Re: problems with soft-nfs when the server goes down

From: Yu-Shun Wang <yushunwa_at_ISI.EDU>
Date: Fri, 08 Jul 2005 15:49:39 -0700
Mohan Srinivasan wrote:

> --- Claus Guttesen <kometen_at_gmail.com> wrote:

>>>Now my NFS-nt is running 6-CURRENT and my NFS-server still runs 5.4
>>>When the server gets unavailable, the proccesses on the client will lock
>>>- OK so far.
>>>The problematic thing is, after the server is back online, I still see no
>>>way to get the locked processes back to continue. They are still in heir
>>>NFS-lock. All I can do is reboot the system. That only happens when I
>>>mount the nfs filesystem with "soft".
>>
>>Do you tcp- or udp-mount? AFAIK tcp-mount's don't work properly in 6.0
>>(current).

> NFS/TCP mounts should work in -current. There was a bug reported
> against NFS/TCP (retransmits were not happening reliably when the 
> server went down). That was fixed several weeks ago.
> 
> If you see something broken with NFS/TCP, please let me know.

Hi,

I am not sure that's the case. We might be talking about differnt
bugs FWIW. I also had NFS-related lockups as recent as CURRENT on
6/27/2005, probably on 7/5/2005 as well. You could search for
'nfs' on the list for the last two months. As far as I could
remember, the problem has to do with nfs pkts or mbuf mis-alignment,
and the realign functions in either the client or the server
couldn't deal with serious mis-alignment.

I could be wrong, check the archive for more details.

Just a data point of my setup: nfs client running CURRENT,
server is solaris, was using NFS/TCP with NFSv3. Simple
TCPDUMP showed that the client kept sending access request
and the server kept answering it, but the client never
accepted (for the lack of better terms) those replies.
So in my case, the server was never down (at least during
those lockups).

I have no hard evidence on how to re-produce the lockups,
some mentioned after or duing heavy IO traffic. But it's
definitely there. It annoyed me enough that I took the
easy way out: switched to NFS/UDP since yesterday. :-)

Thanks,

yushun
-- 
Yu-Shun Wang <yushunwa_at_isi.edu>  http://www.isi.edu/~yushunwa
USC Information Sciences Institute

Received on Fri Jul 08 2005 - 20:50:59 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:38 UTC