#! /usr/bin/env perl

# This script calls xCAT command 'nodestat' and 'xdsh' to
# implement a status monitoring tool like CSM command 'csmstat
#
# Currently, it only runs as passing the flag:
#   -d: -s Status,Network-Interfaces for 'csmstat' command:
# #/opt/csm/bin/csmstat -d: -s Status,Network-Interfaces

use Getopt::Long;
Getopt::Long::Configure("bundling");
Getopt::Long::Configure("pass_through");

my $checkpowerstate = 0;

my %nodestat;
my @sshdnodes;
my @nopingnodes;

# Parse the arguments
my $DILIMITER;
my $SELECTOR;

$::USAGE = "Usage: xcatstat [noderange]

This command can be used to display the OS and network interface status (On or Off status) for compute nodes which are managed by current xCAT MN.

The output is similar as the CSM command csmstat: /opt/csm/bin/csmstat -d: -s Status,Network-Interfaces

noderange can be any valid xCAT noderange. If it's ignored, the nodes in the all group will be operated.

Options:
  -h : Display this usage message.

Output:
  For OS status: 1 means on; 0 means off; 127 means unknown
  For nic interface status: the format is <nicname-<status>>. 1 means on; 0 means off; 127 means unknown

Example: 
  Get the status of three nodes: rh6mn,rh7mn,rh7cn
   # ./xcatstat rh6mn,rh7mn,rh7cn
     # Hostname:Status:Network-Interfaces:
       rh6mn:1:eth0-1:
       rh7cn:0:127:
       rh7mn:1:docker0-2:eth0-1:

Author:  Wang, Xiao Peng\n";

if (!GetOptions(
        'h|help' => \$::HELP,)) {
    print $::USAGE;
    exit 1;
}

if ($::HELP) { print $::USAGE; exit 0; }

# Set the default value
$DILIMITER = ':';
$SELECTOR  = "Status,Network-Interfaces";

# Get the noderange
my $noderange;
if ($#ARGV < 0) {
    $noderange = "all";
} else {
    $noderange = $ARGV[0];
}

# Execute 'nodestat' to get the node status
my $cmd    = "/opt/xcat/bin/nodestat $noderange 2>/dev/null";
my @output = `$cmd`;

foreach my $line (@output) {
    if ($line =~ /^(.+): +(.+)$/) {
        my $nodename = $1;
        my $value    = $2;
        if ($value =~ /sshd/) {    # it means the node has ssh enabled
            $nodestat{$nodename}{status}      = "on";
            $nodestat{$nodename}{powerstatus} = "on";
            push @sshdnodes, $nodename;
        }
        elsif ($value =~ /ping/) {    # it means the node is pingable
            $nodestat{$nodename}{status}      = "off";
            $nodestat{$nodename}{powerstatus} = "on";
        }
        else {
            $nodestat{$nodename}{status}      = "unknown";
            $nodestat{$nodename}{powerstatus} = "unknown";
            push @nopingnodes, $nodename;
        }
    }
}

# Check the Power Status for the nodes which are noping
if ($checkpowerstate) {
    $cmd    = "/opt/xcat/bin/rpower $nopingnodes stat 2>/dev/null";
    @output = `$cmd`;

    foreach my $line (@output) {
        if ($line =~ /^(.+): +(.+)$/) {
            my $nodename = $1;
            my $value    = $2;
            if ($vlaue =~ /^on/) {
                $nodestat{$nodename}{powerstatus} = "on";
            } elsif ($value =~ /^off/) {
                $nodestat{$nodename}{powerstatus} = "off";
            } else {
                $nodestat{$nodename}{powerstatus} = "unknown";
            }
        }
    }
}

# Check the network interfaces for the nodes which are sshd
if (@sshdnodes) {
    $cmd = "/opt/xcat/bin/xdsh " . join(',', @sshdnodes) . " ifconfig -a 2>/dev/null";
    @output = `$cmd`;

    my $nodename;
    my $ifname;
    foreach my $line (@output) {
        if ($line =~ /^(.+): +(.+): (flags=.*)$/) {    # for aix and rh7
            $nodename = $1;
            $ifname   = $2;
            if ($3 =~ /RUNNING/) {
                $nodestat{$nodename}{netif}{$ifname} = "on";
            } else {
                $nodestat{$nodename}{netif}{$ifname} = "off";
            }
        } elsif ($line =~ /^(.+): ([^ ]+) +Link/) {    # for rh6
            $nodename                            = $1;
            $ifname                              = $2;
            $nodestat{$nodename}{netif}{$ifname} = "off";
        } elsif ($line =~ /UP.*RUNNING/) {
            $nodestat{$nodename}{netif}{$ifname} = "on";
        }
    }
}

my @statusoutput;
if ($DILIMITER) {
    foreach my $node (keys %nodestat) {
        my $statusstring;
        if (defined($nodestat{$node}{status})
            && $nodestat{$node}{status} eq "on") {
            $statusstring = "$node$DILIMITER" . "1"; # 1 means on; 0 means off; 127 means unknown
        } elsif (defined($nodestat{$node}{status})
            && $nodestat{$node}{status} eq "unknown") {
            $statusstring = "$node$DILIMITER" . "127";
        } else {
            $statusstring = "$node$DILIMITER" . "0";
        }

        if (defined($nodestat{$node}{netif})) {
            foreach my $if (sort (keys %{ $nodestat{$node}{netif} })) {
                if ($if =~ /^lo[\d]*/) {
                    next;
                }
                if ($nodestat{$node}{netif}{$if} eq "on") {
                    $statusstring .= "$DILIMITER" . "$if-1"; # 1 means online; 2 means offline
                } else {
                    $statusstring .= "$DILIMITER" . "$if-2";
                }
            }
        } else {
            if ($statusstring) {
                $statusstring .= "$DILIMITER" . "127";
            } else {
                $statusstring = "$node$DILIMITER" . "127";
            }
        }

        if ($statusstring) {
            push @statusoutput, $statusstring;
        }
    }
}

print "# Hostname:Status:Network-Interfaces:\n";
foreach (sort @statusoutput) {
    print "$_ :\n";
}

exit 0;