Wietse Venema:
> Jose Perez:
> > Hi:
> >
> > On Thu, Mar 26, 2009 at 6:05 PM, Wietse Venema <wie...@porcupine.org> wrote:
> > > carconni:
> > >> Hi,
> > >>
> > >> I've been digging around and I haven't been able to find what I'm
> > >> really looking for so I thought I'd go straight to the ones "who know".
> > >>
> > >> Can Postfix be run in a clustered environment (ie: multiple servers
> > >> running postfix utilizing one data store) under any OS?
> > >
> > > Each Postfix instance must have its own config_directory,
> > > queue_directory and data_directory. These cannot be shared.
> >
> > So how can I make sure that a queue file (stored in one node of a
> > cluster) it will be managed by another postfix instance (running in a
> > different node of a cluster)?
>
> I don't have to solve that problem.
>
> You can use any file/disk sharing mechanism as long as there is no
> simultaneous access to queue_directory and data_directory, and as
> long as the sharing mechanism respects the RFC 5321 transaction
> requirement:
>
> Once the MTA replies with 2XX to END-OF-DATA, mail must not
> disappear just because of some crash or power failure.
In terms of system implementations, this means that fsync() must
not return until the queue file's data and metadata are stored on
the disk (or on their way to the disk, in battery-backed buffers).
> Thus, running rsync from one disk to another does not qualify,
> because there is a time window where a file exists only in one
> place.
I mention rsync here because it is an example where replication
happens after fsync(), resulting in a time window where data
exists in one place but not in the other.
I don't know if DRBD propagates data before fsync() returns, or
whether its updates happen later. If the updates happen later, the
backup may never learn that mail was queued because the update
still sits in the sender's DRBD queue.
NFS does not have this propagation problem, but of course you
need a server with high-quality RAID.
Wietse
> > I thought that using DRBD (network mirrored data) over the postfix
> > queue directory would be the solution but apparently I'm wrong...
> >
> > >> I know NFS is not really recommended - is there any alternative?
> > >
> > > For support statement, see http://www.postfix.org/NFS_README.html
>
>
