anvil/tools/anvil-shutdown-server

#!/usr/bin/perl
# 
# This program shuts downs a server (or servers). It can be called as either a job from the webui or directly
# from another program or a terminal.
# 
# Exit codes;
# 0 = Normal exit.
# 1 = No database connection.
# 
# TODO: 
# - We need to support shutdown ordering (inverese of boot ordering)
# 

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();

# Read switches (target ([user@]host[:port]) and the file with the target's password.
$anvil->Get->switches({list => [
	"immediate",
	"no-db", 
	"no-wait", 
	"server", 
	"server-uuid", 
	"wait"], 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__, level => 2, key => "log_0115", variables => { program => $THIS_FILE }});

# Connect to DBs.
if ($anvil->data->{switches}{'no-db'})
{
	$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, secure => 0, key => "log_0743"});
	
	# If there was a job-uuid, clear it.
	$anvil->data->{sys}{database}{connections} = 0;
	$anvil->data->{switches}{'job-uuid'}       = "";
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
		'sys::database::connections' => $anvil->data->{sys}{database}{connections},
		'switches::job-uuid'         => $anvil->data->{switches}{'job-uuid'},
	}});
}
else
{
	$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 =~ /server=(.*?)$/)
		{
			$anvil->data->{switches}{'server'} = $1;
			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
				'switches::server' => $anvil->data->{switches}{'server'},
			}});
		}
		if ($line =~ /server-uuid=(.*?)$/)
		{
			$anvil->data->{switches}{'server-uuid'} = $1;
			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
				'switches::server-uuid' => $anvil->data->{switches}{'server-uuid'},
			}});
		}
	}
}

# Now check that we have a server. If it's a server_uuid, read the server name.
if ($anvil->data->{switches}{'server-uuid'})
{
	# DO we have DB connection(s)?
	if (not $anvil->data->{sys}{database}{connections})
	{
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0265"});
		$anvil->Job->update_progress({progress => 100, message => "error_0265"});
		$anvil->nice_exit({exit_code => 1});
	}
	
	# Convert the server_uuid to a server_name.
	my $query = "SELECT server_name FROM servers WHERE server_uuid = ".$anvil->Database->quote($anvil->data->{switches}{'server-uuid'}).";";
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { query => $query }});
	
	my $server_name = $anvil->Database->query({query => $query, source => $THIS_FILE, line => __LINE__})->[0]->[0];
	   $server_name = "" if not defined $server_name;
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { server_name => $server_name }});
	
	if ($server_name)
	{
		$anvil->data->{switches}{'server'} = $server_name;
		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
			'switches::server' => $anvil->data->{switches}{'server'},
		}});
	}
	else
	{
		# Invalid server UUID. 
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0269", variables => {
			server_uuid => $anvil->data->{switches}{'server-uuid'}, 
		}});
		$anvil->Job->update_progress({progress => 100, message => "error_0269,!!server_uuid!".$anvil->data->{switches}{'server-uuid'}."!!"});
		$anvil->nice_exit({exit_code => 1});
	}
}

# Do we have a server name?
if (not $anvil->data->{switches}{'server'})
{
	# Unable to proceed.
	$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0263"});
	$anvil->Job->update_progress({progress => 100, message => "error_0263"});
	$anvil->nice_exit({exit_code => 1});
}

# If we're forcing off a server, make sure '--server' isn't 'all'.
if (($anvil->data->{switches}{'server'} =~ /all/i) && ($anvil->data->{switches}{immediate}))
{
	# Unable to proceed.
	$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0462"});
	$anvil->Job->update_progress({progress => 100, message => "error_0462"});
	$anvil->nice_exit({exit_code => 1});
}

# Are we a node or DR host?
$anvil->data->{sys}{host_type} = $anvil->Get->host_type();
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
	'sys::host_type' => $anvil->data->{sys}{host_type},
}});
if (($anvil->data->{sys}{host_type} ne "node") && ($anvil->data->{sys}{host_type} ne "dr"))
{
	$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0264"});
	$anvil->Job->update_progress({progress => 100, message => "error_0264"});
	$anvil->nice_exit({exit_code => 1});
}

# Make sure that we're in an Anvil! system.
$anvil->data->{sys}{anvil_uuid} = "";
if (($anvil->data->{sys}{host_type} eq "node") && ($anvil->data->{sys}{database}{connections}))
{
	$anvil->data->{sys}{anvil_uuid} = $anvil->Cluster->get_anvil_uuid();
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
		'sys::anvil_uuid' => $anvil->data->{sys}{anvil_uuid},
	}});
	if (not $anvil->data->{sys}{anvil_uuid})
	{
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0260"});
		$anvil->Job->update_progress({progress => 100, message => "error_0260"});
		$anvil->nice_exit({exit_code => 1});
	}
}

# This is copied from anvil-boot-server, but it works here as well. We can't use 'pcs' without pacemaker 
# being up.
if ($anvil->data->{sys}{host_type} eq "node")
{
	wait_for_pacemaker($anvil);
}

# If 'server' is 'all', shut down all servers.
if (lc($anvil->data->{switches}{'server'}) eq "all")
{
	shutdown_all_servers($anvil);
}
else
{
	my $wait = $anvil->data->{switches}{'no-wait'} ? 0 : 1;
	shutdown_server($anvil, $anvil->data->{switches}{'server'}, $wait, 50);
}

$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0281"});
$anvil->Job->update_progress({progress => 100, message => "job_0281"}) if $anvil->data->{switches}{'job-uuid'};

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


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

sub wait_for_pacemaker
{
	my ($anvil) = @_;

	# Shutdown the server using pcs, but of course, wait for the node to be up.
	my $waiting = 1;
	while($waiting)
	{
		my $problem = $anvil->Cluster->parse_cib({debug => 2});
		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { problem => $problem }});
		if (not $problem)
		{
			my $node_name = $anvil->data->{cib}{parsed}{'local'}{name};
			my $ready     = $anvil->data->{cib}{parsed}{data}{node}{$node_name}{node_state}{ready};
			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { ready => $ready }});
			if ($ready)
			{
				# We're good. 
				$waiting = 0;
				$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { waiting => $waiting }});
				$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0279"});
				$anvil->Job->update_progress({progress => 15, message => "job_0279"}) if $anvil->data->{switches}{'job-uuid'};
			}
			else
			{
				# Node isn't ready yet.
				$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0278"});
				$anvil->Job->update_progress({progress => 10, message => "job_0278"}) if $anvil->data->{switches}{'job-uuid'};
			}
		}
		else
		{
			# Cluster hasn't started.
			$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0277"});
			$anvil->Job->update_progress({progress => 5, message => "job_0277"}) if $anvil->data->{switches}{'job-uuid'};
		}
		if ($waiting)
		{
			sleep 10;
		}
	}
	
	return(0);
}

sub shutdown_server
{
	my ($anvil, $server, $wait, $progress) = @_;
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
		server   => $server,
		'wait'   => $wait, 
		progress => $progress, 
	}});
	
	if (($wait) && ($anvil->data->{switches}{immediate}))
	{
		$wait = 0;
		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 'wait' => $wait }});
	}
	
	# Is the server in the cluster?
	if (not exists $anvil->data->{cib}{parsed}{data}{server}{$server})
	{
		# Nope.
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "log_0548", variables => { server => $server }});
		$anvil->Job->update_progress({progress => 100, message => "log_0548,!!server!".$server."!!"}) if $anvil->data->{switches}{'job-uuid'};
		$anvil->nice_exit({exit_code => 1});
	}
	
	my $status = $anvil->data->{cib}{parsed}{data}{server}{$server}{status};
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { status => $status }});
	if ($status eq "off")
	{
		# It's off already
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0284", variables => { server => $server }});
		$anvil->Job->update_progress({progress => $progress, message => "job_0284,!!server!".$server."!!"}) if $anvil->data->{switches}{'job-uuid'};
		return(0);
	}
	
	# Now shut down. Forcibly?
	if ($anvil->data->{switches}{immediate})
	{
		# Kill it!
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0474", variables => { server => $server }});
		$anvil->Job->update_progress({progress => $progress, message => "job_0474,!!server!".$server."!!"}) if $anvil->data->{switches}{'job-uuid'};
	}
	else
	{
		# Lets be gentle
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0289", variables => { server => $server }});
		$anvil->Job->update_progress({progress => $progress, message => "job_0289,!!server!".$server."!!"}) if $anvil->data->{switches}{'job-uuid'};
	}
	
	my $problem = 0;
	if ($anvil->Get->host_type eq "dr")
	{
		# Shut down using virsh. Invert the return.
		my $force   = $anvil->data->{switches}{immediate} ? 1 : 0;
		my $success = $anvil->Server->shutdown_virsh({
			debug     => 2,
			server    => $server, 
			force     => $force, 
			wait_time => $wait ? 0 : 1,
		});
		$problem = $success ? 0 : 1;
		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
			force   => $force, 
			success => $success,
			problem => $problem, 
		}});
	}
	else
	{
		$problem = $anvil->Cluster->shutdown_server({
			debug  => 2, 
			server => $server, 
			'wait' => $wait,
		});
		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { problem => $problem }});
		
		# If we're being asked to force the server off, do it now regardless of what pacemaker returned.
		if ($anvil->data->{switches}{immediate})
		{
			my $success = $anvil->Server->shutdown_virsh({
				debug     => 2,
				server    => $server, 
				force     => 1, 
				wait_time => 0,
			});
			$problem = $success ? 0 : 1;
			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
				success => $success,
				problem => $problem, 
			}});
		}
	}
	
	if ($problem)
	{
		# Failed, abort.
		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0268", variables => { server => $server }});
		$anvil->Job->update_progress({progress => 100, message => "error_0268,!!server!".$server."!!"}) if $anvil->data->{switches}{'job-uuid'};
		$anvil->nice_exit({exit_code => 1});
	}
	else
	{
		if ($wait)
		{
			# Stopped!
			$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0285", variables => { server => $server }});
			$anvil->Job->update_progress({progress => $progress, message => "job_0285,!!server!".$server."!!"}) if $anvil->data->{switches}{'job-uuid'};
		}
		else
		{
			# Stop requested.
			$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0286", variables => { server => $server }});
			$anvil->Job->update_progress({progress => $progress, message => "job_0286,!!server!".$server."!!"}) if $anvil->data->{switches}{'job-uuid'};
		}
	}
	
	return(0);
}

sub shutdown_all_servers
{
	my ($anvil) = @_;
	
	### TODO: Manage the stop order here, inverse of boot order.
	# We top out at 90, bottom is 20. 
	my $server_count = keys %{$anvil->data->{cib}{parsed}{data}{server}};
	my $increment    = $server_count ? int(70 / $server_count) : 70;
	my $percent      = 15;
	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
		server_count => $server_count, 
		increment    => $increment,
	}});
	foreach my $server (sort {$a cmp $b} keys %{$anvil->data->{cib}{parsed}{data}{server}})
	{
		my $status    = $anvil->data->{cib}{parsed}{data}{server}{$server}{status};
		my $host_name = $anvil->data->{cib}{parsed}{data}{server}{$server}{host_name};
		my $role      = $anvil->data->{cib}{parsed}{data}{server}{$server}{role};
		my $active    = $anvil->data->{cib}{parsed}{data}{server}{$server}{active};
		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
			's1:server'    => $server,
			's2:status'    => $status,
			's2:host_name' => $host_name,
			's4:role'      => $role,
			's5:active'    => $active, 
		}});
		
		if ($status ne "off")
		{
			# Shut it down (don't wait).
			my $wait    =  $anvil->data->{switches}{'wait'} ? 1 : 0;
			   $percent += $increment;
			shutdown_server($anvil, $server, $wait, $percent);
		}
	}
	
	return(0);
}