found location of hang; make annotations; added more log lines

[scpubgit/Object-Remote.git] / lib / Object / Remote / ReadChannel.pm

diff --git a/lib/Object/Remote/ReadChannel.pm b/lib/Object/Remote/ReadChannel.pm
index 5de5bf1..6402bdd 100644 (file)
--- a/lib/Object/Remote/ReadChannel.pm
+++ b/lib/Object/Remote/ReadChannel.pm
@@ -27,10 +27,24 @@ has on_line_call => (is => 'rw');
 
 has _receive_data_buffer => (is => 'ro', default => sub { my $x = ''; \$x });
 
+#TODO confirmed this is the point of the hang - sysread() is invoked on a 
+#socket inside the controller that blocks and deadlocks the entire system.
+#The remote nodes are all waiting to receive data at that point.
+#Validated this behavior exists in an unmodified Object::Remote from CPAN 
+#by wrapping this sysread() with warns that have the pid in them and pounding 
+#my local machine with System::Introspector via ssh and 7 remote perl instances
+#It looks like one of the futures is responding to an event regarding the ability
+#to read from a socket and every once in a while an ordering issue means that
+#there is no actual data to read from the socket
 sub _receive_data_from {
   my ($self, $fh) = @_;
   log_trace { "Preparing to read data" };
+  #use Carp qw(cluck); cluck();
   my $rb = $self->_receive_data_buffer;
+  #TODO is there a specific reason sysread() and syswrite() aren't
+  #a part of ::MiniLoop? It's one spot to handle errors and other
+  #logic involving filehandles
+  #TODO why are the buffers so small? BUFSIZ is usually 32768
   my $len = sysread($fh, $$rb, 1024, length($$rb));
   my $err = defined($len) ? '' : ": $!";
   if (defined($len) and $len > 0) {