mirror of
https://github.com/xcat2/xcat-core.git
synced 2025-10-31 11:22:27 +00:00
350 lines
13 KiB
Perl
Executable File
350 lines
13 KiB
Perl
Executable File
#! /usr/bin/perl
|
|
# IBM(c) 2016 EPL license http://www.eclipse.org/legal/epl-v10.html
|
|
|
|
BEGIN { $::XCATROOT = $ENV{'XCATROOT'} ? $ENV{'XCATROOT'} : -d '/opt/xcat' ? '/opt/xcat' : '/usr'; }
|
|
|
|
use lib "$::XCATROOT/probe/lib/perl";
|
|
use probe_utils;
|
|
use File::Basename;
|
|
use Net::Ping;
|
|
use Getopt::Long qw(:config no_ignore_case);
|
|
|
|
#use Data::Dumper;
|
|
use warnings;
|
|
|
|
my $program_name = basename("$0");
|
|
my $help;
|
|
my $noderange = "";
|
|
my $test;
|
|
my $output = "stdout";
|
|
my $verbose = 0;
|
|
my $rst = 0;
|
|
|
|
$::USAGE = "Usage:
|
|
$program_name -h
|
|
$program_name {-c|-d} [-n noderange] [-V]
|
|
|
|
Description:
|
|
Use this command to check if diskless, pingable compute nodes have the same images installed as defines in xCAT DB.
|
|
Use this command to check if all diskless, pingable compute nodes have the same identical installed.
|
|
|
|
Options:
|
|
-h : Get usage information of $program_name
|
|
-n : Range of nodes to check
|
|
-d : To verify diskless, pingable compute nodes have the same images installed as defines in xCAT DB.
|
|
-c : To verify all diskless, pingable compute nodes have the identical images installed.
|
|
-V : To print additional debug information.
|
|
";
|
|
|
|
#-------------------------------------
|
|
# main process
|
|
#-------------------------------------
|
|
if (
|
|
!GetOptions("--help|h" => \$help,
|
|
"T" => \$test,
|
|
"V" => \$VERBOSE,
|
|
"n=s" => \$noderange,
|
|
"c" => \$CONSISTENCY_CHECK,
|
|
"d" => \$DEFINITION_CHECK))
|
|
{
|
|
probe_utils->send_msg("$output", "f", "Invalid parameter for $program_name");
|
|
probe_utils->send_msg("$output", "d", "$::USAGE");
|
|
exit 1;
|
|
}
|
|
|
|
if ($help) {
|
|
if ($output ne "stdout") {
|
|
probe_utils->send_msg("$output", "d", "$::USAGE");
|
|
} else {
|
|
print "$::USAGE";
|
|
}
|
|
exit 0;
|
|
}
|
|
|
|
if ($test) {
|
|
probe_utils->send_msg("$output", "o", "Use this command to check if specified compute nodes have the same images installed or if compute nodes are installed with the same image as defined on MN.");
|
|
exit 0;
|
|
}
|
|
|
|
unless (defined($CONSISTENCY_CHECK) || defined($DEFINITION_CHECK)) {
|
|
probe_utils->send_msg("$output", "f", "At least one of -c or -d flags is required");
|
|
probe_utils->send_msg("$output", "d", "$::USAGE");
|
|
exit 1;
|
|
}
|
|
|
|
if (scalar(@ARGV) >= 1) {
|
|
|
|
# After processing all the expected flags and arguments,
|
|
# there is still left over stuff on the command line
|
|
probe_utils->send_msg("$output", "f", "Invalid flag or parameter: @ARGV");
|
|
probe_utils->send_msg("$output", "d", "$::USAGE");
|
|
exit 1;
|
|
}
|
|
|
|
my @pingable_nodes;
|
|
my @diskless_nodes;
|
|
my $na = "N/A";
|
|
|
|
my $defined_UUID = $na;
|
|
my %node_defined_image_uuid_hash;
|
|
my %node_defined_image_name_hash;
|
|
my %osimage_defined_provmethod_hash;
|
|
|
|
my $all_nodes_provmethod = `lsdef -i provmethod -c $noderange`;
|
|
my $all_osimage_provmethod = `lsdef -t osimage -i provmethod,rootimgdir -c`;
|
|
chomp($all_nodes_provmethod);
|
|
my @all_nodes_provmethod_lines = split("[\n\r]", $all_nodes_provmethod);
|
|
my @all_osimage_provmethod_lines = split("[\n\r]", $all_osimage_provmethod);
|
|
|
|
if ($all_nodes_provmethod =~ /Usage:/) {
|
|
|
|
# lsdef command displayed a Usage message. Must be some noderange formatting problem.
|
|
# Issue a warning and exit.
|
|
probe_utils->send_msg("$output", "w", "Can not get a list of nodes from specified noderange.");
|
|
exit 1;
|
|
}
|
|
|
|
if (scalar(@all_nodes_provmethod_lines) <= 0) {
|
|
|
|
# There were no nodes matching the noderange. Issue a warning and exit.
|
|
probe_utils->send_msg("$output", "w", "No nodes matching the noderange were found.");
|
|
exit 1;
|
|
}
|
|
|
|
# Build a hash of key="osimage name + attribute name" value="provmethod and rootimgdir attribute value"
|
|
foreach (@all_osimage_provmethod_lines) {
|
|
my ($osimage_name, $values) = split "=", $_;
|
|
$osimage_defined_provmethod_hash{$osimage_name} = $values;
|
|
}
|
|
|
|
# First, extract diskless nodes
|
|
foreach (@all_nodes_provmethod_lines) {
|
|
|
|
# Get osimage name for the node
|
|
my ($node_name, $junk, $node_osimage_name) = split "[:=]", $_;
|
|
chomp($node_osimage_name);
|
|
|
|
if (length($node_osimage_name) > 0) {
|
|
|
|
# Get provmethod and rootimgdir for the osimage
|
|
my $osimage_provmethod_type = $osimage_defined_provmethod_hash{ $node_osimage_name . ": provmethod" };
|
|
my $rootimagedir = $osimage_defined_provmethod_hash{ $node_osimage_name . ": rootimgdir" };
|
|
chomp($osimage_provmethod_type) if ($osimage_provmethod_type);
|
|
chomp($rootimagedir) if ($rootimagedir);
|
|
|
|
# Check if it is netboot, meaning diskless
|
|
if ($osimage_provmethod_type && $osimage_provmethod_type eq 'netboot') {
|
|
push(@diskless_nodes, $node_name);
|
|
probe_utils->send_msg("$output", "o", "$node_name is diskless") if ($VERBOSE);
|
|
if (length($rootimagedir) > 0) {
|
|
|
|
# For this diskless node, get UUID from rootimg directory xcatinfo file of the provmethod osimage
|
|
my $xcatinfo_file = $rootimagedir . "/rootimg/opt/xcat/xcatinfo";
|
|
if (-r $xcatinfo_file) {
|
|
$defined_UUID = `awk -F"'" '/IMAGEUUID/ {print \$2}' $xcatinfo_file`;
|
|
chomp($defined_UUID);
|
|
if (length($defined_UUID) < 1) {
|
|
$defined_UUID = $na;
|
|
}
|
|
}
|
|
}
|
|
$node_defined_image_uuid_hash{$node_name} = $defined_UUID;
|
|
$node_defined_image_name_hash{$node_name} = $node_osimage_name;
|
|
print "Node $node_name has defined image $node_osimage_name at $rootimagedir with UUID $defined_UUID\n" if ($VERBOSE);
|
|
}
|
|
else {
|
|
probe_utils->send_msg("$output", "w", "$node_name is not diskless. No image consistency verification will be performed.");
|
|
}
|
|
}
|
|
else {
|
|
probe_utils->send_msg("$output", "w", "$node_name has no provision method defined. No image consistency verification will be performed.");
|
|
}
|
|
}
|
|
|
|
if (scalar(@diskless_nodes) <= 0) {
|
|
|
|
# There were no diskless nodes found. Issue a warning and exit.
|
|
probe_utils->send_msg("$output", "w", "No diskless compute nodes were found.");
|
|
exit 1;
|
|
}
|
|
|
|
if (scalar(@diskless_nodes) <= 0) {
|
|
|
|
# There were no diskless nodes found. Issue a warning and exit.
|
|
probe_utils->send_msg("$output", "w", "No diskless compute nodes were found");
|
|
exit 1;
|
|
}
|
|
|
|
# Next, check if all diskless nodes are pingable
|
|
my $ping_hosts = join ",", @diskless_nodes;
|
|
my $pping_output = `pping $ping_hosts`;
|
|
chomp($pping_output);
|
|
my @pping_lines = split("[\n\r]", $pping_output);
|
|
foreach (@pping_lines) {
|
|
my ($hostname, $result) = split ":", $_;
|
|
my ($token, $status) = split ' ', $result;
|
|
chomp($token);
|
|
if ($token ne 'ping') {
|
|
probe_utils->send_msg("$output", "f", "Pinging $hostname");
|
|
}
|
|
else {
|
|
probe_utils->send_msg("$output", "o", "Pinging $hostname") if ($VERBOSE);
|
|
push(@pingable_nodes, $hostname);
|
|
}
|
|
}
|
|
|
|
if (scalar(@pingable_nodes) <= 0) {
|
|
|
|
# There were no pingable, diskless nodes found. Issue a warning and exit.
|
|
probe_utils->send_msg("$output", "w", "No diskless, pingable compute nodes were found");
|
|
exit 1;
|
|
}
|
|
|
|
if ((scalar(@pingable_nodes) == 1) && ($CONSISTENCY_CHECK)) {
|
|
|
|
# There was only one node in noderange and comparison check was requested.
|
|
# Nothing to compare the single node to.
|
|
probe_utils->send_msg("$output", "w", "Comparison check for a single diskless pingable node will not be performed. Minimum of 2 nodes are needed for that.");
|
|
exit 1;
|
|
}
|
|
|
|
# Next, from all pingable nodes get the IMAGENAME and IMAGEUUID entries from xcatinfo file
|
|
probe_utils->send_msg("$output", "d", "---- Gathering information from all diskless pingable compute nodes ----");
|
|
|
|
my $pingable_hostname_list = join ",", @pingable_nodes;
|
|
my $all_xdsh_output = `xdsh $pingable_hostname_list "cat /opt/xcat/xcatinfo"`;
|
|
my $xcatinfo_image_UUID = ` echo "$all_xdsh_output" | awk -F"=" '/IMAGEUUID/ {gsub(/IMAGEUUID/,"",\$1); gsub(/'"'"'/,"",\$2);; print \$1 \$2}'`;
|
|
|
|
# Check to verify xdsh worked and returned some usefull information
|
|
if (length($xcatinfo_image_UUID) <= 1) {
|
|
probe_utils->send_msg("$output", "w", "Unable to extract image UUID information from compute nodes using xdsh command. No image consistency verification will be performed.");
|
|
exit 1;
|
|
}
|
|
my @xdsh_UUID_lines = split("[\n\r]", $xcatinfo_image_UUID);
|
|
|
|
my $xcatinfo_image_name = ` echo "$all_xdsh_output" | awk -F"=" '/IMAGENAME/ {gsub(/IMAGENAME/,"",\$1); gsub(/'"'"'/,"",\$2); print \$1 \$2}'`;
|
|
|
|
# Check to verify xdsh worked and returned some usefull information
|
|
if (length($xcatinfo_image_name) <= 1) {
|
|
probe_utils->send_msg("$output", "w", "Unable to extract image name information from compute nodes using xdsh command. No image consistency verification will be performed.");
|
|
exit 1;
|
|
}
|
|
my @xdsh_name_lines = split("[\n\r]", $xcatinfo_image_name);
|
|
|
|
my %node_running_image_uuid_hash;
|
|
my %node_running_image_name_hash;
|
|
|
|
# Build a hash of key=hostname, value=running UUID
|
|
foreach (@xdsh_UUID_lines) {
|
|
my ($hostname, $uuid) = split ": ", $_;
|
|
chomp($hostname);
|
|
chomp($uuid);
|
|
if (length($uuid) > 0) {
|
|
$node_running_image_uuid_hash{$hostname} = $uuid;
|
|
}
|
|
else {
|
|
$node_running_image_uuid_hash{$hostname} = $na;
|
|
}
|
|
}
|
|
|
|
# It is possible that some older version xCAT compute nodes will not have an IMAGEUUID line in
|
|
# the xcatinfo file, for those nodes insert $na as the running UUID value
|
|
foreach (@pingable_nodes) {
|
|
unless (exists($node_running_image_uuid_hash{$_})) {
|
|
$node_running_image_uuid_hash{$_} = $na;
|
|
}
|
|
}
|
|
|
|
# Build a hash of key="hostname", value="running OS image name"
|
|
foreach (@xdsh_name_lines) {
|
|
my ($hostname, $osimage_name) = split ": ", $_;
|
|
chomp($hostname);
|
|
chomp($osimage_name);
|
|
if (length($osimage_name) > 0) {
|
|
$node_running_image_name_hash{$hostname} = $osimage_name;
|
|
}
|
|
else {
|
|
$node_running_image_name_hash{$hostname} = $na;
|
|
}
|
|
}
|
|
|
|
# print Dumper(\%node_running_image_uuid_hash);
|
|
# print Dumper(\%node_running_image_name_hash);
|
|
|
|
# Information gathering is done. Now do veification checking.
|
|
|
|
# Probe verification step 1 - make sure all nodes are installed with the osimage name and imageUUID as defined on MN
|
|
if ($DEFINITION_CHECK) {
|
|
my $success_nodes = 0;
|
|
my $msg;
|
|
foreach (@pingable_nodes) {
|
|
if (($node_running_image_name_hash{$_} eq $node_defined_image_name_hash{$_}) &&
|
|
($node_running_image_uuid_hash{$_} eq $node_defined_image_uuid_hash{$_})) {
|
|
if ($node_running_image_uuid_hash{$_} eq $na) {
|
|
$msg = "$_: Not able to determine installed os image name or uuid";
|
|
}
|
|
else {
|
|
$msg = "OS image installed on compute node $_ matches the image defined for it on management node";
|
|
probe_utils->send_msg("$output", "o", "$msg") if ($VERBOSE);
|
|
$success_nodes++;
|
|
next;
|
|
}
|
|
}
|
|
else {
|
|
$msg = "$_: Unmatched os image name or image UUID.\n Defined: name = $node_defined_image_name_hash{$_}" .
|
|
" uuid = $node_defined_image_uuid_hash{$_}\n Installed: name = $node_running_image_name_hash{$_}" .
|
|
" uuid = $node_running_image_uuid_hash{$_}";
|
|
}
|
|
probe_utils->send_msg("$output", "f", "$msg");
|
|
}
|
|
if (scalar(@pingable_nodes) eq $success_nodes) {
|
|
|
|
# All pingable nodes were tested with success
|
|
probe_utils->send_msg("$output", "o", "OS image installed on each diskless compute node matches the image defined for it on management node");
|
|
}
|
|
}
|
|
|
|
# Probe verification step 2 - make sure all nodes are installed with the same osimage name and imageUUID
|
|
if ($CONSISTENCY_CHECK) {
|
|
my $msg = "Undefined";
|
|
my $status = "f";
|
|
my $image_name_and_uuid;
|
|
my $image_uuid;
|
|
my %unique_image_hash;
|
|
|
|
# Go throug the nodes and build a hash of key=image_name+image_uuid and value of nodename
|
|
foreach (@pingable_nodes) {
|
|
$image_name_and_uuid = $node_running_image_name_hash{$_} . ":" . $node_running_image_uuid_hash{$_};
|
|
unless (exists $unique_image_hash{$image_name_and_uuid}) {
|
|
$unique_image_hash{$image_name_and_uuid} = $_;
|
|
}
|
|
}
|
|
|
|
# print Dumper(\%unique_image_hash);
|
|
# If there is more then one key in the hash, nodes have multiple images.
|
|
my $number_of_keys = keys %unique_image_hash;
|
|
if ($number_of_keys == 1) {
|
|
my @image_names = keys %unique_image_hash;
|
|
if ($image_names[0] =~ /$na/) {
|
|
$msg = "Not able to determine os image name or uuid of the image installed on any compute node.";
|
|
$status = "f";
|
|
}
|
|
else {
|
|
$msg = "All compute nodes have the same os image installed: @image_names.";
|
|
$status = "o";
|
|
}
|
|
}
|
|
else {
|
|
my $node_image_table;
|
|
foreach $compute_node (keys %node_running_image_name_hash) {
|
|
$node_image_table .= sprintf(" %-15s %-30s : %-20s\n", $compute_node, $node_running_image_name_hash{$compute_node}, $node_running_image_uuid_hash{$compute_node});
|
|
}
|
|
$msg = "Not all compute nodes are installed with the same os image.\n" . $node_image_table;
|
|
$status = "f";
|
|
}
|
|
|
|
probe_utils->send_msg("$output", "$status", "$msg");
|
|
}
|
|
|
|
exit 0;
|