[comp.mail.sendmail] Sharing a mailq directory

jf@ap.co.umist.ac.uk (John Forrest) (03/02/91)

We're finally seeing one of those problems I knew was there but
I hoped would never occur in real life. Basically all the
machines in our cluster share a common /usr/spool/mailq - not
as bad as it seems because it's on the same disk as the
maildrops. Anyway, there is a finite problem that occurs if two
sendmail processes on different machines have the same pid.
This appears to have happened. One way round is to give each
machine a separate mailq, but I don't really want to do this
because we will have to run a queue flush process periodically
on each. Another way might be to add the creating node's name
to the file. Has anybody got a fix that does the latter, before
I try to write one? I am guessing this is the fault, but it
does ring true - one machine was trying to add something to the
queue while another complained it couldn't with the same queue
numbers. Another way might be to check the software that does
not duplicate an existing mail number - I presume this exists
for when mail fills up.

We run Sendmailv5.65a (IDA version) on a network of Apollo's
OS10.1.

John Forrest
Dept of Computation
UMIST