Hi,
I have a mailing list host with many subscribers unreachable due to
internet cable problem. Those for which there is authoritative DNS
available go into the tcp timeout table and are not retried for 1 h.
Those for which authoritative DNS is not available (all on the other
side of the break) will be retried and hog a "concurrency" until DNS
times out. Those for which no DNS also yield the same result.
Thus, I have several 100 list messages in the queue all constantly
retrying the same set of addresses, all of which clog the system
waiting for DNS replies. This adversely affects deliveries to reachable
addresses.
Would it make sense to have a similar table (or put into the same
table) also host for which authoritative DNS is not available? Usually,
this is due to a network problem (as for the tcpto entries) and it
affects all messages to the host (same as for tcpto table entries). If
the failure is transient, a 1h delay would be acceptable.
The host with complete DNS failure are more difficult, since it would
require storing the host name or a hash thereof, so there it may make
more sense to just keep retrying.
Of course, increasing concurrency would be a solution. However, the
system runs at an average concurrency of 5 normally and due to the
network problems it is now pegged at 120.
Thanks!
-Sincerely, Fred
(Frederik Lindberg, Infectious Diseases, WashU, St. Louis, MO, USA)