As a copy-on-write filesystem, ZFS provides the capability to take snapshots. However, taking and managing them is completely manual, so maintaining a rolling list of snapshots requires either custom bash scripts + cron, installing external tools such as sanoid, or using an entire OS in the case of TrueNAS.
However, this functionality can be easily achieved by utilizing systemd units and timers.
unit file
The goal of this unit file is to keep a week’s worth of daily snapshots. Each snapshot is named in a way that is easily referenced:
1 | autosnapshot-7-days-ago |
Since ZFS snapshots can be renamed, the strategy for rolling them is to delete the oldest one, rename all the subsequent ones, and take a new one.
1 | zfs destroy %I@autosnapshot-7-days-ago |
Conveniently, systemd units have the ExecStartPre directive in the [Service] section, allowing for commands to be run before the actual target command., eg., the latest snapshot.
Additionally, it is possible to prepend the entire command with a -, which prevents the unit from failing if the command fails. This is great for a first-time run where none of the previous snapshots exist, causing the ZFS commands to fail.
Finally, this unit file has the ability to pause snapshots instantly. This is done by requiring a specific path not exist with a negated ConditionPathExists:
1 | ConditionPathExists=!/home/gadget/pause-snapshots |
Pausing is as simple as a touch ~/pause-snapshots, and once the file is deleted, snapshots resume whenever the timer triggers the unit file again. This could even be used to pause multiple (or all) autosnapshots on a system if they are all pointed to the same file.
zfs-7-daily-autosnapshots@.service
1 | [Unit] |
The remainder of the directives were taken from the zfs-scrub@.service bundled along with the ZFS installation on Fedora 41. No [Install] section is required since this unit file will be triggered by a timer.
timer file
The timer is even simpler. The OnCalendar directive is set to daily, and the variable window of time when it runs is reduced to 10 seconds from the default of 1 minutes using AccuracySec:
zfs-7-daily-autosnapshots@.timer
1 | [Unit] |
enabling the service
There’s a gotcha when trying to pass the full ZFS pool+dataset(s) path to the timer as an instance variable: a ZFS path contains invalid characters that can’t be passed on the command line - namely, forward slashes and dashes (/, -).
Luckily, systemd has a way to handle this: systemd-escape, and the variables %i and %I.
First, pass the full ZFS path to systemd-escape:
1 | $ systemd-escape "pool-1/dataset-2" |
This is the safely escaped ZFS path, and the timer can be enabled by using it:
1 | sudo systemctl enable --now 'zfs-7-daily-autosnapshots@pool\x2d1-dataset\x2d2.timer' |
Or in one command:
1 | sudo systemctl enable --now "zfs-7-daily-autosnapshots@$(systemd-escape 'pool-1/dataset-2').timer" |
The timer unit file will use %i to enable the respective unit service file with the escaped string. However, the service will use the unescaped string with %I to run the ZFS snapshot commands.
The systemd timers can be listed with:
1 | systemctl list-timers |
And the snapshots, their creation dates, and the amount of space they use can be listed with:
1 | zfs list -t snapshot -o name,creation,used |