nfs server freezes

kwiqsilver · Guru Joined: 19 Mar 2003 Posts: 360

I have my /home directory nfs exported from one box to another (both run gentoo 1.4 installed the last week of March). When ripping CDs to the nfs mounted /home dir on the client, the nfs connection will just stop (my guess is it's the huge traffic). I can ssh, ping, and ftp between the boxes, but anything like "ls /home", or trying to log in as a normal user on the client freezes, waiting for a response from the server, I presume.
Neither box will shut down properly because of the NFS error.
I've looked through the client's /var/log directory and found nfs timeout error entries, but on the server, I can't find any log entry that helps.
Does anybody have an idea of what's going on here, or some advice to point me in the right direction?
Thanks.

plate · Posted: Mon Aug 18, 2003 6:56 pm Post subject:

Resurrecting this thread with high hopes that either you have gotten it fixed by now and can tell me what's wrong, or somebody else jumps on the occasion and makes this dreadful thing go away.

My problem is very similar to yours, I'm serving /usr/portage/distfiles from my main desktop as a central repository. The adjacent lesser desktop never had problems, it sees the directory as if it was on its own filesystem, but my notebook chokes on a simple emerge operation. See for yourself:

'mount /usr/portage/distfiles' works quickly and without error message. My line in /etc/fstab:

neysx · Retired Dev Joined: 27 Jan 2003 Posts: 795

I once had to use the nolock option when mounting nfs volumes.
Without it, the only way out was to kill rpciod from another console. Ctrl-C would not work.

Simply add nolock in the list of options in your fstab or add it on your cmd line mount -o rw,nolock....

From the depths of my memory I think you need portmap to be started before you mount anything, but the scripts do handle that properly, don't they ?

I am not using it anymore and I also share my distfiles across several machines and also have a home-built cluster of C3 with root-over-nfs and have no problem with nfs.

Hth.

plate · Posted: Mon Aug 18, 2003 9:42 pm Post subject:

Thanks! Much luckier with nolock. Haven't tried it on a heavier package yet, but emerge does MD5 checksums and unpacks files from a remote NFS-mounted /usr/portage/distfiles now. I suspect this has something to do with my WLAN card in the notebook, since I don't need nolock on the other desktop.

Speaking about scripts, all I have in /etc/runlevels/default is nfs. Any idea what nfsmount and netmount could be needed for?

neysx · Retired Dev Joined: 27 Jan 2003 Posts: 795

plate · Posted: Tue Aug 19, 2003 8:18 am Post subject:

Thanks for wrapping this up for me. When it comes to NFS I have this mental block, possibly due to my first job assignment more than a decade ago when our Unix sysadmin struggled with an obscure PC-NFS (and lost, mostly). :lol:

plate · Posted: Wed Aug 20, 2003 1:20 pm Post subject:

Or so I thought. :cry:

Apparently it's rather picky about what works and what doesn't. Emerging the world today worked just fine for the first three packages from the NFS-mounted /usr/portage/distfiles directory (bash, elfutils, prelink), but stalled on the fourth:

neysx · Retired Dev Joined: 27 Jan 2003 Posts: 795

It would be nice to know if it is your client or your server which causes problems. Is your other box able to see your /usr/portage/distfiles through NFS after this one hangs ?
Have you checked with another kernel ? net card kernel options ?
Have tried to export some directory from your other box to connect to it ?
Your problem could be linked to your WLAN card.
Maybe you could try the following options when mounting:
rsize=1024,soft,timeo=75,retrans=1
to reduce to block size which migh be a good idea on error-prone net and to have timeouts faster if they occur

If all fails with NFS, could you try smbfs ?
I could not help with samba, I am not using it as I live in a 100% windown-free environment.

plate · Posted: Wed Aug 20, 2003 4:14 pm Post subject:

Hey, you're my new hero. Next time I'm in Groningen prepare for a drink or two.

Must be the WLAN connection, I suppose. Tuning NFS down to the settings you suggested did it - took forever and had me worried for a bit, but worked. What do you think, is that because of "soft" instead of "intr", the reduced rsize/wsize or the timeout?

In any case, thanks a lot. In the long run that doesn't save me from putting Samba on this machine, too, though: Even our FreeBSD fileservers run Samba, so why shouldn't I...

neysx · Retired Dev Joined: 27 Jan 2003 Posts: 795

Glad I could help.

A smaller rsize means less chance of corrupted packets and those that need to be resent are smaller and timeouts being detected faster means less lost time.

http://www.sallyobriens.nl/groningenindex.htm is the place ;-)

Looks like you are based in a great location yourself.

TheCoop · Posted: Fri Jun 11, 2004 10:05 pm Post subject:

thanks, this solved the problem of my entire wireless connection locking up when i tried to use nfs, specifically /usr/portage/distfiles and /media (music files) mounted over nfs. Thanks!
_________________
95% of all computer errors occur between chair and keyboard (TM)

"One World, One web, One program" - Microsoft Promo ad.
"Ein Volk, Ein Reich, Ein Führer" - Adolf Hitler

Change the world - move a rock