Re: [Exim] Large-scale deployment of exim (Solaris/FC-AL/etc…

On Fri, Mar 29, 2002 at 08:37:28PM +0000, Philip Hazel wrote:
> On Thu, 28 Mar 2002, Elwood Blues wrote:
>
> > Where server-A and server-B have the drive array, which will be
> > set up as a single partition, mounted read-write as /var/mail
> > for mail-spool storage. The servers will have their own internal
> > system disks. Assuming that any program accessing the mail
> > spools locks the spool via fcntl(), would this be possible?

No, it won't...

> I don't know anything about Fibre Channel. Is it purely hardware? If so,
> the locking will apply only within the host that does it, I think.
>
> However, using a lock file should (?) work.

The problem here is that the filesystem implementation inside the
kernel can't use lock files to synchronize access to directories and
metadata. Your filesystem would quickly become garbage if two systems
are trying to update the same filesystem at the same time.

Basically what you need to do is to partition the drive space into two
partitions, one for each server, and then spread mail accounts between
the two servers. To then achieve some resiliency you'll need some
scheme for one healthy server to take over the services provided by a
failed server. The cheap way is to manually remount the partitions
but there are any number of clustering products out there that will
make failover nearly transparet.

Jeff