On Tue, Apr 26, 2005 at 12:56:51PM +0100, Michael Smith wrote:
> This is probably not relevant, and you may have encountered it before,
> but there is an entry in either the bogofilter or berkeley db README
> files which warn you not to put the bogofilter database on an NFS
> mounted volume.

http://bogofilter.sourceforge.net/faq.shtml#nfs

I figured I could dance around the problem by having all the primary
activity only occurring on my mail server (which is mounting home
directories remotely).  I know that locking an NFS file via fcntl() at
least works when both lockers are on the same host.  but now you've got
me checking my db4 compile configuration to make sure it's even doing
that.

> I had a lot of problems using it on a linux NFS client and I
> eventually moved the database to a local disk. I have a vague idea
> that some NFS clients will work with berkeley db, but I don't know if
> NetBSD is one of them.

NetBSD still doesn't handle locking via fcntl() across hosts, which is
something even ultrix did...  10 years ago.

but this is all aside the point which is that my client is losing track
of my server and generating bogus "nfs server foo:/bar: not responding"
messages, and not reawakening until HOURS later:

Apr 29 10:39:43 radbug /netbsd: nfs server arwen:/usr/home: not responding
[...]
Apr 29 12:39:32 radbug /netbsd: nfs server arwen:/usr/home: is alive again

there's got to be some timer somewhere that isn't being updated
correctly.  I'd suspect that it's MP related, but I'm seeing similar
things on my uniprocessor alpha.

-- 
  Aaron J. Grier | "Not your ordinary poofy goof." | agrier@poofygoof.com