Fix bug when workers crash totally unexpectedly, capture and report t…

…he exception when that happens. Pass parent pid down into child so that it can verify it's parent is still there (and exit if not) - there should be a better fix than this, but this will do at a pinch right now..
bobtfish · Dec 3, 2010 · 8328d26 · 8328d26
1 parent 58782ab
commit 8328d26
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 3 deletions.
diff --git a/lib/CatalystX/JobServer/JobRunner/Forked/Worker.pm b/lib/CatalystX/JobServer/JobRunner/Forked/Worker.pm
@@ -1,7 +1,7 @@
 package CatalystX::JobServer::JobRunner::Forked::Worker;
 use CatalystX::JobServer::Moose;
 use MooseX::Types::LoadableClass qw/ LoadableClass /;
-#use MooseX::Types::Moose qw/ ArrayRef /;
+use MooseX::Types::Moose qw/ Int /;
 use JSON;
 use Try::Tiny;
 use IO::Handle;
@@ -25,6 +25,12 @@ foreach my $type (qw/ before after /) {
     );
 }
 
+has ppid => (
+    isa => Int,
+    is => 'ro',
+    required => 1,
+);
+
 method run {
     $0 = 'perl jobserver_worker [idle]';
     STDOUT->autoflush(1);
@@ -49,6 +55,7 @@ method run {
         }
         confess("Got EOF from parent: $!") if (!defined $bytes && $! != EAGAIN);
         sleep 1; # FIXME - Cheesy hack!
+        exit 0 if (!kill 0, $self->ppid);
     }
 }
 

diff --git a/lib/CatalystX/JobServer/JobRunner/Forked/WorkerState.pm b/lib/CatalystX/JobServer/JobRunner/Forked/WorkerState.pm
@@ -16,6 +16,7 @@ use aliased 'CatalystX::JobServer::JobRunner::Forked::WorkerStatus::Complete';
 use Try::Tiny;
 use POSIX ();
 use namespace::autoclean;
+use CatalystX::JobServer::JobRunner::Forked::WorkerStatus::Complete;
 no warnings 'syntax'; # "Statement unlikely to be reached"
 
 with 'CatalystX::JobServer::Role::Storage';
@@ -178,7 +179,7 @@ method __on_error ($hdl, $fatal, $msg) {
     $self->_clear_pid;
     $self->_clear_ae_handle;
     if ($self->working_on) {
-        $self->job_finished($error);
+        $self->job_finished(CatalystX::JobServer::JobRunner::Forked::WorkerStatus::Complete->new( ok => 0, output => $error));
     }
     $self->_clear_working_on;
     $self->_clear_worker_started_at;
@@ -288,7 +289,7 @@ method _spawn_worker_if_needed {
                 push(@cmd, '-I', $lib);
             }
             push (@cmd, '-M' . $self->worker_class);
-            push(@cmd, '-e', $self->worker_class . '->new->run');
+            push(@cmd, '-e', $self->worker_class . '->new(ppid => ' . $$ . ')->run');
             push(@cmd, $self->eval_before_job, $self->eval_after_job);
             exec( @cmd );
         }

diff --git a/lib/CatalystX/JobServer/JobRunner/Forked/WorkerStatus/Complete.pm b/lib/CatalystX/JobServer/JobRunner/Forked/WorkerStatus/Complete.pm
@@ -13,5 +13,10 @@ has ok => (
     traits => ['Serialize'],
 );
 
+has output => (
+    is => 'ro',
+    traits => ['Serialize'],
+);
+
 __PACKAGE__->meta->make_immutable;
 1;