#!/usr/bin/perl  

#ModulePurpose: check ntp stratum of all boxes
#ModuleName:    check_timesync
#TimeRequired:   minutes

# CS and BuCS should be 1
# RSPs should be 2

#Problems Report:
#  --- 22 boxes at incorrect stratum - 08.15.2009 
#     m001      timesync'd to m7server at stratum 4 -- not at stratum 2
#     m002      timesync'd to m7server at stratum 4 -- not at stratum 2



use lib '/h/bmetzger/s/p/hc/';
use HC qw(:ALL);  #functions available to all modules 

use English;
use Carp;
use Cwd;
use DBI;
use Net::Ping;
use File::Basename;
use File::Path;
use File::Glob;
use File::Copy;

###############3###
# define location of, and open files
##################

  my ($TmpDir,$sitelist,$global_problems,$global_details,$global_summary) = InitVars();

  $SSH_OPTIONS=" -o BatchMode=yes -o ConnectTimeout=2";
  $ENV{'PATH'} = 'misc_tools:/usr/bin:/usr/local/bin:/usr/sbin/:/usr/ucb:/etc/:/opt/platform7/lbin/:.';
  &initvars();

  #create list of boxes to check
#configurable
  qx[/opt/platform7/lbin/list_processors -sw > $sitelist];
  qx[cat /etc/opt/OSSInstall/servers.conf >> $sitelist];
#  qx[list_processors -s | grep m005 > $sitelist];

### checks
 ($header, $local_problems, $local_details)= &check(); 
  summarize($header, $local_problems, $local_details);

###################33
sub initvars {

   ($sys_day,$sys_mon,$sys_year) = (localtime)[3,4,5];
   $sys_mon += 1;  #jan is 0
   $sys_year += 1900;
   $today = sprintf "%02d.%02d.%d",$sys_mon,$sys_day,$sys_year;  # zero fill date

   $SSH_OPTIONS=" -o BatchMode=yes -o ConnectTimeout=2";
   $ENV{'PATH'} = '/usr/bin:/usr/local/bin:/usr/sbin/:/usr/ucb:/etc/:.';
} #initvars

####################
sub line_count {
# perform a  wc -l

  my $file = shift;
    open my $fh, "<", $file or die "could not open $file:$!";
    my $lines = 0;
    $lines++ while <$fh>;
    return $lines;
    close ($fh);
} #line_count

#########
sub check {
     print "check timesync\n";

#configurable
     my $THRESHOLD = 60; 

     my ($p, $box);
     my $local_problems = "$TmpDir/swap_problems";
     my $local_details  = "$TmpDir/swap_details";
     my $local_junk1    = "$TmpDir/junk1";    # temp file for problems
     my $local_junk2    = "$TmpDir/junk2";    # temp file for details

    # open local output files
    open PROBLEMS, ">>$local_problems" or croak "Can't open local_problems for write: $!\n";
    open DETAILS, ">>$local_details" or croak "Can't open local_details for write: $!\n";

    open SITE, "<$sitelist" or croak "Can't open sitelist file for read $!\n";

    # check each box 
    while (defined ($box = <SITE>)) {
        @arr = ();
        my ($server, $startum) = "";

        chomp($box);
        open PROBLEMS, ">>$local_problems" or croak "Can't open local_problems for write: $!\n";

        $p = Net::Ping->new("icmp");   #need icmp or fails to tra1 and tra2
        if ( $p->ping($box) ) {  #pingable

              qx[remsh $box '/usr/sbin/ntpq -p | grep -E "\\*" ' > $local_junk1];

            open JUNK1, "<$local_junk1" or croak "Can't open junk1 for read $!\n";

             while (my $line = <JUNK1>) {
                     chomp $line;

                     $line =~ s/\*//g,$line;
                     $line =~ s/\s+/ /g,$line;
                     my @arr = split(/\s+/,$line);

                     my $server  = $arr[0];
                     my $stratum = $arr[2];
                     if ( ($box eq "m7server") || ($box eq "m7altsrv")) {
                           if ($stratum != 1) {
                          printf PROBLEMS  "   %-8s  timesync'd to %-8s  at stratum %s -- not at stratum 1\n", $box, $server, $stratum;
          printf   "   %-8s  timesync'd to %-8s at stratum %s -- not at stratum 1\n", $box, $server, $stratum;
                           }
                     }

                     if ( ($box ne "m7server") && ($box ne "m7altsrv") && ($stratum != 2)){
                          printf PROBLEMS  "   %-8s  timesync'd to %-8s at stratum %s -- not at stratum 2\n", $box, $server, $stratum;
        printf   "   %-8s  timesync'd to %-8s  at stratum %s -- not at stratum 1\n", $box, $server, $stratum;
                     }
             }
        } else {
               printf PROBLEMS "    $10s NOT pingable\n",$box ;
               printf  "$10s NOT pingable\n",$box ;
        }
        $p->close();
        close(PROBLEMS); 
        unlink("$local_junk1") || print $!;
  } #for each box in sitelist

    #gotta close, so lines can be counted
    close(DETAILS);  
    close(PROBLEMS); 

    my $badcount = line_count($local_problems);

    # open local output files
    open JUNK1, ">>$local_junk1" or croak "Can't open file $local_junk1 for write: $!\n";
    open JUNK2, ">>$local_junk2" or croak "Can't open file $local_junk2 for write: $!\n";

    my $header = "\n--- $badcount boxes at incorrect stratum -$today \n";

    if ( $badcount > 0 ){  # at least one problem
         #format and header
         print JUNK1 $header;
             qx[grep -v ping $local_problems >> $local_junk1];
             qx[grep    ping $local_problems >> $local_junk1];
         close(JUNK1);

         print JUNK2 $header;
         close(JUNK2);
    }else {
         print JUNK2 $header;
         close(JUNK2);
    }

  copy($local_junk1,$local_problems) or die "junk1 cannot be copied";
  copy($local_junk2,$local_details) or die "junk2 cannot be copied";

   unlink("$local_junk1") || print $!;
   unlink("$local_junk2") || print $!;
   unlink("$sitelist") || print $!;

  return($header, $local_problems, $local_details);
} #check

