#!/usr/bin/perl

# This script runs a set of sanity checks on the system
# and reports status.

## Reporting thresholds
$disk_threshold = 95; # percentage of capacity
$load_threshold = 1.5;
$miserver_log_threshold = 20000000; # 20M
$cpu_threshold = 90; # percent
$mem_threshold = 10000000; # 10M
# swap thresholds signify minimums
# $swap_total_threshold = 1243384; # in kilobytes
$swap_total_threshold = 984000; # in kilobytes
$swap_available_threshold = 300000; # should have at least 300M available


## Check disk capacity.
open (DF, "/usr/local/bin/df -k |") || die $!;
while (<DF>) {
  next if m,/cdrom,;
  next unless /(\d+)%/;
  if ($1 >= $disk_threshold) {
    print "BAD: disk nearing capacity\n$_\n";
  }
}
close DF;


## Check system load.
$uptime = `/bin/uptime`;
($load1, $load5, $load15) = ($uptime =~ m/load average: (\S+) (\S+) (\S+)$/);
if ($load15 > $load_threshold) {
  print "BAD: load average over threshold\n$uptime";
} else {
  print "load average: $load1 $load5 $load15\n\n";
}


## Check miserver.log size.
$miserver_logsize = -s "/home/miadmin/data/miserver.log";
if ($miserver_logsize > $miserver_log_threshold) {
  printf "BAD: miserver.log getting too big to be useful (%.1fM)\n", $miserver_logsize / 1000000;
}

## Try to catch runaway processes.
open (PS, "/usr/ucb/ps wwwaux |") || die $!;
$_ = <PS>; # throw away header line
while (<PS>) {
  $cpu = substr($_, 14, 5);
  $mem = substr($_, 24, 5);
  if ($cpu > $cpu_threshold && $mem > $mem_threshold) {
    print "BAD: high memory and CPU usage: $_";
  } elsif ($cpu > $cpu_threshold) {
    print "BAD: high CPU usage:            $_";
  } elsif ($mem > $mem_threshold) {
    print "BAD: $high memory usage:        $_";
  }
}
close PS;

## Check on swap.
chop($swap = `/usr/sbin/swap -s`);
($swap_used, $swap_available) = ($swap =~ m/(\d+)k used, (\d+)k available/);
$swap_total = $swap_used + $swap_available;
if ($swap_total < $swap_total_threshold) {
  print "BAD: Not all swap is allocated\n";
}
if ($swap_available < $swap_available_threshold) {
  printf "BAD: Available swap down to %dM\n", $swap_available/1000;
}

## Check status of disk mirrors.
open (M, "/usr/opt/SUNWmd/sbin/metastat -p |") || die $!;
while (<M>) {
  next unless /^d(\d) -m/;
  if (! /^d$1 -m d\d\d d\d\d (d\d\d )?1/) {
    print "BAD: malformed mirror: $_\n";
  }
}

## Report illustra dumptimes.
$select_string = "select database_name, database_dumptimes from databases;";
open (MSQL, "/usr/local/bin/msql -U miadmin -P ifmxsux template1 -c \"$select_string\" |") || die $!;
while (<MSQL>) {
  print;
}
close MSQL;


$metastat_output = `/usr/opt/SUNWmd/sbin/metastat`;
if ($metastat_output =~ /maintenance/i) {
    print "MUCHO BAD: drives need maintenance:\n$metastat_output";
}

open(SYSLOG, "/var/adm/messages") || die $!;
while (<SYSLOG>) {
    next unless /Sense/;
    print "MUCHO BAD: possible imminent drive failure, check syslog\n";
}
close SYSLOG;

exit 0;