Snapshots basically put a pin in all the data at that moment and say “these blocks are not going to be physically deleted as long as I exist”, so the “additional” data use of the snapshots is equal to the data contained within the snapshot that doesn’t exist at the current moment. I.e., if I have two 50GB files, take a snapshot, and delete one, I will still have 100GB physical disk usage. I can also take 400 more snapshots and disk usage will remain at 100GB, as the snapshots are just virtual. Then I can either bring that deleted file back from the snapshot, or I can delete the snapshot itself and my disk usage will adjust to the “true” 50GB as the snapshot releases its hold on the blocks.

What sanoid and other snapshot managers do is they repeatedly take snapshots at specified intervals and delete old snapshots past a certain date, which in practice results in a “rolling” system where you can access snapshots from e.g. every hour in the past month. Then once a snapshot becomes older than a month, sanoid will auto-delete it and free up the disk space that it’s holding onto. The exact settings are all configurable to what you’re comfortable with in terms of trading additional physical disk usage of potential “dead” data for the convenience of being able to resurrect that data for a certain amount of time.

I really like the “data comet” visual from this Ars Technica article.