added beekeeper tag to aid in tracking dead workers

modules/Bio/EnsEMBL/Hive/Queen.pm

@@ -81,7 +81,9 @@ our @ISA = qw(Bio::EnsEMBL::DBSQL::BaseAdaptor);
 =cut
 
 sub create_new_worker {
-  my ($self,$analysis_id) = @_;
+  my $self        = shift;
+  my $analysis_id = shift;
+  my $beekeeper   = shift;
 
   my $analStatsDBA = $self->_analysisStatsAdaptor;
   return undef unless($analStatsDBA);
@@ -101,10 +103,12 @@ sub create_new_worker {
 
   my $host = hostname;
   my $ppid = getppid;
+  $beekeeper = '' unless($beekeeper);
 
   my $sql = "INSERT INTO hive SET born=now(), last_check_in=now()".
             ",process_id='$ppid' ".
             ",analysis_id='$analysis_id' ".
+            ",beekeeper='$beekeeper' ".
             ",host='$host'";
 
   my $sth = $self->prepare($sql);
@@ -410,6 +414,7 @@ sub _columns {
 
   return qw (h.hive_id
              h.analysis_id
+             h.beekeeper
              h.host
              h.process_id
              h.work_done
@@ -433,6 +438,7 @@ sub _objs_from_sth {
     $worker->init;
 
     $worker->hive_id($column{'hive_id'});
+    $worker->beekeeper($column{'beekeeper'});
     $worker->host($column{'host'});
     $worker->process_id($column{'process_id'});
     $worker->work_done($column{'work_done'});

modules/Bio/EnsEMBL/Hive/Worker.pm

@@ -87,6 +87,11 @@ sub db {
   $self->{'_db'} = shift if(@_);
   return $self->{'_db'};
 }
+sub beekeeper {
+  my $self = shift;
+  $self->{'_beekeeper'} = shift if(@_);
+  return $self->{'_beekeeper'};
+}
 
 =head2 analysis
   Arg [1] : (optional) Bio::EnsEMBL::Analysis $value

scripts/local_beekeeper.pl

@@ -7,6 +7,7 @@ use Bio::EnsEMBL::Compara::DBSQL::DBAdaptor;
 use Bio::EnsEMBL::Hive::Worker;
 use Bio::EnsEMBL::Hive::Queen;
 use Bio::EnsEMBL::Hive::URLFactory;
+use Sys::Hostname;
 
 # ok this is a hack, but I'm going to pretend I've got an object here
 # by creating a blessed hash ref and passing it around like an object
@@ -44,6 +45,7 @@ my $DBA;
 
 if($url) {
   $DBA = Bio::EnsEMBL::Hive::URLFactory->fetch($url);
+  die("Unable to connect to $url\n") unless($DBA);
 } else {
   if($host)   { $self->{'db_conf'}->{'-host'}   = $host; }
   if($port)   { $self->{'db_conf'}->{'-port'}   = $port; }
@@ -63,9 +65,11 @@ if($url) {
   # connect to database specified
   $DBA = new Bio::EnsEMBL::Compara::DBSQL::DBAdaptor(%{$self->{'db_conf'}});
 }
-
+print("$DBA\n");
 $self->{'queen'} = new Bio::EnsEMBL::Hive::Queen($DBA);
 
+check_for_dead_workers($self, $self->{'queen'});
+
 run_beekeeper($self);
 
 Bio::EnsEMBL::Hive::URLFactory->cleanup;
@@ -120,12 +124,6 @@ sub run_beekeeper
   my $self = shift;
   my $queen = $self->{'queen'};
   
-  my $overdueWorkers = $queen->fetch_overdue_workers(3600);  #overdue by 1hr
-  print(scalar(@{$overdueWorkers}), " overdue workers\n");
-  foreach my $worker (@{$overdueWorkers}) {
-    printf("%10d %20s    analysis_id=%d\n", $worker->hive_id,$worker->host, $worker->analysis->dbID);
-    $queen->register_worker_death($worker);
-  }
 
   $queen->update_analysis_stats();
   my($analysis_id, $count) = $queen->next_clutch();
@@ -137,4 +135,26 @@ sub run_beekeeper
 }
 
 
+sub check_for_dead_workers {
+  my $self = shift;
+  my $queen = shift;
+
+  my $host = hostname;
+
+  my $overdueWorkers = $queen->fetch_overdue_workers(3600);  #overdue by 1hr
+  print(scalar(@{$overdueWorkers}), " overdue workers\n");
+  foreach my $worker (@{$overdueWorkers}) {
+    if(($worker->beekeeper eq 'local') and ($worker->host eq $host)) {
+      my $cmd = "ps -p ". $worker->process_id;
+      my $check = qx/$cmd/;
+
+      print("check : $check\n");
+      
+      printf("%10d %20s    analysis_id=%d\n", $worker->hive_id,$worker->host, $worker->analysis->dbID);
+      #$queen->register_worker_death($worker);
+    }
+  }
+
+}
+
 

sql/tables.sql

@@ -14,6 +14,7 @@
 CREATE TABLE hive (
   hive_id          int(10) NOT NULL auto_increment,
   analysis_id      int(10) NOT NULL,
+  beekeeper        varchar(80) DEAFULT '' NOT NULL,
   host	           varchar(40) DEFAULT '' NOT NULL,
   process_id       int(10) NOT NULL,
   work_done        int(11) DEFAULT '0' NOT NULL,