Merge lp:~y-trudeau/mysql-mmm/NonBlocking into lp:mysql-mmm

[Yves Trudeau]

This patch prevents MMM from locking when the agent is stuck. Such situations happen when the local disk subsystem is having issues. The new code set an alarm to allow continuous if the remote agent is blocked.

Unmerged revisions

149. By Yves Trudeau on 2010-09-28

tested version, behave as expected

148. By Yves Trudeau on 2010-09-24

addition of the unblocking code

=== modified file 'lib/Monitor/Agent.pm'
--- lib/Monitor/Agent.pm	2009-10-27 12:48:10 +0000
+++ lib/Monitor/Agent.pm	2010-11-09 21:32:49 +0000
@@ -5,6 +5,7 @@
 use Log::Log4perl qw(:easy);
 use MMM::Common::Socket;
 use MMM::Monitor::ChecksStatus;
+use POSIX ':signal_h';
 
 our $VERSION = '0.01';
 
@@ -69,7 +70,6 @@
 	my $cmd		= shift;
 	my @params	= @_;
 
-
 	my $checks_status = MMM::Monitor::ChecksStatus->instance();
 	unless ($checks_status->ping($self->host)) {
 		return 0;
@@ -77,43 +77,59 @@
 
 	DEBUG sprintf("Sending command '%s(%s)' to %s (%s:%s)", $cmd, join(', ', @params), $self->host, $self->ip, $self->port);
 
-	my $socket;
-CONNECT: {
-	$socket = MMM::Common::Socket::create_sender($self->ip, $self->port, 10);
-	unless ($socket && $socket->connected) {
-		redo CONNECT if ($!{EINTR});
-		return 0;
-	}
-}
-
-
-	print $socket join('|', $cmd, main::MMM_PROTOCOL_VERSION, $self->host, @params), "\n";
-
-	my $res;
-READ: {
-	$res = <$socket>;
-	redo READ if !$res && $!{EINTR};
-}
-	close($socket);
-
-	unless (defined($res)) {
-		WARN sprintf('Received undefined answer from host %s. $!: %s', $self->host, $!);
-		return 0;
-	}
-
-	DEBUG "Received Answer: $res";
-
-	if ($res =~ /(.*)\|UP:(.*)/) {
-		$res = $1;
-		my $uptime = $2;
-		$self->uptime($uptime);
-		$self->last_uptime($uptime) if ($self->state eq 'ONLINE');
-	}
-	else {
-		WARN sprintf('Received bad answer \'%s\' from host %s. $!: %s', $res, $self->host, $!);
-	}
+    my $mask = POSIX::SigSet->new( SIGALRM );
+    my $action = POSIX::SigAction->new(
+        sub {   DEBUG "Alarm triggered while sending command";
+		die 'Timeout'; },
+        $mask,
+    );
+    my $oldaction = POSIX::SigAction->new();
+    sigaction( SIGALRM, $action, $oldaction );
+
+    eval {
+
+		alarm(15);  # Alarm defined at 15s
+		my $socket;
 	
-	return $res;
+		CONNECT: {
+			$socket = MMM::Common::Socket::create_sender($self->ip, $self->port, 10);
+			unless ($socket && $socket->connected) {
+				redo CONNECT if ($!{EINTR});
+				alarm(0);
+				return 0;
+			}
+		}
+
+		print $socket join('|', $cmd, main::MMM_PROTOCOL_VERSION, $self->host, @params), "\n";
+
+		my $res = '';
+		READ: {
+			$res = <$socket>;
+			redo READ if !$res && $!{EINTR};
+		}
+
+		unless (defined($res)) {
+			WARN sprintf('Received undefined answer from host %s. $!: %s', $self->host, $!);
+			alarm(0);
+			return 0;
+		}
+		close($socket);
+		alarm(0);
+
+		DEBUG "Received Answer: $res";
+
+		if ($res =~ /(.*)\|UP:(.*)/) {
+			$res = $1;
+			my $uptime = $2;
+			$self->uptime($uptime);
+			$self->last_uptime($uptime) if ($self->state eq 'ONLINE');
+		}
+		else {
+			WARN sprintf('Received bad answer \'%s\' from host %s. $!: %s', $res, $self->host, $!);
+		}
+		
+		return $res;
+	}
 }
 
 sub _send_command_retry {