Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!mrfloods.cc.umich.edu!paul
From: paul@mrfloods.cc.umich.edu ('da Kingfish)
Newsgroups: comp.mail.sendmail
Subject: Re: NFS mail
Message-ID: <968@mailrus.cc.umich.edu>
Date: 3 Mar 89 22:28:06 GMT
References: <17946@genrad.UUCP> <8853@polyslo.CalPoly.EDU>
Sender: usenet@mailrus.cc.umich.edu
Reply-To: paul@mrfloods.cc.umich.edu ('da Kingfish)
Organization: University of Michigan Computing Center, Ann Arbor
Lines: 15
UUCP-Path: mailrus!mrfloods.cc.umich.edu!paul

We do this on a large apollo ring (not with NFS).  I wrote versions of
open and flock that use Apollo's ios_$ calls and take care of locking
and concurrency control correctly in the Apollo environment.  There,
all the machines just call themselves the same thing
(caen.engin.umich.edu) and we depend on looking at received-by lines to
see when and where the message was generated (we use Apollo uids for
message ids).

For NFS setups we have all the drone machines forward all mail to the
master.  The master has the /usr/spool/mail directory, the drones have
mqueues.  Then the normal sendmail queueing will take care of retries,
etc.  It also makes things easier relative to locking and root access
over NFS, etc.

--paul