burzio@mmlai.UUCP (Anthony Burzio) (08/10/89)
I found an interesting problem with the lp program and sendmail on my 300 HP-UX 6.5 cluster. When a print job is cancelled by root, the user gets a mail message back informing him. If the print job was queued from a discless node, the mail gets sent to this node. On a discless node, however, the sendmail program is not supposed to be run, according to the documentation. The mail message sits round in the mailq for three days after which root gets a returned mail message. Is this a feature, or is there a way around it? ********************************************************************* Tony Burzio * You poor deluded woman! Martin Marietta Labs * - Batman mmlai!burzio@uunet.uu.net * *********************************************************************
diamant@hpfclp.SDE.HP.COM (John Diamant) (08/14/89)
> The > mail message sits round in the mailq for three days after > which root gets a returned mail message. Is this a feature, > or is there a way around it? Well, I don't think anyone will argue that this is a feature :-) As a side node, this situation only happens as a result of the lpdaemon (or possibly other automatically generated mail). Mail sent from any node in a diskless client has a return address of the server, so replies to normal mail will be delivered correctly (even if the reply is automatically generated by a mailer). In any case, you can make the message bounce right away instead of three days later quite easily. I believe the 6.5 versio of HP-UX has a reference in the sendmail.cf to a file called /etc/hosts.smtp (I'm not positive). If it does, you can set that file up to list the machines you talk to with SMTP mail (not including any diskless nodes). This would cause the message to have bounced immediately as it would recognize that the message was destined for a node that didn't accept mail. With HP-UX 7.0, there will be a much better solution. You would be able to set up your domain server to list MX records for the diskless node to receive all mail for diskless clients, in which case the mail would be sent to the rootserver (setting up the MX records may not be the easiest thing to do if you have lots of diskless nodes to worry about). Alternately, it is possible to run sendmail on all nodes in a cluster. HP doesn't recommend this because it takes extra work to set it up, but if the lp mail problem bothers you enough, it is an option. For sendmail to run correctly on each node, the sendmail.cf would have to be made to understand all nodenames in the cluster to be equivalent (local). Once that were done (this requires modifying ruleset 0, as well as adding to the definition of the class w, if I remember correctly), a sendmail daemon could be run on each node with only the rootserver processing the queue (-q30m). This is pretty messy unless you're a real sendmail jock. I believe the standard sendmail.cf shipped with 7.0 has the necessary support for this. The other alternative (also not recommended by HP, I believe) is to make /usr/spool/mqueue a CDF and run completely independent sendmail's on each node (which means mail from one diskless node to another would be sent via SMTP instead of simply written to the filesystem). John Diamant Software Engineering Systems Division Hewlett-Packard Co. Internet: diamant@hpfclp.sde.hp.com Fort Collins, CO UUCP: {hplabs,hpfcla}!hpfclp!diamant Disclaimer: The opinions expressed above are mine alone and may or may not represent the views of Hewlett-Packard Company.