http://tracker.ceph.com/issues/10411

gistfile1.txt

# Collect all down/incomplete PGs and create "inventory" files for playbook_remove_pg.yml per PG (separated by empty lines)
ceph health detail |
  grep -P "is (down\+)?incomplete, acting" |
  cut -d' ' -f2 |
  while read pg
  do
    ceph pg $pg query |
      jq -r '.recovery_state[0].probing_osds[]' |
        while read osd
        do
          host=$(ceph osd find $osd | jq -r '.crush_location.host')
          echo $host $osd
        done |
        sort |
        awk '{h=$1;if(a[h])a[h]=a[h] OFS $2;else{a[h]=$0;b[++i]=h}}END{for(x=1;x<=i;x++)print a[b[x]]}' |
        while read host osds
        do
          echo "$host osds=[$(echo $osds | tr ' ' ,)] pg=$pg"
        done
    echo
  done

playbook_remove_pg.yml

---
- hosts: all
  become: True
  tasks:
    - name: Check that PG exists
      stat:
        path: "/var/lib/ceph/osd/ceph-{{ item }}/current/{{ pg }}_head"
      register: pgstat
      with_items: "{{ osds }}"
    - debug: var=item
      with_items: "{{ pgstat.results }}"
    - name: Stop OSD
      service:
        name: "ceph-osd@{{ item.item }}"
        state: stopped
      with_items: "{{ pgstat.results }}"
      when: item.stat.exists
    - name: Move PG away
      shell: "mv /var/lib/ceph/osd/ceph-{{ item.item }}/current/{{ pg }}_head /root/backup_{{ item.item }}_{{ pg }}; sleep 10"
      args:
        creates: "/root/backup_{{ item.item }}_{{ pg }}"
      with_items: "{{ pgstat.results }}"
      when: item.stat.exists
    - name: Restart OSD
      service:
        name: "ceph-osd@{{ item.item }}"
        state: restarted
      with_items: "{{ pgstat.results }}"
      when: item.stat.exists
    - name: Restart OSD (post-op)
      service:
        name: "ceph-osd@{{ item.item }}"
        state: restarted
      with_items: "{{ pgstat.results }}"