Set block distribution when srun requests 0 memory.

Bug 1810

NEWS

@@ -33,6 +33,7 @@ documents those changes that are of interest to users and administrators.
  -- Correct sacct.a man pages describing -i option.
  -- Capture salloc/srun information in sdiag statistics.
  -- Fix bug in node selection with topology optimization.
+ -- Set block distribution when srun requests 0 memory.
 
 * Changes in Slurm 14.11.8
 ==========================

src/slurmctld/step_mgr.c

@@ -2128,6 +2128,13 @@ step_create(job_step_create_request_msg_t *step_specs,
 	    (job_ptr->end_time <= time(NULL)))
 		return ESLURM_ALREADY_DONE;
 
+	/* Set to block in the case that mem is 0. srun leaves the dist
+	 * set to unknown if mem is 0.
+	 * ex. SallocDefaultCommand=srun -n1 -N1 --mem=0 ... */
+	if ((step_specs->task_dist == SLURM_DIST_UNKNOWN) &&
+	    (!(step_specs->pn_min_memory &(~MEM_PER_CPU))))
+		step_specs->task_dist = SLURM_DIST_BLOCK;
+
 	if ((step_specs->task_dist != SLURM_DIST_CYCLIC) &&
 	    (step_specs->task_dist != SLURM_DIST_BLOCK) &&
 	    (step_specs->task_dist != SLURM_DIST_CYCLIC_CYCLIC) &&

src/srun/libsrun/launch.c

@@ -264,9 +264,16 @@ extern int launch_common_create_job_step(srun_job_t *job, bool use_all_cpus,
 		job->ctx_params.plane_size = opt.plane_size;
 		break;
 	default:
-		job->ctx_params.task_dist = (job->ctx_params.task_count <=
-					     job->ctx_params.min_nodes)
-			? SLURM_DIST_CYCLIC : SLURM_DIST_BLOCK;
+		/* Leave distribution set to unknown if taskcount <= nodes and
+		 * memory is set to 0. step_mgr will handle the 0mem case.
+		 * ex. SallocDefaultCommand=srun -n1 -N1 --mem=0 ... */
+		if (!opt.mem_per_cpu || !opt.pn_min_memory)
+			job->ctx_params.task_dist = SLURM_DIST_UNKNOWN;
+		else
+			job->ctx_params.task_dist =
+				(job->ctx_params.task_count <=
+				 job->ctx_params.min_nodes)
+				? SLURM_DIST_CYCLIC : SLURM_DIST_BLOCK;
 		if (opt.ntasks_per_node != NO_VAL)
 			job->ctx_params.plane_size = opt.ntasks_per_node;
 		opt.distribution = job->ctx_params.task_dist;