On Wed, 7 Apr 1999, Nigel Metheringham wrote:
> eximlist@??? said:
> > Well, there is only one machine that is the MX for the domain. But
> > other machines occasionally originate and deliver local mail (such as
> > system crontasks sending to root.. etc) to /var/spool/mail which is
> > NFS mounted on all machines.
>
> I would reconfigure that cluster so that only one machine delivers and
> all the rest deliver to it - this has many advantages - single point of
> management (the pain caused by mislaligned alias files is horrid),
> single delivery process etc.
Yes, I've gotten that part. That is not feasible at this moment in time. I
need to do best-case locking for multiple machines delivering to
/var/spool/mail, with one machine primarily handling the mail, but other
machines occasionally delivering mail that originated on that machine
(such as cron reports, etc), where some machines might be sendmail calling
procmail, and some machines running exim (hopefully also calling procmail
as soon as I get that straightened out)
exim will be reading an NIS map for /etc/aliases, so aliasfile
misalignment is not an issue.
> > Does procmail properly lock over NFS?
>
> Yes, within the limits of NFS. NFS locking is not reliable enough to
> stake your career/life/marriage/testicles on. It just ain't no matter
> what the sales people tell you.
>
> > I am under the impression that exim does handle locking NFS-mounted
> > mailspools properly - am I incorrect?
>
> [duplicate previous answer]
>
> > Why is locking over NFS a problem, especially if lockfiles are used?
> > (eg /var/spool/mail/root.lock)? Is it only a problem for flock/lockf
> > types of locking? (which I don't wholly understand anyway)
>
> > Better yet, is there an FAQ or some other document discussing locking
> > mailspools over NFS that I could refer to?
>
> I used to run a mail system with NFS mounted spool on Sun systems (Suns
> do good NFS locking). It occasionally lost/corrupt mail.
So far all I'm getting is religious opposition to locking mailspools over
NFS. While I do accept that, having some religious type beliefs about
systems myself, I really want to know the details of the mechanics and
issues involved. If the delivering processes use a seperate lockfile, and
the presence of that lockfile indicates that another process is delivering
to that file, why is that not reliable? Is it possible that creation of a
file by one process might not be immediately visible on another system?
I've heard of something called lockd - what is that, does it help, should
I run it? How does procmail interact?
The sendmail->exim conversion conflicts aside, ideally once this is an
all-exim system I would really like to be able to reliably do local
deliveries on multiple machines. Single point of management is cool.
Single point of failure sucks.
Like I said, a pointer to a FAQ or something would be fantastic. I will of
course continue looking for such information from various search engines
as well.
--
*** Exim information can be found at
http://www.exim.org/ ***
