diff options
| -rw-r--r-- | contrib/check_snmp_disk_monitor.pl | 185 | ||||
| -rw-r--r-- | contrib/check_snmp_process_monitor.pl | 227 |
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 00000000..7a70f71a --- /dev/null +++ b/contrib/check_snmp_disk_monitor.pl @@ -0,0 +1,185 @@ +#!/usr/local/bin/perl +# author: Al Tobey <albert.tobey@priority-health.com> +# what: monitor diskspace using the host-resources mib +# license: GPL - http://www.fsf.org/licenses/gpl.txt +# +# Todo: + +use strict; +require 5.6.0; +use lib qw( /opt/nagios/libexec ); +use utils qw(%ERRORS $TIMEOUT &print_revision &support &usage); +use SNMP 5.0; +use Getopt::Long; +use 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 $TIMEOUT %mounts ); + +$PROGNAME = "snmp_disk_monitor.pl"; +$opt_verbose = undef; +$opt_host = undef; +$opt_community = 'public'; +$opt_command = undef; +$opt_warning = 99; +$opt_critical = 100; +$opt_port = 161; +$opt_stats = undef; +$message = undef; +$exit = 'OK'; +%mounts = (); + +sub process_options { + my( $opt_crit, $opt_warn ) = (); + Getopt::Long::Configure( 'bundling' ); + GetOptions( + 'V' => \$opt_version, 'version' => \$opt_version, + 'v' => \$opt_verbose, 'verbose' => \$opt_verbose, + 'h' => \$opt_help, 'help' => \$opt_help, + 's' => \$opt_stats, 'statistics' => \$opt_stats, + 'H:s' => \$opt_host, 'hostname:s' => \$opt_host, + 'p:i' => \$opt_port, 'port:i' => \$opt_port, + 'C:s' => \$opt_community, 'community:s' => \$opt_community, + 'c:i' => \$opt_crit, 'critical:i' => \$opt_crit, + 'w:i' => \$opt_warn, 'warning:i' => \$opt_warn, + 't:i' => \$TIMEOUT, 'timeout:i' => \$TIMEOUT, + 'm:s' => \$opt_mountpoint, 'mountpoint:s'=> \$opt_mountpoint + ); + if ( defined($opt_version) ) { local_print_revision(); } + if ( defined($opt_verbose) ) { $SNMP::debugging = 1; } + if ( !defined($opt_host) || defined($opt_help) || !defined($opt_mountpoint) ) { + print_help(); + exit $ERRORS{UNKNOWN}; + } + $opt_mountpoint = [ split(/,/, $opt_mountpoint) ]; +} + +sub local_print_revision { + print_revision( $PROGNAME, '$Revision$ ' ) +} + +sub print_usage { + print "Usage: $PROGNAME -H <host> -C <snmp_community> [-s] [-w <low>,<high>] [-c <low>,<high>] [-t <timeout>] -m <mountpoint>\n"; +} + +sub print_help { + local_print_revision(); + print "Copyright (c) 2002 Al Tobey <albert.tobey\@priority-health.com>\n\n", + "SNMP Disk Monitor plugin for Nagios\n\n"; + print_usage(); + print <<EOT; +-v, --verbose + print extra debugging information +-h, --help + print this help message +-H, --hostname=HOST + name or IP address of host to check +-C, --community=COMMUNITY NAME + community name for the host's SNMP agent +-m, --mountpoint=MOUNTPOINT + a mountpoint, or a comma delimited list of mountpoints +-w, --warning=INTEGER + percent of disk used to generate WARNING state (Default: 99) +-c, --critical=INTEGER + percent of disk used to generate CRITICAL state (Default: 100) +-s, --statistics + output statistics in Nagios format +EOT +} + +sub verbose (@) { + return if ( !defined($opt_verbose) ); + print @_; +} + +sub check_for_errors { + if ( $snmp_session->{ErrorNum} ) { + print "UNKNOWN - error retrieving SNMP data: $snmp_session->{ErrorStr}\n"; + exit $ERRORS{UNKNOWN}; + } +} + +# =========================================================================== # +# =====> MAIN +# =========================================================================== # +process_options(); + +alarm( $TIMEOUT ); # make sure we don't hang Nagios + +$snmp_session = new SNMP::Session( + DestHost => $opt_host, + Community => $opt_community, + RemotePort => $opt_port, + Version => '2c' +); + +# retrieve the data from the remote host +my( $mps, $alloc, $size, $used ) = $snmp_session->bulkwalk( 0, 4, [['hrStorageDescr'],['hrStorageAllocationUnits'],['hrStorageSize'],['hrStorageUsed']] ); +check_for_errors(); + +alarm( 0 ); # all done with the network connection + +# move all the data into a nice, convenient hash for processing +foreach my $mp ( @$mps ) { $mounts{$mp->iid}->{mountpoint} = $mp->val; } +foreach my $a ( @$alloc ) { $mounts{$a->iid}->{alloc_units} = $a->val; } +foreach my $si ( @$size ) { + if ( exists($mounts{$si->iid}->{alloc_units}) ) { + $mounts{$si->iid}->{size} = $si->val * $mounts{$si->iid}->{alloc_units}; + } + else { + $mounts{$si->iid}->{size} = $si->val; + } +} +foreach my $us ( @$used ) { + if ( exists($mounts{$us->iid}->{alloc_units}) ) { + $mounts{$us->iid}->{used} = $us->val * $mounts{$us->iid}->{alloc_units}; + } + else { + $mounts{$us->iid}->{used} = $us->val; + } +} + +# now find the mountpoint or mountpoints that were actually requested and push onto an array for output +my @matches = (); +foreach my $mp ( @$opt_mountpoint ) { + my $found = scalar(@matches); # count all matches + foreach my $key ( keys(%mounts) ) { + if ( $mounts{$key}->{mountpoint} eq $mp ) { + + # find the percentate - eval to avoid divide by zero errors + eval { $mounts{$key}->{percent_used} = $mounts{$key}->{used} / $mounts{$key}->{size} }; + $mounts{$key}->{percent_used} =~ s/^0\.([0-9]{1,2})([0-9]?).*/\1/; # truncate + if ( $2 >= 5 ) { $mounts{$key}->{percent_used}++ }; # round the number number + + verbose "mountpoint $mp has ", $mounts{$key}->{percent_used}, "% used, ", + $mounts{$key}->{size}, " bytes and ",$mounts{$key}->{used}, " used\n"; + + push( @matches, $mounts{$key} ); + } + } + if ( scalar(@matches) == $found ) { + print "UNKNOWN - could not locate mountpoint $mp on host\n"; + exit $ERRORS{UNKNOWN}; + } +} + +# now run through and check the thresholds +foreach my $mp ( @matches ) { + if ( $mp->{percent_used} >= $opt_warning ) { + $exit = 'WARNING'; + if ( $mp->{percent_used} >= $opt_critical ) { $exit = 'CRITICAL'; } + } + $message .= $mp->{percent_used}.'% used on '.$mp->{mountpoint}.', '; +} +$message =~ s/,\s*$//; + +# append statistics if requested +if ( defined($opt_stats) ) { + my @tmp = (); + foreach my $mp ( @matches ) { + push( @tmp, join(',',$mp->{mountpoint},$mp->{size},$mp->{used}) ); + } + $message .= '|'.join( ':', @tmp ); +} + +print "Disk $exit - $message\n"; +exit $ERRORS{$exit}; + + diff --git a/contrib/check_snmp_process_monitor.pl b/contrib/check_snmp_process_monitor.pl new file mode 100644 index 00000000..263255b5 --- /dev/null +++ b/contrib/check_snmp_process_monitor.pl @@ -0,0 +1,227 @@ +#!/usr/local/bin/perl +# author: Al Tobey <albert.tobey@priority-health.com> +# what: monitor a process using the host-resources mib +# license: GPL - http://www.fsf.org/licenses/gpl.txt +# +# Todo: +# * implement memory and cpu utilization checks +# * maybe cache pids in DBM files if snmp agents get overworked +############################################################################### +# to get a list of processes over snmp try this command: +# snmptable -v2c -c public hostname hrSWRunTable +# for just a list of valid arguments for the '-e' option: +# snmpwalk -v2c -c public hostname hrSWRunName |perl -pe 's:.*/::' +############################################################################### + +use strict; +require 5.6.0; +use lib qw( /opt/nagios/libexec /usr/local/libexec ); +use utils qw(%ERRORS $TIMEOUT &print_revision &support &usage); +use SNMP 5.0; +use Getopt::Long; +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 $TIMEOUT ); + +$PROGNAME = "snmp_process_monitor.pl"; +$opt_verbose = undef; +$opt_host = undef; +$opt_community = 'public'; +$opt_command = undef; +$opt_warning = [ 1, -1 ]; +$opt_critical = [ 1, -1 ]; +$opt_memory = undef; +$opt_cpu = undef; +$opt_port = 161; +%processes = (); +$exit = 'OK'; + +sub process_options { + my( $opt_crit, $opt_warn ) = (); + Getopt::Long::Configure( 'bundling' ); + GetOptions( + 'V' => \$opt_version, 'version' => \$opt_version, + 'v' => \$opt_verbose, 'verbose' => \$opt_verbose, + 'h' => \$opt_help, 'help' => \$opt_help, + 's' => \$opt_stats, 'statistics' => \$opt_stats, + 'H:s' => \$opt_host, 'hostname:s' => \$opt_host, + 'p:i' => \$opt_port, 'port:i' => \$opt_port, + 'C:s' => \$opt_community, 'community:s' => \$opt_community, + 'c:s' => \$opt_crit, 'critical:s' => \$opt_crit, + 'w:s' => \$opt_warn, 'warning:s' => \$opt_warn, + 't:i' => \$TIMEOUT, 'timeout:i' => \$TIMEOUT, + 'e:s' => \$opt_command, 'command:s' => \$opt_command, + 'r:s' => \$opt_regex, 'regex:s' => \$opt_regex, + 'cpu:i' => \$opt_cpu, 'memory:i' => \$opt_memory, + ); + if ( defined($opt_version) ) { local_print_revision(); } + if ( defined($opt_verbose) ) { $SNMP::debugging = 1; } + if ( !defined($opt_host) || defined($opt_help) || (!defined($opt_command) && !defined($opt_regex)) ) { + print_help(); + exit $ERRORS{UNKNOWN}; + } + + if ( defined($opt_crit) ) { + if ( $opt_crit =~ /,/ ) { + $opt_critical = [ split(',', $opt_crit) ]; + } + else { + $opt_critical = [ $opt_crit, -1 ]; + } + } + if ( defined($opt_warn) ) { + if ( $opt_warn =~ /,/ ) { + $opt_warning = [ split(',', $opt_warn) ]; + } + else { + $opt_warning = [ $opt_crit, -1 ]; + } + } +} + +sub local_print_revision { + print_revision( $PROGNAME, '$Revision$ ' ) +} + +sub print_usage { + print "Usage: $PROGNAME -H <host> -C <snmp_community> -e <command> [-w <low>,<high>] [-c <low>,<high>] [-t <timeout>]\n"; +} + +sub print_help { + local_print_revision(); + print "Copyright (c) 2002 Al Tobey <albert.tobey\@priority-health.com>\n\n", + "SNMP Process Monitor plugin for Nagios\n\n"; + print_usage(); + print <<EOT; +-v, --verbose + print extra debugging information +-h, --help + print this help message +-H, --hostname=HOST + name or IP address of host to check +-C, --community=COMMUNITY NAME + community name for the host's SNMP agent +-e, --command=COMMAND NAME (ps -e style) + what command should be monitored? +-r, --regex=Perl RE + use a perl regular expression to find your process +-w, --warning=INTEGER[,INTEGER] + minimum and maximum number of processes before a warning is issued (Default 1,-1) +-c, --critical=INTEGER[,INTEGER] + minimum and maximum number of processes before a critical is issued (Default 1,-1) +EOT +} + +sub verbose (@) { + return if ( !defined($opt_verbose) ); + print @_; +} + +sub check_for_errors { + if ( $snmp_session->{ErrorNum} ) { + print "UNKNOWN - error retrieving SNMP data: $snmp_session->{ErrorStr}\n"; + exit $ERRORS{UNKNOWN}; + } +} + +# =========================================================================== # +# =====> MAIN +# =========================================================================== # +process_options(); + +alarm( $TIMEOUT ); # make sure we don't hang Nagios + +$snmp_session = new SNMP::Session( + DestHost => $opt_host, + Community => $opt_community, + RemotePort => $opt_port, + Version => '2c' +); + +my $process_count = SNMP::Varbind->new( ['hrSystemProcesses', 0] ); +$snmp_session->get( $process_count ); +check_for_errors(); + +# retrieve the data from the remote host +my( $names, $index ) = $snmp_session->bulkwalk( 0, $process_count->val, [['hrSWRunName'], ['hrSWRunIndex']] ); +check_for_errors(); + +alarm( 0 ); # all done with the network connection + +my %namecount = (); +foreach my $row ( @$names ) { + $processes{$row->iid}->{name} = $row->val; + $processes{$row->iid}->{name} =~ s#.*/##; # strip path + + if ( defined($opt_regex) || + ($row->val =~ /(perl|\/usr\/bin\/sh|\/bin\/bash|\/bin\/sh)$/ + && $opt_command !~ /(perl|\/usr\/bin\/sh|\/bin\/bash|\/bin\/sh)$/) ) { + + # fetch the runtime parameters of the process + my $parm_var = SNMP::Varbind->new( ['hrSWRunParameters', $row->iid] ); + $snmp_session->get( $parm_var ); + check_for_errors(); + + # only strip if we're looking for a specific command + if ( defined($opt_command) ) { + verbose "process ",$row->iid," uses $1 as an interpreter - getting parameters\n"; + $processes{$row->iid}->{name} = $parm_var->val; + # strip path name off the front + $processes{$row->iid}->{name} =~ s#.*/##; + # strip everything from the first space to the end + $processes{$row->iid}->{name} =~ s/\s+.*$//; + } + else { + # get the longer full-path style listing + my $path_var = SNMP::Varbind->new( ['hrSWRunPath', $row->iid] ); + $snmp_session->get( $path_var ); + check_for_errors(); + + # use the full 'ps -efl' style listing for regular expression matching + $processes{$row->iid}->{name} = $path_var->val.' '.$parm_var->val; + } + } +} +foreach my $row ( @$index ) { + $processes{$row->iid}->{pid} = $row->val; +} + +my @pids = (); +my @matches = (); +foreach my $key ( keys(%processes) ) { + if ( defined($opt_command) && $processes{$key}->{name} eq $opt_command ) { + push( @matches, $processes{$key} ); + push( @pids, $processes{$key}->{pid} ); + verbose "process '$processes{$key}->{name}' has pid ", + "$processes{$key}->{pid} and index $key\n"; + } + elsif ( defined($opt_regex) && $processes{$key}->{name} =~ /$opt_regex/o ) { + push( @matches, $processes{$key} ); + push( @pids, $processes{$key}->{pid} ); + verbose "process '$processes{$key}->{name}' has pid ", + "$processes{$key}->{pid} and index $key\n"; + } +} +my $count = @matches; + +# warning, critical +if ( ($opt_warning->[0] > 0 && $opt_warning->[0] > $count) + || ($opt_warning->[1] > 0 && $opt_warning->[1] <= $count) ) { + $exit = 'WARNING'; +} +if ( ($opt_critical->[0] > 0 && $opt_critical->[0] > $count) + || ($opt_critical->[1] > 0 && $opt_critical->[1] <= $count) ) { + $exit = 'CRITICAL'; +} + +print "$exit - $count processes with pid(s) ",join(',',@pids); + +# print the number of processes if statistics are requested +if ( defined($opt_stats) ) { + print "|count:$count\n"; +} +else { + print "\n"; +} + +exit $ERRORS{$exit}; + + |