On Wed, Aug 30, 2006 at 10:04:40PM -0400, Robert A. Rosenberg wrote:
thus assume that it is being tried WITHOUT first trying MX#1 [which
is NOT a 100% bullet-proof assumption since the problem triggering a
valid fall-back COULD be a connectivity problem between sender ISP
and MX#1 while sender ISP can still connect to MX#2 due to using a
different path - the whole idea of having separate locations for your
MXs]).
Even if the first and last MX are on the same host but with a different
IP address this can be true. I have seen a layer-3 switch behave
weird; some addresses on the subnet had connectivity while others
did not. I forgot the exact details but say .1, .2 and .3 had
connectivity, .4 to .7 had not, .8 to .11 did, and so on.
Or the almost impossible coincidence that the network fails, but just
before the client is going to try that last MX it suddenly works.
It may not happen more than once in 10,000 but still it happens.
Have ten MXes, the last one being such a spamtrap, and you still can
see valid messages end up there. Chances may be slim but not zero.
Alex