slevy@nic.MR.NET (Stuart Levy) (01/08/89)
We're running sendmail 5.59, and have a batch of aliases for redistributing
Internet mailing lists locally. We don't control all the redistribuees,
and so sometimes messages bounce. Although we have "owner-xxx" aliases
for each mailing list "xxx", the owner- aliases often don't receive the
error reports.
There seem to be two problems:
1) mail passed downstream from such a redistribution is not tagged
for error reporting to the given owner-* alias, e.g. by changing the
SMTP Mail From: field or by adding an Errors-to: line to the message.
That would be nice for limiting redistribution of downstream errors.
Can anyone see why this shouldn't be done?
2) (More serious) If the mail can't be delivered on the first try,
it's queued of course; *however*, there is no indication in the
queued copy that the mail was originally sent through an alias
nor that delivery errors should be sent to owner-alias.
This appears to be hard to fix.
So, if say a redistribuee is down for a while and all their mail
starts timing out, the owner- alias will never see the replies!
Does anybody know how this might be fixed? Or, how do *you* redistribute
mailing lists?
Stuart Levy, Minnesota Supercomputer Center
slevy@uc.msc.umn.edu, slevy@nic.mr.netmatt@oddjob.uchicago.edu (Matt Crawford) (01/10/89)
I distribute mailing lists the same way you do, Stuart, and I've grumbled at the same problems. I looked into fixing them inside sendmail and it didn't look easy, nor was it perfecly obvious that the result would be "Correct". I've seen some other people's cute tricks involving piping the outgoing list mail to "sendmail -f list-owner ...". Messy, but it does do the job. Matt Crawford
slevy@nic.MR.NET (Stuart Levy) (01/15/89)
Brian Kantor (brian@ucsd.edu) offered the following well-thought-out
series of /usr/lib/aliases entries for mailing-list redistribution with
errors, including downstream errors, sent to a specified address
("brian@ucsd" here). I don't think "-odq" is necessary, it just trades
distribution latency for possibly reduced peak load. Downstream messages
have a 'From ' and SMTP 'MAIL FROM:' address of 'tcp-group-request',
so errors are sent there, but the 'From:' line remains the original sender.
tcp-group-request:brian@ucsd
tcp-group: "|/usr/lib/sendmail -odq -oi -ftcp-group-request tcp-group-list"
tcp-group-list: :include: /usr/net/mail/maillists/tcp-group
owner-tcp-group:postmaster@ucsd
That way the mail appears to come from the -request address, so all the
errors go there instead of to the message originator.
- Brian
-----------
Stuart Levy