A poor man's Time Machine backup

Back in the day¹ when I was still using a Mac, I always enjoyed the simplicity of Time Machine backups that Apple introduced almost 20 years ago. Simply plug in your dedicated drive, let Time Machine do its thing and hey presto, snapshots of your precious data. My wife’s MacBook Pro from 2013 is still backing up to a Drobo 5N of a similar vintage over SMB to this day.

I tend to synchronise my important folders such as documents, photos and videos via Syncthing and that works without issue. However, I also have a large ~/Development folder that I don’t synchronise. To a degree this is due to the fact that I push most source code I care about to GitHub or Codeberg. I also had bad experiences with Dropbox in the past where its synchronisation mechanism would get confused with operations in the .git folder across multiple hosts. And because of that experience, I never used to back up my Development folder, which was a shame, really.

So the other day, while I was looking at a disused 256 GB drive lying around on my desk, a thought popped up in my head. What if I asked Clyde² to whip up something real quick in that regard, with the following constraints³:

It should run when I plug in an external drive
It should run as a non-privileged process
It should eject the drive when it is done
It should have an easy to understand config file
It should use rsync
It should use hard links like the original Time Machine

Lo and behold, a little prompting later, I got my first version. It worked flawlessly on the first try and backed up my 87 GB worth of stuff in under 10 minutes. At first, the process was silent, so I didn’t know what was going on, but I then remembered that my flavour of Linux comes with notify-send which made it simple to add toast notifications to my desktop, telling me what was going on.

The beauty of the solution is that you can simply browse the snapshots on disk as a file tree. No special format, no index database. Just files! Thanks to hard links, each snapshot looks like a full backup, but only files that changed since the last one take up extra space.

How it works

systemd tracks every block device as a device unit
when a drive is plugged in, udev creates /dev/disk/by-uuid/<UUID> which systemd turns into a <device> unit
the install script adds a ~/.config/systemd/user/<device>.wants/rsync-snapshot.service symlink that triggers the service when the drive is plugged in
the backup script handles the mounting and unmounting of the drive

Here’s a little demo of it working:

Check out its code: codeberg.org/tsak/rsync-snapshot

It comes with an extensive README.

Or you can just follow the install instructions (after you’ve cloned its repository):

Install:
```
./install.sh
```
On the first run this creates ~/.config/rsync-snapshot/config with an empty UUID and stops before setting up the plug-in trigger, listing the attached drives so you can pick yours (same as lsblk -f).

Edit ~/.config/rsync-snapshot/config: set UUID and list the folders to back up in SOURCES:

UUID="2155b0cb-13a2-452b-a336-f3a309f79b63"
SOURCES=(
    ~/Documents
    ~/Pictures
    ~/Development
)
EXCLUDES=(
    .cache/
    node_modules/
)

Run ./install.sh again. With UUID set it enables the trigger. Do the same whenever you change UUID later, so the trigger follows the new drive.

Footnotes

¹ I started using Linux full-time in 2018. These days I use ~~just a bunch of scripts on top of Arch~~ Omarchy btw

² Naming it Claude instead was a missed opportunity on Anthropic’s part in my opinion

³ The real discussion went a little differently. I asked Clyde about how to get a similar experience to Time Machine on Linux and it suggested restic or borg - the complexities of each were not what I was longing for in my fond memories of Time Machine

A poor man's Time Machine backup.

How it works

Footnotes