ceph-ansible/roles/ceph-handler/templates/restart_osd_daemon.sh.j2

#!/bin/bash

DELAY="{{ handler_health_osd_check_delay }}"
CEPH_CLI="--name client.bootstrap-osd --keyring /var/lib/ceph/bootstrap-osd/{{ cluster }}.keyring --cluster {{ cluster }}"

check_pgs() {
  num_pgs=$($container_exec ceph $CEPH_CLI -s -f json | "{{ discovered_interpreter_python }}" -c 'import sys, json; print(json.load(sys.stdin)["pgmap"]["num_pgs"])')
  if [[ "$num_pgs" == "0" ]]; then
    return 0
  fi
  while [ $RETRIES -ne 0 ]; do
    test "$($container_exec ceph $CEPH_CLI -s -f json | "{{ discovered_interpreter_python }}" -c 'import sys, json; print(json.load(sys.stdin)["pgmap"]["num_pgs"])')" -eq "$($container_exec ceph $CEPH_CLI -s -f json | "{{ discovered_interpreter_python }}" -c 'import sys, json; print(sum ( [ i["count"] for i in json.load(sys.stdin)["pgmap"]["pgs_by_state"] if "active+clean" in i["state_name"]]))')"
    RET=$?
    test $RET -eq 0 && return 0
    sleep $DELAY
    let RETRIES=RETRIES-1
  done
  # PGs not clean, exiting with return code 1
  echo "Error while running 'ceph $CEPH_CLI -s', PGs were not reported as active+clean"
  echo "It is possible that the cluster has less OSDs than the replica configuration"
  echo "Will refuse to continue"
  $container_exec ceph $CEPH_CLI -s
  $container_exec ceph $CEPH_CLI osd dump
  $container_exec ceph $CEPH_CLI osd tree
  $container_exec ceph $CEPH_CLI osd crush rule dump
  exit 1
}

wait_for_socket_in_container() {
  osd_mount_point=$({{ container_binary }} exec "$1" df --output=target | grep '/var/lib/ceph/osd/')
  whoami=$({{ container_binary }} exec "$1" cat $osd_mount_point/whoami)
  if ! {{ container_binary }} exec "$1" timeout 10 bash -c "while [ ! -e /var/run/ceph/{{ cluster }}-osd.${whoami}.asok ]; do sleep 1 ; done"; then
    echo "Timed out while trying to look for a Ceph OSD socket."
    echo "Abort mission!"
    exit 1
  fi
}

get_dev_name() {
  echo $1 | sed -r 's/ceph-osd@([a-z]{1,4})\.service/\1/'
}

get_container_id_from_dev_name() {
  local id
  local count
  count=10
  while [ $count -ne 0 ]; do
    id=$({{ container_binary }} ps | grep -E "${1}$" | cut -d' ' -f1)
    test "$id" != "" && break
    sleep $DELAY
    let count=count-1
  done
  echo "$id"
}

# For containerized deployments, the unit file looks like: ceph-osd@sda.service
# For non-containerized deployments, the unit file looks like: ceph-osd@NNN.service where NNN is OSD ID
for unit in $(systemctl list-units | grep -E "loaded * active" | grep -oE "ceph-osd@([0-9]+).service"); do
  # First, restart daemon(s)
  systemctl restart "${unit}"
  # We need to wait because it may take some time for the socket to actually exists
  COUNT=10
  # Wait and ensure the socket exists after restarting the daemon
  {% if containerized_deployment | bool %}
  osd_id=$(echo ${unit#ceph-osd@} | grep -oE '[0-9]+')
  container_id=$(get_container_id_from_dev_name "ceph-osd-${osd_id}")
  container_exec="{{ container_binary }} exec $container_id"
  {% else %}
  osd_id=$(echo ${unit#ceph-osd@} | grep -oE '[0-9]+')
  {% endif %}
  SOCKET=/var/run/ceph/{{ cluster }}-osd.${osd_id}.asok
  while [ $COUNT -ne 0 ]; do
    RETRIES="{{ handler_health_osd_check_retries }}"
    $container_exec test -S "$SOCKET" && check_pgs && continue 2
    sleep $DELAY
    let COUNT=COUNT-1
  done
  # If we reach this point, it means the socket is not present.
  echo "Socket file ${SOCKET} could not be found, which means the osd daemon is not running. Showing ceph-osd unit logs now:"
  journalctl -u "${unit}"
  exit 1
done