#!/opt/vdops/bin/perl

# This script queries status variables in BLUEARC-SERVER-MIB and 
# produces a report identifying salient issues.

# V      Who       When        What
# ---------------------------------------------------------------------------
# 1.1.0  skendric  2010-01-10  Upgrade to perl 5.10.1
# 1.0.6  skendric  2009-04-19  Distinguish between silent and unresponsive
# 1.0.5  skendric  2009-03-20  Add @down_for_maintenance
# 1.0.4  skendric  2007-12-07  Add owner
# 1.0.3  skendric  2007-06-07  Add clusterQuorumDeviceStatus
# 1.0.2  skendric  2007-03-21  Stylistic mods
# 1.0.1  skendric  2007-02-09  Add logging, separate storage into components
# 1.0.0  skendric  2007-01-23  First Version
#

# Author:  Stuart Kendrick, sbk {put at sign here} skendric {put dot here} com
#
# Source:  http://www.skendric.com/device
#
# This software is available under the GNU GENERAL PUBLIC LICENSE, see
# http://www.fsf.org/licenses/gpl.html
#

#
# This script takes the following approach:
#      -Parses the hosts table for a list of targets (or accepts a command-
#       line list)
#      -Queries a bunch of BlueArc-specific variables
#      -Produces a report
#
#
# Requirements:
#     -The target(s) must be pingable
#
#     -PERL modules:  the FHCRC::Netops collection
#
#
# Assumptions:
#
#
# Tested on:
#        -perl-5.10.1
#        -net-snmp-5.5
#
#
# Instructions:
#       -Customize the script for your site:  find the 'user-configurable
#        variables' section and modify as appropriate
#       -Type "bluearc-alarm" to see the command-line options
#       -Try it out
#
#
#
# Caveats:
#
#
# Known Bugs:
#
#
# To do:
#

# Begin script

# Load modules
use strict;
use warnings;
use feature 'say';
use feature 'switch';
use Carp qw(carp cluck croak confess);
use Data::Dumper;
use English qw( -no_match_vars );
use Getopt::Std;
use List::MoreUtils qw(uniq);
use FHCRC::Netops::HostTools 1.0.3;
use FHCRC::Netops::NetopsTools 2.0.7;
use FHCRC::Netops::NetopsData 1.3.0;
use FHCRC::Netops::PingTools 1.1.5;
use FHCRC::Netops::SNMPTools 1.3.9;
use FHCRC::Netops::Utilities 1.3.9;

# Declare global variables
my %nic;                       # ifOperStatus
my %temperature;               # temperatureSensorStatus
my %fan;                       # fanFittedStatus + fanSpeedStatus
my %power;                     # psuStatus
my %clusterQuorum;             # clusterQuorumDeviceStatus
my %clusterVNode;              # clusterVNodeStatus
my %sysDrive;                  # sysDriveStatus
my %volume;                    # volumeStatus
my @silent;                    # Targets which did not answer pings
my @unresponsive;              # Targets which did not answer SNMP GETs

# Define global variables
$debug = 0;
            # 10 = Logging
            #  9 = Database SELECT operations
            #  8 = Per IP/MAC/Port processing
            #  7 = Database INSERT/UPDATE/DELETE
            #  6 = Dump SNMP var
            #  5 = Dump snmp_packets
            #  4 = Grody:    print big var
            #  3 = Verbose:  print mid var
            #  2 = Simple:   print small var
            #  1 = Basic:    subroutine trace
            #  0 = Disable debugging
$program_name = 'bluearc-alarm';
$usage = 'Usage:  bluearc-alarm -s {yes|no} [-d {integer}] [-r] [-a | -e {expr} | -f {filename} | target1 target2 target3 ...]'; 
$version = '1.1.0';

# Define user-configurable variables

# Binaries
$grab_hosts = '/bin/cat /etc/hosts';

# Report stuff
$institution       = 'Widgets International';
$owner             = 'Susan Way';
$owner_backup      = 'Rick Bawaan';
$report_file       = '/home/netops/rpts/bluearc-alarm.txt';
$report_queries    = 'bsmith@widgets.com';
$report_recipients = 'it-server@fhcrc.org';
$report_subject    = 'BlueArc Alarm Report';

# Pause parameters
$long  = 30;
$mid   = 10;
$short = 5;

# Ping Stuff
$ping_count   = 3;
$ping_timeout = 1;

# SNMP Stuff
# Optimize performance by sorting your community strings and SNMP version
# list, most frequently used to the left, least frequently used to the right
@snmp_read_list    = qw/public public/;
@snmp_version_list = qw/2/;
$snmp_port         = 161;
$snmp_retries      = 3;
$snmp_timeout      = 2000000;

# Syslog stuff
$syslog_facility  = 'local5';
$syslog_host      = 'localhost';
$syslog_port      = 514;        
$syslog_priority  = 'info';
$syslog_socket    = 'unix';    # Other possibilites include 'udp' and
                               # 'stream'; depending on the flavor of Unix,
                               # I've employed each of these

# Target details
@down_for_maintenance = qw//;
@skip_name = qw/swamp/;
@suffixes  = qw/-bluearc/;


# Grab arguments
getopts('ad:e:f:rs:', \%option);
@target = @ARGV;

# Set mode
if      ($option{r})            { $mode = 'report'      }
elsif   (-t STDIN)              { $mode = 'interactive' }
else                            { $mode = 'batch'       }


### Begin Main Program ###############################################
{

  check_args();                         # Check arguments

  # compile_mibs();                     # Compile MIB files

  build_target();                       # Populate @target

  push @silent, target_check();         # Look for errors in @target

  push @unresponsive, basic_info();     # Gather information

  sanity_check();                       # Check for major errors

  do_the_work();                        # Do the work

  identify_alarms();                    # Count devices with alarms

  print_report();                       # Print report

  notify_staff();                       # Mail report

}
##### End Main Program ###############################################


########################################################################
# Do the work
########################################################################
sub do_the_work {
  my $answer;
  my $val;            # Result of snmpGet/Walk

  # Debug trace
  trace_location('begin') if $debug;

  # Notify operator
  print_it('Querying targets...');
  unless ($dome) {
    sleep $short;
    return 1;
  }

  # Loop through targets
  TARGET:
  for my $target (@target) {

    # Walk ifOperStatus
    say 'Walking ifOperStatus' if $debug > 3;
      $val = snmpWalk( {host => $target, oid => 'ifOperStatus'} );
      $answer = 'ok';
      for my $varbind (@$val) {
        my ($iid, $val);
        $iid = $varbind->{iid};
        $val = $varbind->{val};
        log_it("$target ifOperStatus.$iid = $val");
        $answer = 'bad' unless $val eq 'up';
      } 
      $alarm_count{$target}++ unless $answer eq 'ok';
      $nic{$target} = $answer;
    
    # Walk temperatureSensorStatus
    say 'Walking temperatureSensorStatus' if $debug > 3;
      $val = snmpWalk( {host => $target, oid => 'temperatureSensorStatus'} );
      $answer = 'ok';
      for my $varbind (@$val) {
        my ($iid, $val);
        $iid = $varbind->{iid};
        $val = $varbind->{val};
        log_it("$target temperatureSensorStatus.$iid = $val");
        $answer = 'bad' unless $val eq 'ok';
      }  
      $alarm_count{$target}++ if $answer eq 'bad';
      $temperature{$target} = $answer;
    
    # Walk fanFittedStatus
    say 'Walking fanFittedStatus' if $debug > 3;
      $val = snmpWalk( {host => $target, oid => 'fanFittedStatus'} );
      $answer = 'ok';
      for my $varbind (@$val) {
        my ($iid, $val);
        $iid = $varbind->{iid};
        $val = $varbind->{val};
        log_it("$target fanFittedStatus.$iid = $val");
        $answer = 'bad' unless $val eq 'ok';
      }
      $alarm_count{$target}++ if $answer eq 'bad';
      $fan{$target} = $answer;
    
    # Walk fanSpeedStatus
    unless ($fan{$target} eq 'bad') {
      say 'Walking fanSpeedStatus' if $debug > 3;
      $val = snmpWalk( {host => $target, oid => 'fanSpeedStatus'} );
      $answer = 'ok';
      for my $varbind (@$val) {
        my ($iid, $val);
        $iid = $varbind->{iid};
        $val = $varbind->{val};
        log_it("$target fanSpeedStatus.$iid = $val");
        $answer = 'bad' unless $val eq 'ok';
      }
      $alarm_count{$target}++ if $answer eq 'bad';
      $fan{$target} = $answer;
    }
      
    # Walk psuStatus
    say 'Walking psuStatus' if $debug > 3;
      $val = snmpWalk( {host => $target, oid => 'psuStatus'} );
      $answer = 'ok';
      for my $varbind (@$val) {
        my ($iid, $val);
        $iid = $varbind->{iid};
        $val = $varbind->{val};
        log_it("$target psuStatus.$iid = $val");
        $answer = 'bad' unless $val eq 'ok';
      }
      $alarm_count{$target}++ if $answer eq 'bad';
      $power{$target} = $answer;
    
    # Walk clusterVNodeStatus
    say 'Walking clusterVNodeStatus' if $debug > 3;
      $val = snmpWalk( {host => $target, oid => 'clusterVNodeStatus'} );
      $answer = 'ok';
      for my $varbind (@$val) {
        my ($iid, $val);
        $iid = $varbind->{iid};
        $val = $varbind->{val};
        log_it("$target clusterVNodeStatus.$iid = $val");
        $answer = 'bad' unless $val eq 'onLine';
      }
      $alarm_count{$target}++ if $answer eq 'bad';
      $clusterVNode{$target} = $answer;

    # Get clusterQuorumDeviceStatus
    say 'Getting clusterQuorumDeviceStatus' if $debug > 3;
      $val = snmpGet( {host => $target, oid => 'clusterQuorumDeviceStatus.0'} );
      given ($val) {
        when ('owned') { $answer = 'ok'  }
        default        { $answer = 'bad' }
      }
      $alarm_count{$target}++ unless $answer eq 'ok';
      $clusterQuorum{$target} = $answer;

    # Walk sysDriveStatus
    say 'Walking sysDriveStatus' if $debug > 3;
      $val = snmpWalk( {host => $target, oid => 'sysDriveStatus'} );
      $answer = 'ok';
      for my $varbind (@$val) {
        my ($iid, $val);
        $iid = $varbind->{iid};
        $val = $varbind->{val};
        log_it("$target sysDriveStatus.$iid = $val");
        $answer = 'bad' unless $val eq 'online';
      }
      $alarm_count{$target}++ if $answer eq 'bad';
      $sysDrive{$target} = $answer;

    # Walk volumeStatus
    say 'Walking volumeStatus' if $debug > 3;
      $val = snmpWalk( {host => $target, oid => 'volumeStatus'} );
      $answer = 'ok';
      for my $varbind (@$val) {
        my ($iid, $val);
        $iid = $varbind->{iid};
        $val = $varbind->{val};
        log_it("$target volumeStatus.$iid = $val");
        $answer = 'bad' unless $val eq 'mounted';
      }
      $alarm_count{$target}++ if $answer eq 'bad';
      $volume{$target} = $answer;
      
    # Entertain operator
    print $BANG if $mode eq 'interactive';

  }

  # Debug info
  if ($debug > 2) {
    for my $target (@target) {
      if ($alarm_count{$target} > 0) {
        say "alarm_count{$target} = $alarm_count{$target}";
      }
    }
  }

  # Make things look pretty
  say "\n" if $mode eq 'interactive';

  # Debug trace
  trace_location('end') if $debug;

  return 1;
}




########################################################################
# Tell the operator what I discovered
########################################################################
sub print_report {
  my $handle;
  my $total = @target;
  my $now = get_now();

  # If we are running in test mode, skip this routine
  unless ($dome) {
    print_it("Running in test mode, cannot print a meaningful report\n");
    return 1;
  }

  # Debug trace
  trace_location('begin') if $debug;

  # Direct output to screen or to file
  if ($mode eq 'interactive') {
    $handle = *STDOUT;
  }
  else {
    open $handle, '>', $report_file or die "Cannot open $report_file: $!\n";
  }


print {$handle} <<EOF;
#
# Title:           BlueArc Alarm Summary
#
# Institution:     $institution
#
# Date of Report:  $now
#
# Description:     This report lists BlueArcs and the status of various alarms
#
# Active:          $total
#
# Alarm Count:     $shit_happens
#
# In Alarm:        @alarm_list
# 
# Questions:       If you have questions or comments regarding this
#                  report, please mail them to $report_queries
#
# Notes:
#
# ok  = Normal
# bad = Problem
#
# nic      = ifOperStatus
# temp     = temperatureSensorStatus
# fan      = fanFittedStatus + fanSpeedStatus
# power    = psuStatus
# vnode    = clusterVNodeStatus
# quorum   = clusterQuorumDeviceStatus
# sysDrive = sysDriveStatus
# volume   = volumeStatus

#
target           nic  temp  fan  power  vnode  quorum  sysDrive  volume
---------------  ---  ----  ---  -----  -----  ------  --------  ------
EOF
  TARGET:
  for my $target (@target) {
  printf {$handle} "%-15s  %-3s  %-4s  %-3s  %-5s  %-5s  %-6s  %-8s  %-6s\n",
      $target, $nic{$target}, $temperature{$target}, $fan{$target},
      $power{$target}, $clusterVNode{$target}, $clusterQuorum{$target}, 
      $sysDrive{$target}, $volume{$target};
  }
  
  # Add silent devices to the report
  if (@silent > 0) {
    @silent = sort @silent;
    for my $silent (@silent) {
      printf {$handle} "%-15s  Not answering pings\n", $silent;
    }
  }

  # Add unresponsive devices to the report
  if (@unresponsive > 0) {
    @unresponsive = sort @unresponsive;
    for my $unresponsive (@unresponsive) {
      printf {$handle} "%-15s  Not answering SNMP GETs\n", $unresponsive;
    }
  }

  # Clean up
  unless ($handle =~ /STDOUT/) {
    close $handle or warn "Cannot close $report_file: $!\n";
  }

  # Debug trace
  trace_location('end') if $debug;

  return 1;
}




########################################################################
# Sanity check
########################################################################
sub sanity_check {
  my @remove;

  # Debug trace
  trace_location('begin') if $debug;

  # Notify operator
  print_it('Sanity check...');

  # Loop through targets, removing non-BlueArc devices
  SANITY:
  for my $target (@target) {
  
    # Identify manufacturer
    unless ($manufacturer{$target} eq 'BlueArc') {
      say "\nManufacturer of $target is not BlueArc, ignoring" if $debug;
      push @remove, $target;
      next SANITY;
    }

    # Entertain operator
    print $BANG if $mode eq 'interactive';

  }

  # Remove entries which failed checks
  prune_basic(@remove);

  # Make things look pretty
  say "\n" if $mode eq 'interactive';

  # Debug trace
  trace_location('end') if $debug;

  return 1;
}




########################################################################
# Output help
########################################################################
sub HELP_MESSAGE {
  print <<EOF;

  $usage

    -s asks the question:  are you serious?  If you answer 'no', then 
       the script will run in demo mode, making no changes

    -d specifies debug level (typically 0-9)

    -r determines whether or not the script writes its report to disk

    -a instructs the script to filter the hosts table using its internally
       defined suffixes variable to produce a list of targets

    -e requires a regex as its argument, with which the script filters
       the hosts table to produce a list of targets.  Enclose the regex in 
       quotes to protect special characters from the shell

    -f points the script to a file containing the list of targets

    target1 target2 target3 ... is the command-line list of targets you
      have specified, in lieu of the -a, -e, or -f options

EOF

  return 1;
}




########################################################################
# Output version
########################################################################
sub VERSION_MESSAGE {
  say "$program_name v$version";
  return 1;
}