1 #!/usr/local/bin/perl
2 # author: Al Tobey <albert.tobey@priority-health.com>
3 # what: monitor a process using the host-resources mib
4 # license: GPL - http://www.fsf.org/licenses/gpl.txt
5 #
6 # Todo:
7 # * implement memory and cpu utilization checks
8 # * maybe cache pids in DBM files if snmp agents get overworked
9 ###############################################################################
10 # to get a list of processes over snmp try this command:
11 # snmptable -v2c -c public hostname hrSWRunTable
12 # for just a list of valid arguments for the '-e' option:
13 # snmpwalk -v2c -c public hostname hrSWRunName |perl -pe 's:.*/::'
14 ###############################################################################
16 use strict;
17 require 5.6.0;
18 use lib qw( /opt/nagios/libexec /usr/local/libexec );
19 use utils qw(%ERRORS $TIMEOUT &print_revision &support &usage);
20 use SNMP 5.0;
21 use Getopt::Long;
22 use vars qw( $exit $opt_version $opt_timeout $opt_help $opt_command $opt_host $opt_community $opt_verbose $opt_warning $opt_critical $opt_memory $opt_cpu $opt_port $opt_regex $opt_stats %processes $snmp_session $PROGNAME $TIMEOUT );
24 $PROGNAME = "snmp_process_monitor.pl";
25 $opt_verbose = undef;
26 $opt_host = undef;
27 $opt_community = 'public';
28 $opt_command = undef;
29 $opt_warning = [ 1, -1 ];
30 $opt_critical = [ 1, -1 ];
31 $opt_memory = undef;
32 $opt_cpu = undef;
33 $opt_port = 161;
34 %processes = ();
35 $exit = 'OK';
37 sub process_options {
38 my( $opt_crit, $opt_warn ) = ();
39 Getopt::Long::Configure( 'bundling' );
40 GetOptions(
41 'V' => \$opt_version, 'version' => \$opt_version,
42 'v' => \$opt_verbose, 'verbose' => \$opt_verbose,
43 'h' => \$opt_help, 'help' => \$opt_help,
44 's' => \$opt_stats, 'statistics' => \$opt_stats,
45 'H:s' => \$opt_host, 'hostname:s' => \$opt_host,
46 'p:i' => \$opt_port, 'port:i' => \$opt_port,
47 'C:s' => \$opt_community, 'community:s' => \$opt_community,
48 'c:s' => \$opt_crit, 'critical:s' => \$opt_crit,
49 'w:s' => \$opt_warn, 'warning:s' => \$opt_warn,
50 't:i' => \$TIMEOUT, 'timeout:i' => \$TIMEOUT,
51 'e:s' => \$opt_command, 'command:s' => \$opt_command,
52 'r:s' => \$opt_regex, 'regex:s' => \$opt_regex,
53 'cpu:i' => \$opt_cpu, 'memory:i' => \$opt_memory,
54 );
55 if ( defined($opt_version) ) { local_print_revision(); }
56 if ( defined($opt_verbose) ) { $SNMP::debugging = 1; }
57 if ( !defined($opt_host) || defined($opt_help) || (!defined($opt_command) && !defined($opt_regex)) ) {
58 print_help();
59 exit $ERRORS{UNKNOWN};
60 }
62 if ( defined($opt_crit) ) {
63 if ( $opt_crit =~ /,/ ) {
64 $opt_critical = [ split(',', $opt_crit) ];
65 }
66 else {
67 $opt_critical = [ $opt_crit, -1 ];
68 }
69 }
70 if ( defined($opt_warn) ) {
71 if ( $opt_warn =~ /,/ ) {
72 $opt_warning = [ split(',', $opt_warn) ];
73 }
74 else {
75 $opt_warning = [ $opt_crit, -1 ];
76 }
77 }
78 }
80 sub local_print_revision {
81 print_revision( $PROGNAME, '$Revision$ ' )
82 }
84 sub print_usage {
85 print "Usage: $PROGNAME -H <host> -C <snmp_community> -e <command> [-w <low>,<high>] [-c <low>,<high>] [-t <timeout>]\n";
86 }
88 sub print_help {
89 local_print_revision();
90 print "Copyright (c) 2002 Al Tobey <albert.tobey\@priority-health.com>\n\n",
91 "SNMP Process Monitor plugin for Nagios\n\n";
92 print_usage();
93 print <<EOT;
94 -v, --verbose
95 print extra debugging information
96 -h, --help
97 print this help message
98 -H, --hostname=HOST
99 name or IP address of host to check
100 -C, --community=COMMUNITY NAME
101 community name for the host's SNMP agent
102 -e, --command=COMMAND NAME (ps -e style)
103 what command should be monitored?
104 -r, --regex=Perl RE
105 use a perl regular expression to find your process
106 -w, --warning=INTEGER[,INTEGER]
107 minimum and maximum number of processes before a warning is issued (Default 1,-1)
108 -c, --critical=INTEGER[,INTEGER]
109 minimum and maximum number of processes before a critical is issued (Default 1,-1)
110 EOT
111 }
113 sub verbose (@) {
114 return if ( !defined($opt_verbose) );
115 print @_;
116 }
118 sub check_for_errors {
119 if ( $snmp_session->{ErrorNum} ) {
120 print "UNKNOWN - error retrieving SNMP data: $snmp_session->{ErrorStr}\n";
121 exit $ERRORS{UNKNOWN};
122 }
123 }
125 # =========================================================================== #
126 # =====> MAIN
127 # =========================================================================== #
128 process_options();
130 alarm( $TIMEOUT ); # make sure we don't hang Nagios
132 $snmp_session = new SNMP::Session(
133 DestHost => $opt_host,
134 Community => $opt_community,
135 RemotePort => $opt_port,
136 Version => '2c'
137 );
139 my $process_count = SNMP::Varbind->new( ['hrSystemProcesses', 0] );
140 $snmp_session->get( $process_count );
141 check_for_errors();
143 # retrieve the data from the remote host
144 my( $names, $index ) = $snmp_session->bulkwalk( 0, $process_count->val, [['hrSWRunName'], ['hrSWRunIndex']] );
145 check_for_errors();
147 alarm( 0 ); # all done with the network connection
149 my %namecount = ();
150 foreach my $row ( @$names ) {
151 $processes{$row->iid}->{name} = $row->val;
152 $processes{$row->iid}->{name} =~ s#.*/##; # strip path
154 if ( defined($opt_regex) ||
155 ($row->val =~ /(perl|\/usr\/bin\/sh|\/bin\/bash|\/bin\/sh)$/
156 && $opt_command !~ /(perl|\/usr\/bin\/sh|\/bin\/bash|\/bin\/sh)$/) ) {
158 # fetch the runtime parameters of the process
159 my $parm_var = SNMP::Varbind->new( ['hrSWRunParameters', $row->iid] );
160 $snmp_session->get( $parm_var );
161 check_for_errors();
163 # only strip if we're looking for a specific command
164 if ( defined($opt_command) ) {
165 verbose "process ",$row->iid," uses $1 as an interpreter - getting parameters\n";
166 $processes{$row->iid}->{name} = $parm_var->val;
167 # strip path name off the front
168 $processes{$row->iid}->{name} =~ s#.*/##;
169 # strip everything from the first space to the end
170 $processes{$row->iid}->{name} =~ s/\s+.*$//;
171 }
172 else {
173 # get the longer full-path style listing
174 my $path_var = SNMP::Varbind->new( ['hrSWRunPath', $row->iid] );
175 $snmp_session->get( $path_var );
176 check_for_errors();
178 # use the full 'ps -efl' style listing for regular expression matching
179 $processes{$row->iid}->{name} = $path_var->val.' '.$parm_var->val;
180 }
181 }
182 }
183 foreach my $row ( @$index ) {
184 $processes{$row->iid}->{pid} = $row->val;
185 }
187 my @pids = ();
188 my @matches = ();
189 foreach my $key ( keys(%processes) ) {
190 if ( defined($opt_command) && $processes{$key}->{name} eq $opt_command ) {
191 push( @matches, $processes{$key} );
192 push( @pids, $processes{$key}->{pid} );
193 verbose "process '$processes{$key}->{name}' has pid ",
194 "$processes{$key}->{pid} and index $key\n";
195 }
196 elsif ( defined($opt_regex) && $processes{$key}->{name} =~ /$opt_regex/o ) {
197 push( @matches, $processes{$key} );
198 push( @pids, $processes{$key}->{pid} );
199 verbose "process '$processes{$key}->{name}' has pid ",
200 "$processes{$key}->{pid} and index $key\n";
201 }
202 }
203 my $count = @matches;
205 # warning, critical
206 if ( ($opt_warning->[0] > 0 && $opt_warning->[0] > $count)
207 || ($opt_warning->[1] > 0 && $opt_warning->[1] <= $count) ) {
208 $exit = 'WARNING';
209 }
210 if ( ($opt_critical->[0] > 0 && $opt_critical->[0] > $count)
211 || ($opt_critical->[1] > 0 && $opt_critical->[1] <= $count) ) {
212 $exit = 'CRITICAL';
213 }
215 print "$exit - $count processes with pid(s) ",join(',',@pids);
217 # print the number of processes if statistics are requested
218 if ( defined($opt_stats) ) {
219 print "|count:$count\n";
220 }
221 else {
222 print "\n";
223 }
225 exit $ERRORS{$exit};