#!/bin/bash

# This script is provided as a reference for the readers of Learning Ceph (second edition)
# It has been used successfully thousands of times but is by no means perfect, general, or future-proof.
# In fact it's a bit dated.  But it does show the overall process for deleting an OSD and cleaning up.
# The typecode on the FileStore partition is adjusted to prevent udev from trying to remount it
# Because XFS can horque the OS if it tries to mount a corrupt filesystem.
# But we might want to poke around later, especially if the OSD was removed by accident.
# The cluster name is assumed to be 'ceph' and FileStore is assumed as well.
# The service management commands may need adjusted for your distribution.
# Quando Omni Flunkus Moritati
# Your mileage may vary.
# Use at your own risk.  Much fault tolerance could be added.

# Anthony would like to thank Michael Kidd and Mike DeSimone for their contributions.



if [ $# -lt 1 ]; then
  echo "Usage: `basename $0` <OSD number>"
  exit 1
fi
typeset -i osdtorm=$1

typeset    MONSTART=""
typeset    OSDCMD=""
typeset    FACTER=$(which facter 2>/dev/null)
typeset    OSFamily=""

# The facter package may not be installed on the system.  Rather than
# failing, when the packge was insalled, we should use the vendor file in
# /etc/ directory to determine the OS family.

if [ "X$FACTER" != "X" ]; then
  OSFamily=$($FACTER osfamily)
else
  if [ -f /etc/lsb-release ]; then
    OSFamily=Debian
  elif [ -f /etc/redhat-release ]; then
    OSFamily=RedHat
  fi
fi

case $OSFamily in
  RedHat)
    MONSTART="sysvinit"
    OSDCMD="/etc/init.d/ceph stop osd.${osdtorm}"
    ;;
  Debian)
    MONSTART="upstart"
    OSDCMD="stop ceph-osd id=${osdtorm}"
    ;;
  *)
    echo "Neither Ubuntu nor Redhat is detected, exiting."
    exit 1
esac

echo -n "Finding host.. "

remhost=$(ceph osd find $osdtorm 2>/dev/null | grep host | awk '{print $NF}' | sed -e s/[\",]//g)
if [ "x${remhost}" == "x" ]; then
  echo Failed: osd.$osdtorm does not exist.
  exit 1
else
  echo $remhost

  # TODO: These commands  should perhaps be done in a subshell or other method to act appropriate in case they hang.

  ssh -o ConnectTimeout=20 $remhost -- "echo -n 'Stopping osd.$osdtorm: '; ${OSDCMD}; sleep 7 ; echo -n 'Removing $MONSTART flag: '; rm -f /var/lib/ceph/osd/ceph-$osdtorm/$MONSTART; echo done; remdrive=\$(mount | grep \"ceph-$osdtorm \" | awk '{print \$1}'); echo -n \"Umounting $remdrive: \"; umount \$remdrive; echo done; echo -n \"Removing mount point : \"; rm -rf /var/lib/ceph/osd/ceph-$osdtorm; echo done; echo -n \"Set partition typecode : \"; sgdisk --typecode=1:89c57f98-2fe5-4dc0-89c1-f3ad0ceff2be --partition-guid=$(uuidgen -r) $remdrive; echo done"

  echo -n "Removing CRUSH entry for osd.$osdtorm: "
  ceph osd crush remove osd.$osdtorm
  # If it isn't down on its own yet, hedge our bet
  ceph osd down osd.$osdtorm
  echo -n "Removing OSD entry for osd.$osdtorm: "
  ceph osd rm osd.$osdtorm
  echo -n "Removing Auth for osd.$osdtorm: "
  ceph auth del osd.$osdtorm

  echo "Please run 'ceph -w' or 'watch ceph -s' to monitor for rebalancing to complete"
fi

exit 0

# Reference:  the typecodes that Ceph uses.

#JOURNAL_UUID =         '45b0969e-9b03-4f30-b4c6-b4b80ceff106'
#DMCRYPT_JOURNAL_UUID = '45b0969e-9b03-4f30-b4c6-5ec00ceff106'
#OSD_UUID =             '4fbd7e29-9d25-41b8-afd0-062c0ceff05d'
#DMCRYPT_OSD_UUID =     '4fbd7e29-9d25-41b8-afd0-5ec00ceff05d'
#TOBE_UUID =            '89c57f98-2fe5-4dc0-89c1-f3ad0ceff2be'
#DMCRYPT_TOBE_UUID =    '89c57f98-2fe5-4dc0-89c1-5ec00ceff2be'