[Mon-commit] mon/mon.d traceroute.monitor,1.2,1.2.2.1

SourceForge Headquarters 1320 Columbia Street Suite 310 San Diego, CA 92101 +1 (858) 422-6466

Update of /cvsroot/mon/mon/mon.d
In directory sc8-pr-cvs16.sourceforge.net:/tmp/cvs-serv32211/mon.d

Modified Files:
      Tag: mon-1-2-branch
	traceroute.monitor 
Log Message:
traceroute.monitor update

Index: traceroute.monitor
===================================================================
RCS file: /cvsroot/mon/mon/mon.d/traceroute.monitor,v
retrieving revision 1.2
retrieving revision 1.2.2.1
diff -C2 -d -r1.2 -r1.2.2.1
*** traceroute.monitor	15 Nov 2004 14:45:19 -0000	1.2
--- traceroute.monitor	3 Jun 2007 13:08:59 -0000	1.2.2.1
***************
*** 14,22 ****
  # Jon Meek
  # Lawrenceville, NJ
! # me...@ie...
  #
  # $Id$
  #
! #    Copyright (C) 2001, Jon Meek
  #
  #    This program is free software; you can redistribute it and/or modify
--- 14,22 ----
  # Jon Meek
  # Lawrenceville, NJ
! # meekj at ieee.org
  #
  # $Id$
  #
! #    Copyright (C) 2001-2005, Jon Meek
  #
  #    This program is free software; you can redistribute it and/or modify
***************
*** 46,50 ****
  =head1 SYNOPSIS

! B<traceroute.monitor -d -t 20 -c /path/to/traceroute.cf -l /usr/local/mon/logs/routes_YYYYMM.log>

  The logfile template is usually specified in the configuration file.
--- 46,50 ----
  =head1 SYNOPSIS

! B<traceroute.monitor -d -t 20 -c -u /path/to/traceroute.cf -l /usr/local/mon/logs/routes_YYYYMM.log>

  The logfile template is usually specified in the configuration file.
***************
*** 54,67 ****
  =over 5

! =item B<-d> Debug/Test

! =item B<-c config.cfg> Configuration file for this monitor, see example below

! =item B<-t timeout> Timeout for traceroute to run in seconds default is 20s

  =item B<-l log_file_template> /path/to/logs/internet_web_YYYYMM.log
  Current year & month are substituted for YYYYMM, that is the only
  possible template at this time.

  =back

--- 54,87 ----
  =over 5

! =item B<-d>

! Debug/Test

! =item B<-c config.cfg>
! 
! Configuration file for this monitor, see example below
! 
! =item B<-t timeout>
! 
! Timeout for traceroute to run in seconds default is 20s

  =item B<-l log_file_template> /path/to/logs/internet_web_YYYYMM.log
+ 
  Current year & month are substituted for YYYYMM, that is the only
  possible template at this time.

+ =item B<-u>
+ 
+ Fail On Unexpected Hop Only - An alert will be triggered if any of the
+ IP addresses specified using the UnexpectedHop item in the
+ configuration file are observed. Any other route change will not
+ create an alert. This option is useful for alerting when a backup
+ circuit becomes part of a default route. This option causes
+ traceroute.monitor to act more like a traditional mon monitor because
+ an alert will be triggered on every monitor run if an UnexpectedHop
+ appears in the path. The use of mon's alertafter N directive can be
+ used to filter out short term glitches that might cause a route to
+ change for just one monitor cycle.
+ 
  =back

***************
*** 81,85 ****
  =head1 CONFIGURATION FILE EXAMPLE

!  # tracreoute.monitor Config File
   RouteLogFile /usr/local/mon/logs/routes_YYYYMM.log
   RouterList /usr/local/mon/rt.list
--- 101,105 ----
  =head1 CONFIGURATION FILE EXAMPLE

!  # traceroute.monitor Config File
   RouteLogFile /usr/local/mon/logs/routes_YYYYMM.log
   RouterList /usr/local/mon/rt.list
***************
*** 91,94 ****
--- 111,120 ----
   StopAt 172.30.124.17 A firewall
   StopAt 172.31.124.17 Another firewall
+  UnexpectedHop 10.22.249.2 S2S VPN Tunnel
+  AlertMessage
+  For more details see:
+  http://$HOSTNAME/cgi-bin/network/traceroute.anal
+  EndAlertMessage
+ 

  Lines with '#' in the first column are ignored.
***************
*** 100,107 ****

  RouterList - Optional IP address to router name translation in
! /etc/hosts format (IP_address router_bame). Supplying this list will
  provide considerably more meaningful alarm messages, especially if the
! router names contain geographical information. Without this list the
! extended alarm is just a list of interface IP addresses.

  Traceroute - Overrides the default of /usr/sbin/traceroute
--- 126,134 ----

  RouterList - Optional IP address to router name translation in
! /etc/hosts format (IP_address router_name). Supplying this list will
  provide considerably more meaningful alarm messages, especially if the
! router names contain geographical information. Without this list, or
! DNS records, the extended alarm is just a list of interface IP
! addresses. DNS results take precedence over this list.

  Traceroute - Overrides the default of /usr/sbin/traceroute
***************
*** 127,130 ****
--- 154,169 ----
  directive. There can be multiple StopAt lines.

+ UnexpectedHop - See the description of the -u option above. The
+ presence of an UnexpectedHop in a path will trigger an alarm
+ regardless of whether the -u option is set, or not.
+ 
+ AlertMessage - A message that is appended to the alert details starts
+ on the next line. $HOSTNAME, if present in the message, is converted
+ to the name of the machine on which the monitor is running.
+ 
+ EndAlertMessage - The AlertMessage can be terminated by this
+ directive, but it is not required if the message is at the end of the
+ configuration file.
+ 
  =head1 BUGS

***************
*** 141,149 ****
  =cut

! use Getopt::Std;
! use POSIX qw(:signal_h WNOHANG);
! use POSIX qw(strftime);

! getopts ("vdt:l:c:");

  # -l file  Log file name with optional YYYYMM part that will be transformed to current month
--- 180,199 ----
  =cut

! use Getopt::Long;
! use Sys::Hostname;
! use POSIX qw(:signal_h WNOHANG strftime);
! use Socket;

! #getopts ("vdt:l:c:");
! 
! GetOptions ('v' => \$opt_v,
! 	    'd' => \$opt_d,
! 	    'debug' => \$opt_d,
! 	    't=i' => \$opt_t,
! 	    'l=s' => \$opt_l,
! 	    'c=s' => \$opt_c,
! 	    'u' => \$FailOnUnexpectedHopOnly,
! 	    'unexpectedhoponly' => \$FailOnUnexpectedHopOnly,
! 	   );

  # -l file  Log file name with optional YYYYMM part that will be transformed to current month
***************
*** 152,156 ****

  # Usual Linux config
! $TRACEROUTE = '/usr/sbin/traceroute';
  #$STATE_DIR = '/usr/local/mon/state.d';

--- 202,206 ----

  # Usual Linux config
! $TRACEROUTE = '/bin/traceroute';
  #$STATE_DIR = '/usr/local/mon/state.d';

***************
*** 214,217 ****
--- 264,282 ----
        }

+       if ($in =~ /^UnexpectedHop\s+/i) {
+ 	($tag, $ip, $name) = split(' ', $in, 3);
+ 	print "UnexpectedHop $ip, $name<br>\n" if $opt_d;
+ 	$UnexpectedHop{$ip}++;
+ 	$UnexpectedHopName{$ip} = $name;
+       }
+ 
+       if ($in =~ /^AlertMessage/i) { # Extra text to add to an alert starts on next line
+ 	while ($in = <C>) {
+ 	  last if ($in =~ /^EndAlertMessage/i);
+ 	  $AlertMessage .= $in;
+ 	}
+ 	$MonitorHostname = hostname;
+ 	$AlertMessage =~ s/\$HOSTNAME/$MonitorHostname/;
+       }
      }
    } else {
***************
*** 242,246 ****
  if ($TestOnly) {
    foreach $h (@Hosts) {
!     print "Host: $h\n";

      if (defined $EquivIP{$h}) {
--- 307,312 ----
  if ($TestOnly) {
    foreach $h (@Hosts) {
!     $name = &HopName($h);
!     print "Host: $h   $name\n";

      if (defined $EquivIP{$h}) {
***************
*** 319,326 ****
  	$ThisHopIP = $1;
  	push(@HopList, $ThisHopIP); # Build route hop list

  	if ($opt_v) {
  	  chomp $in;
! 	  print "$in $RouterByIP{$ThisHopIP}\n";
  	}

--- 385,396 ----
  	$ThisHopIP = $1;
  	push(@HopList, $ThisHopIP); # Build route hop list
+ 	if ($UnexpectedHop{$ThisHopIP}) {
+ 	  $UnexpectedHopSeen{$TargetHost}++;
+ 	}

  	if ($opt_v) {
  	  chomp $in;
! 	  $name = &HopName($h);
! 	  print "$in  $name  $RouterByIP{$ThisHopIP}\n";
  	}

***************
*** 396,400 ****
      }

!     if (&RouteChanged($current_route, $prev_route)) { # Route changed, alarm and record
        if ($RunningUnderMon) { # Write results
  	open(S, ">$state_file") || warn "Can't open $state_file for writing\n";
--- 466,475 ----
      }

!     $ThisTargetFailed = 0;
!     $ThisTargetFailed++ if $UnexpectedHopSeen{$k};
! 
!     if (&RouteChanged($current_route, $prev_route)) { # Route changed, record, alarm if not using only unexpected hops
!       $ThisTargetFailed++ unless $FailOnUnexpectedHopOnly;
!       $LogString{$k} = $ResultString{$k}; # Use separate string for logging route changes so that we don't always log unexpected hops
        if ($RunningUnderMon) { # Write results
  	open(S, ">$state_file") || warn "Can't open $state_file for writing\n";
***************
*** 402,405 ****
--- 477,483 ----
  	close S;
        }
+     }
+ 
+     if ($ThisTargetFailed) {
        push (@Failures, $k);
        print "  Alarm\n" if $opt_d;
***************
*** 413,416 ****
--- 491,495 ----
        close S;
      }
+ #    $LogString{$k} = $ResultString{$k}; # Always log the first instance
      push (@Failures, $k); # Call it a failure so it will be logged and notification will be sent
      print "  New route added to check: $k\n" if $opt_d;
***************
*** 419,425 ****
  }

! # Write results to logfile, if -l
! 
! #if ($RunningUnderMon && $LogFile) {
  if ($LogFile) {

--- 498,504 ----
  }

! #
! # Write results to logfile
! #
  if ($LogFile) {

***************
*** 445,452 ****
      }

!     if (($NewLogFile == 0) && (@Failures > 0)) { # Just record changes
!       foreach $host (sort @Failures) {
! 	print LOG "$ResultString{$host}\n";
        }
      }

--- 524,532 ----
      }

!     if (($NewLogFile == 0) && (@Failures > 0)) { # Just record changes, not repeated unexpected hops
!       foreach $host (sort keys %LogString) {
! 	print LOG "$LogString{$host}\n";
        }
+ 
      }

***************
*** 459,463 ****
  }

! 
  if (defined $RouterListFile) { # Read the router names if we have a failure
    open(F, $RouterListFile);
--- 539,544 ----
  }

! # Note that we might have already read this file if --v was specified
! #
  if (defined $RouterListFile) { # Read the router names if we have a failure
    open(F, $RouterListFile);
***************
*** 479,487 ****
    (@hop_ips) = split(/\-/, $rest);
    foreach $hop_ip (@hop_ips) {
!     printf " %-15s %s\n", $hop_ip, $RouterByIP{$hop_ip};
    }
    print "\n";
  }

  exit 1;

--- 560,575 ----
    (@hop_ips) = split(/\-/, $rest);
    foreach $hop_ip (@hop_ips) {
!     $name = &HopName($hop_ip);
!     printf " %-15s %s", $hop_ip, $name;
!     print "  Unexpected Hop: $UnexpectedHopName{$hop_ip}" if ($UnexpectedHop{$hop_ip});
!     print "\n";
    }
    print "\n";
  }

+ if (defined $AlertMessage) {
+   print "\n$AlertMessage\n";
+ }
+ 
  exit 1;

***************
*** 515,516 ****
--- 603,634 ----
  }

+ sub HopName {
+   my ($h_ip) = @_;
+   my ($lookupname, $ha);
+ #  print "HopName: $h_ip<br>\n"  if $opt_d;
+   if ($h_ip =~ /^\*/) {
+     $name = 'Not reached';
+     return $name;
+   }
+ 
+   if (exists $HopNameCache{$h_ip}) { # Already have the name
+     $name = $HopNameCache{$h_ip};
+     return $name;
+   }
+   $name = '';
+   $ha = inet_aton($h_ip);
+   $lookupname = gethostbyaddr($ha, &AF_INET);
+ 
+   if (length($lookupname) > 0) { # Attempt lookup of address via "normal" system methods (DNS, hosts, etc)
+     $name = $lookupname;
+   } elsif ($UseRouterList) {
+     $name = $RouterByIP{$h_ip} if (exists $RouterByIP{$h_ip});
+   }
+ 
+   # Have an auxilliary name?   such as "London Internet Router", prepend it
+ #  $name = qq{$RouteHopName{$h_ip}  $name} if (exists $RouteHopName{$h_ip});
+ 
+   $HopNameCache{$h_ip} = $name;
+ #  print "$h_ip -- $name<br>\n" if $opt_d;
+   return $name;
+ }