UPDATE: Files in fs cache gets corrupted.

I got some replies, but the problem remains unsolved.=20
The server is one out of 7 "identical" servers in a cluster, and only
this machine has errors (we think.). The build is the same on all the =
hosts
in the cluster. The nfs mounts are tcp, so it's not corrupted udp
packets. I did a od on one of the files that has been corrupted twice,
and in one instance only 1 bit was different. In the other two bits
diffrered.
The errors are rare.


> We have a strange problem here.
> It seems like sometime a file may be corrupted in the filesystem =
cache,
> ie.
> memory.
> So far it has only happened with files from nfs file systems.
>=20
> The host is a E4500 running solaris 2.6 Generic_105181-23
> If we do a sum of the file on the host, it will be different than on =
the
> nfs
> server. Size
> remains the same. Copying/removing/copying back so it's a "new" file =
will
> clear the
> error since the new inode will trigger a reread of the file from the =
nfs
> server. The same
> will happen if the file isn't read for quite a while.
> We don't use cachefs, so I guess this may happen on ordinary files as
> well.
>=20
> Any hints as to what may be the problem is welcome. We suspect some =
sort
> of
> hardware
> error, but no hardware errors are reported in any logs.
>=20
>   regards
>    -asbj=F8rn
>=20
>=20