gistfile1.sh

#!/usr/bin/bash

if [ "$1" != '--really' ]; then
  echo "command must be run as: $0 --really"
  exit 1
fi

# ensure the cluster has graceful_shutdown enabled
if ! /usr/bin/etcdctl get /deis/graceful_shutdown; then
  exit 0
fi

# procedure requires the store-admin
ADMIN_RUNNING=$(docker inspect --format="{{ .State.Running }}" deis-store-admin)
if [ $? -eq 1 ] || [ "$ADMIN_RUNNING" == "false" ]; then
  echo "deis-store-admin container is required for graceful shutdown"
  exit 1
fi

set -e -x -o pipefail

CURRENT_STATUS=$(/usr/bin/docker exec deis-store-admin ceph health | awk '{print $1}')
OSD_HOSTS=($(/usr/bin/etcdctl ls /deis/store/hosts/| awk -F'/' '{print $5}'))

for HOST in "${OSD_HOSTS[@]}"
do
  PUBLIC_IP=$(/usr/bin/etcdctl get /_coreos.com/fleet/machines/`cat /etc/machine-id`/object | /opt/bin/jq -r '.PublicIP')
  if [ "$HOST" = "$PUBLIC_IP" ] ; then
    OSD_ID=$(/usr/bin/etcdctl get /deis/store/osds/$PUBLIC_IP)
    break
  fi
done

# if we own an osd and its healthy, try to gracefully remove it
if [ -z "$OSD_ID" ] && [ "$CURRENT_STATUS" == "HEALTH_OK"]; then
  /usr/bin/docker exec deis-store-admin ceph osd out $OSD_ID

  sleep 30

  TIMEWAITED=0
  until /usr/bin/docker exec deis-store-admin ceph health | grep HEALTH_OK
  do
    if [ $TIMEWAITED -gt "600" ]
    then
      echo "ceph graceful removal timeout exceeded"
      break
    fi

    echo "waiting" && sleep 5
    TIMEWAITED=$((TIMEWAITED+5))
  done

  /usr/bin/docker stop deis-store-daemon

  /usr/bin/docker exec deis-store-admin ceph osd crush remove osd.$OSD_ID

  /usr/bin/docker exec deis-store-admin ceph auth del osd.$OSD_ID

  /usr/bin/docker exec deis-store-admin ceph osd rm $OSD_ID

  /usr/bin/etcdctl rm /deis/store/osds/$PUBLIC_IP

  etcdctl rm /deis/store/hosts/$PUBLIC_IP && sleep 10
  # remove ceph mon
  /usr/bin/docker stop deis-store-monitor

  /usr/bin/docker exec deis-store-admin ceph mon remove `hostname -f` # fixme

  /usr/bin/docker stop deis-store-metadata
fi

NODE=$(curl -L http://127.0.0.1:7001/v2/admin/machines/`cat /etc/machine-id`)

# remove from etcd cluster
if [ $NODE != 'null' ]; then
  /usr/bin/curl -L -XDELETE http://127.0.0.1:7001/v2/admin/machines/`cat /etc/machine-id`
fi

sudo systemctl stop fleet.service