mom_mach.c

openPBS的开放源代码

		if (strcmp(pname, "walltime") == 0)	return (TRUE);

		pres = (resource *)GET_NEXT(pres->rs_link);
	}

	return (FALSE);
}

/*
 * Setup for polling.
 *
 *	Open kernel device and get namelist info.
 */
int mom_open_poll()
{
	static char	*id = "mom_open_poll";

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

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

	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);

	DBPRT(("%s: entered %s\n", id, pjob->ji_qs.ji_jobid))
	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, "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.
 */
int mom_set_use(pjob)
    job			*pjob;
{
	static char		*id = "mom_set_use";
	resource		*pres;
	attribute		*at;
	resource_def		*rd;
	long			*lp, num;

	assert(pjob != NULL);
	at = &pjob->ji_wattr[(int)JOB_ATR_resc_used];
	assert(at->at_type == ATR_TYPE_RESC);

	if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_Suspend) != 0)
		return (PBSE_NONE);	/* job suspended, don't track it */

	DBPRT(("%s: entered %s\n", id, pjob->ji_qs.ji_jobid))

	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);
		pres->rs_value.at_flags |= ATR_VFLAG_SET;
		pres->rs_value.at_type = ATR_TYPE_LONG;

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

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

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

	rd = find_resc_def(svr_resc_def, "mppt", svr_resc_size);
	assert(rd != NULL);
	pres = find_resc_entry(at, rd);
	assert(pres != NULL);
	lp = &pres->rs_value.at_val.at_long;
	num = mppt_sum(pjob);
	*lp = max(*lp, num);

	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);

	return (PBSE_NONE);
}

/*
 *	Kill a task session.
 *	Call with the task pointer and a signal number.
 */
int kill_task(ptask, sig)
    task	*ptask;
    int		sig;
{
	static char	*id = "kill_task";
	int		ct = 1;
        int		sesid;

	sesid = ptask->ti_qs.ti_sid;
	if (sesid > 1) {
		if (killm(C_JOB, sesid, sig) == -1) {
		    if (errno != ESRCH) {
			sprintf(log_buffer, "killm: sid=%d sig=%d", sesid, sig);
			log_err(errno, id, log_buffer);
		    } else {
			ct = 0;
			sprintf(log_buffer, "killm: sid=%d sig=%d", sesid, sig);
			log_record(PBSEVENT_DEBUG, PBS_EVENTCLASS_JOB,
				   ptask->ti_job->ji_qs.ji_jobid, log_buffer);
		    }
		}
	}
	return ct;
}

/*
 * Clean up everything related to polling.
 */
int mom_close_poll()
{
	return (PBSE_NONE);
}

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

int mom_does_chkpnt()
{
	return (1);
}

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

int mach_checkpoint(ptask, path, abort)
    task	*ptask;
    char	*path;
    int		abort;
{
	int	cprtn;
	long	flags = 0;

	if (abort)
		flags = CHKPNT_KILL;
	cprtn = chkpnt( C_JOB, ptask->ti_qs.ti_sid, path, flags );
	return cprtn;
}

/*
 * Restart the job from the checkpoint file.
 *
 *	Return the session/job id
 */

long mach_restart(ptask, path)
    task	*ptask;
    char	*path;
{
	int	sid;

	sid = restart(path, 0);
	return sid;
}

char	*
cput_job(jobid)
pid_t	jobid;
{
	char		*id = "cput_job";
	struct	resclim		rl;

	rl.resc_category = C_JOB;
	rl.resc_resource = L_CPU;
	if (getlim(jobid, &rl) == -1) {
		rm_errno = RM_ERR_EXIST;
		return NULL;
	}

	sprintf(ret_string, "%.2f", cputfactor * rl.resc_used);
	return ret_string;
}

char	*
cput_proc(pid)
pid_t	pid;
{
	char		*id = "cput_pid";
	struct	resclim		rl;

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

	rl.resc_category = C_PROC;
	rl.resc_resource = L_CPU;
	if (getlim(pid, &rl) == -1) {
		rm_errno = RM_ERR_EXIST;
		return NULL;
	}

	sprintf(ret_string, "%.2f", cputfactor * rl.resc_used);
	return ret_string;
}

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";

	rm_errno = RM_ERR_EXIST;
	return NULL;
}

char	*
mem_proc(pid)
pid_t	pid;
{
	char		*id = "mem_proc";

	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;
		return NULL;
	}

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

char	*
sessions(attrib)
struct	rm_attribute	*attrib;
{
	char			*id = "sessions";

	if (attrib) {
		log_err(-1, id, extra_parm);
		rm_errno = RM_ERR_BADPARAM;
		return NULL;
	}
	rm_errno = RM_ERR_EXIST;
	return NULL;
}

char	*
nsessions(attrib)
struct	rm_attribute	*attrib;
{
	char	*result, *ch;
	int	num = 0;

	if ((result = sessions(attrib)) == NULL)
		return result;

	for (ch=result; *ch; ch++) {
		if (*ch == ' ')		/* count blanks */
			num++;
	}
	sprintf(ret_string, "%d", num);
	return ret_string;
}

char	*
nusers(attrib)
struct	rm_attribute	*attrib;
{
	char			*id = "nusers";

	if (attrib) {
		log_err(-1, id, extra_parm);
		rm_errno = RM_ERR_BADPARAM;
		return NULL;
	}
	rm_errno = RM_ERR_EXIST;
	return NULL;
}

char	*
pids(attrib)
struct	rm_attribute	*attrib;
{
	char		*id = "pids";
	pid_t		jobid;

	if (attrib == NULL) {
		log_err(-1, id, no_parm);
		rm_errno = RM_ERR_NOPARAM;
		return NULL;
	}
	if ((jobid = (pid_t)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) {
		rm_errno = RM_ERR_BADPARAM;
		return NULL;
	}
	rm_errno = RM_ERR_EXIST;
	return NULL;
}

static char	*
totmem(attrib)
struct	rm_attribute	*attrib;
{
	char	*id = "totmem";
	struct	statfs	fsbuf;

	if (attrib) {
		log_err(-1, id, extra_parm);
		rm_errno = RM_ERR_BADPARAM;
		return NULL;
	}

	if (statfs(procfs, &fsbuf, sizeof(struct statfs), 0) == -1) {
		log_err(errno, id, "statfs");
		rm_errno = RM_ERR_SYSTEM;
		return NULL;
	}
	DBPRT(("%s: bsize=%d blocks=%d\n", id, fsbuf.f_bsize, fsbuf.f_blocks))
	sprintf(ret_string, "%d", fsbuf.f_bsize * fsbuf.f_blocks);
	return ret_string;
}

static char	*
availmem(attrib)
struct	rm_attribute	*attrib;
{
	char	*id = "availmem";
	struct	statfs	fsbuf;

	if (attrib) {