[Xymon] SMTP test needs much more capability

[Xymon User in Richmond]

On Fri, May 20, 2011 12:31, Shawn Heisey wrote:
> I have been chasing down the following.  The first is the smtp status
> reported on the xymon status page (which is green), debian package
> version 4.3.0-0.beta2. The second is a corresponding message from
> mail.log on the Postfix (2.7.1 debian package) server.
>
> 220 nexus2.example.com ESMTP Postfix (Debian/GNU)
> 503 5.5.1 Error: send HELO/EHLO first
> 221 2.0.0 Bye
>
> May 20 10:08:15 nexus2 postfix/smtpd[8689]: improper command pipelining
> after MAIL from mcp.example.com[10.2.1.39]
>
> After a lengthy exchange with Wietse Venema on the postfix-users mailing
> list, I got to the bottom of it, and there's no problem with postfix.
> The 503 error is because Xymon does not send a HELO.  I've determined
> that if I send the following, I can avoid the 503 error:
>
> helo xymon
> mail from: xymon at example.com
> quit
>
> If I modify the smtp test to send these strings, the "improper
> pipelining" message is still logged.  This is because Xymon is breaking
> SMTP rules by not waiting for the response to each command before
> sending the next one.  There are no longer any errors logged during the
> actual SMTP session, because I have added the xymon server to a list
> that allows it to get away with improper pipelining, and to another one
> that lets it send invalid HELO strings and FROM addresses.
>
> This would be a problem if I had to monitor a remote postfix server
> whose configuration I could not control, though.  The only reason that
> the test passes is that the string it looks for (220) is returned in the
> initial banner, before any SMTP commands are sent.  It is not actually a
> real test.  Any number of postfix configuration errors would keep the
> server from working, yet still return the banner on initial connect.
>
> Does anyone know how to make Xymon smart enough to wait for a defined
> response on each command before sending the next, and to put custom
> commands into a config file, rather than built in at compile time via
> protocols.cfg?  Ideally it would be able to understand that any 4nn or
> 5nn code is a problem as well.  I will take a look at the code, but my
> programming skills are rusty.
>

There have been conversations on the list about this in the past, and I
think Henrik said that getting rid of the issue entirely would require
recoding the network tester to be able to use a series of expects.  I
think that the cleanest you can get currently is to remove the send
section from the smtp test, letting xymon be happy with the initial 220,
but the server will still log a NOQUEUE because no commands were issued by
the client.