02_tzvolsync

#!/bin/bash
# vim: sw=5
###########################################################################
#
#       Program name:   tzvolsync
#       File name:      tzvolsync.sh
#       Author:         Peter Blajev
#       First revision: 12/20/2010
#
#       Description:
#         Sends incr. ZFS data streams to a remote backup system.
#
#       Designed to be ran from cron:
#       0,30 * * * 1,3,5,0 /usr/local/sbin/tzvolsync run >> /var/log/tzvolsync.log 2>&1
#       15,45 * * * 2,4,6 /usr/local/sbin/tzvolsync run >> /var/log/tzvolsync.log 2>&
#
###########################################################################

# ---------------------------------------------------------------------------
# Things we can setup

# USER that should be used to connect to the remote system
USER=zfshelp

# How many snapshots to keep
TOKEEP=3

# Config file
configFile="/usr/local/etc/tzvolsync.conf"

# ---------------------------------------------------------------------------
# Do not edit below this line

# Tailored for Solaris
PATH="/usr/gnu/bin:/usr/bin:/bin:/usr/X11/bin:/usr/sbin:/sbin"

HOSTNAME=`hostname`
self="`basename $0`"

myid=`id -u`
myname=`id -un`

# ===========================================================================
# START the INITIAL Run of the recursive sequence
# ===========================================================================
if [ "$1" == "run" ]; then
     # This is the initial run

     # Is root the one who ran tzvolsync?
     if [ $myid -ne 0 ]; then
	  echo "
     ${self}:
	     Must be run as root (uid=0).
	     You are currently $myname (uid=$myid).
     "
	  exit 1
     fi

     # Make sure no other process is running
     RUNFILE=/var/run/$self.pid

     if [ -f $RUNFILE ]; then
	  echo "(`date +%H:%M:%S`) Old $self is running. PID in $RUNFILE: \
	       `cat $RUNFILE`"
	  exit 1
     fi

     # Save the process ID.
     echo $$ > $RUNFILE

     # Parse the config file
     if [ ! -r $configFile ]; then
	  echo "${self}: Config file $configFile not found."
	  rm -f $RUNFILE
	  exit 1
     fi

     cat $configFile | \
	  grep -v "^\#" | \
	  grep -v "^ " | \
	  grep -v "^$" |
     while read configLine; do
	  MASTER=`echo $configLine | awk -F "|" '{print $1}'`
	  VOLUME=`echo $configLine | awk -F "|" '{print $2}'`
	  SLAVE=`echo $configLine | awk -F "|" '{print $3}'`

	  if [ "$MASTER" != "$HOSTNAME" ]; then
	       # this entry is not for us
	       continue
	  fi

	  # run recursive
	  $0 $MASTER $VOLUME $SLAVE
     done

     rm -f $RUNFILE

     echo ""
     echo "--------------------------------------------------------------------"
     echo "==> [`date +\"%m/%d/%Y - %H:%M\"`] Done"

     exit 0
fi
# ===========================================================================
# END the INITIAL Run of the recursive sequence
# ===========================================================================

# ---------------------------------------------------------------------------
# Make sure the system is good to run the command
goodtogo ()
{
EXITSTAT=0

# Did we get a dataset
[ "$2" == "" ] && EXITSTAT=1

# Is the DATASETIN a ZFS pool?
POOLS=`zpool list       \
     | grep -v NAME     \
     | awk '{print $1}'`

for POOL in $POOLS; do
     [ "$2" == "$POOL" ] && EXITSTAT=2
done

# Does the FS exit?
zfs list $2 > /dev/null 2>&1
[ $? -ne 0 ] && EXITSTAT=3

# Did we get a slave server
[ "$3" == "" ] && EXITSTAT=4

# Is the slave server running any replication at this time?
ISRUNNING=`ssh -n $USER@$3 ps -ef | grep -E 'zfs send|zfs receive' | grep -v grep`
[ -n "$ISRUNNING" ] && EXITSTAT=5

echo $EXITSTAT

} # END goodtogo

# ---------------------------------------------------------------------------
# START main program
# ---------------------------------------------------------------------------

MASTER=$1
VOLUME=$2
SLAVE=$3


echo ""
echo "--------------------------------------------------------------------"
echo "==> [`date +\"%m/%d/%Y - %H:%M\"`] $MASTER:$VOLUME -> $SLAVE"

case `goodtogo $MASTER $VOLUME $SLAVE` in
1)
     echo ""
     echo "    ERROR: Dataset empty."
     exit 1
     ;;
2)
     echo ""
     echo "    ERROR: Looks like $VOLUME is a pool."
     exit 1
     ;;
3)
     echo ""
     echo "    ERROR: ZFS dataset $VOLUME is invalid."
     exit 1
     ;;
4)
     echo ""
     echo "    ERROR: Destination host not set."
     exit 1
     ;;
5)
     echo ""
     echo "    ERROR: Process \"zfs [send,receive]\" found on $SLAVE"
     exit 1
     ;;
esac

# At this point we are ready to start replication

# ---------------------------------------------------------------------------
# Find the last snapshots on both servers
LASTSNAP=`zfs list -r -t snapshot -o name $VOLUME | \
     grep $VOLUME@$self | tail -1 | tr '@' ' ' | awk '{print \$2}'`

if [ "$LASTSNAP" == "" ]; then
     echo ""
     echo "    No old snapshots for $VOLUME. Running for the first time."
     INCREMENTAL=no

     # There should be no dataset existing on the remote server
     ssh -n $USER@$SLAVE "sudo /usr/sbin/zfs list $VOLUME" > /dev/null 2>&1
     if [ $? -eq 0 ]; then
	  echo "    ERROR: Dataset $VOLUME exists on the remote server."
	  echo "    Destroy it first and then rerun $self."
	  exit 1
     fi
else
     # Make sure the snapshot exists on the slave
     LASTRSNAP=`ssh -n $USER@$SLAVE "sudo /usr/sbin/zfs list -r -t snapshot -o name $VOLUME 2>&1 | grep $VOLUME@$self" | \
	  tail -1 | tr '@' ' ' | awk '{print $2}'`
     if [ "$LASTSNAP" == "$LASTRSNAP" ]; then
	  echo "    Last snapshot $VOLUME@$LASTSNAP"
	  echo "    found on both hosts."
	  INCREMENTAL=yes
     else
	  echo "    Snapshot $VOLUME@$LASTSNAP not found or"
	  echo "    not the most recent on the remote server."
	  exit 1
     fi
fi

# ---------------------------------------------------------------------------
# New snapshot

NEWSNAP=`date +$self%Y%m%d%H%M`
if [ "$NEWSNAP" == "$LASTSNAP" ]; then
     echo "   Snapshot $NEWSNAP exists. Wait at least 1 min. between runs."
     exit 1
fi

CMD="zfs snapshot $VOLUME@$NEWSNAP"
echo ""
#echo "    (`date +%H:%M:%S`) New snapshot"
echo "    (`date +%H:%M:%S`) New snapshot: $VOLUME@$NEWSNAP"
echo "      Run: $CMD"

$CMD

if [ $? -ne 0 ]; then
     echo "    ERROR: Creating sanpshot failed."
     exit 1
fi

# ---------------------------------------------------------------------------
# Send data stream to the slave

CMDOPT=""
ROPT=""

[ "$INCREMENTAL" == "yes" ] && CMDOPT="-i $VOLUME@$LASTSNAP"

CMD="zfs send $CMDOPT $VOLUME@$NEWSNAP | ssh $USER@$SLAVE 'sudo /usr/sbin/zfs receive $ROPT $VOLUME'"

echo ""
echo "    (`date +%H:%M:%S`) Sending data to the slave"
echo "      Run: $CMD"

echo $CMD | bash

if [ $? -ne 0 ]; then
     echo "    Something went wrong. Removing the snapshot."
     zfs destroy $VOLUME@$NEWSNAP
else
     # Set ReadOnly flag on the remote dataset
     [ "$INCREMENTAL" == "no" ] \
	  && ssh -n $USER@$SLAVE "sudo /usr/sbin/zfs set readonly=on $VOLUME"
fi

# ---------------------------------------------------------------------------
# Remove old snapshots

NUMSNAPS=`zfs list -r -t snapshot -H -o name $VOLUME | \
	  grep $VOLUME@$self | wc -l`

NUMTODEL=`expr $NUMSNAPS - $TOKEEP`

echo ""
echo "    (`date +%H:%M:%S`) Removing old snapshots"

if [ $NUMTODEL -lt 1 ]; then
     echo "    No snapshots to delete."
else
     SNAPSTODEL=`zfs list -r -t snapshot -H -o name $VOLUME | \
	  grep $VOLUME@$self | head -${NUMTODEL}`

     echo "    Removing: $SNAPSTODEL"

     for SNAP in $SNAPSTODEL; do
	  zfs destroy $SNAP
	  ssh -n $USER@$SLAVE "sudo /usr/sbin/zfs destroy $SNAP"
     done
fi

# ---------------------------------------------------------------------------

exit 0