Uploading to NFS server crashes/hangs system

dennisn

When I try to upload large files to my NFS server, after a few minutes of transferring, my system hangs -- the keyboard no longer responds. Sometimes, before the complete hang, the mouse is operational, but my window manager doesn't respond.

From my server's /etc/exports file I have the options (rw,no_subtree_check,sync).

From my /etc/fstab I have "rw,hard,intr".

This problem is pretty reproduceable for me. So far, it hasn't occurred when downloading large files.

I tried upgrading to the latest nfs-utils-1.0.12, and the problem still exists.

My server is currently running linux 2.6.19.1 -- My box is running 2.6.20.3

Abraxas · l33t Joined: 25 May 2003 Posts: 814

What version of NFS are you using?

kashani · Posted: Tue Mar 20, 2007 4:41 pm Post subject:

Are you actually exporting a whole partition? no_subtree_check should only be used when the NFS dir you're exporting is the top level of a partition. Assuming /var is /dev/sda5 or whatever using no_subtree_check would be fine. In the same case exporting /var/nfsmounts would not be fine.

kashani
_________________
Will personally fix your server in exchange for motorcycle related shop tools in good shape.

dennisn · Posted: Wed Mar 21, 2007 4:24 am Post subject:

When I use "subtree_check", after 10-20min the NFS share hangs

. When I "ls" the mounted directory, it displays "??? ????" for the user, group, creation time, etc.

When I use "no_subtree_check", sometimes I get "Stale NFS file handle" when I try to mount some directories. For others it does mount fine, even though they're not root dirs

-- although I'll get the problem that I originally mentioned

Sigh.

dennisn · Posted: Wed Mar 21, 2007 4:17 pm Post subject:

Hrm. Upon further examination, my NFS mounts now consistently (hang/crash/produce ????? ??? on ls) after ~15min, regardless of whether i have subtree_check or no_subtree check on the server

I have NFSD_V3 in the kernel (not V4). (although nfsstat shows activity for nfs v2 as well).

Should I be using V4?

dennisn · Posted: Wed Mar 21, 2007 6:52 pm Post subject:

Upon even further examination, in particular after reading about this nfs-utils bug,

https://bugs.gentoo.org/show_bug.cgi?id=168170

, i found that deleting a rogue/obsolete rpc.mountd process from my server (there were 2 running) fixed the timeout problems.

However, I'm still experiencing problems when uploading large files. Except it doesn't crash my system, but rather renders it VERY unresponsive. Applications take a ~10seconds to start. Sometimes the mouse will freeze for a few seconds. Same with keyboard. Windows may take a few seconds to redraw, etc.

Downloading large files from the server doesn't cause these issues.

(nfs-utils-1.0.12)

JeroenV · Posted: Mon Mar 26, 2007 7:56 pm Post subject:

I've been having the same problem, consistently upgrading nfs-utils on all my machines to 1.0.12-r2 didn't help.

When I copy a large file (> several hundred MB), it is transferred in bursts of about 2 seconds, followed by a stall of a few seconds, untill it reaches a permanent stall and then after a few seconds freezes my desktop. I can only recover from this by logging in via ssh and killing (-9) the process that was copying the file.
After I kill the process, it takes a short while before the system becomes responsive again.

This behaviour seems quite peculiar to me, isn't it "defective by design" that one hanging process can suspend all other processes, even if they have nothing to do with this particular I/O?

Anyway, any info from NFS experts would be most welcome, or should I start looking for alternatives (coda?).
_________________
Cheers

Jeroen
~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
May The Source be with you!