ceph-ansible/infrastructure-playbooks/switch-from-non-containerized-to-containerized-ceph-daemons.yml

---
# This playbook switches from non-containerized to containerized Ceph daemons

- name: confirm whether user really meant to switch from non-containerized to containerized ceph daemons

  hosts: localhost

  gather_facts: false
  any_errors_fatal: true
  vars_prompt:
    - name: ireallymeanit
      prompt: Are you sure you want to switch from non-containerized to containerized ceph daemons?
      default: 'no'
      private: no

  tasks:
    - import_role:
        name: ceph-defaults

    - name: fail when less than three monitors
      fail:
        msg: "This playbook requires at least three monitors."
      when: groups[mon_group_name] | length | int < 3

    - name: exit playbook, if user did not mean to switch from non-containerized to containerized daemons?
      fail:
        msg: >
          "Exiting switch-from-non-containerized-to-containerized-ceph-daemons.yml playbook,
           cluster did not switch from non-containerized to containerized ceph daemons.
           To switch from non-containerized to containerized ceph daemons, either say 'yes' on the prompt or
           or use `-e ireallymeanit=yes` on the command line when
           invoking the playbook"          
      when: ireallymeanit != 'yes'


- name: gather facts

  hosts:
    - "{{ mon_group_name|default('mons') }}"
    - "{{ mgr_group_name|default('mgrs') }}"
    - "{{ osd_group_name|default('osds') }}"
    - "{{ mds_group_name|default('mdss') }}"
    - "{{ rgw_group_name|default('rgws') }}"
    - "{{ rbdmirror_group_name|default('rbdmirrors') }}"
    - "{{ nfs_group_name|default('nfss') }}"

  become: true

  vars:
    delegate_facts_host: True

  tasks:
    - import_role:
        name: ceph-defaults

    - name: gather and delegate facts
      setup:
        gather_subset:
          - 'all'
          - '!facter'
          - '!ohai'
      delegate_to: "{{ item }}"
      delegate_facts: True
      with_items: "{{ groups['all'] | difference(groups.get(client_group_name, [])) }}"
      run_once: true
      when: delegate_facts_host | bool
      tags: always

    - import_role:
        name: ceph-facts
    - import_role:
        name: ceph-validate

- name: switching from non-containerized to containerized ceph mon
  vars:
    containerized_deployment: true
    switch_to_containers: True
    mon_group_name:       mons
  hosts: "{{ mon_group_name|default('mons') }}"
  serial: 1
  become: true
  pre_tasks:
    - name: select a running monitor
      set_fact: mon_host={{ item }}
      with_items: "{{ groups[mon_group_name] }}"
      when: item != inventory_hostname

    - name: stop non-containerized ceph mon
      service:
        name: "ceph-mon@{{ ansible_facts['hostname'] }}"
        state: stopped
        enabled: no

    - name: remove old systemd unit files
      file:
        path: "{{ item }}"
        state: absent
      with_items:
        - /usr/lib/systemd/system/ceph-mon@.service
        - /usr/lib/systemd/system/ceph-mon.target
        - /lib/systemd/system/ceph-mon@.service
        - /lib/systemd/system/ceph-mon.target

    - import_role:
        name: ceph-defaults

    - import_role:
        name: ceph-facts

    # NOTE: changed from file module to raw find command for performance reasons
    # The file module has to run checks on current ownership of all directories and files. This is unnecessary
    # as in this case we know we want all owned by ceph user
    - name: set proper ownership on ceph directories
      command: "find /var/lib/ceph/mon /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown -h {{ ceph_uid }}:{{ ceph_uid }} {} +"
      changed_when: false

    - name: check for existing old leveldb file extension (ldb)
      shell: stat /var/lib/ceph/mon/*/store.db/*.ldb
      changed_when: false
      failed_when: false
      register: ldb_files

    - name: rename leveldb extension from ldb to sst
      shell: rename -v .ldb .sst /var/lib/ceph/mon/*/store.db/*.ldb
      changed_when: false
      failed_when: false
      when: ldb_files.rc == 0

    - name: copy mon initial keyring in /etc/ceph to satisfy fetch config task in ceph-container-common
      command: cp /var/lib/ceph/mon/{{ cluster }}-{{ ansible_facts['hostname'] }}/keyring /etc/ceph/{{ cluster }}.mon.keyring
      args:
        creates: /etc/ceph/{{ cluster }}.mon.keyring
      changed_when: false
      failed_when: false

  tasks:
    - import_role:
        name: ceph-handler

    - import_role:
        name: ceph-container-engine

    - import_role:
        name: ceph-container-common

    - import_role:
        name: ceph-mon

  post_tasks:
    - name: waiting for the monitor to join the quorum...
      command: "{{ container_binary }} run --rm  -v /etc/ceph:/etc/ceph:z --entrypoint=ceph {{ ceph_docker_registry }}/{{ ceph_docker_image }}:{{ ceph_docker_image_tag }} --cluster {{ cluster }} quorum_status --format json"
      register: ceph_health_raw
      changed_when: false
      until: ansible_facts['hostname'] in (ceph_health_raw.stdout | trim | from_json)["quorum_names"]
      retries: "{{ health_mon_check_retries }}"
      delay: "{{ health_mon_check_delay }}"

- name: switching from non-containerized to containerized ceph mgr

  hosts: "{{ mgr_group_name|default('mgrs') }}"

  vars:
    containerized_deployment: true
    mgr_group_name: mgrs

  serial: 1
  become: true
  pre_tasks:
    # failed_when: false is here because if we're
    # working with a jewel cluster then ceph mgr
    # will not exist
    - name: stop non-containerized ceph mgr(s)
      service:
        name: "ceph-mgr@{{ ansible_facts['hostname'] }}"
        state: stopped
        enabled: no
      failed_when: false

    - name: remove old systemd unit files
      file:
        path: "{{ item }}"
        state: absent
      with_items:
        - /usr/lib/systemd/system/ceph-mgr@.service
        - /usr/lib/systemd/system/ceph-mgr.target
        - /lib/systemd/system/ceph-mgr@.service
        - /lib/systemd/system/ceph-mgr.target

    - import_role:
        name: ceph-defaults

    - import_role:
        name: ceph-facts

    # NOTE: changed from file module to raw find command for performance reasons
    # The file module has to run checks on current ownership of all directories and files. This is unnecessary
    # as in this case we know we want all owned by ceph user
    - name: set proper ownership on ceph directories
      command: "find /var/lib/ceph/mgr /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown -h {{ ceph_uid }}:{{ ceph_uid }} {} +"
      changed_when: false

  tasks:
    - import_role:
        name: ceph-handler

    - import_role:
        name: ceph-container-engine

    - import_role:
        name: ceph-container-common

    - import_role:
        name: ceph-mgr


- name: set osd flags
  hosts: "{{ mon_group_name | default('mons') }}[0]"
  become: True
  tasks:
    - import_role:
        name: ceph-defaults
    - import_role:
        name: ceph-facts

    - name: get pool list
      command: "{{ ceph_cmd }} --cluster {{ cluster }} osd pool ls detail -f json"
      register: pool_list
      changed_when: false
      check_mode: false

    - name: get balancer module status
      command: "{{ ceph_cmd }} --cluster {{ cluster }} balancer status -f json"
      register: balancer_status_switch
      changed_when: false
      check_mode: false

    - name: set_fact pools_pgautoscaler_mode
      set_fact:
        pools_pgautoscaler_mode: "{{ pools_pgautoscaler_mode | default([]) | union([{'name': item.pool_name, 'mode': item.pg_autoscale_mode}]) }}"
      with_items: "{{ pool_list.stdout | default('{}') | from_json }}"

    - name: disable balancer
      command: "{{ ceph_cmd }} --cluster {{ cluster }} balancer off"
      changed_when: false
      when: (balancer_status_switch.stdout | from_json)['active'] | bool

    - name: disable pg autoscale on pools
      command: "{{ ceph_cmd }} --cluster {{ cluster }} osd pool set {{ item.name }} pg_autoscale_mode off"
      with_items: "{{ pools_pgautoscaler_mode }}"
      when:
        - pools_pgautoscaler_mode is defined
        - item.mode == 'on'

    - name: set osd flags
      ceph_osd_flag:
        name: "{{ item }}"
        cluster: "{{ cluster }}"
      environment:
        CEPH_CONTAINER_IMAGE: "{{ ceph_docker_registry + '/' + ceph_docker_image + ':' + ceph_docker_image_tag if containerized_deployment | bool else None }}"
        CEPH_CONTAINER_BINARY: "{{ container_binary }}"
      with_items:
        - noout
        - nodeep-scrub


- name: switching from non-containerized to containerized ceph osd

  vars:
    containerized_deployment: true
    osd_group_name: osds
    switch_to_containers: True

  hosts: "{{ osd_group_name|default('osds') }}"

  serial: 1
  become: true
  pre_tasks:

    - import_role:
        name: ceph-defaults

    - name: collect running osds
      shell: |
        set -o pipefail;
        systemctl list-units | grep -E "loaded * active" | grep -Eo 'ceph-osd@[0-9]+.service|ceph-volume'        
      register: running_osds
      changed_when: false
      failed_when: false

    # systemd module does not support --runtime option
    - name: disable ceph-osd@.service runtime-enabled
      command: "systemctl disable --runtime {{ item }}"  # noqa 303
      changed_when: false
      failed_when: false
      with_items: "{{ running_osds.stdout_lines | default([]) }}"
      when: item.startswith('ceph-osd@')

    - name: stop/disable/mask non-containerized ceph osd(s) (if any)
      systemd:
        name: "{{ item }}"
        state: stopped
        enabled: no
      with_items: "{{ running_osds.stdout_lines | default([]) }}"
      when: running_osds != []

    - name: disable ceph.target
      systemd:
        name: ceph.target
        enabled: no

    - name: remove old ceph-osd systemd units
      file:
        path: "{{ item }}"
        state: absent
      with_items:
        - /usr/lib/systemd/system/ceph-osd.target
        - /usr/lib/systemd/system/ceph-osd@.service
        - /usr/lib/systemd/system/ceph-volume@.service
        - /lib/systemd/system/ceph-osd.target
        - /lib/systemd/system/ceph-osd@.service
        - /lib/systemd/system/ceph-volume@.service

    - import_role:
        name: ceph-facts

    # NOTE: changed from file module to raw find command for performance reasons
    # The file module has to run checks on current ownership of all directories and files. This is unnecessary
    # as in this case we know we want all owned by ceph user
    - name: set proper ownership on ceph directories
      command: "find /var/lib/ceph/osd /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown -h {{ ceph_uid }}:{{ ceph_uid }} {} +"
      changed_when: false

    - name: check for existing old leveldb file extension (ldb)
      shell: stat /var/lib/ceph/osd/*/current/omap/*.ldb
      changed_when: false
      failed_when: false
      register: ldb_files

    - name: rename leveldb extension from ldb to sst
      shell: rename -v .ldb .sst /var/lib/ceph/osd/*/current/omap/*.ldb
      changed_when: false
      failed_when: false
      when: ldb_files.rc == 0

    - name: check if containerized osds are already running
      command: >
        {{ container_binary }} ps -q --filter='name=ceph-osd'        
      changed_when: false
      failed_when: false
      register: osd_running

    - name: get osd directories
      command: >
        find /var/lib/ceph/osd {% if dmcrypt | bool %}/var/lib/ceph/osd-lockbox{% endif %} -maxdepth 1 -mindepth 1 -type d        
      register: osd_dirs
      changed_when: false
      failed_when: false

    - name: unmount all the osd directories
      command: >
        umount {{ item }}        
      changed_when: false
      failed_when: false
      with_items: "{{ osd_dirs.stdout_lines }}"
      when: osd_running.rc != 0 or osd_running.stdout_lines | length == 0

  tasks:
    - import_role:
        name: ceph-handler

    - import_role:
        name: ceph-container-engine

    - import_role:
        name: ceph-container-common

    - import_role:
        name: ceph-osd

  post_tasks:
    - name: get num_pgs
      command: >
        {{ container_binary }} exec ceph-mon-{{ hostvars[groups[mon_group_name][0]]['ansible_facts']['hostname'] }} ceph --cluster {{ cluster }} pg stat --format json        
      register: ceph_pgs
      delegate_to: "{{ groups[mon_group_name][0] }}"
      changed_when: false

    - name: container - waiting for clean pgs...
      command: >
        {{ container_binary }} exec ceph-mon-{{ hostvars[groups[mon_group_name][0]]['ansible_facts']['hostname'] }} ceph --cluster {{ cluster }} pg stat --format json        
      register: ceph_health_post
      until: >
        (((ceph_health_post.stdout | from_json).pg_summary.num_pg_by_state | length) > 0)
        and
        (((ceph_health_post.stdout | from_json).pg_summary.num_pg_by_state | selectattr('name', 'search', '^active\\+clean') | map(attribute='num') | list | sum) == (ceph_pgs.stdout | from_json).pg_summary.num_pgs)        
      delegate_to: "{{ groups[mon_group_name][0] }}"
      retries: "{{ health_osd_check_retries }}"
      delay: "{{ health_osd_check_delay }}"
      when: (ceph_pgs.stdout | from_json).pg_summary.num_pgs != 0
      changed_when: false


- name: unset osd flags
  hosts: "{{ mon_group_name | default('mons') }}[0]"
  become: True
  tasks:
    - import_role:
        name: ceph-defaults
    - import_role:
        name: ceph-facts

    - name: re-enable pg autoscale on pools
      command: "{{ ceph_cmd }} --cluster {{ cluster }} osd pool set {{ item.name }} pg_autoscale_mode on"
      with_items: "{{ pools_pgautoscaler_mode }}"
      when:
        - pools_pgautoscaler_mode is defined
        - item.mode == 'on'

    - name: unset osd flags
      ceph_osd_flag:
        name: "{{ item }}"
        cluster: "{{ cluster }}"
        state: absent
      environment:
        CEPH_CONTAINER_IMAGE: "{{ ceph_docker_registry + '/' + ceph_docker_image + ':' + ceph_docker_image_tag if containerized_deployment | bool else None }}"
        CEPH_CONTAINER_BINARY: "{{ container_binary }}"
      with_items:
        - noout
        - nodeep-scrub

    - name: re-enable balancer
      command: "{{ ceph_cmd }} --cluster {{ cluster }} balancer on"
      changed_when: false
      when: (balancer_status_switch.stdout | from_json)['active'] | bool


- name: switching from non-containerized to containerized ceph mds

  hosts: "{{ mds_group_name|default('mdss') }}"

  vars:
    containerized_deployment: true
    mds_group_name: mdss

  serial: 1
  become: true
  pre_tasks:

    - name: stop non-containerized ceph mds(s)
      service:
        name: "ceph-mds@{{ ansible_facts['hostname'] }}"
        state: stopped
        enabled: no

    - name: remove old systemd unit files
      file:
        path: "{{ item }}"
        state: absent
      with_items:
        - /usr/lib/systemd/system/ceph-mds@.service
        - /usr/lib/systemd/system/ceph-mds.target
        - /lib/systemd/system/ceph-mds@.service
        - /lib/systemd/system/ceph-mds.target

    - import_role:
        name: ceph-defaults

    - import_role:
        name: ceph-facts

    # NOTE: changed from file module to raw find command for performance reasons
    # The file module has to run checks on current ownership of all directories and files. This is unnecessary
    # as in this case we know we want all owned by ceph user
    - name: set proper ownership on ceph directories
      command: "find /var/lib/ceph/mds /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown {{ ceph_uid }}:{{ ceph_uid }} {} +"
      changed_when: false

  tasks:
    - import_role:
        name: ceph-handler

    - import_role:
        name: ceph-container-engine

    - import_role:
        name: ceph-container-common

    - import_role:
         name: ceph-mds


- name: switching from non-containerized to containerized ceph rgw

  hosts: "{{ rgw_group_name|default('rgws') }}"

  vars:
    containerized_deployment: true
    rgw_group_name: rgws

  serial: 1
  become: true
  pre_tasks:
    - import_role:
        name: ceph-defaults

    - import_role:
        name: ceph-facts

    - import_role:
        name: ceph-config
        tasks_from: rgw_systemd_environment_file.yml

    # NOTE: changed from file module to raw find command for performance reasons
    # The file module has to run checks on current ownership of all directories and files. This is unnecessary
    # as in this case we know we want all owned by ceph user
    - name: set proper ownership on ceph directories
      command: "find /var/lib/ceph/radosgw /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown {{ ceph_uid }}:{{ ceph_uid }} {} +"
      changed_when: false

  tasks:
    - name: stop non-containerized ceph rgw(s)
      service:
        name: "ceph-radosgw@rgw.{{ ansible_facts['hostname'] }}.{{ item.instance_name }}"
        state: stopped
        enabled: no
      with_items: "{{ rgw_instances }}"

    - name: remove old systemd unit files
      file:
        path: "{{ item }}"
        state: absent
      with_items:
        - /usr/lib/systemd/system/ceph-radosgw@.service
        - /usr/lib/systemd/system/ceph-radosgw.target
        - /lib/systemd/system/ceph-radosgw@.service
        - /lib/systemd/system/ceph-radosgw.target

    - import_role:
        name: ceph-handler

    - import_role:
        name: ceph-container-engine

    - import_role:
        name: ceph-container-common

    - import_role:
        name: ceph-rgw


- name: switching from non-containerized to containerized ceph rbd-mirror

  hosts: "{{ rbdmirror_group_name|default('rbdmirrors') }}"

  vars:
    containerized_deployment: true
    rbdmirror_group_name: rbdmirrors

  serial: 1
  become: true
  pre_tasks:
    - name: stop non-containerized ceph rbd mirror(s)
      service:
        name: "ceph-rbd-mirror@rbd-mirror.{{ ansible_facts['hostname'] }}"
        state: stopped
        enabled: no

    - name: remove old systemd unit files
      file:
        path: "{{ item }}"
        state: absent
      with_items:
        - /usr/lib/systemd/system/ceph-rbd-mirror@.service
        - /usr/lib/systemd/system/ceph-rbd-mirror.target
        - /lib/systemd/system/ceph-rbd-mirror@.service
        - /lib/systemd/system/ceph-rbd-mirror.target

    - import_role:
        name: ceph-defaults

    - import_role:
        name: ceph-facts

    # NOTE: changed from file module to raw find command for performance reasons
    # The file module has to run checks on current ownership of all directories and files. This is unnecessary
    # as in this case we know we want all owned by ceph user
    - name: set proper ownership on ceph directories
      command: "find /var/lib/ceph /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown {{ ceph_uid }}:{{ ceph_uid }} {} +"
      changed_when: false

  tasks:
    - import_role:
        name: ceph-handler

    - import_role:
        name: ceph-container-engine

    - import_role:
        name: ceph-container-common

    - import_role:
        name: ceph-rbd-mirror


- name: switching from non-containerized to containerized ceph nfs

  hosts: "{{ nfs_group_name|default('nfss') }}"

  vars:
    containerized_deployment: true
    nfs_group_name: nfss

  serial: 1
  become: true
  pre_tasks:

    # failed_when: false is here because if we're
    # working with a jewel cluster then ceph nfs
    # will not exist
    - name: stop non-containerized ceph nfs(s)
      service:
        name: nfs-ganesha
        state: stopped
        enabled: no
      failed_when: false

    - import_role:
        name: ceph-defaults

    - import_role:
        name: ceph-facts

    # NOTE: changed from file module to raw find command for performance reasons
    # The file module has to run checks on current ownership of all directories and files. This is unnecessary
    # as in this case we know we want all owned by ceph user
    - name: set proper ownership on ceph directories
      command: "find /var/lib/ceph /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown {{ ceph_uid }}:{{ ceph_uid }} {} +"
      changed_when: false

  tasks:
    - import_role:
        name: ceph-handler

    - import_role:
        name: ceph-container-engine

    - import_role:
        name: ceph-container-common

    - import_role:
        name: ceph-nfs

- name: switching from non-containerized to containerized iscsigws
  hosts: "{{ iscsi_gw_group_name|default('iscsigws') }}"
  vars:
    containerized_deployment: true
    iscsi_gw_group_name: iscsigws
  become: true
  serial: 1
  pre_tasks:
    - import_role:
        name: ceph-defaults

    - name: stop iscsigw services
      service:
        name: "{{ item }}"
        state: stopped
        enabled: no
      with_items:
        - tcmu-runner
        - rbd-target-gw
        - rbd-target-api

    - name: remove old systemd unit files
      file:
        path: "/usr/lib/systemd/system/{{ item }}.service"
        state: absent
      with_items:
        - tcmu-runner
        - rbd-target-gw
        - rbd-target-api
  tasks:
    - import_role:
        name: ceph-facts

    - import_role:
        name: ceph-handler

    # NOTE: changed from file module to raw find command for performance reasons
    # The file module has to run checks on current ownership of all directories and files. This is unnecessary
    # as in this case we know we want all owned by ceph user
    - name: set proper ownership on ceph directories
      command: "find /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown -h {{ ceph_uid }}:{{ ceph_uid }} {} +"
      changed_when: false

    - import_role:
        name: ceph-container-engine

    - import_role:
        name: ceph-container-common

    - import_role:
        name: ceph-iscsi-gw

- name: switching from non-containerized to containerized ceph-crash

  hosts:
    - "{{ mon_group_name | default('mons') }}"
    - "{{ osd_group_name | default('osds') }}"
    - "{{ mds_group_name | default('mdss') }}"
    - "{{ rgw_group_name | default('rgws') }}"
    - "{{ rbdmirror_group_name | default('rbdmirrors') }}"
    - "{{ mgr_group_name | default('mgrs') }}"

  vars:
    containerized_deployment: true
  become: true
  tasks:
    - name: stop non-containerized ceph-crash
      service:
        name: ceph-crash
        state: stopped
        enabled: no

    - import_role:
        name: ceph-defaults

    - import_role:
        name: ceph-facts
        tasks_from: container_binary.yml

    - import_role:
        name: ceph-handler

    - import_role:
        name: ceph-crash

- name: final task
  hosts:
    - "{{ mon_group_name|default('mons') }}"
    - "{{ mgr_group_name|default('mgrs') }}"
    - "{{ osd_group_name|default('osds') }}"
    - "{{ mds_group_name|default('mdss') }}"
    - "{{ rgw_group_name|default('rgws') }}"
  vars:
    containerized_deployment: true
  become: true
  tasks:
    - import_role:
        name: ceph-defaults
    # NOTE: changed from file module to raw find command for performance reasons
    # The file module has to run checks on current ownership of all directories and files. This is unnecessary
    # as in this case we know we want all owned by ceph user
    - name: set proper ownership on ceph directories
      command: "find /var/lib/ceph /etc/ceph -not -( -user {{ ceph_uid }} -or -group {{ ceph_uid }} -) -execdir chown {{ ceph_uid }}:{{ ceph_uid }} {} +"
      changed_when: false