#!/usr/bin/perl -w

use strict;
use POSIX;
use Getopt::Long;
use LWP;

my $send_email = 0;
#my $email_target = 'backupandrecovery2@radian.biz';
my $email_target = 'gabriel.rosenkoetter@radian.biz';
my $send_opcmsg = 0;
my $help = 0;
my $debug = 0;

my $pp_def = $pol_prefix;
my $et_def = $email_target;

GetOptions('policy=s' => \$pol_prefix, 'email' => \$send_email,
	'recipient=s' => \$email_target, 'openview' => \$send_opcmsg,
	'help' => \$help, 'debug' => \$debug);

if ($help) {
	print "  --policy, -p ..... Specify policy [prefix] to watch.\n",
				"                     [Default: $pp_def]\n",
				"  --email, -e ...... Send email notification.\n",
				"  --recipient, -r .. Specify email recipient.\n",
				"                     [Default: $et_def]\n",
				"  --openview, -o ... Use opcmsg to raise an OpenView alert.\n",
				"  --help, -h ....... Display this message.\n\n",
				"  In absence of -e or -o, output is to STDOUT.\n";
	exit 0;
}

$ENV{PATH} .= ':/usr/openv/netbackup/bin:/usr/openv/netbackup/bin/admincmd:/usr/openv/netbackup/bin/goodies:/usr/openv/volmgr/bin:/usr/openv/volmgr/bin/goodies:/usr/openv/java:/opt/OV/bin';

my $date = strftime('%c', localtime());
# %c is "locale appropriate"

# XXX we want to:
# 1 check the last time we ran (lock file with $date as
#   sec-since-epoch?
# 2 only check log messages since then (INCLUDING ROTATED LOGS!)
#
# considering 1 more carefully later...

my $email_body = '';

# XXX
# 1 figure out what MM uses for logging
# 2 parse the appropriate log file out of syslog.conf
# 3 bonus points: figure out what log rotation we're using and parse
#   "yesterday's" file out of THAT
my $syslog_file = "/var/adm/syslog/syslog.log";
open SYSLOG, 'r', $syslog_file;

while (<SYSLOG>) {
	# Drives can be downed by bptm like this:
  # Feb 13 08:54:44 pa2ibks2 bptm[10416]: Application (NetBackup) has DOWN'ed drive index 2, see application error log for further information
	# (for more than 3 I/O failures in the last N hours)
	#
	# or more directly by robot daemons (tldd, acsd, ...?) like this:
	# Feb 16 09:10:58 pa2ibks2 tldd[17447]: TLD(1) drive 3 (device 14) is being DOWNED, status: Robotic mount failure
	# (could also be dismount failure)
	#
	# We care about these.
	#
	# They can also get downed by vmoprcmd through ltid like this:
	# Feb 15 10:32:56 pa2ibks2 ltid[17353]: Operator has DOWN'ed drive P2R1E4-9940b-00-ndmp (device 12)
	# 
	# We don't care about these.
	#
	# The consistent ID here is "(device X)", where X is the index
	# number for that host, so we want to use that and tpconfig to
	# find the rest of the drive's information.
	#
	# Robots can be downed by robot daemons like this:
	# Feb 12 22:21:49 pa2ibks2 tldd[1886]: TLD(1) going to DOWN state, status: Unable to sense robotic device
	# (need some more samples... ACS especially may give more info on the cause)

	if ($_ = m/DOWN/) {
		my $proc = (split(' ', $_))[4];
		# XXX ugh, portabiliwhat?

		if ($proc = m/^bptm/) {
			# definitely a drive
		} elsif ($proc = m/^tldd/ || 
						 $proc = m/^acsd/ ||
						 $proc = m/^tl4d/ || 
						 $proc = m/^tl8d/ ||
						 $proc = m/^tlhd/ ||
						 $proc = m/^tlmd/ ||
						 $proc = m/^ts8d/ ||
		         # XXX relevant? (below)
						 $proc = m/^odld/ ||
						 $proc = m/^tsdd/) {
			# drive or robot
		}

	my $message = sprintf
		"NetBackup job ID %s still running on %s for %s under policy %s at %s.\n",
		$job->{'jobid'}, $job->{master_server}, $job->{client}, $job->{class},
		$date;
	$email_body .= $message;
	chomp($message);
	if ($send_opcmsg) {
		my $command = 'opcmsg severity=major application=Netbackup '
			. 'msg_grp=SAN object="Long-running Backup" msg_text="'
			. $message . '"';
		$email_body .= "Notifying OpenView with command:\n  $command\n";
# XXX uncomment when OpenView's ready
		my $output = qx{$command 2>&1};
		$email_body .= $output if ($? > 0);
	}
	$email_body .= "\n";
}

chomp(my $hostname = qx{uname -n});
if ($send_email && length($email_body) > 0) {
	my $ua = LWP::UserAgent->new;
	
	my $req = HTTP::Request->new(POST => "mailto:$email_target");
	$req->header(Subject => "Long-running backup jobs for $hostname");
	$req->content($email_body);
	
	my $res = $ua->request($req);

	print STDERR $res->status_line, "\n" unless ($res->is_success);
} else {
	print $email_body;
}

exit 0;