> I guess their idea was to pipe the copy into a compression tool right away. Yes that is exactly the use case. I mean the backup is nothing fancy really, just `tar czf mybackup.tar.gz path/to/sqlite/` > I am wondering though, if using a specific filesystem is a possibility, could that be achieved with something like ZFS (a COW filesystem)? Where the copying process is fast and the resulting file is using minimal desk space in reality? As for the active/main database, using a compressed file system might hurt performance too much. But it probably would be an option for the backup location, if that is a dedicated partition (it is not in my case, at least right now). Another idea I had is to use a named pipe (`mkfifo`): The Backup API could then use the pipe as target (or the `VACUUM INTO path/to/pipe` command). On the other side of the pipe would be the backup tool which reads the stream, compresses it and writes it into the compressed backup file. Not sure if that really works, but it should in theory as far as I can tell. In the end the first approach just seems very straight forward and was close to what has been in use so far.