fixed bug in bluegene plugin to fail job if the partition is freed while running the job

src/plugins/select/bluegene/plugin/bg_block_info.c

@@ -79,10 +79,11 @@ static int _block_is_deallocating(bg_record_t *bg_record)
 				  bg_record->user_name)) {
 				error("Partition %s was in a ready state "
 				      "for user %s but is being freed. "
-				      "Job was lost.",
+				      "Job %s was lost.",
 				      bg_record->bg_block_id,
-				      bg_record->user_name);
-				term_jobs_on_block(bg_record->bg_block_id);
+				      bg_record->user_name,
+				      bg_record->job_running);
+				(void) slurm_fail_job(bg_record->job_running);
 			} else {
 				debug("Partition %s was in a ready state "
 				      "but is being freed. No job running.",

src/plugins/select/bluegene/plugin/bg_job_place.c

@@ -112,7 +112,7 @@ static int _find_best_block_match(struct job_record* job_ptr,
 	   we want to fall through to tell the scheduler that it is runnable
 	   just not right now. 
 	*/
-	if(full_system_block->job_running && checked<2) {
+	if((full_system_block->job_running != -1) && checked<2) {
 		checked++;
 		select_g_set_jobinfo(job_ptr->select_jobinfo,
 				     SELECT_DATA_CHECKED, &checked);
@@ -149,7 +149,7 @@ static int _find_best_block_match(struct job_record* job_ptr,
 		*/
 		slurm_mutex_lock(&block_state_mutex);
 		debug3("job_running = %d", record->job_running);
-		if(record->job_running && checked<2) {
+		if((record->job_running != -1) && checked<2) {
 			job_running++;
 			debug("block %s in use by %s", 
 			      record->bg_block_id,

src/plugins/select/bluegene/plugin/bg_job_run.c

@@ -186,7 +186,9 @@ static void _sync_agent(bg_update_t *bg_update_ptr)
 		error("No block %s", bg_update_ptr->bg_block_id);
 		return;
 	}
+	slurm_mutex_lock(&block_state_mutex);
 	bg_record->job_running = bg_update_ptr->job_id;
+	slurm_mutex_unlock(&block_state_mutex);
 	
 	if(bg_record->state==RM_PARTITION_READY) {
 		if(bg_record->user_uid != bg_update_ptr->uid) {
@@ -301,7 +303,7 @@ static void _start_agent(bg_update_t *bg_update_ptr)
 			sleep(1);
 		}
 		
-		if(bg_record->job_running == 0) 
+		if(bg_record->job_running == -1) 
 			return;
 		if((rc = boot_block(bg_record))
 		   != SLURM_SUCCESS) {
@@ -447,7 +449,7 @@ static void _term_agent(bg_update_t *bg_update_ptr)
 		}
 			
 		slurm_mutex_lock(&block_state_mutex);
-		bg_record->job_running = 0;
+		bg_record->job_running = -1;
 		
 		/*remove user from list */
 		if(bg_record->target_name) {
@@ -755,7 +757,7 @@ int term_job(struct job_record *job_ptr)
 		     job_ptr->job_id, 
 		     bg_record->bg_block_id);
 		bg_record->state = RM_PARTITION_FREE;
-		bg_record->job_running = 0;
+		bg_record->job_running = -1;
 		last_bg_update = time(NULL);		
 		xfree(block_id);
 	}

src/plugins/select/bluegene/plugin/block_sys.c

@@ -286,8 +286,8 @@ int read_bg_blocks()
 				break;
 			}
 		} else {
-			if ((rc = rm_get_data(block_list, RM_PartListFirstPart, 
-					&block_ptr)) != STATUS_OK) {
+			if ((rc = rm_get_data(block_list, RM_PartListFirstPart,
+					      &block_ptr)) != STATUS_OK) {
 				error("rm_get_data(RM_PartListFirstPart): %s",
 					bg_err_str(rc));
 				break;
@@ -330,6 +330,7 @@ int read_bg_blocks()
 
 		bg_record->state = -1;
 		bg_record->quarter = -1;
+		bg_record->job_running = -1;
 				
 		if ((rc = rm_get_data(block_ptr, 
 				      RM_PartitionBPNum,