Excerpts from transarc.external.info-afs: 22-Nov-94 Re: news server in
AFS? "Perry E. Metzger"@imsi. (2251)
> The other method requires extensive hacking and yields, IMHO mediocre
> results, especially with respect to reliability, load, and security.
I have to wonder what experience you base this on.
> I can name dozens of reasons to run pop instead of having a single
> /var/spool/mail or even dropping the mail in users boxes. The former
> is like shooting fish in a barrel, but the latter is full of problems,
> too. As one example, what happens when the user's server goes down? Do
> they stop getting mail or does your radically hacked sendmail just
> hang? Or does it requeue?
When you go to create a file in a directory resident on a server that's
down or unreachable, your AFS CM will return an error rather than
complete the call. It will do this after a minute or so, unless it's
already discovered that it can't talk to the given server, in which case
the error return is immediate. All a delivery process has to do is to
translate the transient filesystem error code into EX_TEMPFAIL, at which
point the STOCK (not hacked) sendmail will queue the message and try
again later. Of course, that user doesn't get mail until their server
becomes available again, but that's life in the distributed-processing
world. Fortunately, other users, with home dirs on other servers,
continue to get their mail.
There's no a priori reason to hack sendmail with AFS-based delivery:
anything I've mentioned is a change or two to sendmail.cf. The new
version-8 one even understands AFS-ish error codes, like ETIMEDOUT, and
will treat those as transient errors internally.
Craig
