#zfs #cli #automatic #snapshot

bin+lib zfs-autosnap

Automatic ZFS snapshot utility

1 unstable release

0.3.1 Oct 9, 2021

#1705 in Command line utilities

MIT license

460 lines


Automatic ZFS snapshot utility.

zfs-autosnap periodically snapshots one or more ZFS datasets and removes old ones according to a retention policy.

Changes From Upstream

In my fork I have:

  • Reduced the number of dependencies
  • Tidied and reorganised the code
  • Added a small number of unit tests


  1. Add the retention policy to the dataset you want to snapshot. E.g. zfs set at.rollc.at:snapkeep=h24d30w8m6y1 tank
  2. Run zfs-autosnap snap hourly via cron.hourly or systemd timer
  3. Run zfs-autosnap gc daily via cron.daily

Try zfs-autosnap status to check what's going on.


See the systemd directory for sample timer files. To use these you need to update the dataset name, then enable and start the timers. E.g.

systemctl enable --now zfs-autosnap-snap.timer
systemctl enable --now zfs-autosnap-gc.timer


Pre-compiled Binaries

Pre-compiled binaries are available for x86_64 Linux and FreeBSD, they have no third-party runtime dependencies:

Example to download and extract a binary:

curl https://releases.wezm.net/zfs-autosnap/0.3.1/zfs-autosnap-0.3.1-x86_64-unknown-linux-musl.tar.gz | tar zxf -

Build from Source

Minimum Supported Rust Version: 1.53.0

zfs-autosnap is implemented in Rust. See the Rust website for [instructions on installing the toolchain][rustup].

From Git Checkout or Release Tarball

Build the binary with cargo build --release --locked. The binary will be in target/release/zfs-autosnap.

From crates.io

cargo install zfs-autosnap


I have a PKGBUILD for a pacman package in my personal package collection, which can be used as the basis of your own:


How It Works

Retention policy is set via the property at.rollc.at:snapkeep, which must be present on any datasets (filesystems or volumes) that you'd like to be managed. The proposed default of h24d30w8m6y1 means to keep 24 hourly, 30 daily, 8 weekly, 6 monthly and 1 yearly snapshots but you can select any value for each time unit, including omitting it.

The garbage collector looks at every snapshot under the managed datasets, and considers its creation time to decide whether to keep it.

Important: The snapshot name does not matter, only the creation time.

If you'd like to ensure a particular snapshot is not removed, set its at.rollc.at:snapkeep property to minus (-).

If in doubt, consider reading the source: it's only about 500ish lines of code.


zfs-autosnap will try not to eat your data; the only destructive operation is contained within a function that will refuse to work on things that are not snapshots - but there's NO WARRANTY. The previous version (written in Python), was in production use since ca 2015 and there were zero incidents; this (Rust) version is basically a source port (prior to wezm's changes).



~22K SLoC