Recent postings from Graeme Wilford and Alan Thew have brought to light
an unintended change that I made at release 3.14 (change 8). I
re-organized the way Exim handles the "ultimate address timeout" to fix
a problem, but this had the following side-effect:
If a bounce message fails and gets frozen because ignore_errmsg_errors
is not set, manual or auto thawing won't get rid of it (assuming it
continues to fail). It just gets frozen again. However, if you use
ignore_errmsg_errors_after, it will be discarded after that has kicked
in and caused another retry.
The previous machinery would ultimately clean such things up even in the
absence of ignore_errmsg_errors_after. After some thought, I have
decided that there is no need to reinstate the (rather inelegant) rule I
had before. It seems to me that the existence of ignore_errmsg_errors_after
is now a sufficient way of dealing with these messages. It might even be
a good idea to put something like ignore_errmsg_errors_after=7d into the
default configuration.
However, this made me think about frozen messages that are not failed
bounces. These can be a result of freezing in the system filter, or
freezing because Exim has hit something unexpected. As long as they
continue to get frozen, these messages will never go away, even if
auto-thawed regularly.
The normal timeout mechanisms ensure that non-frozen messages will
ultimately fail. It seems to me that perhaps there should be an explicit
way of timing out frozen messages. The simplest would be to introduce a
new option, e.g.
timeout_frozen_after = 5d
This would operate like this: if a queue runner found a frozen messsage
that was that old, it would get rid of it, in one of two ways:
(a) If the message is not a bounce message, it will generate a bounce,
exactly as if the -Mg option had been used to force Exim to give
up.
(b) If the message is a bounce message, it will just discard it.
Anybody like to comment?
--
Philip Hazel University of Cambridge Computing Service,
ph10@??? Cambridge, England. Phone: +44 1223 334714.
