@dvl I see what you're doing here, but it's not quite what I was referring to. This is simply doing a snapshot and reading the files from the snapdir so there's (relative) consistency on the filesystem during the backup run.

What I was referring to was having Bacula run "zfs diff" on snapshots.

Each Bacula run would recursively snapshot all of the filesystems before starting the backup job, and then it would run "zfs diff" to get a list of all files that have changed since the last Bacula job. First job run would have no snapshots to compare to, so it would do the normal full backup.

This eliminates the need for Bacula to do any filesystem walking, checksumming, etc. It just gets a perfect list of all files that are known to have changed or created since the last backup directly from ZFS and then uses that as the file list.