* improve the test suite to run a slow response handler (sleep 20) and
  check that we actually spot and kill that process

* if it's polling the server every so often, it might as well generate
  a periodic report of what it finds.  How many processes, how long
  each has been running, how much memory, etc.

* add a regex for URLs not to be watched, since these can be lengthy
upload/download operations. Note that you have only 64 chars of the URI
provided by Apache::Scoreboard

* Eric has proposed to add a support for virtual hosts, based on
vhostrec->server_name

* From: Brad Morris <brad.morris@turner.com>
Subject: Apache-Watchdog-RunAway suggestion/help

I like the idea of watchdog very much.  We have various script that
sometimes run away, and it is pretty hard to debug properly because we
are dealing with unreliable feed data.

However, a weak point of your current implementation seems to be the
kill -9.

The problem for us is that although we want to kill the runaway
processes, we also want a record of what was killed doing what so we
can go back later and fix.

Have you thought about using the advise from the Debugging mod_perl -
Hanging Processes: Detection and Diagnostics - Determination of the
reason - Using the Perl Trace section of your excellent modperl guide?

Installing a signal handler in the startup.pl and killing that signal
inside the Apache-Watchdog-RunAway gives the added bonus of a logged
stack trace every time a dead process is killed.

Also, it would be nice to log the request that caused this problem.
This can be gotten from image->request;.  This is pretty simple.

Anyway, I have implemented the kill -USR2 in my startup.pl and things
seem to work (it logs the carp nicely to the error log).  Then I need
to change your kill to do a kill -USR2.