mom_mach.c

来自「OpenPBS」· C语言 代码 · 共 2,037 行 · 第 1/3 页

			if (retval != PBSE_NONE)
			        return (error(pname, retval));
			if ((mem_limit == 0) || (value < mem_limit))
				mem_limit = value;
		} else if (strcmp(pname, "pvmem") == 0) {	/* set */
			if (set_mode == SET_LIMIT_SET)  {
			    retval = getsize(pres, &value);
			    if (retval != PBSE_NONE)
			        return (error(pname, retval));
			    if (value > INT_MAX)
			        return (error(pname, PBSE_BADATVAL));
			if ((mem_limit == 0) || (value < mem_limit))
				mem_limit = value;
			}
		} else if (strcmp(pname, "pmem") == 0) {	/* set */
			if (set_mode == SET_LIMIT_SET)  {
			    retval = getsize(pres, &value);
			    if (retval != PBSE_NONE)
			        return (error(pname, retval));
			    reslim.rlim_cur = reslim.rlim_max = value;
			    if (setrlimit(RLIMIT_RSS, &reslim) < 0)
	        		return (error("RLIMIT_RSS", PBSE_SYSTEM));
			}
		} else if (strcmp(pname, "walltime") == 0) {	/* Check */
			retval = gettime(pres, &value);
			if (retval != PBSE_NONE)
			        return (error(pname, retval));
		} else if (strcmp(pname, "nice") == 0) {	/* set nice */
			if (set_mode == SET_LIMIT_SET)  {
			    errno = 0;
			    if ((nice((int)pres->rs_value.at_val.at_long) == -1)
			        && (errno != 0))
				return (error(pname, PBSE_BADATVAL));
			}
		} else if ((pres->rs_defin->rs_flags & ATR_DFLAG_RMOMIG) == 0)
			/* don't recognize and not marked as ignore by mom */
			return (error(pname, PBSE_UNKRESC));
		pres = (resource *)GET_NEXT(pres->rs_link);
	}
	if (set_mode == SET_LIMIT_SET)  {
	    /* if either of vmem or pvmem was given, set sys limit to lesser */
	    if (mem_limit != 0) {
		reslim.rlim_cur = reslim.rlim_max = mem_limit;
		if (setrlimit(RLIMIT_DATA, &reslim) < 0)
	        	return (error("RLIMIT_DATA", PBSE_SYSTEM));
		if (setrlimit(RLIMIT_STACK, &reslim) < 0)
	        	return (error("RLIMIT_STACK", PBSE_SYSTEM));
	    }
	}

	return (PBSE_NONE);
}

/*
 * State whether MOM main loop has to poll this job to determine if some
 * limits are being exceeded.
 *
 *	Sets flag TRUE if polling is necessary, FALSE otherwise.  Actual
 *	polling is done using the mom_over_limit machine-dependent function.
 */

int mom_do_poll(pjob)
    job			*pjob;
{
	char		*id = "mom_do_poll";
	char		*pname;
	resource	*pres;

	DBPRT(("%s: entered\n", id))
	assert(pjob != NULL);
	assert(pjob->ji_wattr[(int)JOB_ATR_resource].at_type == ATR_TYPE_RESC);
	pres = (resource *)
	    GET_NEXT(pjob->ji_wattr[(int)JOB_ATR_resource].at_val.at_list);

	while (pres != NULL) {
		assert(pres->rs_defin != NULL);
		pname = pres->rs_defin->rs_name;
		assert(pname != NULL);
		assert(*pname != '\0');

		if (strcmp(pname, "walltime") == 0 ||
		    strcmp(pname, "cput") == 0 ||
		    strcmp(pname, "pvmem") == 0 ||
		    strcmp(pname, "vmem") == 0)
			return (TRUE);
		pres = (resource *)GET_NEXT(pres->rs_link);
	}

	return (FALSE);
}

/*
 * Setup for polling.
 *
 */
int mom_open_poll()
{
	char		*id = "mom_open_poll";

	DBPRT(("%s: entered\n", id))

	proc_tbl = malloc(ASIZE*sizeof(struct procsinfo));
	proctot = ASIZE;

	return (PBSE_NONE);
}


/*
 * Declare start of polling loop.
 *
 *	Until the next call to mom_get_sample, all mom_over_limit calls will
 *	use the same data.  Returns a PBS error code.
 */

int mom_get_sample()
{
	char		*id = "mom_get_sample";
	struct	procsinfo	*pp;
	int		num, addnum;
	pid_t		pid;

	DBPRT(("%s: entered\n", id))

	addnum = proctot;
	nproc = 0;
	pid = 0;
	pp = proc_tbl;

	while ((num = getprocs(pp, sizeof(struct procsinfo),
			NULL, sizeof(struct fdsinfo),
			&pid, addnum)) > 0) {
		DBPRT(("%s: loop start: got %d\n", id, num))

		nproc += num;
		if (num < addnum)
			break;

		proctot += ASIZE;
		addnum = ASIZE;
		proc_tbl = realloc(proc_tbl, proctot*sizeof(struct procsinfo));
		pp = &proc_tbl[nproc];
	}
	if (num == -1) {
		log_err(errno, id, "getprocs");
		return PBSE_SYSTEM;
	}
	DBPRT(("%s: nproc = %d\n", id, nproc))


	return (PBSE_NONE);
}

/*
 * Measure job resource usage and compare with its limits.
 *
 *	If it has exceeded any well-formed polled limit return TRUE.
 *	Otherwise, return FALSE.
 */

int mom_over_limit(pjob)
    job			*pjob;
{
	char		*id = "mom_over_limit";
	char		*pname;
	int		retval;
	unsigned long	value, num;
	resource	*pres;

	assert(pjob != NULL);
	assert(pjob->ji_wattr[(int)JOB_ATR_resource].at_type == ATR_TYPE_RESC);

	DBPRT(("%s: entered\n", id))


	pres = (resource *)
	    GET_NEXT(pjob->ji_wattr[(int)JOB_ATR_resource].at_val.at_list);
	for ( ; pres != NULL; pres = (resource *)GET_NEXT(pres->rs_link)) {
		assert(pres->rs_defin != NULL);
		pname = pres->rs_defin->rs_name;
		assert(pname != NULL);
		assert(*pname != '\0');
		if (strcmp(pname, "cput") == 0) {
			retval = gettime(pres, &value);
			if (retval != PBSE_NONE)
				continue;
			if ((num = cput_sum(pjob)) > value) {
				sprintf(log_buffer,
					"cput %lu exceeded limit %lu",
					num, value);
				return (TRUE);
			}
		} else if (strcmp(pname, "vmem") == 0) {
			retval = getsize(pres, &value);
			if (retval != PBSE_NONE)
				continue;
			if ((num = mem_sum(pjob)) > value) {
				sprintf(log_buffer,
					"vmem %lu exceeded limit %lu",
					num, value);
				return (TRUE);
			}
		} else if (strcmp(pname, "pvmem") == 0) {
			retval = getsize(pres, &value);
			if (retval != PBSE_NONE)
				continue;
			if (overmem_proc(pjob, value)) {
				sprintf(log_buffer,
					"pvmem exceeded limit %lu", value);
				return (TRUE);
			}
		} else if (strcmp(pname, "walltime") == 0) {
			retval = gettime(pres, &value);
			if (retval != PBSE_NONE)
				continue;
			num = (unsigned long)((double)(time_now - pjob->ji_qs.ji_stime) * wallfactor);
			if (num > value) {
				sprintf(log_buffer,
					"walltime %d exceeded limit %d",
					num, value);
				return (TRUE);
			}
		}
	}

	return (FALSE);
}

/*
 * Update the job attribute for resources used.
 *
 *	The first time this is called for a job, set up resource entries for
 *	each resource that can be reported for this machine.  Fill in the
 *	correct values.  Return an error code.
 *
 *	Assumes that the session ID attribute has already been set.
 */
int mom_set_use(pjob)
    job			*pjob;
{
	char			*id = "mom_set_use";
	resource		*pres;
	attribute		*at;
	resource_def		*rd;
	unsigned long		*lp, lnum;

	DBPRT(("%s: entered\n", id))
	assert(pjob != NULL);
	at = &pjob->ji_wattr[(int)JOB_ATR_resc_used];
	assert(at->at_type == ATR_TYPE_RESC);

	at->at_flags |= ATR_VFLAG_MODIFY;
	if ((at->at_flags & ATR_VFLAG_SET) == 0) {
		at->at_flags |= ATR_VFLAG_SET;

		rd = find_resc_def(svr_resc_def, "cput", svr_resc_size);
		assert(rd != NULL);
		pres = add_resource_entry(at, rd);
		assert(pres != NULL);
		pres->rs_value.at_flags |= ATR_VFLAG_SET;
		pres->rs_value.at_type = ATR_TYPE_LONG;
		pres->rs_value.at_val.at_long = 0;

		rd = find_resc_def(svr_resc_def, "vmem", svr_resc_size);
		assert(rd != NULL);
		pres = add_resource_entry(at, rd);
		assert(pres != NULL);
		pres->rs_value.at_flags |= ATR_VFLAG_SET;
		pres->rs_value.at_type = ATR_TYPE_SIZE;
		pres->rs_value.at_val.at_size.atsv_shift = 10; /* KB */
		pres->rs_value.at_val.at_size.atsv_units = ATR_SV_BYTESZ;
		pres->rs_value.at_val.at_size.atsv_num = 0;

		rd = find_resc_def(svr_resc_def, "walltime", svr_resc_size);
		assert(rd != NULL);
		pres = add_resource_entry(at, rd);
		assert(pres != NULL);
		pres->rs_value.at_flags |= ATR_VFLAG_SET;
		pres->rs_value.at_type = ATR_TYPE_LONG;
		pres->rs_value.at_val.at_long = 0;

		rd = find_resc_def(svr_resc_def, "mem", svr_resc_size);
		assert(rd != NULL);
		pres = add_resource_entry(at, rd);
		assert(pres != NULL);
		pres->rs_value.at_flags |= ATR_VFLAG_SET;
		pres->rs_value.at_type = ATR_TYPE_SIZE;
		pres->rs_value.at_val.at_size.atsv_shift = 10; /* KB */
		pres->rs_value.at_val.at_size.atsv_units = ATR_SV_BYTESZ;
		pres->rs_value.at_val.at_size.atsv_num = 0;
	}

	rd = find_resc_def(svr_resc_def, "cput", svr_resc_size);
	assert(rd != NULL);
	pres = find_resc_entry(at, rd);
	assert(pres != NULL);
	lp = (unsigned long *)&pres->rs_value.at_val.at_long;
	lnum = cput_sum(pjob);
	*lp = MAX(*lp, lnum);

	rd = find_resc_def(svr_resc_def, "vmem", svr_resc_size);
	assert(rd != NULL);
	pres = find_resc_entry(at, rd);
	assert(pres != NULL);
	lp = &pres->rs_value.at_val.at_size.atsv_num;
	lnum = (mem_sum(pjob) + 1023) >> 10;	/* as KB */
	*lp = MAX(*lp, lnum);

	rd = find_resc_def(svr_resc_def, "walltime", svr_resc_size);
	assert(rd != NULL);
	pres = find_resc_entry(at, rd);
	assert(pres != NULL);
	pres->rs_value.at_val.at_long = (long)((double)(time_now - pjob->ji_qs.ji_stime) * wallfactor);

	rd = find_resc_def(svr_resc_def, "mem", svr_resc_size);
	assert(rd != NULL);
	pres = find_resc_entry(at, rd);
	assert(pres != NULL);
	lp = &pres->rs_value.at_val.at_size.atsv_num;
	lnum = (resi_sum(pjob) + 1023) >> 10;	/* as KB */
	*lp = MAX(*lp, lnum);

	return (PBSE_NONE);
}

/*
 *	Kill a task session.
 *	Call with the job pointer and a signal number.
 */
int kill_task(ptask, sig)
    task	*ptask;
    int  	sig;
{
	int	ct = 0;
	int	i, err;
	int	sesid;

	sesid = ptask->ti_qs.ti_sid;
	if (sesid <= 1)
		return 0;

	if ((err = mom_get_sample()) != PBSE_NONE)
		return 0;

	for (i=0; i<nproc; i++) {
		register struct procsinfo	*pp = &proc_tbl[i];

		if (pp->pi_state == SNONE)
			continue;

		if (sesid != pp->pi_sid)
			continue;

		DBPRT(("kill_task: send signal %d to pid %d\n",
			sig, pp->pi_pid))
		(void)kill(pp->pi_pid, sig);
		++ct;
	}
	return ct;
}

/*
 * Clean up everything related to polling.
 *
 */
int mom_close_poll()
{
	DBPRT(("mom_close_poll entered\n"))

	if (proc_tbl) {
		free(proc_tbl);
		proc_tbl = NULL;
	}

	return (PBSE_NONE);
}

/*
 * mom_does_chkpnt - return 1 if mom supports checkpoint
 *			    0 if not
 */

int mom_does_chkpnt()
{
	return (0);
}

/*
 * Checkpoint the job.
 *
 *	If abort is true, kill it too.
 */

int mach_checkpoint(ptask, file, abort)
    task	*ptask;
    char	*file;
    int		abort;
{
       	return (-1);
}

/*
 * Restart the job from the checkpoint file.
 *
 *	Return a -1 on error or sid.
 */

long mach_restart(ptask, file)
    task	*ptask;
    char	*file;
{
	return (-1);
}

int
kvm_read(fd, addr, buf, size)
    int		fd;
    long	addr;
    char	*buf;
    int		size;
{
	int	ret;

	if (lseek(fd, addr, SEEK_SET) != addr)
		return -1;
	if ((ret = read(fd, buf, size)) == -1)
		return -1;
	return ret;
}

int
getproctab()
{
	static	uint	lastproc = 0;
	char		*id = "getproctab";

	if (lastproc == reqnum)		/* don't need new proc table */
		return nproc;

	if (mom_get_sample() != PBSE_NONE)
		return 0;

	lastproc = reqnum;
	return(nproc);
}

double
dsecs(val)
struct	timeval	*val;
{
        return ( (double)val->tv_sec + (double)val->tv_usec*1e-6 );
}

char	*
cput_job(jobid)
pid_t	jobid;
{
	char		*id = "cput_job";
	int		i, nproc;
	int		found = 0;
	double		cputime, addtime;

	if ((nproc = getproctab()) == 0) {
		rm_errno = RM_ERR_SYSTEM;
		return NULL;
	}
	cputime = 0;
	for (i=0; i<nproc; i++) {
		register struct procsinfo	*pp = &proc_tbl[i];

		if (pp->pi_state == SNONE)
			continue;

		if (jobid != pp->pi_sid)
			continue;

		found = 1;
		DBPRT(("%s: pid=%d", id, pp->pi_pid))
		if (pp->pi_state == SZOMB) {
			DBPRT((" (zombie)"))
			addtime = dsecs(&pp->pi_utime) +
				dsecs(&pp->pi_stime);
		}
		else {
			DBPRT((" (active)"))
			addtime = dsecs(&pp->pi_ru.ru_utime) +
				dsecs(&pp->pi_ru.ru_stime) +
				dsecs(&pp->pi_cru.ru_utime) +
				dsecs(&pp->pi_cru.ru_stime);
		}
		cputime += addtime;
		DBPRT((" %.2f total=%.2f\n", addtime, cputime))

	}
	if (found) {
		sprintf(ret_string, "%.2f", cputime * cputfactor);
		return ret_string;
	}

	rm_errno = RM_ERR_EXIST;
	return NULL;
}

char	*
cput_proc(pid)
pid_t	pid;
{
	char			*id = "cput_proc";
	int			i, nproc;
	int			found = 0;
	double			cputime;

	if ((nproc = getproctab()) == 0) {
		rm_errno = RM_ERR_SYSTEM;
		return NULL;
	}
	for (i=0; i<nproc; i++) {
		register struct procsinfo	*pp = &proc_tbl[i];

		if (pp->pi_state == SNONE)
			continue;

		if (pid != pp->pi_pid)
			continue;

		DBPRT(("%s: pid=%d", id, pp->pi_pid))
		if (pp->pi_state == SZOMB) {
			DBPRT((" (zombie)"))
			cputime = dsecs(&pp->pi_utime) +
				dsecs(&pp->pi_stime);
		}
		else {
			DBPRT((" (active)"))
			cputime = dsecs(&pp->pi_ru.ru_utime) +
				dsecs(&pp->pi_ru.ru_stime) +
				dsecs(&pp->pi_cru.ru_utime) +
				dsecs(&pp->pi_cru.ru_stime);
		}
		DBPRT((" %.2f\n", cputime))
		found = 1;
		break;
	}
	if (found) {
		sprintf(ret_string, "%.2f", cputime * cputfactor);
		return ret_string;
	}

	rm_errno = RM_ERR_EXIST;
	return NULL;
}

char	*
cput(attrib)
struct	rm_attribute	*attrib;
{
	char			*id = "cput";
	int			value;

	if (attrib == NULL) {
		log_err(-1, id, no_parm);
		rm_errno = RM_ERR_NOPARAM;
		return NULL;
	}
	if ((value = atoi(attrib->a_value)) == 0) {
		sprintf(log_buffer, "bad param: %s", attrib->a_value);
		log_err(-1, id, log_buffer);
		rm_errno = RM_ERR_BADPARAM;
		return NULL;
	}
	if (momgetattr(NULL)) {
		log_err(-1, id, extra_parm);
		rm_errno = RM_ERR_BADPARAM;
		return NULL;
	}

	if (strcmp(attrib->a_qualifier, "session") == 0)
		return (cput_job((pid_t)value));
	else if (strcmp(attrib->a_qualifier, "proc") == 0)
		return (cput_proc((pid_t)value));
	else {
		rm_errno = RM_ERR_BADPARAM;
		return NULL;
	}
}

char	*
mem_job(jobid)
pid_t	jobid;
{
	char			*id = "mem_job";
	int			i, nproc;
	int			memsize;
	int			found = 0;

	if ((nproc = getproctab()) == 0) {
		rm_errno = RM_ERR_SYSTEM;
		return NULL;
	}
	memsize = 0;
	for (i=0; i<nproc; i++) {
		register struct procsinfo	*pp = &proc_tbl[i];

		if (pp->pi_state == SNONE)
			continue;

		if (jobid != pp->pi_sid)
			continue;

		found = 1;
		memsize += pp->pi_size;
		DBPRT(("%s: pid %d memsize %d pi_size %d\n", id, pp->pi_pid,
				memsize, pp->pi_size))
	}
	if (found) {
		sprintf(ret_string, "%ukb", ctob(memsize) >> 10); /* KB */
		return ret_string;
	}

	rm_errno = RM_ERR_EXIST;
	return NULL;
}

char	*
mem_proc(pid)
pid_t	pid;
{
	char			*id = "mem_proc";
	int			i, nproc;
	int			memsize;
	int			found = 0;

	if ((nproc = getproctab()) == 0) {
		rm_errno = RM_ERR_SYSTEM;
		return NULL;
	}
	memsize = 0;
	for (i=0; i<nproc; i++) {
		register struct procsinfo	*pp = &proc_tbl[i];

		if (pp->pi_state == SNONE)
			continue;

		if (pid != pp->pi_pid)
			continue;

		found = 1;
		memsize = pp->pi_size;
		break;
	}
	if (found) {
		sprintf(ret_string, "%ukb", ctob(memsize) >> 10); /* KB */
		return ret_string;
	}

	rm_errno = RM_ERR_EXIST;
	return NULL;
}

char	*
mem(attrib)
struct	rm_attribute	*attrib;
{
	char			*id = "mem";
	int			value;

	if (attrib == NULL) {
		log_err(-1, id, no_parm);
		rm_errno = RM_ERR_NOPARAM;
		return NULL;
	}
	if ((value = atoi(attrib->a_value)) == 0) {
		sprintf(log_buffer, "bad param: %s", attrib->a_value);
		log_err(-1, id, log_buffer);
		rm_errno = RM_ERR_BADPARAM;
		return NULL;
	}
	if (momgetattr(NULL)) {
		log_err(-1, id, extra_parm);
		rm_errno = RM_ERR_BADPARAM;