changes to replication so that if a replicant is offline when we do the initial conne...
[dbsrgits/DBIx-Class.git] / lib / DBIx / Class / Storage / DBI / Replicated / Pool.pm
index b1cdc82..a3db265 100644 (file)
@@ -168,14 +168,31 @@ and return it.
 sub connect_replicant {
   my ($self, $schema, $connect_info) = @_;
   my $replicant = $self->create_replicant($schema);
-    
-  $replicant->connect_info($connect_info);    
-  $replicant->ensure_connected;
-  DBIx::Class::Storage::DBI::Replicated::Replicant->meta->apply($replicant);
-    
+  $replicant->connect_info($connect_info);
+  $self->_safely_ensure_connected($replicant);
+  DBIx::Class::Storage::DBI::Replicated::Replicant->meta->apply($replicant);  
   return $replicant;
 }
 
+=head2 _safely_ensure_connected ($replicant)
+
+The standard ensure_connected method with throw an exception should it fail to
+connect.  For the master database this is desirable, but since replicants are
+allowed to fail, this behavior is not desirable.  This method wraps the call
+to ensure_connected in an eval in order to catch any generated errors.  That
+way a slave to go completely offline (ie, the box itself can die) without
+bringing down your entire pool of databases.
+
+=cut
+
+sub _safely_ensure_connected {
+  my ($self, $replicant, @args) = @_;
+  my $return; eval {
+    $return = $replicant->ensure_connected(@args);
+  };  
+  return $return;
+}
+
 =head2 connected_replicants
 
 Returns true if there are connected replicants.  Actually is overloaded to
@@ -248,9 +265,9 @@ sub validate_replicants {
   my $self = shift @_;
   foreach my $replicant($self->all_replicants) {
     if(
+      $self->_safely_ensure_connected($replicant) &&  
       $replicant->is_replicating &&
-      $replicant->lag_behind_master <= $self->maximum_lag &&
-      $replicant->ensure_connected
+      $replicant->lag_behind_master <= $self->maximum_lag
     ) {
       $replicant->active(1)
     } else {