summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--contrib/check_snmp_disk_monitor.pl185
-rw-r--r--contrib/check_snmp_process_monitor.pl227
2 files changed, 412 insertions, 0 deletions
diff --git a/contrib/check_snmp_disk_monitor.pl b/contrib/check_snmp_disk_monitor.pl
new file mode 100644
index 0000000..7a70f71
--- /dev/null
+++ b/contrib/check_snmp_disk_monitor.pl
@@ -0,0 +1,185 @@
1#!/usr/local/bin/perl
2# author: Al Tobey <albert.tobey@priority-health.com>
3# what: monitor diskspace using the host-resources mib
4# license: GPL - http://www.fsf.org/licenses/gpl.txt
5#
6# Todo:
7
8use strict;
9require 5.6.0;
10use lib qw( /opt/nagios/libexec );
11use utils qw(%ERRORS $TIMEOUT &print_revision &support &usage);
12use SNMP 5.0;
13use Getopt::Long;
14use vars qw( $exit $message $opt_version $opt_timeout $opt_help $opt_command $opt_host $opt_community $opt_verbose $opt_warning $opt_critical $opt_port $opt_mountpoint $opt_stats $snmp_session $PROGNAME $TIMEOUT %mounts );
15
16$PROGNAME = "snmp_disk_monitor.pl";
17$opt_verbose = undef;
18$opt_host = undef;
19$opt_community = 'public';
20$opt_command = undef;
21$opt_warning = 99;
22$opt_critical = 100;
23$opt_port = 161;
24$opt_stats = undef;
25$message = undef;
26$exit = 'OK';
27%mounts = ();
28
29sub process_options {
30 my( $opt_crit, $opt_warn ) = ();
31 Getopt::Long::Configure( 'bundling' );
32 GetOptions(
33 'V' => \$opt_version, 'version' => \$opt_version,
34 'v' => \$opt_verbose, 'verbose' => \$opt_verbose,
35 'h' => \$opt_help, 'help' => \$opt_help,
36 's' => \$opt_stats, 'statistics' => \$opt_stats,
37 'H:s' => \$opt_host, 'hostname:s' => \$opt_host,
38 'p:i' => \$opt_port, 'port:i' => \$opt_port,
39 'C:s' => \$opt_community, 'community:s' => \$opt_community,
40 'c:i' => \$opt_crit, 'critical:i' => \$opt_crit,
41 'w:i' => \$opt_warn, 'warning:i' => \$opt_warn,
42 't:i' => \$TIMEOUT, 'timeout:i' => \$TIMEOUT,
43 'm:s' => \$opt_mountpoint, 'mountpoint:s'=> \$opt_mountpoint
44 );
45 if ( defined($opt_version) ) { local_print_revision(); }
46 if ( defined($opt_verbose) ) { $SNMP::debugging = 1; }
47 if ( !defined($opt_host) || defined($opt_help) || !defined($opt_mountpoint) ) {
48 print_help();
49 exit $ERRORS{UNKNOWN};
50 }
51 $opt_mountpoint = [ split(/,/, $opt_mountpoint) ];
52}
53
54sub local_print_revision {
55 print_revision( $PROGNAME, '$Revision$ ' )
56}
57
58sub print_usage {
59 print "Usage: $PROGNAME -H <host> -C <snmp_community> [-s] [-w <low>,<high>] [-c <low>,<high>] [-t <timeout>] -m <mountpoint>\n";
60}
61
62sub print_help {
63 local_print_revision();
64 print "Copyright (c) 2002 Al Tobey <albert.tobey\@priority-health.com>\n\n",
65 "SNMP Disk Monitor plugin for Nagios\n\n";
66 print_usage();
67 print <<EOT;
68-v, --verbose
69 print extra debugging information
70-h, --help
71 print this help message
72-H, --hostname=HOST
73 name or IP address of host to check
74-C, --community=COMMUNITY NAME
75 community name for the host's SNMP agent
76-m, --mountpoint=MOUNTPOINT
77 a mountpoint, or a comma delimited list of mountpoints
78-w, --warning=INTEGER
79 percent of disk used to generate WARNING state (Default: 99)
80-c, --critical=INTEGER
81 percent of disk used to generate CRITICAL state (Default: 100)
82-s, --statistics
83 output statistics in Nagios format
84EOT
85}
86
87sub verbose (@) {
88 return if ( !defined($opt_verbose) );
89 print @_;
90}
91
92sub check_for_errors {
93 if ( $snmp_session->{ErrorNum} ) {
94 print "UNKNOWN - error retrieving SNMP data: $snmp_session->{ErrorStr}\n";
95 exit $ERRORS{UNKNOWN};
96 }
97}
98
99# =========================================================================== #
100# =====> MAIN
101# =========================================================================== #
102process_options();
103
104alarm( $TIMEOUT ); # make sure we don't hang Nagios
105
106$snmp_session = new SNMP::Session(
107 DestHost => $opt_host,
108 Community => $opt_community,
109 RemotePort => $opt_port,
110 Version => '2c'
111);
112
113# retrieve the data from the remote host
114my( $mps, $alloc, $size, $used ) = $snmp_session->bulkwalk( 0, 4, [['hrStorageDescr'],['hrStorageAllocationUnits'],['hrStorageSize'],['hrStorageUsed']] );
115check_for_errors();
116
117alarm( 0 ); # all done with the network connection
118
119# move all the data into a nice, convenient hash for processing
120foreach my $mp ( @$mps ) { $mounts{$mp->iid}->{mountpoint} = $mp->val; }
121foreach my $a ( @$alloc ) { $mounts{$a->iid}->{alloc_units} = $a->val; }
122foreach my $si ( @$size ) {
123 if ( exists($mounts{$si->iid}->{alloc_units}) ) {
124 $mounts{$si->iid}->{size} = $si->val * $mounts{$si->iid}->{alloc_units};
125 }
126 else {
127 $mounts{$si->iid}->{size} = $si->val;
128 }
129}
130foreach my $us ( @$used ) {
131 if ( exists($mounts{$us->iid}->{alloc_units}) ) {
132 $mounts{$us->iid}->{used} = $us->val * $mounts{$us->iid}->{alloc_units};
133 }
134 else {
135 $mounts{$us->iid}->{used} = $us->val;
136 }
137}
138
139# now find the mountpoint or mountpoints that were actually requested and push onto an array for output
140my @matches = ();
141foreach my $mp ( @$opt_mountpoint ) {
142 my $found = scalar(@matches); # count all matches
143 foreach my $key ( keys(%mounts) ) {
144 if ( $mounts{$key}->{mountpoint} eq $mp ) {
145
146 # find the percentate - eval to avoid divide by zero errors
147 eval { $mounts{$key}->{percent_used} = $mounts{$key}->{used} / $mounts{$key}->{size} };
148 $mounts{$key}->{percent_used} =~ s/^0\.([0-9]{1,2})([0-9]?).*/\1/; # truncate
149 if ( $2 >= 5 ) { $mounts{$key}->{percent_used}++ }; # round the number number
150
151 verbose "mountpoint $mp has ", $mounts{$key}->{percent_used}, "% used, ",
152 $mounts{$key}->{size}, " bytes and ",$mounts{$key}->{used}, " used\n";
153
154 push( @matches, $mounts{$key} );
155 }
156 }
157 if ( scalar(@matches) == $found ) {
158 print "UNKNOWN - could not locate mountpoint $mp on host\n";
159 exit $ERRORS{UNKNOWN};
160 }
161}
162
163# now run through and check the thresholds
164foreach my $mp ( @matches ) {
165 if ( $mp->{percent_used} >= $opt_warning ) {
166 $exit = 'WARNING';
167 if ( $mp->{percent_used} >= $opt_critical ) { $exit = 'CRITICAL'; }
168 }
169 $message .= $mp->{percent_used}.'% used on '.$mp->{mountpoint}.', ';
170}
171$message =~ s/,\s*$//;
172
173# append statistics if requested
174if ( defined($opt_stats) ) {
175 my @tmp = ();
176 foreach my $mp ( @matches ) {
177 push( @tmp, join(',',$mp->{mountpoint},$mp->{size},$mp->{used}) );
178 }
179 $message .= '|'.join( ':', @tmp );
180}
181
182print "Disk $exit - $message\n";
183exit $ERRORS{$exit};
184
185
diff --git a/contrib/check_snmp_process_monitor.pl b/contrib/check_snmp_process_monitor.pl
new file mode 100644
index 0000000..263255b
--- /dev/null
+++ b/contrib/check_snmp_process_monitor.pl
@@ -0,0 +1,227 @@
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###############################################################################
15
16use strict;
17require 5.6.0;
18use lib qw( /opt/nagios/libexec /usr/local/libexec );
19use utils qw(%ERRORS $TIMEOUT &print_revision &support &usage);
20use SNMP 5.0;
21use Getopt::Long;
22use 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 $TIMEOUT );
23
24$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';
36
37sub 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 $ERRORS{UNKNOWN};
60 }
61
62 if ( defined($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}
79
80sub local_print_revision {
81 print_revision( $PROGNAME, '$Revision$ ' )
82}
83
84sub print_usage {
85 print "Usage: $PROGNAME -H <host> -C <snmp_community> -e <command> [-w <low>,<high>] [-c <low>,<high>] [-t <timeout>]\n";
86}
87
88sub print_help {
89 local_print_revision();
90 print "Copyright (c) 2002 Al Tobey <albert.tobey\@priority-health.com>\n\n",
91 "SNMP Process Monitor plugin for Nagios\n\n";
92 print_usage();
93 print <<EOT;
94-v, --verbose
95 print extra debugging information
96-h, --help
97 print this help message
98-H, --hostname=HOST
99 name or IP address of host to check
100-C, --community=COMMUNITY NAME
101 community name for the host's SNMP agent
102-e, --command=COMMAND NAME (ps -e style)
103 what command should be monitored?
104-r, --regex=Perl RE
105 use a perl regular expression to find your process
106-w, --warning=INTEGER[,INTEGER]
107 minimum and maximum number of processes before a warning is issued (Default 1,-1)
108-c, --critical=INTEGER[,INTEGER]
109 minimum and maximum number of processes before a critical is issued (Default 1,-1)
110EOT
111}
112
113sub verbose (@) {
114 return if ( !defined($opt_verbose) );
115 print @_;
116}
117
118sub 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}
124
125# =========================================================================== #
126# =====> MAIN
127# =========================================================================== #
128process_options();
129
130alarm( $TIMEOUT ); # make sure we don't hang Nagios
131
132$snmp_session = new SNMP::Session(
133 DestHost => $opt_host,
134 Community => $opt_community,
135 RemotePort => $opt_port,
136 Version => '2c'
137);
138
139my $process_count = SNMP::Varbind->new( ['hrSystemProcesses', 0] );
140$snmp_session->get( $process_count );
141check_for_errors();
142
143# retrieve the data from the remote host
144my( $names, $index ) = $snmp_session->bulkwalk( 0, $process_count->val, [['hrSWRunName'], ['hrSWRunIndex']] );
145check_for_errors();
146
147alarm( 0 ); # all done with the network connection
148
149my %namecount = ();
150foreach my $row ( @$names ) {
151 $processes{$row->iid}->{name} = $row->val;
152 $processes{$row->iid}->{name} =~ s#.*/##; # strip path
153
154 if ( defined($opt_regex) ||
155 ($row->val =~ /(perl|\/usr\/bin\/sh|\/bin\/bash|\/bin\/sh)$/
156 && $opt_command !~ /(perl|\/usr\/bin\/sh|\/bin\/bash|\/bin\/sh)$/) ) {
157
158 # fetch the runtime parameters of the process
159 my $parm_var = SNMP::Varbind->new( ['hrSWRunParameters', $row->iid] );
160 $snmp_session->get( $parm_var );
161 check_for_errors();
162
163 # only strip if we're looking for a specific command
164 if ( defined($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();
177
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}
183foreach my $row ( @$index ) {
184 $processes{$row->iid}->{pid} = $row->val;
185}
186
187my @pids = ();
188my @matches = ();
189foreach 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}
203my $count = @matches;
204
205# warning, critical
206if ( ($opt_warning->[0] > 0 && $opt_warning->[0] > $count)
207 || ($opt_warning->[1] > 0 && $opt_warning->[1] <= $count) ) {
208 $exit = 'WARNING';
209}
210if ( ($opt_critical->[0] > 0 && $opt_critical->[0] > $count)
211 || ($opt_critical->[1] > 0 && $opt_critical->[1] <= $count) ) {
212 $exit = 'CRITICAL';
213}
214
215print "$exit - $count processes with pid(s) ",join(',',@pids);
216
217# print the number of processes if statistics are requested
218if ( defined($opt_stats) ) {
219 print "|count:$count\n";
220}
221else {
222 print "\n";
223}
224
225exit $ERRORS{$exit};
226
227