feature: jobless workers will now leave module compilation errors in the...

feature: jobless workers will now leave module compilation errors in the job_message table (thanks, Kathryn!)

modules/Bio/EnsEMBL/Hive/AnalysisJob.pm

@@ -220,7 +220,7 @@ sub warning {
     my ($self, $msg) = @_;
 
     if( my $job_adaptor = $self->adaptor ) {
-        $job_adaptor->db->get_JobMessageAdaptor()->register_message($self->dbID, $msg, 0);
+        $job_adaptor->db->get_JobMessageAdaptor()->store_job_message($self->dbID, $msg, 0);
     } else {
         print STDERR "Warning: $msg\n";
     }

modules/Bio/EnsEMBL/Hive/DBSQL/AnalysisJobAdaptor.pm

@@ -621,7 +621,7 @@ sub release_undone_jobs_from_worker {
             }
         }
 
-        $self->db()->get_JobMessageAdaptor()->register_message($job_id, $msg, not $passed_on );
+        $self->db()->get_JobMessageAdaptor()->store_job_message($job_id, $msg, not $passed_on );
 
         unless($passed_on) {
             $self->release_and_age_job( $job_id, $max_retry_count, not $resource_overusage );

modules/Bio/EnsEMBL/Hive/DBSQL/JobMessageAdaptor.pm

@@ -6,7 +6,9 @@
 
 =head1 SYNOPSIS
 
-    $dba->get_JobMessageAdaptor->register_message($job_id, $msg, $is_error);
+    $dba->get_JobMessageAdaptor->store_job_message($job_id, $msg, $is_error);
+
+    $dba->get_JobMessageAdaptor->store_worker_message($worker_id, $msg, $is_error);
 
 =head1 DESCRIPTION
 
@@ -31,14 +33,14 @@ sub default_table_name {
 }
 
 
-sub register_message {
+sub store_job_message {
     my ($self, $job_id, $msg, $is_error) = @_;
 
     chomp $msg;   # we don't want that last "\n" in the database
 
     my $table_name = $self->table_name();
 
-        # (the timestamp 'time' column will be set automatically)
+        # Note: the timestamp 'time' column will be set automatically
     my $sql = qq{
         INSERT INTO $table_name (job_id, worker_id, retry, status, msg, is_error)
                            SELECT job_id, worker_id, retry_count, status, ?, ?
@@ -50,5 +52,24 @@ sub register_message {
     $sth->finish();
 }
 
+
+sub store_worker_message {
+    my ($self, $worker_id, $msg, $is_error) = @_;
+
+    chomp $msg;   # we don't want that last "\n" in the database
+
+    my $table_name = $self->table_name();
+
+        # Note: the timestamp 'time' column will be set automatically
+    my $sql = qq{
+        INSERT INTO $table_name (worker_id, status, msg, is_error)
+                           SELECT worker_id, status, ?, ?
+                             FROM worker WHERE worker_id=?
+    };
+    my $sth = $self->prepare( $sql );
+    $sth->execute( $msg, $is_error ? 1 : 0, $worker_id );
+    $sth->finish();
+}
+
 1;
 

modules/Bio/EnsEMBL/Hive/Worker.pm

@@ -533,6 +533,7 @@ sub run {
         } or do {
             my $msg = "Could not compile Runnable '".$self->analysis->module."' :\n\t".$@;
             warn "$msg\n";
+            $self->adaptor->db->get_JobMessageAdaptor()->store_worker_message($self->dbID, $msg, 1 );
 
             $self->cause_of_death('CONTAMINATED');
         }
@@ -689,7 +690,7 @@ sub run_one_batch {
             my $job_status_at_the_moment = $job->status();
             my $action = $job->incomplete ? 'died' : 'exited';
             $job_completion_line = "\njob $job_id : $action in status '$job_status_at_the_moment' for the following reason: $msg_thrown\n";
-            $self->adaptor->db->get_JobMessageAdaptor()->register_message($job_id, $msg_thrown, $job->incomplete );
+            $self->adaptor->db->get_JobMessageAdaptor()->store_job_message($job_id, $msg_thrown, $job->incomplete );
         }
 
         print STDERR $job_completion_line if($self->log_dir and ($self->debug or $job->incomplete));            # one copy goes to the job's STDERR

sql/patch_2012-11-13.sql

+    # allow job_id and retry to be NULL by default, to make it possible for jobless workers to leave messages:
+ALTER TABLE job_message MODIFY COLUMN job_id                    int(10) DEFAULT NULL;
+ALTER TABLE job_message MODIFY COLUMN retry                     int(10) DEFAULT NULL;
+
+    # re-route getting analysis_id/logic_name via worker table, so that it would work in the absence of the job:
+CREATE OR REPLACE VIEW msg AS
+    SELECT a.analysis_id, a.logic_name, m.*
+    FROM job_message m
+    JOIN worker w USING (worker_id)
+    JOIN analysis_base a USING (analysis_id)
+    LEFT JOIN job j USING (job_id);
+

sql/procedures.mysql

@@ -34,8 +34,9 @@ CREATE OR REPLACE VIEW progress AS
 CREATE OR REPLACE VIEW msg AS
     SELECT a.analysis_id, a.logic_name, m.*
     FROM job_message m
-    JOIN job j USING (job_id)
-    JOIN analysis_base a USING (analysis_id);
+    JOIN worker w USING (worker_id)
+    JOIN analysis_base a USING (analysis_id)
+    LEFT JOIN job j USING (job_id);
 
 
 #### time an analysis or group of analyses (given by a name pattern) ######################################

sql/procedures.sqlite

@@ -34,6 +34,7 @@ CREATE VIEW IF NOT EXISTS progress AS
 CREATE VIEW IF NOT EXISTS msg AS
     SELECT a.analysis_id, a.logic_name, m.*
     FROM job_message m
-    JOIN job j USING (job_id)
-    JOIN analysis_base a USING (analysis_id);
+    JOIN worker w USING (worker_id)
+    JOIN analysis_base a USING (analysis_id)
+    LEFT JOIN job j ON (m.job_id=j.job_id);
 

sql/tables.sql

@@ -231,10 +231,10 @@ CREATE TABLE job (
 
 CREATE TABLE job_message (
   job_message_id            int(10) NOT NULL AUTO_INCREMENT,
-  job_id                    int(10) NOT NULL,
+  job_id                    int(10) DEFAULT NULL,
   worker_id                 int(10) unsigned NOT NULL,
   time                      timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
-  retry                     int(10) DEFAULT 0 NOT NULL,
+  retry                     int(10) DEFAULT NULL,
   status                    enum('UNKNOWN','COMPILATION','PRE_CLEANUP','FETCH_INPUT','RUN','WRITE_OUTPUT','POST_CLEANUP','PASSED_ON') DEFAULT 'UNKNOWN',
   msg                       text,
   is_error                  TINYINT,

sql/tables.sqlite

@@ -215,10 +215,10 @@ CREATE        INDEX IF NOT EXISTS worker_idx ON job (worker_id);
 
 CREATE TABLE job_message (
   job_message_id            INTEGER NOT NULL PRIMARY KEY AUTOINCREMENT,
-  job_id                    INTEGER NOT NULL,
+  job_id                    INTEGER DEFAULT NULL,
   worker_id                 INTEGER NOT NULL,
   time                      timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
-  retry                     int(10) DEFAULT 0 NOT NULL,
+  retry                     int(10) DEFAULT NULL,
   status                    TEXT DEFAULT 'UNKNOWN', /* enum('UNKNOWN', 'COMPILATION', 'FETCH_INPUT', 'RUN', 'WRITE_OUTPUT', 'PASSED_ON') DEFAULT 'UNKNOWN', */
   msg                       TEXT,
   is_error                  BOOLEAN