Пример #1
0
Файл: print.c Проект: Cray/slurm
static int _get_node_cnt(job_info_t * job)
{
	int node_cnt = 0;

	/*  For PENDING jobs, return the maximum of the requested nodelist,
	 *   requested maximum number of nodes, or requested CPUs rounded
	 *   to nearest node.
	 *
	 *  For COMPLETING jobs, the job->nodes nodelist has already been
	 *   altered to list only the nodes still in the comp state, and
	 *   thus we count only those nodes toward the total nodes still
	 *   allocated to this job.
	 */

	if (IS_JOB_PENDING(job)) {
		node_cnt = _nodes_in_list(job->req_nodes);
		node_cnt = MAX(node_cnt, job->num_nodes);
		if ((node_cnt == 1) && (job->num_cpus > 1)
		    && job->ntasks_per_node
		    && (job->ntasks_per_node != (uint16_t) NO_VAL)) {
			int num_tasks = job->num_cpus;
			if (job->cpus_per_task != (uint16_t) NO_VAL)
				num_tasks /= job->cpus_per_task;
			node_cnt = (num_tasks + 1) / job->ntasks_per_node;
			if (node_cnt > num_tasks)
				node_cnt = num_tasks;
			else if (!node_cnt)
				node_cnt = 1;
		}
	} else
		node_cnt = _nodes_in_list(job->nodes);
	return node_cnt;
}
Пример #2
0
Файл: print.c Проект: IFCA/slurm
static int _get_node_cnt(job_info_t * job)
{
	int node_cnt = 0, round;

	/*  For PENDING jobs, return the maximum of the requested nodelist,
	 *   requested maximum number of nodes, or requested CPUs rounded
	 *   to nearest node.
	 *
	 *  For COMPLETING jobs, the job->nodes nodelist has already been
	 *   altered to list only the nodes still in the comp state, and
	 *   thus we count only those nodes toward the total nodes still
	 *   allocated to this job.
	 */

	if (IS_JOB_PENDING(job)) {
		node_cnt = _nodes_in_list(job->req_nodes);
		node_cnt = MAX(node_cnt, job->num_nodes);
		round  = job->num_cpus + params.max_cpus - 1;
		round /= params.max_cpus;	/* round up */
		node_cnt = MAX(node_cnt, round);
	} else
		node_cnt = _nodes_in_list(job->nodes);
	return node_cnt;
}
Пример #3
0
/*
 * slurm_sprint_job_step_info - output information about a specific Slurm
 *	job step based upon message as loaded using slurm_get_job_steps
 * IN job_ptr - an individual job step information record pointer
 * IN one_liner - print as a single line if true
 * RET out - char * containing formatted output (must be freed after call)
 *           NULL is returned on failure.
 */
char *
slurm_sprint_job_step_info ( job_step_info_t * job_step_ptr,
			    int one_liner )
{
	char tmp_node_cnt[40];
	char time_str[32];
	char limit_str[32];
	char tmp_line[128];
	char *out = NULL;
	uint32_t cluster_flags = slurmdb_setup_cluster_flags();

	/****** Line 1 ******/
	slurm_make_time_str ((time_t *)&job_step_ptr->start_time, time_str,
		sizeof(time_str));
	if (job_step_ptr->time_limit == INFINITE)
		sprintf(limit_str, "UNLIMITED");
	else
		secs2time_str ((time_t)job_step_ptr->time_limit * 60,
				limit_str, sizeof(limit_str));
	snprintf(tmp_line, sizeof(tmp_line),
		 "StepId=%u.%u UserId=%u StartTime=%s TimeLimit=%s",
		 job_step_ptr->job_id, job_step_ptr->step_id,
		 job_step_ptr->user_id, time_str, limit_str);
	out = xstrdup(tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 2 ******/
	snprintf(tmp_line, sizeof(tmp_line),
		 "State=%s ",
		 job_state_string(job_step_ptr->state));
	xstrcat(out, tmp_line);
	if (cluster_flags & CLUSTER_FLAG_BG) {
		char *io_nodes;
		select_g_select_jobinfo_get(job_step_ptr->select_jobinfo,
					    SELECT_JOBDATA_IONODES,
					    &io_nodes);
		if (io_nodes) {
			snprintf(tmp_line, sizeof(tmp_line),
				 "Partition=%s MidplaneList=%s[%s] Gres=%s",
				 job_step_ptr->partition,
				 job_step_ptr->nodes, io_nodes,
				 job_step_ptr->gres);
			xfree(io_nodes);
		} else
			snprintf(tmp_line, sizeof(tmp_line),
				 "Partition=%s MidplaneList=%s Gres=%s",
				 job_step_ptr->partition,
				 job_step_ptr->nodes,
				 job_step_ptr->gres);
	} else {
		snprintf(tmp_line, sizeof(tmp_line),
			"Partition=%s NodeList=%s Gres=%s",
			job_step_ptr->partition, job_step_ptr->nodes,
			job_step_ptr->gres);
	}
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 3 ******/
	if (cluster_flags & CLUSTER_FLAG_BGQ) {
		uint32_t nodes = 0;
		select_g_select_jobinfo_get(job_step_ptr->select_jobinfo,
					    SELECT_JOBDATA_NODE_CNT,
					    &nodes);
		convert_num_unit((float)nodes, tmp_node_cnt,
				 sizeof(tmp_node_cnt), UNIT_NONE);
	} else {
		convert_num_unit((float)_nodes_in_list(job_step_ptr->nodes),
				 tmp_node_cnt, sizeof(tmp_node_cnt),
				 UNIT_NONE);
	}

	snprintf(tmp_line, sizeof(tmp_line),
		"Nodes=%s Tasks=%u Name=%s Network=%s",
		 tmp_node_cnt, job_step_ptr->num_tasks, job_step_ptr->name,
		job_step_ptr->network);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 4 ******/
	snprintf(tmp_line, sizeof(tmp_line),
		"ResvPorts=%s Checkpoint=%u CheckpointDir=%s",
		 job_step_ptr->resv_ports,
		 job_step_ptr->ckpt_interval, job_step_ptr->ckpt_dir);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 5 ******/
	if (job_step_ptr->cpu_freq == NO_VAL) {
		snprintf(tmp_line, sizeof(tmp_line), 
			 "CPUFreqReq=Default\n\n");
	} else if (job_step_ptr->cpu_freq & CPU_FREQ_RANGE_FLAG) {
		switch (job_step_ptr->cpu_freq) 
		{
		case CPU_FREQ_LOW :
			snprintf(tmp_line, sizeof(tmp_line),
				 "CPUFreqReq=Low\n\n");
			break;
		case CPU_FREQ_MEDIUM :
			snprintf(tmp_line, sizeof(tmp_line),
				 "CPUFreqReq=Medium\n\n");
			break;
		case CPU_FREQ_HIGH :
			snprintf(tmp_line, sizeof(tmp_line),
				 "CPUFreqReq=High\n\n");
			break;
		default :
			snprintf(tmp_line, sizeof(tmp_line),
				 "CPUFreqReq=Unknown\n\n");
		}
	} else {
		snprintf(tmp_line, sizeof(tmp_line),
			 "CPUFreqReq=%u\n\n", job_step_ptr->cpu_freq);
	}
	xstrcat(out, tmp_line);

	return out;
}
Пример #4
0
/*
 * slurm_sprint_job_step_info - output information about a specific Slurm
 *	job step based upon message as loaded using slurm_get_job_steps
 * IN job_ptr - an individual job step information record pointer
 * IN one_liner - print as a single line if true
 * RET out - char * containing formatted output (must be freed after call)
 *           NULL is returned on failure.
 */
char *
slurm_sprint_job_step_info ( job_step_info_t * job_step_ptr,
			    int one_liner )
{
	char tmp_node_cnt[40];
	char time_str[32];
	char limit_str[32];
	char tmp_line[128];
	char *out = NULL;
	uint32_t cluster_flags = slurmdb_setup_cluster_flags();

	/****** Line 1 ******/
	slurm_make_time_str ((time_t *)&job_step_ptr->start_time, time_str,
		sizeof(time_str));
	if (job_step_ptr->time_limit == INFINITE)
		sprintf(limit_str, "UNLIMITED");
	else
		secs2time_str ((time_t)job_step_ptr->time_limit * 60,
				limit_str, sizeof(limit_str));
	if (job_step_ptr->array_job_id) {
		if (job_step_ptr->step_id == INFINITE) {	/* Pending */
			snprintf(tmp_line, sizeof(tmp_line),
				 "StepId=%u_%u.TBD ",
				 job_step_ptr->array_job_id,
				 job_step_ptr->array_task_id);
		} else {
			snprintf(tmp_line, sizeof(tmp_line), "StepId=%u_%u.%u ",
				 job_step_ptr->array_job_id,
				 job_step_ptr->array_task_id,
				 job_step_ptr->step_id);
		}
		out = xstrdup(tmp_line);
	} else {
		if (job_step_ptr->step_id == INFINITE) {	/* Pending */
			snprintf(tmp_line, sizeof(tmp_line), "StepId=%u.TBD ",
				 job_step_ptr->job_id);
		} else {
			snprintf(tmp_line, sizeof(tmp_line), "StepId=%u.%u ",
				 job_step_ptr->job_id, job_step_ptr->step_id);
		}
		out = xstrdup(tmp_line);
	}
	snprintf(tmp_line, sizeof(tmp_line),
		 "UserId=%u StartTime=%s TimeLimit=%s",
		 job_step_ptr->user_id, time_str, limit_str);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 2 ******/
	snprintf(tmp_line, sizeof(tmp_line),
		 "State=%s ",
		 job_state_string(job_step_ptr->state));
	xstrcat(out, tmp_line);
	if (cluster_flags & CLUSTER_FLAG_BG) {
		char *io_nodes = NULL;
		select_g_select_jobinfo_get(job_step_ptr->select_jobinfo,
					    SELECT_JOBDATA_IONODES,
					    &io_nodes);
		if (io_nodes) {
			snprintf(tmp_line, sizeof(tmp_line),
				 "Partition=%s MidplaneList=%s[%s] Gres=%s",
				 job_step_ptr->partition,
				 job_step_ptr->nodes, io_nodes,
				 job_step_ptr->gres);
			xfree(io_nodes);
		} else
			snprintf(tmp_line, sizeof(tmp_line),
				 "Partition=%s MidplaneList=%s Gres=%s",
				 job_step_ptr->partition,
				 job_step_ptr->nodes,
				 job_step_ptr->gres);
	} else {
		snprintf(tmp_line, sizeof(tmp_line),
			"Partition=%s NodeList=%s Gres=%s",
			job_step_ptr->partition, job_step_ptr->nodes,
			job_step_ptr->gres);
	}
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 3 ******/
	if (cluster_flags & CLUSTER_FLAG_BGQ) {
		uint32_t nodes = 0;
		select_g_select_jobinfo_get(job_step_ptr->select_jobinfo,
					    SELECT_JOBDATA_NODE_CNT,
					    &nodes);
		convert_num_unit((float)nodes, tmp_node_cnt,
				 sizeof(tmp_node_cnt), UNIT_NONE,
				 CONVERT_NUM_UNIT_EXACT);
	} else {
		convert_num_unit((float)_nodes_in_list(job_step_ptr->nodes),
				 tmp_node_cnt, sizeof(tmp_node_cnt),
				 UNIT_NONE, CONVERT_NUM_UNIT_EXACT);
	}

	snprintf(tmp_line, sizeof(tmp_line),
		"Nodes=%s CPUs=%u Tasks=%u Name=%s Network=%s",
		 tmp_node_cnt, job_step_ptr->num_cpus, job_step_ptr->num_tasks,
		 job_step_ptr->name, job_step_ptr->network);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 4 ******/
	snprintf(tmp_line, sizeof(tmp_line), "TRES=%s",
		 job_step_ptr->tres_alloc_str);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 5 ******/
	snprintf(tmp_line, sizeof(tmp_line),
		"ResvPorts=%s Checkpoint=%u CheckpointDir=%s",
		 job_step_ptr->resv_ports,
		 job_step_ptr->ckpt_interval, job_step_ptr->ckpt_dir);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 6 ******/
	if (cpu_freq_debug(NULL, NULL, tmp_line, sizeof(tmp_line),
			   job_step_ptr->cpu_freq_gov,
			   job_step_ptr->cpu_freq_min,
			   job_step_ptr->cpu_freq_max, NO_VAL) != 0) {
		xstrcat(out, tmp_line);
	} else {
		xstrcat(out, "CPUFreqReq=Default");
	}
	xstrfmtcat(out, " Dist=%s",
		   slurm_step_layout_type_name(job_step_ptr->task_dist));
	xstrcat(out, "\n\n");

	return out;
}