You can use the multi-master replication as a performance booster.
You just have to make sure that all traffic for a particular user (SMTP
and IMAP) goes to one server (not necessarily the same one) for the
duration of their session.
The easiest way is to just assign a user to a server permanently, use
perdition to proxy their imap and setup some mojo with postfix so it
delivers it to the right box.
If your in a really high load environment this architecture seems like
it might work.
Stick 2 big boxes running just postfix + dbmail-lmtpd + mysql, again
keep the postfix rules so that particular users mail goes to a
particular server.
Add a bunch of little boxes running just dbmail-imap + mysql, use
perdition/ultra-monkey to direct users such that they sit on one server.
Your little boxes are going to be doing mostly read only access, and
they can cache the stuff for their local users so you get to keep your
performance, the big boxes are going to be handling all your inserts etc.
Could work ;->
Gordan Bobic wrote:
It depends on how you use it. I have 2 servers in a cluster and each
is set up to use it's local instance of the DB, with the DB in
master-master configuration. The front end is DNS load balanced, but
dbmail itself always uses the local database. This works around most
of the scope for race conditions to occur. Either way, you should
stress-test any such configuration extensively before you deploy it.
Gordan
Jesse Norell wrote:
The more dbmail-specific (or mail-specific, really) reason is it breaks
the imap protocol. Aside from any issues like Gordan alludes to here, I
believe you could use it with pop3 and lmtp without a problem.
On Thu, 2008-06-12 at 18:56 +0100, Gordan Bobic wrote:
James Greig wrote:
Hi,
I've probably asked this before however can someone remind me why
mysql
master-master (innodb) replication is not recommend with dbmail?
It's not "recommended" with anything. By the very nature of it,
there is a potential very nasty race condition inherent in
multi-master MySQL replication. In practice you'd have to be
extremely unlucky to encounter it, but since there is no
cluster-wide locking available in this setup, it is a possibility.
FWIW, I use it with multi-master replication. YMMV.
Gordan
_______________________________________________
DBmail mailing list
[email protected]
https://mailman.fastxs.nl/mailman/listinfo/dbmail
_______________________________________________
DBmail mailing list
[email protected]
https://mailman.fastxs.nl/mailman/listinfo/dbmail
