Scott Morizot wrote:

As far as I understand Sam expected that DNS caches (specially bind) would return randomized lists of MXs servers as a way to guarantee that, in case of temporary delivery failures, Courier would try others top priority MXs. Unfortunatelly, bind randomizes it's answers in a way that the goal of Courier trying other top priority MXs in a near equally distributed fashion isn't accomplished.

I might agree with you that this might be the best way to setup one's domain MX records. I haven't really thought about it much. But the point here is that Courier have to deal with domains that eventually have a mix of a lot of same *and* different priority MX records in the MX list. There aren't much options on this point. It might be good to Courier be able to deal with it better than it's doing right now.

BTW let me say that you have just clarified to me an important detail of this issue. It will only happen if and only if there are a mix of a lot of same *and* different priority MX records in the MX list. This makes it an issue in fewer cases than I previously saw.

About the question *where* it should be solved, I really think the right answer is in Courier's side. It's Courier that's expecting answers in a way that no current DNS cache that I am aware of provide.

There is also the question *if* it should be solved. Remembering that this issue affects a smaller set of setups than I previously imagined I'm wondering if it might be better to:

a) not to deal with it at all because there might be so few cases that there is no reason to bother (I not sure this is a good line of thinking but a possible one for sure) or

b) try to detect the specific situation when this problem occurs with simple non-obstrusive code and use a more convoluted strategy to choose MXs only then;

instead of changing the way Courier chooses MXs for every situation as it's current implementation is quite appropriate most of the times.

Regards,