#1 nfsd: fd cache inconsistency!

[Anonymous]

Hi folks,

I'm trying to use ClusterNFS to setup a MOSIX cluster
of Linux boxes with some of them diskless. At present I
have one node with 2 NICs and real filesystem and two
diskless nodes that boot from the first one (DHCP and
ClusterNFS). Linux is Red Hat 7.0, kernel 2.4.4,
MOSIX-1.0.1 up and running.
I see the message
nfsd[pid]: fd cache inconsistency!
repeated many times on /var/log/messages of the machine
running your rpc.nfsd
Universal NFS Server 2.2.beta47
TRANSLATE-NAMES support version 0.91alpha
Now moy question is:

WHAT DOES THIS MESSAGE MEAN?
What did I wrong?

Thanks for your work, I'm sure that ClusterNFS IS the
solution.

Roberto Vidmar

[Gregory Warnes]

Logged In: YES
user_id=9316

Hi Roberto,

Actually, I've never seen this message, but I haven't tried
running it under 2.4.x kernels yet, either.

Searching the web only turns up one person who has reported
a similar error with (plain) nfs-server, but there was no
information about the error except the report.

I'll look through the code to see what this error message
means. In the mean time, can you try running the standard
user-mode NFS server to see if you get the same message?

BTW, what distribution are you using?

-Greg

[Gregory Warnes]

Logged In: YES
user_id=9316

Roberto,

Try running the standard user-mode nfs server that goes with
your distribution. Let me know if that works.

Looking quickly at the code, I see that the error message
you are seeing is generated by code in fh.c. Unfortunately,
this message can be generated by failure of one of several
tests. I'm attaching a modified fh.c that makes these
messages distinct. This should help me figure out what is
happening.

-Greg

[Gregory Warnes]

modified fh.c file that makes "fd cache inconsistency!" messages distinct.

[Jim Stewart]

Logged In: YES
user_id=524498

This may not be ClusterNFS's fault. I have been
experimenting with MOSIX and ClusterNFS machines here, and
I've had problems where all the file descriptors on a
machine get used up. It doesn't appear to be related to
ClusterNFS, though; even on client machines running MOSIX,
all the file descriptors disappear sometimes, affecting
pretty much everything on the client. You might just be
noticing this in rpc.nfsd because it tends to use a buttload
of file descriptors.

Jim