[net.mail] the mail must go through

dave@uwvax.ARPA (05/13/84)

> Erik Fair says that if ARPANET mail tranfer fails (due to unavailable
> ARPANET service), a message sent to
>     ...seismo!hao!hplabs!....!sri-unix!foo@ucl-cs.arpa
> will be returned to the sender.
>
> This is unacceptable if UUCP service from seismo to hao is available.
>
> Remember the adage of the mailman ...
> "Neither rain nor snow nor dead of night ...
>  Shall keep these couriers from their flight."  (paraphrased)
>
> This should apply to electronic mail also.  Do YOU think that if
> airplane service is unavailable at the moment, transatlantic mail is
> returned?  Is your local mail returned if the mailman's mail truck
> breaks down?  Of course not.

Sendmail (and ARPAnet) have two ways for mail to fail, temporary (analgous
to your mail truck breaking down) and permanent (like mailing to the
empty lot down the street).  If the mail fails because of an incorrect
address, it should be returned.  Period.

> Solution:  Have sendmail take the other route if the higher-priority
> route fails,

It doesn't matter if seismo can talk to hao, as that isn't what the person
said to do (does the mailman know that when someone mailed to that empty
lot, they really meant to send it to you?).

> or (better idea) have the message queued at seismo for
> later delivery until seismo's ARPANET service comes back up.

This is the way ARPAnet mail works already.  There is generally a three
day timeout (sometimes longer) on a message before mail is returned because
of a transmission problem.  That timeout is necessary too -- what if the
receiving host is permanently down?
-- 
Dave Cohrs @ wisconsin
...!{allegra,heurikon,ihnp4,seismo,sfwin,ucbvax,uwm-evax}!uwvax!dave
dave@wisc-rsch.arpa

chris@umcp-cs.UUCP (05/14/84)

Saying ``use whichever one works'' is not a solution either.
If umcp-cs receives mail to ``seismo!username@washington''
and there is a user called ``username'' at washington, and
we send it to seismo because our ACC is down at the moment,
even though the intended destination was the user ``username''
at seismo, then someone is going to get upset....
-- 
In-Real-Life: Chris Torek, Univ of MD Comp Sci (301) 454-7690
UUCP:	{seismo,allegra,brl-bmd}!umcp-cs!chris
CSNet:	chris@umcp-cs		ARPA:	chris@maryland

gds@mit-eddie.UUCP (Greg Skinner) (05/14/84)

Arpanet mail is not queued if a sending host (seismo in this case) has
disabled ARPAnet service, to my knowledge.

Also, it is not the case of a wrong address here ...
   
   ...seismo!hao!...!sri-unix!foo@arpasite.arpa

... it is the case that UUCP sites on the ARPAnet choose to interpret
the address differently than as the reply to a message sent from
arpasite.arpa to
 
   ....!hao!seismo!....!endhost!enduser@sri-unix.arpa

Instead of introducing state machines in sendmail/SMTP, perhaps a
quoting feature could be implemented so that the address in quotes is
interpreted by the current mail agent for his particular context.   For
example, 

   ...!seismo!hao!...!sri-unix!enduser@endarpahost.arpa

means deliver me to seismo, then hao, etc.  until you get to sri-unix,
then SMTP automagically takes over, whereas

   ...!seismo!"hao!...!sri-unix!enduser@endarpahost.arpa"

means deliver me to seismo, THEN use SMTP to deliver me to endarpahost,
etc. 

Hopefully, when domains are implemented, these problems will be
resolved, but in the meantime, back to header munging ...
-- 
					Be ye moby,
					for I am moby.

Greg Skinner (gregbo)
{decvax!genrad, eagle!mit-vax, ihnp4}!mit-eddie!gds

Joy is in the ears that hear.

gds@mit-eddie.UUCP (05/17/84)

Erik Fair says that if ARPANET mail tranfer fails (due to unavailable
ARPANET service), a message sent to

    ...seismo!hao!hplabs!....!sri-unix!foo@ucl-cs.arpa

will be returned to the sender.

This is unacceptable if UUCP service from seismo to hao is available.

Remember the adage of the mailman ...
"Neither rain nor snow nor dead of night ...
 Shall keep these couriers from their flight."  (paraphrased)

This should apply to electronic mail also.  Do YOU think that if
airplane service is unavailable at the moment, transatlantic mail is
returned?  Is your local mail returned if the mailman's mail truck
breaks down?  Of course not.

Soultion:  Have sendmail take the other route if the higher-priority
route fails, or (better idea) have the message queued at seismo for
later delivery until seismo's ARPANET service comes back up.

The mail must go through ...
-- 
					Be ye moby,
					for I am moby.

Greg Skinner (gregbo)
{decvax!genrad, eagle!mit-vax, ihnp4}!mit-eddie!gds

You can't trust anyone around here with the su password these days.