Re: [reiserfs-list] Re: Apparent instability of reiserfs on 2.4.1

[Andi Kleen]

Rogerio Brito <[e-mail]> writes:

> On Feb 11 2001, Andi Kleen wrote:
> > The reiserfs nfs problem in standard 2.4 is very simple -- it'll
> > barf as soon as you run out of file handle/inode cache. Any workload
> > that accesses enough files in parallel can trigger it.
> 
> 	I'm just trying to evaluate if I should use reiserfs here or
> 	not: is this phenomenon that you describe above happening
> 	independently of whether I choose the knfsd or userspace nfsd?

This should be all covered extensively in the reiserfs FAQ and list archives, 
here a last time:

It only applies to knfsd, but unfsd unfortunately has different problems
with reiserfs. It makes assumptions about the inode space by the underlying
filesystem by assuming that it can encode a dev_t in upper bits. Reiserfs
unlike ext2 periodically cycles through the full 31bit of inode values, and
after some weeks on a busy file system unfsd starts to complain about 
conflicts. There is a patch at ftp.suse.com:/pub/people/ak/nfs/unfsd*
that works around the problem when you specify --no-cross-mounts (but 
you cannot export trees of multiple file systems then with a single mount
anymore) 

Please also note that the patch also adds a rather obscure bug, which triggers
very seldom (patch partly exists, but not really tested yet)

Another alternative is to use knfsd with Chris Mason's 2.4 knfsd patches.

-Andi