[Gluster-devel] Feature suggestion

[Brent A Nelson]

Sage Weil with the Ceph filesystem came up with a clever idea, and it 
might be a tempting optional feature on GlusterFS.  He calls it rbytes 
accounting, if I recall correctly.

Basically, the idea is to just keep a running tally of how much data is 
stored in each directory.  I believe he did this originally for quota 
support, but the neat part is that he decided to make the collected 
statistics available simply through ls.  If you do an ls -l, the size 
associated with any directory listed is the current estimated amount of 
data contained in that directory (and any subdirectories within), rather 
than the size of the directory inode itself (and who really cares about 
the size of the directory inode?).  Neat!

In filesystems where du can take a very long time to run, this would be 
rather convenient.  It's possible that it might break some applications, 
so it certainly should be optional, but such applications are likely quite 
rare.

Perhaps this would fit in nicely with a per-directory quota translator, as 
such a translator might store usage data by directory in an extended 
attribute, anyway?

Thanks,

Brent