Can check Lustre mount and Lustre space used (-s flag). Segrey

diff --git a/plugins/check_lustre.pl b/plugins/check_lustre.pl
index cc36e5a3cdd3d57f92fd3612c08a0fbbaa5532bd..3d8ba4fd6a245d38b060d1a9bc041faba08ec880 100755 (executable)
--- a/plugins/check_lustre.pl
+++ b/plugins/check_lustre.pl
@@ -1,5 +1,11 @@
 #!/usr/bin/perl -w
 
+########################################################
+# plugin script for Nagios to monitor Lustre           #
+#                                                      #
+# Sergey Yurevich                                      #
+########################################################
+
 use strict;
 use warnings;
 
@@ -9,6 +15,7 @@ use Data::Dumper;
 use IO::Handle;
 use Time::Local;
 use Time::localtime;
+use Getopt::Std;
 
 # the following is for the status server
 # to communicate with Nagios plugin
@@ -17,9 +24,17 @@ use threads::shared;
 use IO::Socket;
 use IO::Select;
 
+our ($opt_s, $opt_h);
+getopts('hs');
+
+if($opt_h){
+    &showHelp();
+    exit(0);
+}
+
 my $status     : shared = "OK";
 my $time_ls    : shared = &getTime();
-my $sleep_time : shared = 120;
+my $sleep_time : shared = 600;         # 10 minutes
 
 our $server_port = '50502';
 our $protocol    = 'tcp';
@@ -38,33 +53,31 @@ sub main {
     while (1)
     {
 
-       if( abs($time_du - $time_ls) > 60*60*24 ){
+       if( $opt_s && abs($time_du - $time_ls) > 60*60*24 ){
            $lustre_size = &getTotalSize();
            $time_du = &getTime();
        }
        
-       my $line =`ls -d /lustre_alpha/hades/beam/sep08`;
+       my $line =`ls -d /lustre_alpha/hades/user`;
        chomp($line);
        
        $time_ls = &getTime();
 
-       if ($line eq "/lustre_alpha/hades/beam/sep08") 
+       if ($line eq "/lustre_alpha/hades/user") 
        {
-           if ($lustre_size > 30) {
-               $status = "CRITICAL - Lustre hades/beam space used: $lustre_size TB";
-           }
-           elsif ($lustre_size > 25) {
-               $status  = "WARNING - Lustre hades/beam space used: $lustre_size TB";
+           if($opt_s){
+               $status  = "OK - Lustre space used: $lustre_size TB";
+               &checkLustreSize($lustre_size);
            }
-           else {
-               $status  = "OK - Lustre hades/beam space used: $lustre_size TB";
+           else{
+               $status  = "OK - Lustre is mounted";
            }
        }
        else {
-           $status  = "CRITICAL - failure of Lustre file system!";
+           $status  = "CRITICAL - Lustre mount failure!";
        }
        
-       sleep $sleep_time; # 2 minutes
+       sleep $sleep_time; # 10 minutes
     }
 }
 
@@ -78,7 +91,7 @@ sub statusServer{
                      IO::Socket::INET->new( LocalPort => $server_port,
                                             Proto     => 'tcp',
                                             Listen    => SOMAXCONN ) ))
-   {
+    {
         print "ERROR: Cannot start status server!\n";
     }
 
@@ -86,14 +99,14 @@ sub statusServer{
 
     while(1) {
 
-        # wait 5 seconds for connections
+        #--- wait 5 seconds for connections
         while (my @file_handles = $selector->can_read( 5 )) {
 
             foreach my $file_handle (@file_handles) {
 
                 if($file_handle == $server_socket) {
 
-                    # create a new socket for this transaction
+                    #--- create a new socket for this transaction
                     unless (defined( $client_socket = $server_socket->accept() ))
                     {
                         print "ERROR: Cannot open socket to send status!\n";
@@ -137,7 +150,7 @@ sub getTotalSize
     #--- get total size of lustre beam dir
 
     my $totsize = 0;
-    my @size = `du -cms /lustre_alpha/hades/beam`;
+    my @size = `du -cms /lustre_alpha/hades`;
     
     foreach my $line (@size){
        if( $line =~ /total/ ){
@@ -150,6 +163,35 @@ sub getTotalSize
     return $totsize;
 }
 
+sub checkLustreSize
+{
+    my $lustre_size = shift;
+
+    if ($lustre_size > 30) {
+       $status = "CRITICAL - Lustre space used: $lustre_size TB";
+    }
+    elsif ($lustre_size > 25) {
+       $status  = "WARNING - Lustre space used: $lustre_size TB";
+    }
+}
+
+sub showHelp
+{
+    print << 'EOF';
+
+    Nagios plugin  
+       
+      This script checks periodically a Lustre mount on 
+      a local machine. The script also runs a status server 
+      in a separate thread for reporting the status to Nagios.
+      
+    Usage: check_lustre.pl [-s] [-h]
+
+      -s : Check also a disk space used on Lustre file system.
+      -h : Print this help.
+
+EOF
+}