Re: nfs_inode_cache not getting pruned

[Andrew Morton <akpm@osdl.org>]

Brad Fitzpatrick <brad@danga.com> wrote:
>
> On Tue, 6 Jul 2004, Andrew Morton wrote:
> 
>  > Brad Fitzpatrick <brad@danga.com> wrote:
>  > >
>  > > I've got a couple boxes here that keep running out of lowmem after about a
>  > >  day of uptime, rendering them almost entirely unusable.  (OOM killer
>  > >  taking out processes left and right, despite 1.5 GB highmem available...)
>  > >
>  > >  From what I can tell, nfs_inode_cache isn't responding to memory pressure
>  > >  and cleaning itself (enough?).
>  >
>  > It works OK here.  Please provide kernel version and full NFS mount options.
> 
>  2.6.7, both client and server.
> 
>  Both machines are exporting 14 filesystems (each ext3 on 3w-9xxx) and
>  mounting the other's 14 over NFSv3:

Sorry, I still cannot reproduce it.  I suspect it's specific to the access
patterns in some way.

Are you able to generate some simple testcase which demonstrates the
problem?  Say, start off from a clean boot, just mount a single filesystem
and then find some access pattern which causes the inode cache to get
stuck?

Also, perhaps the entire VFS cache reclaim mechanism got stuck - check the
logs for any oopses, because that can do it for sure.

And do a huge ls -lR on a local filesystem, check that the local filesystem
VFS caches are being correctly reclaimed.