On Wed, Aug 08, 2012 at 12:58:41AM +0300, Nikolaos Milas wrote:
> A question: Could we use the format:
>
> /etc/postfix/transportmap
> example.com relay:[mail.example.com], relay:[mail2.example.com]
No. As documented.
> to force relaying to *both* those servers (mail AND mail2) at the
> same time? According to the documentation, I don't see that's
> possible (it seems only one transport:nexthop definition is
> acceptable), but can we achieve this somehow (except using
> always_bcc on mail.example.com)?
You can use virtual_alias_maps to rewrite each user to two mailbox
addresses and deliver to both, this is likely unwise. The two
mailboxes are no longer identical, messages arrive out of order
with slightly different content, deletion is difficult to synchronize,
...
> A second question: Is anyone aware of any (Linux) "smart" script
> automating (at least to some extent) reliably enough detection of a
> mail server failure (to avoid re-inventing the wheel), so that we
> can subsequently switch to a different transportmap?
This decision is best not automated. If you want HA, design a
resilient IMAP store, that supports LMTP delivery from multiple
MTAs. The store will support replication of data in some fashion
(block level with SANs, FS layer with NAS, mail store as a replicated
database, ...)
This is not a Postfix issue, you need to look at how this is done
in Zimbra, Exchange, large-scale Cyrus IMAP deployments, ... The
problem is not trivial, the architectures of the largest deployments
(Gmail, ...) are not publically documented.
--
Viktor.
