A while ago, I posted a question asking what the deal was with having
a single /var/spool/mail mounted via NFS on many machines. I was
curious whether lockd bugs could cause mail to be lost. Specifically,
I was interested in how SunOS and Ultrix dealt with this, and how elm
and sendmail fit into the picture.
I apologize for taking so long to summarize; there are two main things
that I found out:
1) Mail user agents (/usr/ucb/mail -- Ultrix & SunOS, elm, etc.)
generally use a second file to lock the mailbox such as
/usr/spool/mail/mosedale.lock. This seems to keep many problems at
bay.
2) It is considered a very good idea to have the mail transfer agent
(eg sendmail) on the subsidiary machines punt all incoming mail to
/var/spool/mail's home machine and have it delivered only on that home
machine. In SunOS, this is done by setting up sendmail.cf to use the
OR command (see /usr/lib/sendmail.subsidiary.cf). Ultrix sendmail
looks like it may a similar capability, but I'm not entirely clear on
how it works. IDA sendmail can do this using the MAILSERVER macro in
the m4 source file for sendmail.cf.
The general consensus seems to be that these measures are reasonable
solutions, and almost no-one has had any lost mail problems once these
are in place.
Thanks to everyone who helped me out.
Dan Mosedale
mosedale@genome.stanford.edu
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:50 CDT