On Wed, 2008-04-02 at 11:17 -0700, Jeroen van Aart wrote:
> There are those, enough of them, who disagree. As I said countless
> discussions can be found, like this one:
> http://mailman1.u.washington.edu/pipermail/imap-uw/2007-September/001574.html
Quite; there are also counter-discussions such as this one:
http://readlist.com/lists/lists.sourceforge.net/courier-imap/0/983.html
> I am sorry but you perfectly point out why NFS would be a bad idea. It
> requires exim to work around NFS' flaws instead of just working, like
> when it'd access a real filesystem. How can you say exim works great
> with NFS if you have to set up your transport to use lock files rather
> than flock, or else files may become corrupted. That to me sounds like
> it does NOT work great, but works sort of, if you're careful and avoid
> the problems.
I've built a number of reasonably big Exim+Courier-IMAP load-balanced
clusters which used NetApp backends for storage with the filesystems
exposed to the IMAP servers over NFS. I don't recall having to tweak
anything in the Exim config at all, certainly no "working around flaws",
but I do recall having to set specific options on the IMAP daemon to do
with concurrent access from multiple clients.
I know of several high-volume Postfix & Courier systems doing the same
thing without issue, too.
Just because UW and Cyrus are alleged to not work properly, or reliably,
over NFS doesn't mean it can't be done properly, or reliably.
> In that same line of thought qmail works great IF you apply the 600 or
> so patches it needs to actually turn it into something resembling an MTA.
That's an entirely different argument altogether, I'm afraid.
Graeme
