Added loading existing corosync/authkey from peer during rebuild.

Signed-off-by: digimer <mkelly@alteeve.ca>

Anvil/Tools.pm

@@ -1061,6 +1061,7 @@ sub _set_paths
 				'autoindex.conf'		=>	"/etc/httpd/conf.d/autoindex.conf", 
 				'cib.xml'			=>	"/var/lib/pacemaker/cib/cib.xml",
 				'corosync.conf'			=>	"/etc/corosync/corosync.conf",
+				'corosync-authkey'		=>	"/etc/corosync/authkey",
 				'dhcpd.conf'			=>	"/etc/dhcp/dhcpd.conf",
 				'dnf.conf'			=>	"/etc/dnf/dnf.conf",
 				'drbd-proxy.license'		=>	"/etc/drbd-proxy.license",

share/words.xml

@@ -1765,6 +1765,7 @@ Note: This is a permanent action! If you protect this server again later, a full
 		<key name="job_0482">Waiting for the peer's hostname to be: [#!variable!peer_name!#] in the database.</key>
 		<key name="job_0483">The peer's hostname is: [#!variable!peer_name!#], proceeding.</key>
 		<key name="job_0484">The peer's hostname is currently: [#!variable!old_peer_name!#], waiting for it to be changed to: [#!variable!peer_name!#]... Will check again shortly.</key>
+		<key name="job_0485">The corosync authkey file does not exist locally, but it does exist on the peer. Copying the file to here.</key>
 
 		<!-- Log entries -->
 		<key name="log_0001">Starting: [#!variable!program!#].</key>

tools/anvil-daemon

@@ -1617,7 +1617,7 @@ sub run_jobs
 			my $job_status          =  $anvil->data->{jobs}{modified_date}{$modified_date}{job_uuid}{$job_uuid}{job_status};
 			my $started_seconds_ago =  $job_picked_up_at ? (time - $job_picked_up_at) : 0;
 			my $updated_seconds_ago =  $job_updated      ? (time - $job_updated)      : 0;
-			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
+			$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 1, list => { 
 				's01:job_uuid'             => $job_uuid,
 				's02:job_command'          => $job_command, 
 				's03:short_command'        => $short_command, 
@@ -1821,21 +1821,22 @@ sub run_jobs
 				{
 					# Is the job_uuid associated with this command done?
 					my $started_job_uuid = $anvil->data->{jobs_started}{$short_command};
-					$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { started_job_uuid => $started_job_uuid }});
+					$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 1, list => { started_job_uuid => $started_job_uuid }});
 					
 					if (exists $anvil->data->{jobs}{running}{$started_job_uuid})
 					{
 						# If the previously running job and this job have the same
 						# UUID, it failed and needs to restart.
-						$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { 
+						$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 1, list => { 
 							job_uuid                                           => $job_uuid, 
 							started_job_uuid                                   => $started_job_uuid,
 							"jobs::running::${started_job_uuid}::job_progress" => $anvil->data->{jobs}{running}{$started_job_uuid}{job_progress}, 
 						}});
+						die if not $job_uuid;
 						if ($started_job_uuid eq $job_uuid)
 						{
 							# We're restarting.
-							$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0741", variables => { 
+							$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0742", variables => { 
 								command  => $job_command,
 								job_uuid => $job_uuid, 
 							}});

tools/anvil-join-anvil

@@ -59,7 +59,7 @@ if (not $anvil->data->{sys}{database}{connections})
 }
 
 # If there isn't a job UUID, see if the user is asking to rejoin this subnode to a cluster.
-if (not $anvil->data->{switches}{rejoin})
+if ($anvil->data->{switches}{rejoin})
 {
 	process_rejoin($anvil);
 	$anvil->nice_exit({exit_code => 0});
@@ -841,6 +841,30 @@ sub configure_pacemaker
 		return(0);
 	}
 	
+	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 1, list => { 'path::configs::corosync-authkey' => $anvil->data->{path}{configs}{'corosync-authkey'} }});
+	if (not -f $anvil->data->{path}{configs}{'corosync-authkey'})
+	{
+		# In case we're rebuilding, see if the peer already has the '/etc/corosync/authkey' file.
+		my $corosync_authkey = $anvil->Storage->read_file({
+			file   => $anvil->data->{path}{configs}{'corosync-authkey'},
+			target => $peer_host_name,
+		});
+		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 1, list => { corosync_authkey => $corosync_authkey }});
+		if ($corosync_authkey ne "!!error!!")
+		{
+			# Write the file out.
+			$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0485"});
+			$anvil->Storage->write_file({
+				debug => 2,
+				body  => $corosync_authkey,
+				file  => $anvil->data->{path}{configs}{'corosync-authkey'}, 
+				user  => "root",
+				group => "root",
+				mode  => "0400",
+			});
+		}
+	}
+	
 	### Run on both nodes.
 	# Enable pcsd and start the pcsd daemon.
 	my ($return_code) = $anvil->System->enable_daemon({daemon => "pcsd.service"});
@@ -887,6 +911,7 @@ sub configure_pacemaker
 	
 	# If there is no corosync.conf, see if the peer has it. If so, copy it. If not, we'll initialize the
 	# cluster shortly.
+	$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, secure => 1, list => { 'path::configs::corosync.conf' => $anvil->data->{path}{configs}{'corosync.conf'} }});
 	if (not -e $anvil->data->{path}{configs}{'corosync.conf'})
 	{
 		my $corosync_conf = $anvil->Storage->read_file({
@@ -907,6 +932,13 @@ sub configure_pacemaker
 				mode  => "0644",
 			});
 		}
+		
+		# Restart corosync
+		my ($return_code) = $anvil->System->restart_daemon({daemon => "corosync.service"});
+		$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => { return_code => $return_code }});
+		$return_code = undef;
+		$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 1, key => "job_0095", variables => { daemon => "corosync.service" }});
+		update_progress($anvil, ($anvil->data->{job}{progress} += 2), "job_0095,!!daemon!corosync.service!!");
 	}
 	
 	# Node 1 initializes, node 2 waits.
@@ -990,6 +1022,7 @@ sub configure_pacemaker
 			}
 			if (not $both_online)
 			{
+				### TODO: Left off here... In case of rebuilding, we'll never get past this, so we need to try 'pcs cluster start' once.
 				my $delay = 5;
 				$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, level => 1, key => "log_0816", variables => { delay => $delay }});
 				sleep $delay;