#!/bin/bash
# Borrows heavily from Seán C. McCord's https://github.com/Ulexus/docker-ceph repository

set -e

ETCD_PORT=${ETCD_PORT:-4001}
ETCD="$HOST:$ETCD_PORT"
ETCD_PATH=${ETCD_PATH:-/deis/store}
HOSTNAME=`hostname`

function etcd_set_default {
  set +e
  etcdctl --no-sync -C $ETCD mk $ETCD_PATH/$1 $2 >/dev/null 2>&1
  if [[ $? -ne 0 && $? -ne 4 ]]; then
    echo "etcd_set_default: an etcd error occurred. aborting..."
    exit 1
  fi
  set -e
}

# set some defaults in etcd - these are templated in ceph.conf
# These defaults are sane for 3 hosts, and may need to be tweaked for larger clusters.
etcd_set_default delayStart 15
etcd_set_default size 3 # maintain 3 copies of all data
etcd_set_default minSize 1 # since we have 3 copies of data, the cluster can operate with just one host up
etcd_set_default pgNum 64 # this gives us a reasonable number of placement groups per host, assuming 3 hosts and 12 pools

# New clusters use 768 PGs per host (12 pools * 64 PGs per pool = 768 PGs per OSD)
# However, upgraded clusters may still use 128 PGs per pool, so we set this to 1536 PGs per host to suppress the
# "too many placement groups per host" warning
etcd_set_default maxPGsPerOSDWarning 1536

if ! etcdctl --no-sync -C $ETCD get ${ETCD_PATH}/monSetupComplete >/dev/null 2>&1 ; then
  echo "store-monitor: Ceph hasn't yet been deployed. Trying to deploy..."
  # let's rock and roll. we need to obtain a lock so we can ensure only one machine is trying to deploy the cluster
  if etcdctl --no-sync -C $ETCD mk ${ETCD_PATH}/monSetupLock $HOSTNAME >/dev/null 2>&1 \
  || [[ `etcdctl --no-sync -C $ETCD get ${ETCD_PATH}/monSetupLock` == "$HOSTNAME" ]] ; then
    echo "store-monitor: obtained the lock to proceed with setting up."

    # Generate administrator key
    ceph-authtool /etc/ceph/ceph.client.admin.keyring --create-keyring --gen-key -n client.admin --set-uid=0 --cap mon 'allow *' --cap osd 'allow *' --cap mds 'allow'

    # Generate the mon. key
    ceph-authtool /etc/ceph/ceph.mon.keyring --create-keyring --gen-key -n mon. --cap mon 'allow *'

    fsid=$(uuidgen)
    etcdctl --no-sync -C $ETCD set ${ETCD_PATH}/fsid ${fsid} >/dev/null

    # Generate initial monitor map
    monmaptool --create --add ${HOSTNAME} ${HOST} --fsid ${fsid} /etc/ceph/monmap

    etcdctl --no-sync -C $ETCD set ${ETCD_PATH}/monKeyring < /etc/ceph/ceph.mon.keyring >/dev/null
    etcdctl --no-sync -C $ETCD set ${ETCD_PATH}/adminKeyring < /etc/ceph/ceph.client.admin.keyring >/dev/null

    # mark setup as complete
    echo "store-monitor: setup complete."
    etcdctl --no-sync -C $ETCD set ${ETCD_PATH}/monSetupComplete youBetcha >/dev/null
  else
    until etcdctl --no-sync -C $ETCD get ${ETCD_PATH}/monSetupComplete >/dev/null 2>&1 ; do
      echo "store-monitor: waiting for another monitor to complete setup..."
      sleep 5
    done
  fi
fi

until confd -onetime -node $ETCD --confdir /app --log-level error; do
  echo "store-monitor: waiting for confd to write initial templates..."
  sleep 5
done

# If we don't have a monitor keyring, this is a new monitor
if [ ! -e /var/lib/ceph/mon/ceph-${HOSTNAME}/keyring ]; then
  if [ ! -f /etc/ceph/monmap ]; then
    ceph mon getmap -o /etc/ceph/monmap
  fi

  # Import the client.admin keyring and the monitor keyring into a new, temporary one
  ceph-authtool /tmp/ceph.mon.keyring --create-keyring --import-keyring /etc/ceph/ceph.client.admin.keyring
  ceph-authtool /tmp/ceph.mon.keyring --import-keyring /etc/ceph/ceph.mon.keyring

  # Make the monitor directory
  mkdir -p /var/lib/ceph/mon/ceph-${HOSTNAME}

  # Prepare the monitor daemon's directory with the map and keyring
  ceph-mon --mkfs -i ${HOSTNAME} --monmap /etc/ceph/monmap --keyring /tmp/ceph.mon.keyring

  # Clean up the temporary key
  rm /tmp/ceph.mon.keyring
fi

exec /usr/bin/ceph-mon -d -i ${HOSTNAME} --public-addr ${HOST}:6789