XFS and LVM

Growing is a fairly simple operation for filesystems, shrinking complexity is a function of metadata complexity, and XFS metadata is very complex, the exception to this is the case of filesystems designed with shrinking in mind, but they will almost certainly have to make compromises to support this.

Shrinking is also a less used operation in production situations than growing, it might be fun to play with, but the type of system XFS comes from you would tend to just buy more disks than move one out of an existing filesystem. The only really useful 'shrink' operation in this scenario is remove a failing drive, and given the volume managers which usually sit under XFS the only thing capable of doing this is mirroring the drive (or using hardware raid).

Oh, and online shrinking would be the preferable implementation in a production situation, which means doing it with transactions.

So these are the reasons XFS does not have a shrink capability. As a technical option it is possible to implement this, but the code would be very complex and we have a lot of other things to do which have a higher priority. As I have said before, if someone wants to try it I would be happy to offer suggestions.

Steve

This what IBM marketing as well always told me (when I was an IBMer) but it turned out over time that I don't always buy a new disk when I need more space, but try to clean up some leftovers and resize my partitioning. Then it's nice if you can shrink. I mean you can always backup a filesystem, delete it, create it smaller, and restore the backup. But that is cumbersome and requires more scheduled downtime than desirable.

Does SGI have plans to eventually implement shrinking, or am I left with the every so often heard marketing statement that I sure don't need it?

(SCNR)

Harald Milz           |     [email protected]     | "Our vision is to speed
Linux Magazin         | phone  +49 (0) 89 993411-20 | up time, eventually
Stefan-George-Ring 24 | fax    +49 (0) 89 993411-99 | eliminating it." --
D-81929 Muenchen      | http://www.linux-magazin.de | Alex Schure

Well, when it comes to working for a company which is supposed to be making money, you cannot just go and implement features which will burn maybe $75,000 to get going correctly (that includes dealing with support calls etc from the customers who find bugs, fixing the bugs etc), unless there is a return on the investment.

So I think the answer is no, we will not be doing it in the forseeable future. But if some large customer was to wave several million in orders at us with this as a requirement it might be a different matter.

Steve