Re: dentry cache leak? Re: rsync out of memory 2.6.8-rc2

[Marcelo Tosatti]

On Fri, Jul 30, 2004 at 12:47:44PM -0700, Andrew Morton wrote:
> Marcelo Tosatti <marcelo.tosatti@xxxxxxxxxxxx> wrote:
> >
> > On Thu, Jul 29, 2004 at 02:07:43PM -0700, Andrew Morton wrote:
> > > kladit@xxxxxxxxxxx (Klaus Dittrich) wrote:
> > > >
> > > > >Can you narrow the onset of the problem down to any particular kernel
> > > > >snapshot?
> > > > 
> > > > Did it and here is the answer.
> > > > 
> > > > kernel-2.6.7 and bk's up to 2.6.7-bk7 survived a du -s,
> > > > kernels starting with 2.6.7-bk8 did not.
> > > 
> > > I can reproduce this oom btw.  Am (very, very slowly) working out what's
> > > causing it.  It's unrelated to the vfs-cache-pressure patch.  I'd hope to
> > > have it fixed up for 2.6.8. 
> > 
> > Odd, because the only thing I can see which affects dcache related code
> > between -bk7 and -bk8 is the vfs-cache-pressure patch.
> 
> It can be triggered with that patch reverted.
> 
> > What are the exact steps you're using to reproduce the leak?
> 
> Just a `du -s' over zillions of files on a 2G machine.
> 
> > And where do you think the problem lies?
> 
> Seems that we reach a state where lowmem pagecache get reclaimed faster
> than dcache/icache.  This causes the number of pages scanned for lowmem
> allocations to fall.  This causes less scanning of the slab and the whole
> thing repeats.  I expect changing nr_used_zone_pages() to ignore highmem
> will fix it, and might be the long-term fix, too.

I'll try making that change to nr_used_zone_pages() tomorrow morning and see 
what happens. 

But, why does Klaus claim he can't trigger the bug with -bk7 and only -bk8 ? 
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/