anvil/tools/striker-boot-machine

#!/usr/bin/perl
# 
# This program will boot a target machine using either it's IPMI interface, if available, or one of the 
# (non-PDU) fence methods, if the target is in an Anvil! and we have a manifest for it.
# 
# Exit codes;
# 0 = Normal exit.
# 1 = No database connection.
# 
# TODO: 
# 

use strict;
use warnings;
use Anvil::Tools;

my $THIS_FILE           =  ($0 =~ /^.*\/(.*)$/)[0];
my $running_directory   =  ($0 =~ /^(.*?)\/$THIS_FILE$/)[0];
if (($running_directory =~ /^\./) && ($ENV{PWD}))
{
	$running_directory =~ s/^\./$ENV{PWD}/;
}

# Turn off buffering so that the pinwheel will display while waiting for the SSH call(s) to complete.
$| = 1;

my $anvil = Anvil::Tools->new();

$anvil->Get->switches({list => [
	"host",
	"host-uuid",
	"job-uuid"], man => $THIS_FILE});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => $anvil->data->{switches}});
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2, secure => 0, key => "log_0115", variables => { program => $THIS_FILE }});

$anvil->Database->connect();
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 3, secure => 0, key => "log_0132"});
if (not $anvil->data->{sys}{database}{connections})
{
	# No databases, update the job, sleep for a bit and then exit. The daemon will pick it up and try 
	# again after we exit.
	$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0075"});
	sleep 10;
	$anvil->nice_exit({exit_code => 1});
}

if ($anvil->data->{switches}{'job-uuid'})
{
	# Load the job data.
	$anvil->Job->clear();
	$anvil->Job->get_job_details();
	$anvil->Job->update_progress({
		progress         => 1,
		job_picked_up_by => $$, 
		job_picked_up_at => time, 
		message          => "job_0283", 
	});
	
	# Pull out the job data.
	foreach my $line (split/\n/, $anvil->data->{jobs}{job_data})
	{
		if ($line =~ /host-uuid=(.*?)$/)
		{
			$anvil->data->{switches}{'host-uuid'} = $1;
			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
				'switches::host-uuid' => $anvil->data->{switches}{'host-uuid'},
			}});
		}
		if ($line =~ /host=(.*?)$/)
		{
			$anvil->data->{switches}{host} = $1;
			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
				'switches::host' => $anvil->data->{switches}{host},
			}});
		}
	}
}

# Get the host info (copy host-uuid to host)
$anvil->data->{switches}{'host-name'} = "" if not defined $anvil->data->{switches}{'host-name'};
if ((not $anvil->data->{switches}{host}) && ($anvil->data->{switches}{'host-uuid'}))
{
	$anvil->data->{switches}{host} = $anvil->data->{switches}{'host-uuid'};
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
		'switches::host' => $anvil->data->{switches}{host},
	}});
}

# If the host is 'all', don't translate to the host_uuid. Otherwise, look for the host_uuid from the host string
if ((not $anvil->data->{switches}{'host-uuid'}) && ($anvil->data->{switches}{host} ne "all"))
{
	$anvil->data->{switches}{'host-uuid'} = $anvil->Database->get_host_uuid_from_string({string => $anvil->data->{switches}{host}});
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
		'switches::host-uuid' => $anvil->data->{switches}{'host-uuid'},
	}});
	
	# host name not found
	if (not $anvil->data->{switches}{'host-uuid'})
	{
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0291", variables => { host_name => $anvil->data->{switches}{'host-name'} }});
		$anvil->Job->update_progress({progress => 100, message => "error_0291,!!host_name!".$anvil->data->{switches}{'host-name'}."!!"});
		$anvil->nice_exit({exit_code => 1});
	}
	
	$anvil->data->{switches}{'host-name'} = $anvil->Get->host_name_from_uuid({host_uuid => $anvil->data->{switches}{'host-uuid'}});
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
		'switches::host-name' => $anvil->data->{switches}{'host-name'},
	}});
}

find_boot_method($anvil);

$anvil->nice_exit({exit_code => 0});


#############################################################################################################
# Functions                                                                                                 #
#############################################################################################################

# This will try to boot the node with host_ipmi data, if available. If not, it will try to find a (non-PDU)
# fence method to try
sub find_boot_method
{
	my ($anvil) = @_;
	
	$anvil->Database->get_hosts_info({debug => 2});
	
	my $hosts = [];
	if ($anvil->data->{switches}{'host-uuid'})
	{
		push @{$hosts}, $anvil->data->{switches}{'host-name'};
	}
	elsif ($anvil->data->{switches}{host} eq "all")
	{
		$anvil->Database->get_hosts({debug => 2});
		foreach my $host_name (sort {$a cmp $b} keys %{$anvil->data->{sys}{hosts}{by_name}})
		{
			my $host_uuid = $anvil->data->{sys}{hosts}{by_name}{$host_name};
			my $host_ipmi = $anvil->data->{hosts}{host_uuid}{$host_uuid}{host_ipmi};
			my $host_key  = $anvil->data->{hosts}{host_uuid}{$host_uuid}{host_key};
			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
				's1:host_name' => $host_name,
				's2:host_uuid' => $host_uuid, 
				's3:host_ipmi' => $anvil->Log->is_secure($host_ipmi),
			}});
			next if $host_key eq "DELETED";
			next if not $host_ipmi;
			
			push @{$hosts}, $host_name;
		}
	}
	
	my $host_count = @{$hosts};
	my $steps      = int((80 / $host_count) / 3);
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
		host_count => $host_count, 
		steps      => $steps,
	}});
	
	$anvil->data->{sys}{progress} = 5;
	foreach my $host_name (sort {$a cmp $b} @{$hosts})
	{
		my $host_uuid = $anvil->Database->get_host_uuid_from_string({string => $host_name});
		my $host_ipmi = $anvil->data->{hosts}{host_uuid}{$host_uuid}{host_ipmi};
		my $host_type = $anvil->data->{hosts}{host_uuid}{$host_uuid}{host_ipmi};
		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
			host_uuid => $host_uuid, 
			host_name => $host_name,
			host_ipmi => $anvil->Log->is_secure($host_ipmi),
			host_type => $host_type, 
		}});
		
		$anvil->data->{sys}{progress} += $steps;
		
		if ($host_ipmi)
		{
			# Got it.
			$anvil->data->{sys}{progress} += $steps;
			$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, key => "job_0327", variables => { host_name => $host_name }});
			$anvil->Job->update_progress({progress => $anvil->data->{sys}{progress}, message => "job_0327,!!host_name!".$host_name."!!"});
			
			# First, is the node already on?
			my $shell_call = $host_ipmi." -o status";
			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 1, list => { shell_call => $shell_call }});
			
			$anvil->data->{sys}{progress} += $steps;
			my $problem = call_fence_agent($anvil, $shell_call, $anvil->data->{sys}{progress});
			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 1, list => { problem => $problem }});
			
			$anvil->data->{sys}{progress} += $steps;
			if ($problem)
			{
				# If I am here, there is no IPMI. Can we boot it using another fence method?
				# The machine would have to be in an Anvil! for this to work.
				my $anvil_uuid = $anvil->Cluster->get_anvil_uuid({host_uuid => $host_uuid});
				$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { anvil_uuid => $anvil_uuid }});
				if ($anvil_uuid)
				{
					my $anvil_name = $anvil->Cluster->get_anvil_name({anvil_uuid => $anvil_uuid});
					$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { anvil_name => $anvil_name }});
					$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, key => "job_0331", variables => { 
						host_name  => $host_name, 
						anvil_name => $anvil_name,
					}});
					$anvil->Job->update_progress({progress => $anvil->data->{sys}{progress}, message => "job_0331,!!host_name!".$host_name."!!,!!anvil_name!".$anvil_name."!!"});
					
					$anvil->Cluster->get_fence_methods({host_uuid => $host_uuid});
					foreach my $target_host_name (sort {$a cmp $b} keys %{$anvil->data->{fence_method}})
					{
						$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { target_host_name => $target_host_name }});
						foreach my $order (sort {$a cmp $b} keys %{$anvil->data->{fence_method}{$target_host_name}{order}})
						{
							$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { order => $order }});
							foreach my $method (sort {$a cmp $b} keys %{$anvil->data->{fence_method}{$target_host_name}{order}{$order}{method}})
							{
								next if $method =~ /pdu/;
								my $shell_call = $anvil->data->{fence_method}{$target_host_name}{order}{$order}{method}{$method}{command};
								$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
									method     => $method,
									shell_call => $anvil->Log->is_secure($shell_call),
								}});
								my $problem = call_fence_agent($anvil, $shell_call, $anvil->data->{sys}{progress});
								$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { problem => $problem }});
								
								if ($problem)
								{
									$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0297"});
									$anvil->Job->update_progress({progress => $anvil->data->{sys}{progress}, message => "error_0297"});
								}
							}
						}
					}
				}
				else
				{
					# Nothing we can do to boot this machine.
					$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0294"});
					$anvil->Job->update_progress({progress => $anvil->data->{sys}{progress}, message => "error_0294"});
				}
				
			}
		}
	}
	
	# Done.
	$anvil->Job->update_progress({progress => 100, message => "message_0025"});
	$anvil->nice_exit({exit_code => 0});
	
	return(0);
}

# This calls a fence agent and exits if it successfully boots the target
sub call_fence_agent
{
	my ($anvil, $shell_call, $progress) = @_;
	
	my ($output, $return_code) = $anvil->System->call({shell_call => $shell_call, secure => 1});
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
		output      => $output, 
		return_code => $return_code,
	}});
	
	if ($return_code eq "0")
	{
		# The machine is already on
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0328"});
		$anvil->Job->update_progress({progress => 100, message => "job_0328"});
		return(0);
	}
	elsif ($return_code eq "1")
	{
		# Unable to connect to the fence device.
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0293", variables => { output => $output }});
		$anvil->Job->update_progress({progress => 100, message => "error_0293,!!output!".$output."!!"});
		return(1);
	}
	elsif ($return_code eq "2")
	{
		# The machine is off, try to start it.
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, key => "job_0329"});
		$anvil->Job->update_progress({progress => $anvil->data->{sys}{progress}+=10, message => "job_0329"});
		
		# First, is the node already on?
		$shell_call .= " -o on";
		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 1, list => { shell_call => $shell_call }});
		
		my ($output, $return_code) = $anvil->System->call({shell_call => $shell_call, secure => 1});
		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
			output      => $output, 
			return_code => $return_code,
		}});
		
		if ($return_code eq "0")
		{
			# Success!
			$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, key => "job_0330"});
			$anvil->Job->update_progress({progress => 100, message => "job_0330"});
			
			# Update the host's status to 'booting' and exit
			$anvil->Database->update_host_status({
				debug       => 2,
				host_uuid   => $anvil->data->{switches}{'host-uuid'}, 
				host_status => "booting", 
			});
			return(0);
		}
		else
		{
			# Failed. 
			$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0292", variables => { output => $output }});
			$anvil->Job->update_progress({progress => 100, message => "error_0292,!!output!".$output."!!"});
			return(1);
		}
	}
	
	return(0);
}