ZFS snapshots are extremely lightweight on the system (actually cheaper to take snapshot than not!) and cheap on storage (only changed blocks are saved) and allow you to restore files from past versions.

The gist

You run these tools from cron to periodically

take ZFS snapshots (zfssnap.sh)
serialize and export them remotely (zfsbk.sh)

The tools take care of limiting the number of snapshots, so you do not have to periodically prune old snapshots.

Snapshots can be named so you can create groups of them, e.g. "hourly snapshots", "newdeplo snapshots" and "exported snapshots".

An additional tool takes care of exporting such snapshots, and uploading them to a remote server.

Installation

Simply place all *.sh files of the package into directory /usr/local/sbin/.

Snapshot management

You run zfssnap.sh for taking snapshots, typically from cron:

vim /etc/crontab ...
# take hourly snapshot, keep 24 of them
@hourly     root    /usr/local/sbin/zfssnap.sh hour 24

This will create, every hour, a new ZFS snapshot tagged 'hour':

# ls /.zfs/snapshot/
zbk-hour-20140318-140000
zbk-hour-20140318-150000
zbk-hour-20140318-160000

Only 24 of these snapshots will be kept (see Rotation below). The name of each snapshot comes with format zbk-[tag]-[date]-[time]. Date is YYYYMMDD and time is hhmmss.

Snapshot groups

Each snapshot is tagged:

# create a snapshot tagged 'foobar'. Maintain 10 at all times
zfssnap.sh foobar 10

Snapshots with the same tag make a snapshot group. For example, the foobar group will count up to 10 members at all times.

Multiple groups can exist, just take snapshots with different tags:

# take 2 'hourly' snaps
zfssnap.sh hourly 10
zfssnap.sh hourly 10
# take 3 'daily' snaps
zfssnap.sh daily 10
zfssnap.sh daily 10
zfssnap.sh daily 10
# ls /.zfs/snapshot
zfs-hourly-20140318-140001
zfs-hourly-20140318-140003
zfs-daily-20140318-140111
zfs-daily-20140318-140112
zfs-daily-20140318-140114

Neither zfssnap.sh nor ZFS put a limit on the number of snaps you can maintain. The tool was tested with over 200. Bear in mind that it's shell scripts, so inherent limitations of arguments length could get in your way.

I recommend staying under 50 snaps per group and 200 snaps total.

Snapshot rotation

zfssnap.sh takes a new snapshot every time it's run. When the number of existing snapshots exceeds the given limit, the oldest snapshot of that group (tag) is removed, so only so many are kept:

# take snap xyz, then keep only last 2 from xyz group
zfssnap.sh xyz 2

This bounds the number of snapshots for each group to 2. To remove all snaps in a group, simply pass ''i'' as limit:

# remove all snaps of group xyz
zfssnap.sh xyz 0

Excluding dataset from backup

zfssnap.sh takes a recursive backup of the zroot pool. If you do not intend to maintain backups for certain datasets, you can exclude them with the EXCLUDES and EXTRA_EXCLUDES environment variables:

# exclude only these datasets
EXCLUDES=“/mydataset/foobar"
# exclude these datasets in addition to default exclusions
EXTRA_EXCLUDES=“/mydataset/foobar"

Notice that these are dataset names, not mountpoints! If dataset zroot/foo is at mountpoint /bar, specify /foo here.

The following datasets, common for FreeBSD users, are excluded by default:

/usr/ports
/usr/src
/backups

If you do not want these excluded, pass an empty EXCLUDES envvar.

Recovering files (local backup)

Lost a file? Find it under:

# list content of michele's home at 2pm (1400)
ls /.zfs/snapshot/zbk-hour-140000/home/michele

Notice that you must look for the /.zfs directory at the root of the dataset actually holding it:

# list content of michele's home, if /home is on zroot/home
ls /home/.zfs/snapshot/zbk-hour-140000/michele

Full snapshot management cron example

# take 15' backups for the last hour
*/15 * * * * root /usr/local/sbin/zfssnap.sh qrt 4
# take hourly backups for the 6 hours
1 * * * * root /usr/local/sbin/zfssnap.sh hourly 6
# take 6-hours backups for the last day
1 */6 * * * root /usr/local/sbin/zfssnap.sh 6hr 4
# take daily backups for the last week
1 1 * * * root /usr/local/sbin/zfssnap.sh day 7
# take weekly backups for the last 2 months
1 1 * * 1 root /usr/local/sbin/zfssnap.sh week 8

Generating remote backups

The zfsbk.sh lets you generate backups and upload them to a remote location.

This takes a snapshot with tag mybk and serializes it in file /backups/zbk-mybk-140000.dump:

# generate ZFS streaming package, save to /backups folder
/usr/local/sbin/zfsbk.sh mybk
ls /backups
zfs-mybk-20140318-061900.dump

Incremental backups

Pass a number to zfsbk.sh and it will create incremental snapshots:

# 1. make full replication if this is the first snap in group
# 2. else make incremental replication wrt latest snap in group
# 3. reset the snap group after 1+9 steps have been made
/usr/local/sbin/zfsbk.sh mybk 10

Incremental packages are named after their snapshot endpoints:

ls /backups
zbk-mybk-20140318-140000--zbk-mybk-20140318-150000.dump

If the given integer is 1, zfsbk.sh sends full replication packages for every run.

Uploading backups remotely

zfsbk.sh can upload each replication package after generating it, at the end of the run.

Pass the destination coordinates with the UPLOAD_PATH environment variable. Currently, rsync:// and scp:// are supported:

# take snap, generate backup, upload it to remote server
UPLOAD_PATH="rsync://user@backup.server.com::server12/" /usr/local/sbin/zfsbk.sh mybk 10

zfsbk.sh relies on zfssnap.sh to take the snapshot to backup. Therefore, you can exclude different datasets from its backup by passing the respective EXCLUDES or EXTRA_EXCLUDES variables:

# take selective backup
EXTRA_EXCLUDES="/jails/test.dom.com" /usr/local/sbin/zfsbk.sh mybk 1

README.md Unescape Escape

zfsbk Overview