Carl S. Gutekunst wrote:
I'd like to think that it wouldn't matter; an accumulation at a single
MTA should just be the trivial case, no?
I think it depends on the MTA queuing methods, its sorting, groupings,
limits set to not overload sites, throughput desires/needs. But one
of the minor points I guess is sending backed up mail to your site
still needs to be aligned with the normal rest of the loads to other
sites (i.e. should it be priorities). Queue residence time (age) to
factors in, at least for our logic.
Its also the unknowns though on how your site will limit the "storm",
even an intentionally small 100 amount example maybe good enough to
trigger it at many sites.
For the period when your are down, I wonder how keeping a server
active just to issue a 421 may be less helpful than just not listening
for connections. This is not greylisting and our retry delays are
higher for socket connection timeouts. Though I can see where a 421
with a wait=time will have a similar benefit and without the
connection delay. That can a plus.