{ docker_exec_cmd }} ceph --cluster {{ cluster }} -s"
      delegate_to: "{{ groups[mon_group_name][0] }}"

    - name: show ceph osd tree
      command: "{{ docker_exec_cmd }} ceph --cluster {{ cluster }} osd tree"
      delegate_to: "{{ groups[mon_group_name][0] }}"