do not cache task objects

Caching the task object means it will survive long enough for Minion to
clean it up entirely using `POSIX::_exit(0)`. Since this is an abrupt,
immediate exit, things are not able to be cleaned up (like database
connections, resulting in spurious log messages).

Committed from Perl Toolchain Summit 2018. Thanks @toddr and @atoomic
for helping to debug this issue!

lib/Beam/Minion/Util.pm

@@ -128,14 +128,14 @@ sub build_mojo_app {
             $minion->add_task( "$container_name:$service_name" => sub {
                 my ( $job, @args ) = @_;
 
-                my $obj = eval { $wire->get( $service_name ) };
+                my $obj = eval { $wire->get( $service_name, lifecycle => 'factory' ) };
                 if ( $@ ) {
                     return $job->fail( { error => $@ } );
                 }
 
                 my $exit = eval { $obj->run( @args ) };
-                if ( $@ ) {
-                    return $job->fail( { error => $@ } );
+                if ( my $err = $@ ) {
+                    return $job->fail( { error => $err } );
                 }
 
                 my $method = $exit ? 'fail' : 'finish';

t/command/worker.t

@@ -96,4 +96,11 @@ subtest 'BEAM_MINION must be set' => sub {
         'BEAM_MINION missing raises exception';
 };
 
+subtest 'test that object is destroyed' => sub {
+    my $id = $minion->enqueue( 'container:success', [] );
+    my $job = $minion->job( $id );
+    $job->execute;
+    is $Local::Service::DESTROYED, 1, 'DESTROY was called';
+};
+
 done_testing;

t/lib/Local/Service.pm

@@ -41,4 +41,10 @@ sub run {
     return $self->exit_code;
 }
 
+our $DESTROYED = 0;
+sub DESTROY {
+    $DESTROYED++;
+}
+
 1;
+