Temporary fix. If slurmstepd dies do not hang srun but abort the

entire parallel job.

src/api/step_launch.c

@@ -1788,6 +1788,16 @@ step_launch_notify_io_failure(step_launch_state_t *sls, int node_id)
 		      node_id);
 		sls->abort = true;
 		pthread_cond_broadcast(&sls->cond);
+	} else {
+		/* FIXME
+		 * If stepd dies or we see I/O error with stepd.
+		 * Do not abort the whole job but collect all
+		 * taks on the node just like if they exited.
+		 */
+		error("%s: aborting, io error with slurmstepd on node %d",
+		      __func__, node_id);
+		sls->abort = true;
+		pthread_cond_broadcast(&sls->cond);
 	}
 
 	pthread_mutex_unlock(&sls->lock);