Viktor Dukhovni:
> Therefore, if this were to be made possible, the right mechanism would
> be to to somehow expedite message expiration, with normal processing
> on message expiration happening earlier than it would otherwise.
I have a list of alternatives. The more reasonable ones reuse the
parallized expiration code in the queue manager, instead of duplicating
it in the non-parallized postsuper command.
> This could be done by adding a place-holder record to the queue file,
> that could be later atomically (one byte write) updated to indicate
> that the message is administratively expired. Such messages would
> expire after the next delivery attempt and would not be deferred
> again.
It can also be done by atomically updating an existing flag in the
size record. That means the postsuper command needs to be careful,
but it can be done.
Wietse
