Added ability to send mail when something goes wrong

This is rate-limited to once per hour, and includes the last
portion of the log file.

diff --git a/scripts/radwatch.in b/scripts/radwatch.in
index a3bd101..90c206b 100644 (file)
--- a/scripts/radwatch.in
+++ b/scripts/radwatch.in
@@ -28,20 +28,41 @@
 #  reading the file doc/supervise-radiusd.txt
 
 #
+#  This simplifies the script, and avoids most issues with (say)
+#  Debian re-naming "radiusd" to "freeradius".
+#
+name=radiusd
+
 prefix=@prefix@
 exec_prefix=@exec_prefix@
 sbindir=@sbindir@
 localstatedir=@localstatedir@
 logdir=@logdir@
-rundir=${localstatedir}/run/radiusd
+rundir=${localstatedir}/run/${name}
 sysconfdir=@sysconfdir@
-pid_file=${rundir}/radiusd.pid
-log_file=${logdir}/radiusd_safe.log
+pid_file=${rundir}/${name}.pid
+log_file=${logdir}/${name}_safe.log
 
-RADIUSD=$sbindir/radiusd
+RADIUSD=$sbindir/${name}
 RADDBDIR=${sysconfdir}/raddb
 
 #
+#  If you want to send email, define this field to be an email address.
+#  This part of the functionality hasn't been well tested, so please
+#  test it before putting it into production.
+#
+#  It also presumes that you have a functioning mail system on
+#  the maching running RADIUS.  You will need to check that the
+#  "mail" command exists, and sends mail to the address below, e.g.:
+#
+#     echo test | mail -s "Testing" $MAILTO
+#
+#  If you receive the message, then enable MAILTO.  Otherwise, fix
+#  your mail system so that it delivers mail.
+#
+MAILTO=
+
+#
 #  Allow "radiusd_safe -X" for testing the radiusd_safe functionality.
 #
 ARGS="$@"
@@ -49,12 +70,6 @@ ARGS="$@"
 test -f $RADIUSD || exit 0
 test -f $RADDBDIR/radiusd.conf || exit 0
 
-#
-#  This simplifies the script, and avoids most issues with (say)
-#  Debian re-naming "radiusd" to "freeradius".
-#
-name=`basename $RADIUSD`
-
 ulimit -c unlimited
 
 #
@@ -89,6 +104,8 @@ then
 fi
 
 started=0
+restarts=0
+last_email=0
 
 #
 #  Loop forever, or until we're told to exit via a signal.
@@ -110,6 +127,43 @@ do
     #
     if test "$started" != "0"
     then
+       #  Send mail when the server starts
+       if test "$MAILTO" != ""
+       then
+           now=`date +"%s"`
+           restarts=`expr $restarts + 1`
+
+           # send email the first time it restarts
+           if test "$last_email" = "0"
+           then
+                   cat |  mail -s "ERROR - $name died, restarting.." $MAILTO <<EOF
+$name has restarted unexpectedly at $now
+
+See $log_file for details.  Last 20 lines are:
+
+----------------------------------------------------------------------
+`tail -n 20 $log_file`
+EOF
+               $last_email="$now"
+               restarts=0
+           else
+               #  Send email only once every hour.
+               if test "$now" -gt `expr $last_email + 3600`
+               then
+                   cat |  mail -s "ERROR - $name died, restarting.." $MAILTO <<EOF
+$name has restarted $restarts times since last email at $last_email
+
+See $log_file for details.  Last 100 lines are:
+
+----------------------------------------------------------------------
+`tail -n 100 $log_file`
+EOF
+                   $last_email="$now"
+                   restarts=0
+               fi
+           fi
+       fi
+
        if test "$started" = `date +"%s"`
        then
            sleep 1