On Monday 25 February 2008, Christopher Chan wrote:
Hmm...it will still build. To really fix it, you need to do one more step:
rpm -e --nodeps sendmail
Now that is a permanent solution.
Like a hand grenade is a "solution". Not likely to help him much, tho. =/ Doesn't even begin to address his situation since sendmail wasn't the problem to begin with.
Seems to me that it's a bad idea to use NFS as a mail store. For example, the RedHat documentation specifically recommends strongly *against* it. Very flatly:
Never put the mail spool directory, /var/spool/mail/, on an NFS shared volume.
http://www.redhat.com/docs/manuals/linux/RHL-9-Manual/security-guide/s1-serv...
Also, NFS has various locking problems which prevent its use in a proper mail cluster. Read up on sendmail's mbox vs qmail's maildir for more details. Not suggesting that you switch to qmail, with it's "recompile the whole !@#! thing every time you change a config option" mentality, but it's useful information nonetheless, especially when you get into having multiple mail receipt hosts.
The additional complexity of NFS is what seems to have caused this gentleman's problem - not only did sendmail itself have to work properly, so did NFS, DNS, and the spam filter.
How to avoid it? Either:
1) Reduce complexity. (get rid of the need for DNS, NFS, etc. or
2) "Beef up" the various pieces so they don't fail - make sure you are using high quality servers and equipment, or
3) Increase redundancy, so that no single point of failure exists.
Why is he depending on a single DNS server? Why is he using NFS, with it's implicit single-point-of-failure rather than GlusterFS, which provides multiple-primary-host redundancy and automatic failover? http://www.gluster.org/
-Ben