I am trying to find a way to display how many blocks a snapshot refers to “uniquely”. Then, I want to find to which files those blocks pertain.
Goal is to manually inspect if a file was (potentially inadvertently) altered, and if so, find which version I really want to keep.
Background for the question:
As a baseline, I’ve read Jim’s space accounting article. But apparently I am not understanding a few things.
I know there is zfs list -t snapshots <filesystem> and that displays USED and REFER columns. Over time, if I am solely adding new files to a dataset, the REFER column should have a monotonically growing number, and REFER should say 0B for the entire snapshot history of the dataset. However, without (knowingly) altering files, some of my datasets’ snapshot histories will have a non-zero value sprinkled in a few times on the USED column.
This might be access times, and I have at some point set atime=off on all my datasets. (But I didn’t document when that was, so I can’t test if that is the cause, to see if I am getting 0B always in the USED column since.)
When I zfs diff the non-zero USED snapshot with the snapshot immediately {prior | post}, the diff returns empty, and I get a new prompt.
So I am wondering if there is a way to print a delta REFER column, so tiny changes don’t get swallowed up in rounding of the reported REFER column, as related in the third to last paragraph of the “One Block, Many Snapshots” section of the linked article.