On Wed, Sep 08, 2004 at 12:26:52PM -0700, Chris Wedgwood wrote:
> On Wed, Sep 08, 2004 at 03:39:55PM +0200, Jakob Oestergaard wrote:
>
> > *) ext3 - no-go, because quota isn't journalled
>
> there are patches for that
Would *you* base a production environment on that ? ;)
> > *) reiserfs - no-go, because of lack of quota
>
> i was pretty sure this was merged, if not there are patches for that
> too
I've been corrected on that one; there *is* quota in reiserfs, but it is
not journalled (but there are rumers of patches for that as well).
We *need* the server to come up quickly after a crash or UPS failure.
Checking quota for the better part of a day is unacceptable.
>
> > Some 4-8 hours after the backup has started, the dreaded
> > 'debug.c:106' message will appear (at some random place thru the
> > filesystem - it is not a consistent error in one specific location
> > in the filesystem), and the server will need a reboot.
>
> have you run vmstat duing the backup to see what is going on here?
> checking /proc/meminfo periodically might also be useful. im guessing
> you're getting massive amounts of slab being used from the dcache and
> runing low on low-memory and coupled with page-cache pressure
> something is getting upset... (this was much more apparent in 2.4.x,
> i've not seen it for a long time in 2.6.x though)
Your guess is correct. More details are in Anders' previous posting to
LKML:
http://lkml.org/lkml/2004/8/24/53
>
> how much ram does the machine have and are you using highmem?
2.5G RAM IIRC - but slab is in lowmem so the machine will barf when it
reaches ~850MB slab.
>
> > Does anyone actually use XFS for serious file-serving? (yes, I run
> > it on my desktop at home and I don't have problems there - such
> > reports are not really relevant).
>
> plenty of people. some of them would claim a few hungred GB isn't
> very serious either.
There seems to be a common trend that NFS+XFS+SMP is causing this. I
suppose that could explain some of that.
Could be, that the problem is not in XFS at all...
Anyway; hopefully Anders will get some testing done in this respect
tomorrow, then we'll know more.
> > Is anyone actually maintaining/bugfixing XFS?
>
> yes
Hey, I had to ask ;)
...
> netapp hardware is nice (and their nfs works great) --- but it costs a
> lot more typically
Thanks for your feedback!
--
/ jakob
|