C++ slurm_get_errno函数代码示例

/* get details about this slurm job: jobid and allocated node */
static void _get_job_info(void)
{
	xassert(params.job_id != NO_VAL);

	if (slurm_sbcast_lookup(params.job_id, params.step_id, &sbcast_cred)
	    != SLURM_SUCCESS) {
		if (params.step_id == NO_VAL) {
			error("Slurm job ID %u lookup error: %s",
			      params.job_id, slurm_strerror(slurm_get_errno()));
		} else {
			error("Slurm step ID %u.%u lookup error: %s",
			      params.job_id, params.step_id,
			      slurm_strerror(slurm_get_errno()));
		}
		exit(1);
	}

	if (params.step_id == NO_VAL)
		verbose("jobid      = %u", params.job_id);
	else
		verbose("jobid      = %u.%u", params.job_id, params.step_id);
	verbose("node_cnt   = %u", sbcast_cred->node_cnt);
	verbose("node_list  = %s", sbcast_cred->node_list);
	/* also see sbcast_cred->node_addr (array) */

	if (params.verbose)
		print_sbcast_cred(sbcast_cred->sbcast_cred);

	/* do not bother to release the return message,
	 * we need to preserve and use most of the information later */
}

/*
 * Test if any BG blocks are in deallocating state since they are
 * probably related to this job we will want to sleep longer
 * RET	1:  deallocate in progress
 *	0:  no deallocate in progress
 *     -1: error occurred
 */
static int _blocks_dealloc(void)
{
	static block_info_msg_t *bg_info_ptr = NULL, *new_bg_ptr = NULL;
	int rc = 0, error_code = 0, i;

	if (bg_info_ptr) {
		error_code = slurm_load_block_info(bg_info_ptr->last_update,
						   &new_bg_ptr, SHOW_ALL);
		if (error_code == SLURM_SUCCESS)
			slurm_free_block_info_msg(bg_info_ptr);
		else if (slurm_get_errno() == SLURM_NO_CHANGE_IN_DATA) {
			error_code = SLURM_SUCCESS;
			new_bg_ptr = bg_info_ptr;
		}
	} else {
		error_code = slurm_load_block_info((time_t) NULL,
						   &new_bg_ptr, SHOW_ALL);
	}

	if (error_code) {
		error("slurm_load_partitions: %s",
		      slurm_strerror(slurm_get_errno()));
		return -1;
	}
	for (i=0; i<new_bg_ptr->record_count; i++) {
		if(new_bg_ptr->block_array[i].state == BG_BLOCK_TERM) {
			rc = 1;
			break;
		}
	}
	bg_info_ptr = new_bg_ptr;
	return rc;
}

/* get details about this slurm job: jobid and allocated node */
static int _get_job_info(struct bcast_parameters *params)
{
	int rc;

	xassert(params->job_id != NO_VAL);

	rc = slurm_sbcast_lookup(params->job_id, params->step_id, &sbcast_cred);
	if (rc != SLURM_SUCCESS) {
		if (params->step_id == NO_VAL) {
			error("Slurm job ID %u lookup error: %s",
			      params->job_id,
			      slurm_strerror(slurm_get_errno()));
		} else {
			error("Slurm step ID %u.%u lookup error: %s",
			      params->job_id, params->step_id,
			      slurm_strerror(slurm_get_errno()));
		}
		return rc;
	}
	if (params->step_id == NO_VAL)
		verbose("jobid      = %u", params->job_id);
	else
		verbose("stepid     = %u.%u", params->job_id, params->step_id);
	verbose("node_cnt   = %u", sbcast_cred->node_cnt);
	verbose("node_list  = %s", sbcast_cred->node_list);
	/* also see sbcast_cred->node_addr (array) */

	if (params->verbose)
		print_sbcast_cred(sbcast_cred->sbcast_cred);

	/* do not bother to release the return message,
	 * we need to preserve and use most of the information later */

	return rc;
}

/*
 * scontrol_requeue - requeue a pending or running batch job
 * IN job_id_str - a job id
 * RET 0 if no slurm error, errno otherwise. parsing error prints
 *              error message and returns 0
 */
extern int
scontrol_requeue(int argc, char **argv)
{
	int rc = SLURM_SUCCESS;
	int i;
	uint32_t *ids;
	uint32_t num_ids;

	if (! argv[0]) {
		exit_code = 1;
		return 0;
	}

	ids = _get_job_ids(argv[0], &num_ids);
	if (ids == NULL) {
		exit_code = 1;
		return 0;
	}

	for (i = 0; i < num_ids; i++) {
		rc = slurm_requeue(ids[i], 0);
		if (rc != SLURM_SUCCESS) {
			fprintf(stderr, "%s  array job_id %u\n",
					slurm_strerror(slurm_get_errno()), ids[i]);
			exit_code = 1;
			break;
		}
	}

	xfree(ids);

	return rc;
}

/*
 * slurm_signal_job - send the specified signal to all steps of an existing job
 * IN job_id     - the job's id
 * IN signal     - signal number
 * RET 0 on success, otherwise return -1 and set errno to indicate the error
 */
extern int
slurm_signal_job (uint32_t job_id, uint16_t signal)
{
	int rc = SLURM_SUCCESS;
	resource_allocation_response_msg_t *alloc_info = NULL;
	signal_job_msg_t rpc;

	if (slurm_allocation_lookup_lite(job_id, &alloc_info)) {
		rc = slurm_get_errno();
		goto fail1;
	}

	/* same remote procedure call for each node */
	rpc.job_id = job_id;
	rpc.signal = (uint32_t)signal;

	rc = _local_send_recv_rc_msgs(alloc_info->node_list,
				      REQUEST_SIGNAL_JOB, &rpc);
	slurm_free_resource_allocation_response_msg(alloc_info);
fail1:
	if (rc) {
		slurm_seterrno_ret(rc);
	} else {
		return SLURM_SUCCESS;
	}
}

/* Load current node table information into *node_buffer_pptr */
extern int
scontrol_load_nodes (node_info_msg_t ** node_buffer_pptr, uint16_t show_flags)
{
	int error_code;
	static int last_show_flags = 0xffff;
	node_info_msg_t *node_info_ptr = NULL;

	if (old_node_info_ptr) {
		if (last_show_flags != show_flags)
			old_node_info_ptr->last_update = (time_t) 0;
		error_code = slurm_load_node (old_node_info_ptr->last_update,
					      &node_info_ptr, show_flags);
		if (error_code == SLURM_SUCCESS)
			slurm_free_node_info_msg (old_node_info_ptr);
		else if (slurm_get_errno () == SLURM_NO_CHANGE_IN_DATA) {
			node_info_ptr = old_node_info_ptr;
			error_code = SLURM_SUCCESS;
			if (quiet_flag == -1)
				printf ("slurm_load_node no change in data\n");
		}
	}
	else
		error_code = slurm_load_node ((time_t) NULL, &node_info_ptr,
					      show_flags);

	if (error_code == SLURM_SUCCESS) {
		old_node_info_ptr = node_info_ptr;
		last_show_flags = show_flags;
		*node_buffer_pptr = node_info_ptr;
	}

	return error_code;
}

static resource_allocation_response_msg_t *
_wait_for_allocation_response(uint32_t job_id, const listen_t *listen,
			      int timeout)
{
	resource_allocation_response_msg_t *resp = NULL;
	int errnum;

	info("job %u queued and waiting for resources", job_id);
	if (_wait_for_alloc_rpc(listen, timeout, &resp) <= 0) {
		errnum = errno;
		/* Maybe the resource allocation response RPC got lost
		 * in the mail; surely it should have arrived by now.
		 * Let's see if the controller thinks that the allocation
		 * has been granted.
		 */
		if (slurm_allocation_lookup_lite(job_id, &resp) >= 0) {
			return resp;
		}
		if (slurm_get_errno() == ESLURM_JOB_PENDING) {
			debug3("Still waiting for allocation");
			errno = errnum;
			return NULL;
		} else {
			debug3("Unable to confirm allocation for job %u: %m",
			       job_id);
			return NULL;
		}
	}
	info("job %u has been allocated resources", job_id);
	return resp;
}

extern int get_new_info_config(slurm_ctl_conf_info_msg_t **info_ptr)
{
	static slurm_ctl_conf_info_msg_t *new_ctl_ptr = NULL;
	int error_code = SLURM_NO_CHANGE_IN_DATA;

	if (g_ctl_info_ptr) {
		error_code = slurm_load_ctl_conf(g_ctl_info_ptr->last_update,
						 &new_ctl_ptr);
		if (error_code == SLURM_SUCCESS)
			slurm_free_ctl_conf(g_ctl_info_ptr);
		else if (slurm_get_errno () == SLURM_NO_CHANGE_IN_DATA) {
			error_code = SLURM_NO_CHANGE_IN_DATA;
			new_ctl_ptr = g_ctl_info_ptr;
		}
	} else {
		new_ctl_ptr = NULL;
		error_code = slurm_load_ctl_conf((time_t) NULL, &new_ctl_ptr);
	}
	g_ctl_info_ptr = new_ctl_ptr;

	if (g_ctl_info_ptr && (*info_ptr != g_ctl_info_ptr))
		error_code = SLURM_SUCCESS;

	*info_ptr = new_ctl_ptr;

	return error_code;

}

/*
 * Send message to stdout of specified job
 * argv[0] == jobid
 * argv[1]++ the message
 */
extern int
scontrol_job_notify(int argc, char *argv[])
{
	int i;
	uint32_t job_id;
	char *message = NULL;

	job_id = atoi(argv[0]);
	if (job_id <= 0) {
		fprintf(stderr, "Invalid job_id %s", argv[0]);
		return 1;
	}

	for (i=1; i<argc; i++) {
		if (message)
			xstrfmtcat(message, " %s", argv[i]);
		else
			xstrcat(message, argv[i]);
	}

	i = slurm_notify_job(job_id, message);
	xfree(message);

	if (i)
		return slurm_get_errno ();
	else
		return 0;
}

/*
 * scontrol_create_part - create a slurm partition configuration per the
 *	supplied arguments
 * IN argc - count of arguments
 * IN argv - list of arguments
 * RET 0 if no slurm error, errno otherwise. parsing error prints
 *			error message and returns 0
 */
extern int
scontrol_create_part (int argc, char *argv[])
{
	int update_cnt = 0;
	update_part_msg_t part_msg;

	slurm_init_part_desc_msg ( &part_msg );
	scontrol_parse_part_options (argc, argv, &update_cnt, &part_msg);

	if (part_msg.name == NULL) {
		exit_code = 1;
		error("PartitionName must be given.");
		return 0;
	} else if (xstrcasecmp(part_msg.name, "default") == 0) {
		exit_code = 1;
		error("PartitionName cannot be \"DEFAULT\".");
		return 0;
	}

	if (update_cnt == 0) {
		exit_code = 1;
		error("No parameters specified");
		return 0;
	}

	if (slurm_create_partition(&part_msg)) {
		exit_code = 1;
		slurm_perror("Error creating the partition");
		return slurm_get_errno ();
	} else
		return 0;
}

/*
 * Load current front_end table information into *node_buffer_pptr
 */
extern int
scontrol_load_front_end(front_end_info_msg_t ** front_end_buffer_pptr)
{
	int error_code;
	front_end_info_msg_t *front_end_info_ptr = NULL;

	if (old_front_end_info_ptr) {
		error_code = slurm_load_front_end (
				old_front_end_info_ptr->last_update,
				&front_end_info_ptr);
		if (error_code == SLURM_SUCCESS)
			slurm_free_front_end_info_msg (old_front_end_info_ptr);
		else if (slurm_get_errno () == SLURM_NO_CHANGE_IN_DATA) {
			front_end_info_ptr = old_front_end_info_ptr;
			error_code = SLURM_SUCCESS;
			if (quiet_flag == -1) {
				printf("slurm_load_front_end no change in "
				       "data\n");
			}
		}
	}
	else
		error_code = slurm_load_front_end((time_t) NULL,
						  &front_end_info_ptr);

	if (error_code == SLURM_SUCCESS) {
		old_front_end_info_ptr = front_end_info_ptr;
		*front_end_buffer_pptr = front_end_info_ptr;
	}

	return error_code;
}

/* Load current partiton table information into *part_buffer_pptr */
extern int
scontrol_load_block (block_info_msg_t **block_info_pptr)
{
	int error_code;
	block_info_msg_t *info_ptr = NULL;
	uint16_t show_flags = 0;

	if (all_flag)
		show_flags |= SHOW_ALL;
	if (old_block_info_ptr) {
		error_code = slurm_load_block_info(
			old_block_info_ptr->last_update, &info_ptr, show_flags);
		if (error_code == SLURM_SUCCESS)
			slurm_free_block_info_msg(old_block_info_ptr);
		else if (slurm_get_errno() == SLURM_NO_CHANGE_IN_DATA) {
			info_ptr = old_block_info_ptr;
			error_code = SLURM_SUCCESS;
			if (quiet_flag == -1)
				printf ("slurm_load_block no "
					"change in data\n");
		}
	} else
		error_code = slurm_load_block_info((time_t)NULL,
						   &info_ptr, show_flags);

	if (error_code == SLURM_SUCCESS) {
		old_block_info_ptr = info_ptr;
		*block_info_pptr = info_ptr;
	}

	return error_code;
}

/*
 * scontrol_update_part - update the slurm partition configuration per the
 *	supplied arguments
 * IN argc - count of arguments
 * IN argv - list of arguments
 * RET 0 if no slurm error, errno otherwise. parsing error prints
 *			error message and returns 0
 */
extern int
scontrol_update_part (int argc, char *argv[])
{
	int update_cnt = 0;
	update_part_msg_t part_msg;

	slurm_init_part_desc_msg ( &part_msg );
	scontrol_parse_part_options (argc, argv, &update_cnt, &part_msg);

	if (part_msg.name == NULL) {
		exit_code = 1;
		error("PartitionName must be given.");
		return 0;
	}
	if (update_cnt <= 1) {
		exit_code = 1;
		error("No changes specified");
		return 0;
	}

	if (slurm_update_partition(&part_msg)) {
		exit_code = 1;
		return slurm_get_errno ();
	} else
		return 0;
}

/* get details about this slurm job: jobid and allocated node */
static void _get_job_info(void)
{
	char *jobid_str;
	uint32_t jobid;

	jobid_str = getenv("SLURM_JOB_ID");
	if (!jobid_str) {
		error("Command only valid from within SLURM job");
		exit(1);
	}
	jobid = (uint32_t) atol(jobid_str);
	verbose("jobid      = %u", jobid);

	if (slurm_sbcast_lookup(jobid, &sbcast_cred) != SLURM_SUCCESS) {
		error("SLURM jobid %u lookup error: %s",
		      jobid, slurm_strerror(slurm_get_errno()));
		exit(1);
	}

	verbose("node_cnt   = %u", sbcast_cred->node_cnt);
	verbose("node_list  = %s", sbcast_cred->node_list);
	/* also see sbcast_cred->node_addr (array) */

	if (params.verbose)
		print_sbcast_cred(sbcast_cred->sbcast_cred);

	/* do not bother to release the return message,
	 * we need to preserve and use most of the information later */
}

/* Load current job table information into *job_buffer_pptr */
extern int
scontrol_load_job(job_info_msg_t ** job_buffer_pptr, uint32_t job_id)
{
	int error_code;
	static uint16_t last_show_flags = 0xffff;
	uint16_t show_flags = 0;
	job_info_msg_t * job_info_ptr = NULL;

	if (all_flag)
		show_flags |= SHOW_ALL;

	if (detail_flag) {
		show_flags |= SHOW_DETAIL;
		if (detail_flag > 1)
			show_flags |= SHOW_DETAIL2;
	}
	if (federation_flag)
		show_flags |= SHOW_FEDERATION;
	if (local_flag)
		show_flags |= SHOW_LOCAL;
	if (sibling_flag)
		show_flags |= SHOW_FEDERATION | SHOW_SIBLING;

	if (old_job_info_ptr) {
		if (last_show_flags != show_flags)
			old_job_info_ptr->last_update = (time_t) 0;
		if (job_id) {
			error_code = slurm_load_job(&job_info_ptr, job_id,
						    show_flags);
		} else {
			error_code = slurm_load_jobs(
				old_job_info_ptr->last_update,
				&job_info_ptr, show_flags);
		}
		if (error_code == SLURM_SUCCESS)
			slurm_free_job_info_msg (old_job_info_ptr);
		else if (slurm_get_errno () == SLURM_NO_CHANGE_IN_DATA) {
			job_info_ptr = old_job_info_ptr;
			error_code = SLURM_SUCCESS;
			if (quiet_flag == -1)
 				printf ("slurm_load_jobs no change in data\n");
		}
	} else if (job_id) {
		error_code = slurm_load_job(&job_info_ptr, job_id, show_flags);
	} else {
		error_code = slurm_load_jobs((time_t) NULL, &job_info_ptr,
					     show_flags);
	}

	if (error_code == SLURM_SUCCESS) {
		old_job_info_ptr = job_info_ptr;
		if (job_id)
			old_job_info_ptr->last_update = (time_t) 0;
		last_show_flags  = show_flags;
		*job_buffer_pptr = job_info_ptr;
	}

	return error_code;
}