[Gluster-devel] Readdir d_off encoding

[Jeff Darcy]

> Apologies, obviously I'm catching up here, but what exactly are the
> current user complaints about?
> 
> (My understanding was that a) there are no known problems in the xfs
> case, b) in the ext4 case with 64-bit cookies we expect problems but
> only in extreme setups not yet common enough to produce numerous
> complaints.)

There's strongly suggestive evidence of two users on the mailing list
hitting this.

http://www.gluster.org/pipermail/gluster-users/2014-December/019934.html
http://www.gluster.org/pipermail/gluster-users/2014-December/019940.html

I've also heard of at least one Red Hat customer escalation related
to this, but of course I couldn't disclose details even if I knew them.
Are these users' setups extreme?  I don't know, but I was seeing 16M
files in a directory before I even came to Red Hat (five years ago)
so I'm inclined to believe it's common enough to worry about today.

More importantly, the frequency of users hitting this problem is only
likely to increase.  Both clusters and directories are likely to
continue getting larger, requiring more bits on both sides of whatever
dividing line we choose.  Whether it's today or two years from today,
sooner or later the numbers are going to catch up with us.