On 02.08.23 17:08, David Wysochanski wrote:
On Wed, Aug 2, 2023 at 3:35 AM Ole Salscheider <ole@xxxxxxxxxxxxxxx> wrote:
Hello,
I noticed some strange behaviour of fscache with kernel 6.4.x in
combination with NFS (v4.2) and the pagecache.
I have multiple clients mounting an NFS v4.2 file system from a server
and caching the data locally using fscache. After booting the clients,
most data is served from fscache and only a few requests hit the NFS
server. But after some minutes of operation (which typically includes a
lot of cyclic data reading for us), fewer and fewer files are served
from fscache and most are served by the NFS server.
It seems that this has something to do with the files being also in the
pagecache. If I periodcally drop the pagecache with "echo 3 >
/proc/sys/vm/drop_caches", then most of the files are served from
fscache again.
Is this a known problem? If not, do you have any suggestion how I can
try to find the root cause?
Yes this is most likely a known problem with a fix now in the
mm-unstable branch at
git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
If it's the same problem, take these 3 patches and that should fix it:
6fbefbe174ca mm: call folio_mapping() inside folio_needs_release()
04cb8ddab902 mm, netfs, fscache: stop read optimisation when folio
removed from pagecache
854fc8a567ab mm: merge folio_has_private()/filemap_release_folio() call pairs
The main patch that fixes it is "04cb8ddab902 mm, netfs, fscache: Stop
read optimisation when folio removed from pagecache" but there's a
prerequisite patch and a fixup patch.
I ran some tests yesterday and today, and it seems that these patches
fix the problem completely. Thanks!
Thank you in advance!
Best regards,
Ole
--
Linux-cachefs mailing list
Linux-cachefs@xxxxxxxxxx
https://listman.redhat.com/mailman/listinfo/linux-cachefs
--
Linux-cachefs mailing list
Linux-cachefs@xxxxxxxxxx
https://listman.redhat.com/mailman/listinfo/linux-cachefs
