The trouble with symbolic links

Posted Jul 19, 2022 13:35 UTC (Tue) by mathstuf (subscriber, #69389)
In reply to: The trouble with symbolic links by ras
Parent article: The trouble with symbolic links

Have there been systems with transactional filesystem APIs?

to post comments

The trouble with symbolic links

Posted Jul 19, 2022 14:59 UTC (Tue) by ras (subscriber, #33059) [Link] (1 responses)

A quick google reveals most desktops are running one: https://en.wikipedia.org/wiki/Transactional_NTFS

It's not clear if that still true, but it blows me away anyway.

The trouble with symbolic links

Posted Jul 19, 2022 16:40 UTC (Tue) by mathstuf (subscriber, #69389) [Link]

Given that Microsoft has declared it deprecated and may have it disappear in the future…it seems like a failed experiment on at least the deployment level.

The trouble with symbolic links

Posted Jul 19, 2022 16:24 UTC (Tue) by kleptog (subscriber, #1183) [Link]

PostgreSQL has support for transactional large objects. To call it a filesystem is a bit of a stretch since objects only have numeric identifiers not names. However, they have permissions, and support open/close/seek/read/write and are all within a transaction. You could certainly build a complete transactional filesystem on top of it.

But this also shows the difficulty, given the amount of work database systems have to do to make transactions work while giving everyone a consistent view and making it perform. Then again, the problem space of filesystems is much simpler.

Even if we supported only transactional metadata changes (rename, move, create, unlink) would be a step up. Certainly being able the work on a fixed snapshot of the filesystem would be good enough for quite a lot of purposes (I think btrfs has this).