zfs_repl.sh

#!/bin/bash

if ! [ $3 ] ; then
  echo "$0 <SOURCE POOL/DATASET> <TARGET_HOST> <TARGET POOL/DATASET>"
  exit 1
fi

SRC=$1
TARGET_HOST=$2
DST=$3
SSH="ssh -C -p 2222 -o ConnectionAttempts=2 -o ConnectTimeout=10 -o BatchMode=yes"
KILLFILE=/root/.zfs_repl_kill
EMAIL="sysadmins"

if [ -f $KILLFILE ] ; then
  echo "$KILLFILE found. zfs_repl will run once."
fi

function die {
  echo $1 | /bin/mailx -s "FAIL: zfs_repl on `hostname` has failed" $EMAIL
  exit 1
}

function quit {
  echo $1 | /bin/mailx -s "FAIL: zfs_repl on `hostname` has stopped" $EMAIL
  exit 0
}

while true ; do
  STAMP=$( date '+%Y%m%d%H%M%S' )
  DATE="[$( date +%Y%m%d-%H:%M:%S )]"

  echo "$DATE [`hostname`] Creating $SRC@repl-$TARGET_HOST-$STAMP"
  zfs snapshot -r $SRC@repl-$TARGET_HOST-$STAMP || die "Could not create snapshot: $SRC@repl-$TARGET_HOST-$STAMP"

  for FS in $( zfs list -t filesystem -rH -o name $SRC ) ; do
    NEWSNAP="$FS@repl-$TARGET_HOST-$STAMP"

    LASTSNAP=$( zfs list -rH -o name -t snapshot $SRC | grep ^$FS@repl-$TARGET_HOST | grep -v $STAMP | tail -1 )
    if [ $LASTSNAP ] ; then
      echo "$DATE [`hostname`] Sending $LASTSNAP -> $NEWSNAP"
      SEND=$( zfs send -i $LASTSNAP $NEWSNAP | $SSH $TARGET_HOST zfs recv -vdF $DST | grep -v receiving )
      echo "$DATE [$TARGET_HOST] $SEND"
    else
      echo "$DATE [`hostname`] Sending INITIAL snapshot $NEWSNAP"
      SEND=$( zfs send $NEWSNAP | $SSH $TARGET_HOST zfs recv -vdF $DST | grep -v receiving )
      echo "$DATE [$TARGET_HOST] $SEND"
    fi
  done

  $SSH $TARGET_HOST zfs list -rH -o name $NEWSNAP > /dev/null
  if [ $? != 0 ] ; then
    ERROR="$DATE [ERROR] snapshot does not exist on $TARGET_HOST:$NEWSNAP; dying"
    echo "$ERROR"
    die $ERROR
  fi

  L_SNAPS=$( zfs list -t snapshot -o name | grep repl | grep $SRC | grep $TARGET_HOST | wc -l )
  R_SNAPS=$( $SSH $TARGET_HOST zfs list -t snapshot -o name | grep repl | grep $SRC | grep $TARGET_HOST | wc -l )

  if [ "$L_SNAPS" != "$R_SNAPS" ] ; then
    ERROR="$DATE [ERROR] Missing snapshots. Local:$L_SNAPS Remote:$R_SNAPS"
    echo "$ERROR"
    die $ERROR
  fi

  for OLD in $( zfs list -t snapshot -rH -o name $SRC | grep "@repl-$TARGET_HOST-" ) ; do
    OLD_DATE=$( echo $OLD | sed -e "s/^.*@repl-$TARGET_HOST-//" )
    if [ $OLD_DATE != $STAMP ] ; then
      echo "$DATE [`hostname`] Destroying $OLD (current: $STAMP)"
      zfs destroy $OLD
      echo "$DATE [$TARGET_HOST] Destroying $OLD (current: $STAMP)"
      $SSH $TARGET_HOST zfs destroy $OLD
    fi
  done

  if [ -f $KILLFILE ] ; then
    ERROR="$DATE $KILLFILE found, exiting."
    echo $ERROR
    rm $KILLFILE
    quit "$ERROR"
  fi

  x=0
  echo -n "$DATE sleeping"
  while [ $x -lt 10 ]; do
    echo -n "."
    x=`expr $x + 1`
    sleep 1
  done

  echo

done