Two parts. The first part is the important part, the database. I run a scheduled database dump for this (the lemmy documentation actually explains how to do a database dump and how to restore it - I just have a bash script triggered by a cron job).

The second part is the images. Uploaded images and the image cache (for thumbnails for posts on other servers, etc) are virtually impossible to tell apart, so it’s all backed up (about 150GB just for images). For this, I just do a tarball of the whole directory. Any in-progress changes get skipped by the process but my theory is that since images are only added and deleted, if the process was run 30 seconds earlier then any being added would have been missed anyway, so they can wait until tomorrow’s backup. The vast, vast majority are cached images rather than image uploads anyway.

These two things are done overnight NZ time as they can be resource intensive, but I’ve done daylight hour backups before when needed, and it doesn’t cause much of an impact to the site performance anyway.