M. wrote, at 11/03/2008 03:41 PM: > On Mon, 2008-11-03 at 15:26 -0500, Jorey Bump wrote: >> Although checking the MX record before provisioning would provide the >> ultimate verification, it would expose the domain to the possibility of >> lost mail, since it requires the customer to change the MX before the >> destination is ready to accept mail (resulting in a permanent error). >> This could cause problems if the customer is migrating from a working >> system (but is obviously less of a problem if this is the first MX for >> the domain and addresses aren't in circulation, yet). It seems to me >> that, for many cases, best practice mandates that the MX record should >> not point to the destination until it is ready. > > Yep, this is exactly what I was thinking about ;) > Maybe there is workaround to this problem using transports, relays or > whatever?
Separating your MX from your submission service still appears to offer the best protection. There are multiple ways to achieve this, depending on the flexibility you require. This makes your MX less likely to ever need to relay to an external domain, so its own configuration wouldn't cause conflicts. Your submission service would always resolve the destination MX, and since it's not an MX itself, it wouldn't interfere, either. I recently had a client experience the type of trouble you're describing. Their developer partially configured DNS and mail at another host, but gave up when it proved to be too complicated. As a result, DNS for the domain was hijacked for all customers of that hosting company (my DNS server was still the SOA, so the rest of the Internet wasn't affected). Now the developer and other important domains could no longer communicate with my client, as they were using the hosting company's resolver, which pointed to their own MX, which rejected the "unknown users" (fortunately). Once I determined the problem, the developer deleted the account, and everything started working again. Separation of the MX and submission service (each using a resolver independent of the hosting company's authoritative SOA servers) would have prevented this mess. This approach will help with your original problem, but you'll still need to improve your provisioning process, so your users can't hijack other domains.
