Icinga 2 plugins

check_boot_kernel

#!/usr/bin/env perl

my $currentkernel = `uname -r`;
chomp($currentkernel);
`ls -t1 /boot/vmlinuz-* | head -1` =~ q!^/boot/vmlinuz-(.+)$!;
my $latestkernel = $1;
chomp($latestkernel);

if ($currentkernel eq '' || $latestkernel eq '') {
  print "UNKNOWN, could not determine kernel version, current=\"$currentkernel\", latest=\"$latestkernel.\"\n";
  exit(3);
}

if ($currentkernel eq $latestkernel) {
  `uname -v` =~ /([0-9]\.[0-9]+\.[0-9]+\-[^ ]+)/;
  my $curexactkernel = $1;
  my $instexactkernel = `/usr/bin/dpkg-query -W -f='\${Version}' linux-image-$currentkernel 2>/dev/null`;
  if ($?) {
    $instexactkernel = $curexactkernel;
  } else {
    chomp($instexactkernel);
  }
  if ($curexactkernel eq $instexactkernel) {
    print "OK, running kernel $currentkernel ($curexactkernel).\n";
    exit(0);
  }
  print "CRITICAL, running kernel $curexactkernel but installed is $instexactkernel.\n";
  exit(2);
}

if ($currentkernel lt $latestkernel) {
  print "CRITICAL, running kernel $currentkernel but latest is $latestkernel.\n";
  exit(2);
}

if ($currentkernel gt $latestkernel) {
  print "CRITICAL, running kernel $currentkernel but latest is $latestkernel.\n";
  exit(1);
}

print "UNKNOWN, current=\"$currentkernel\", latest=\"$latestkernel\".\n";
exit(3);

check_ceph

#!/usr/bin/env perl

use JSON qw(decode_json);
use Data::Dumper;

my $ignored = "noout.+";

my $cephoutput = `\$(which sudo) ceph -s --format=json`;
$cephoutput =~ s/^\s+|\s+$//g;
my $ceph = decode_json($cephoutput);
my @health_issues = ();
my $max_severity = 0;
my $status = 3;

#print Dumper($ceph);

foreach $hi (keys %{$ceph->{health}->{checks}}) {
  my $item = $ceph->{health}->{checks}->{$hi};
  my $message = $item->{summary}->{message};
  next if $message =~ m/$ignored/;
  #print '# '.$hi.": ".$message." ".$item->{severity}."\n";
  $max_severity = 1 if $item->{severity} eq 'HEALTH_WARN';
  $max_severity = 2 if $item->{severity} eq 'HEALTH_CRIT';
  push(@health_issues, $message);
}

my $osdmap = $ceph->{osdmap}->{osdmap};
if (($osdmap->{num_in_osds} < $osdmap->{num_osds} || $osdmap->{num_up_osds} < $osdmap->{num_osds}) && $status < 2) {
  $max_severity = 2;
  push(@health_issues, 'not all OSDs up or in');
}

if ($osdmap->{full}) {
  $max_severity = 2;
  push(@health_issues, 'full OSDs');
}

if ($osdmap->{nearfull}) {
  $max_severity = 1 if $max_severity < 2;
  push(@health_issues, 'nearfull OSDs');
}

if (@health_issues == 0 and $ceph->{health}->{overall_status}) {
  $status = 0;
}

my $pgmap = $ceph->{pgmap};
my %pgstates;
foreach $pg (@{$pgmap->{pgs_by_state}}) {
  #print Dumper($pg)."\n";
  foreach $pgstate (split(/\+/, $pg->{state_name})) {
    $pgstates{$pgstate} += $pg->{count};
  }
}
#print Dumper(%pgstates);
if ($pgstates{'degraded'} + $pgstates{'backfill_toofull'} + $pgstates{'recovery_toofull'} + $pgstates{'down'} > 1) {
  $max_severity = 2;
}

if ($max_severity > 1) {
  $status = 2;
} elsif ($max_severity > 0) {
  $status = 1;
}

my $message = '';

if ($status == 0) {
  $message .= 'OK';
} elsif ($status == 1) {
  $message .= 'WARNING';
} elsif ($status == 2) {
  $message .= 'CRITICAL';
} else {
  $message .= 'UNKNOWN';
}

if (@health_issues == 0) {
  push(@health_issues, $ceph->{health}->{overall_status});
}

$message .= ': '.join(', ', @health_issues);
$message .= ', OSDs: '.$osdmap->{num_in_osds}.'/'.$osdmap->{num_up_osds}.'/'.$osdmap->{num_osds};

$message .= "|";
foreach $st (keys(%pgstates)) {
  $message .= "pg_$st=".$pgstates{$st}.";;; ";
}
foreach $pgd (keys(%{$pgmap})) {
  next if ref($pgmap->{$pgd});
  $message .= "$pgd=".$pgmap->{$pgd};
  $message .= 'B' if $pgd =~ m/bytes$/ || $pgd =~ m/^bytes/;
  $message .= '%' if $pgd =~ m/ratio$/;
  $message .= ';;; ';
}

$message .= "\n";
print $message;

exit($status);

check_mem_full

#!/usr/bin/awk -E
BEGIN {
  while ((getline <"/proc/meminfo") > 0 ) {
    gsub(/:/,"");
    gsub(/\(/,"_");
    gsub(/\)/,"");
    a[$1]=$2;
    u[$1]=toupper($3);
  }
  swapfactor=(ENVIRON["swapfactor"]~/^[0-9]+$/)?ENVIRON["swapfactor"]:0.5;
  critical=(ENVIRON["critical"]~/^[0-9]+$/)?ENVIRON["critical"]:10;
  warning=(ENVIRON["warning"]~/^[0-9]+$/)?ENVIRON["warning"]:20;
}
END {
  ma=a["MemAvailable"];
  mt=a["MemTotal"];
  sf=a["SwapFree"];
  if (mt<10000 || ma<10) exit 3;
  mr=100*(ma+(sf*swapfactor))/mt;
  ec=0;
  printf "Memory ";
  if (mr<critical) { printf "CRITICAL"; ec=2; } else
  if (mr<warning) { printf "WARNING"; ec=1; } else
  printf "OK";
  printf ", %.1f%% (%d/%d MB + %d MB swap) available|",(ma*100/mt),(ma/1024),(mt/1024),(sf/1024);
  for (k in a) printf k"="a[k]u[k]";;; ";
  print "";
  exit ec;
}

check_puppet5

#!/opt/puppetlabs/puppet/bin/ruby

require 'yaml'
require 'puppet'
require 'optparse'

options = {
  :lastrun_warn => 3600,
  :lastrun_crit => 36000,
  :runtime_warn => 25,
  :runtime_crit => 60,
}
OptionParser.new do |opt|
  opt.banner = "Usage: #{$0} [options]"
  opt.on('-l', '--lastrun w,c', Array, 'Time since last Puppet run') { |w| options[:lastrun_warn] = w[0].to_i; options[:lastrun_crit] = w[1].to_i }
  opt.on('-r', '--runtime w,c', Array, 'Total Puppet run time') { |w| options[:runtime_warn] = w[0].to_i; options[:runtime_crit] = w[1].to_i }
end.parse!

report = YAML.load_file('/opt/puppetlabs/puppet/cache/state/last_run_report.yaml')

overall = 3;

if (report.status == 'unchanged')
  overall = 0
  message = 'No changes'
elsif (report.status == 'failed')
  overall = 2
  message = 'Failed resources'
else
  overall = 1
  message = 'Changed resources'
end

lastrun = Time.now.to_i - Time.at(report.time).to_i
if (overall < 2)
  if (lastrun > options[:lastrun_crit])
    overall = 2
    message << ", Last run: #{lastrun}"
  end
  if (report.metrics['time']['total'] > options[:runtime_crit])
    overall = 2
    message << ", Total run time: #{report.metrics['time']['total']}"
  end
end

if (overall < 1)
  if (lastrun > options[:lastrun_warn])
    overall = 1
    message << ", Last run: #{lastrun}"
  end
  if (report.metrics['time']['total'] > options[:runtime_warn])
    overall = 1
    message << ", Total run time: #{report.metrics['time']['total']}"
  end
end

case overall
when 2
  print 'CRITICAL'
when 1
  print 'WARNING'
when 0
  print 'OK'
else
  print 'UNKNOWN'
end

print ', %s|lastrun=%i;%i;%i;0; ' % [ message, lastrun, options[:lastrun_warn], options[:lastrun_crit] ]

if !report.metrics.empty?
  for met in [ 'total', 'skipped', 'failed', 'failed_to_restart', 'restarted', 'changed', 'out_of_sync', 'scheduled' ]
    print "res_#{met}=#{report.metrics['resources'][met]};;;0; "
  end
  puts "runtime=#{report.metrics['time']['total']}s;#{options[:runtime_warn]};#{options[:runtime_crit]};0; "
end

exit overall