improved. Sergey.

diff --git a/plugins/check_lustre.pl b/plugins/check_lustre.pl
index ed868a98cb58b35a6195a678fb5e6b7623e73157..cc36e5a3cdd3d57f92fd3612c08a0fbbaa5532bd 100755 (executable)
--- a/plugins/check_lustre.pl
+++ b/plugins/check_lustre.pl
@@ -8,6 +8,7 @@ use Net::FTP;
 use Data::Dumper;
 use IO::Handle;
 use Time::Local;
+use Time::localtime;
 
 # the following is for the status server
 # to communicate with Nagios plugin
@@ -16,69 +17,60 @@ use threads::shared;
 use IO::Socket;
 use IO::Select;
 
-my $status : shared = "OK";
+my $status     : shared = "OK";
+my $time_ls    : shared = &getTime();
+my $sleep_time : shared = 120;
 
 our $server_port = '50502';
 our $protocol    = 'tcp';
 
-threads->new( \&statusServer);
+threads->new( \&statusServer );
 &main();
 
 exit(0);
 
 sub main {
-    my $counter=0;
-    my @total;
-    my $statsize = "0";
     
+    my $time_ls = &getTime();
+    my $time_du = 0;
+    my $lustre_size = 0;
+
     while (1)
     {
-       if ($counter == 0)     
-       {
-            # every 24 hours
-           my @size = `du -cms /lustre_alpha/hades`;
-           
-           foreach my $li (@size) 
-           {
-               if($li =~/total/) 
-               {
-                   @total = split(" ", $li);
-                   $statsize = $total[0];
-                   $statsize =sprintf ("%.2f",$statsize/1024/1024);
-               }
-           }
-           $counter = 720;
-           
-           print "/lustre_alpha/hades total size: $statsize TB\n";
-       }   # should be 720
+
+       if( abs($time_du - $time_ls) > 60*60*24 ){
+           $lustre_size = &getTotalSize();
+           $time_du = &getTime();
+       }
        
        my $line =`ls -d /lustre_alpha/hades/beam/sep08`;
        chomp($line);
        
+       $time_ls = &getTime();
+
        if ($line eq "/lustre_alpha/hades/beam/sep08") 
        {
-           if ($statsize > 30) {
-               $status = "CRITICAL - Lustre disk space used: $statsize TB";
+           if ($lustre_size > 30) {
+               $status = "CRITICAL - Lustre hades/beam space used: $lustre_size TB";
            }
-           elsif ($statsize > 25) {
-               $status  = "WARNING - Lustre disk space used: $statsize TB";
+           elsif ($lustre_size > 25) {
+               $status  = "WARNING - Lustre hades/beam space used: $lustre_size TB";
            }
            else {
-               $status  = "OK - Lustre disk space used: $statsize TB";
+               $status  = "OK - Lustre hades/beam space used: $lustre_size TB";
            }
        }
        else {
            $status  = "CRITICAL - failure of Lustre file system!";
        }
        
-       $counter --;
-       
-       #print "$status \n";
-       sleep 120; #should be 120
+       sleep $sleep_time; # 2 minutes
     }
 }
 
-sub statusServer{ my $server_socket;
+sub statusServer{ 
+
+    my $server_socket;
     my $client_socket;
     my $selector;
 
@@ -102,25 +94,61 @@ sub statusServer{ my $server_socket;
                 if($file_handle == $server_socket) {
 
                     # create a new socket for this transaction
-                    unless (defined( $client_socket = $server_socket->accept() )
-)
+                    unless (defined( $client_socket = $server_socket->accept() ))
                     {
                         print "ERROR: Cannot open socket to send status!\n";
                     }
 
-
-                    print $client_socket $status;
+                   #--- report the status
+                   my $current_time = &getTime();
+                   if( abs($current_time - $time_ls) < 3*$sleep_time ){
+                       print $client_socket $status;
+                   }
+                   else{
+                       my $time_diff = int((abs($current_time - $time_ls))/60);
+                       print $client_socket "time out! Last check done: $time_diff min ago";
+                   }
 
                     close( $client_socket );
                 }
             }
         }
     }
+}
+
+sub getTime
+{
+    #--- get local time in seconds
 
+    my $tm = localtime;
 
+    my $sec = timegm($tm->sec,       # sec    0-60
+                    $tm->min,       # min    0-59
+                    $tm->hour,      # hours  0-23
+                    $tm->mday,      # mday   0-31
+                    $tm->mon,       # month  0-11
+                    $tm->year);     # year   1-138 (since 1900)
+    
+    return $sec;
 }
 
+sub getTotalSize
+{
+    #--- get total size of lustre beam dir
+
+    my $totsize = 0;
+    my @size = `du -cms /lustre_alpha/hades/beam`;
+    
+    foreach my $line (@size){
+       if( $line =~ /total/ ){
+           my @total = split(" ", $line);
+           $totsize  = $total[0];
+           $totsize  = sprintf ("%.2f",$totsize/1024/1024);
+       }
+    }
 
+    return $totsize;
+}