Each jail is in its own fileset. Snapshot each fileset, back it up. Destroy the filesets.

1 - The FileSet

# the filesets will be different on each jail, thus, we'll always be doing a full unless 
# we define this for each jail server
#
FileSet {
  Name = "zuul jail snapshots"
  Include {
    Options {
      signature = MD5 
      Exclude = yes
    }

    Exclude Dir Containing = .NOBACKUP

    File = "\\|/usr/home/dan/bin/jail-snapshots-for-backup.sh list"
  }
}

2 - The backup job

Job {
  Name            = "zuul jail snapshots"
  JobDefs         = "DefaultJob"
  Client          = zuul-fd
  FileSet         = "zuul jail snapshots"
  Write Bootstrap = "/usr/local/bacula/bsr/zuul-fd-jail-snapshots.bsr"

  RunScript {
    RunsWhen       = Before
    FailJobOnError = Yes
    Command        = "/usr/home/dan/bin/jail-snapshots-for-backup.sh create"
  }

  RunScript {
    RunsWhen       = After
    FailJobOnError = No
    Command        = "/usr/home/dan/bin/jail-snapshots-for-backup.sh destroy"
  }
}

3 - jail-snapshots-for-backup.sh

#!/bin/sh

ACTION=$1

JLS=/usr/sbin/jls

JAILS=`${JLS} -h path`

SNAPSHOTDIRECTORY='/.zfs/snapshot/'

SNAPNAME="snapshot-for-backup"

for jail in ${JAILS}
do
  # ignore the one value of path, usually the first, because that's the header line.
  if [ ${jail} = 'path' ]
  then
     continue
  fi
  
  DATASETNAME=`/sbin/zfs get -H -o name mountpoint ${jail}`
  
  # the snapshot name is of the form: system/usr/local/jails/fedex@snapshot-for-backup
  SNAPSHOTFORBACKUP="${DATASETNAME}@${SNAPNAME}"
  
  # the backup dir is of the form: /usr/local/jails/fedex/.zfs/snapshot/snapshot-for-backup
  BACKUPDIR="${jail}${SNAPSHOTDIRECTORY}${SNAPNAME}"

  case ${ACTION} in
    "create")
       zfs snapshot ${SNAPSHOTFORBACKUP}
       ;;

    "list")
       # echo back out the directory for backup...
       echo ${BACKUPDIR}
       ;;

    "destroy")
       zfs destroy ${SNAPSHOTFORBACKUP}
       ;;
  esac    
done

4 - jail-snapshots-for-backup.sh

#!/bin/sh

# this script caters for multiple jail locations

ACTION=$1

SNAPSHOTDIRECTORY='/.zfs/snapshot/'

SNAPNAME="snapshot-for-backup"

ZFSJAILROOTS="system/iocage/jails system/jails"

for ZFSJAILROOT in $ZFSJAILROOTS
do

  DATASETS="/sbin/zfs get -r -H -o name -t filesystem name ${ZFSJAILROOT}"

  for dataset in `${DATASETS}`
  do
    # Ignore the jail root.  Nothing in there to backup.
    if [ "${dataset}" = "${ZFSJAILROOT}" ]
    then
       continue
    fi

    MOUNTPOINT=`/sbin/zfs get -H -o value mountpoint ${dataset}`
  
    # the snapshot name is of the form: main_tank/iocage/jails/fedex@snapshot-for-backup
    SNAPSHOTFORBACKUP="${dataset}@${SNAPNAME}"

    # the backup dir is of the form: /usr/local/jails/fedex/.zfs/snapshot/snapshot-for-backup
    BACKUPDIR="${MOUNTPOINT}${SNAPSHOTDIRECTORY}${SNAPNAME}"

    case ${ACTION} in
      "create")
         zfs snapshot ${SNAPSHOTFORBACKUP}
         ;;

      "list")
         # echo back out the directory for backup...
         echo ${BACKUPDIR}
         ;;

      "destroy")
         zfs destroy ${SNAPSHOTFORBACKUP}
         ;;
    esac    
  done

done

I know we talked about this via IRC, and I think we concluded recursive had a down-side, but I cannot recall what that was. I'll check logs when I'm on my other laptop.

I know now why 'zfs snapshot -r' doesn't help much. Well, it help for creating the snapshots, but you still (for now) have to remove them one at a time. You also have to iterate through the datasets (zfs list -r) to find the mountpoints.

I realized this as I was rewriting the script to work with iocage (https://github.com/pannon/iocage).

It's been a while, zfs destroy -r has worked for a while now :-) I'd like to adapt this for backing up my pools as well, bacula gets upset if things vanish from under it in a multi-day backup

Today I amended the script to cater for multiple jail locations.

Both the original and the new script are listed above.