If I only knew what others use when they do this successfully then all
my problems would be solved. Just kidding.
However here is a try. In 1998 I was a sysadmin and our setup was with
sendmail and postfix not smtpd. We had 3 Sun Ultra Sparc computers
connected with a small switch to a Netapp NFS filer, the MX priority was
all the same in DNS so as to achieve a load-share (probably not a true
share). The problems back then were NFS related particularily locking
meaning the Sun computers ran Solaris because it at the time did this
best. Back then it was proper to have backup MX's too I think with a
higher priority field in DNS, this ran postfix and all they did was
queue the mail and wait for the main mail hosts to come back from
whatever caused them to be down, then they'd deliver the mail there. It
was just a relayer.
You can try making a NFS solution, but the NFS filer is a single point
of failure itself, ours had multiple power supplies and I was there too
short a time to see it work over a few years. I don't know how well
CARP'ed NFS servers behave, I don't know if it creates problems, that's
for you to find out. I also don't know how well the locking works with
NFS and OpenBSD these days. A solution is to use dot-lock files which
gets its atomicity around an NFS stat operation (if that even exists, if
not then open operation).
The removal of the single point of failure is the tough part, it isn't
simple, IMO. Now that I've tried giving you insights to the holy grail
of mail, I wish you luck building your own system.
Regards,
-peter
On 11/24/18 1:15 AM, Mik J wrote:
Hello,
I'm wondering how to do a proper mail server failover.
Let's say smtp1 is down, the internet client resolves the other mx
with a lower priority and the mail goes to smtp2.
Now smtp2 writes the message on the disk in order to store it.
What do you people do in order to have a common storage for both smtp
which can be correct regardless whether a smtp goes up or down.
How do you manage the failover ?
Thank you
--
You received this mail because you are subscribed to [email protected]
To unsubscribe, send a mail to: [email protected]
