On 6/2/13 9:12 AM, Wietse Venema wrote:
> For example, periodically send email to mailboxname+timest...@example.com,
> and parse the "to=<mailboxname+timest...@example.com>" and
> "status=delivered" out of the logfile record stream.
Or, even better, try to retrieve those messages from the actual mailbox
using POP3. That way you've simulated everything a user does, end to end.
This is not hard to do. Using the Perl modules Net::SMTP and Net::POP3,
it's easy to write a Nagios/Icinga check that operates thusly:
1. Using POP3, check for a message with a timestamp in the
subject that indicates that it was sent in the last X
minutes (where X is the maximum delay you're willing to
tolerate). If not found, the result will be CRITICAL.
2. Using POP3, delete every message in the mailbox.
3. Using SMTP, send a new message to the mailbox with a
timestamp in the subject.
That's all it takes. Each time the check runs, (1) should find the
message sent by (3) of the previous check. The "timestamp in the
subject" can be as simple as the epoch seconds.
As I wrote on this list 11 years ago(!), "In general, the thing I've
learned about monitoring is that when possible, check the system by
using it, rather [than] looking for changes in side effects (such as
logs, or number of processes running, etc.). That's not to say that the
other things can't give useful information, too, but if you can check
the system by using it, you don't have to worry so much about whether
you've added a regexp for every possible log entry and so forth."
--
Robert L Mathews, Tiger Technologies, http://www.tigertech.net/
