Wietse Venema wrote:
Miles Fidelman:
To find solutions, open your favorite search engine and try "cyrus
mailbox replication", "dovecot meailbox replication", and so on.
I've been wondering about this too, and it strikes me that "mailbox
replication" is only relevant to local delivery. What about replicating
the various intermediate mail queues? (My current HA setup is brute
force - a failover virtual machine, with a completely replicated file
system. But I've been looking for ways that are more granular, and that
are easier to do across two separate data centers.)
Have you considered the following:
- Inbound mail spends a fraction of a second in the queue.
- Inbound mail spends days or weeks or more in the mailbox.
- If an MTA goes down, mail flows via alternate MX hosts.
- If the mailbox store goes down, then you have no mail.
That's why high availability focuses on the mailbox store,
not on the MTA in the middle.
Well yes, in theory - but in practice we run a bunch of email lists, and
I find that there are always cases where one or more destinations are
temporarily unavailable - so there are various messages that will hang
around for a while. So HA for the queues is not unreasonable to think
about.
Miles
--
In theory, there is no difference between theory and practice.
In practice, there is. .... Yogi Berra
