getrsrcs.c

openPBS的开放源代码

	goto bail;
    }

    /* Receive AVAILMEM response from resource monitor. */
    response = getreq(rm);
    if (response != NULL) {
	new_rsrcs->freemem = schd_val2byte(response);
	new_rsrcs->freemem *= schd_FAKE_MACH_MULT;
	(void)free(response);
    } else {
	(void)sprintf(log_buffer, "bad return from getreq(freemem), %d, %d", 
		      pbs_errno, errno);
	log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
	badreply = 1;
	goto bail;
    }

    /* Receive PHYSMEM response from resource monitor. */
    response = getreq(rm);
    if (response != NULL) {
	pmem_avail = schd_val2byte(response);
	pmem_avail *= schd_FAKE_MACH_MULT;
	(void)free(response);
    } else {
	(void)sprintf(log_buffer, "bad return from getreq(realmem), %d, %d", 
		      pbs_errno, errno);
	log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
	badreply = 1;
	goto bail;
    }

    /* Receive NCPUS response from resource monitor. */
    response = getreq(rm);
    if (response != NULL) {
	cpus_avail = atoi(response) * schd_FAKE_MACH_MULT;
	(void)free(response);
    } else {
	(void)sprintf(log_buffer, "bad return from getreq(ncpus), %d, %d", 
		      pbs_errno, errno);
	log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
	badreply = 1;
	goto bail;
    }

#ifdef NODEMASK
    /* Receive available nodes from resource monitor. */
    response = getreq(rm);
    if (response == NULL) {
	(void)sprintf(log_buffer, "bad return from getreq(availmask), %d, %d", 
	    pbs_errno, errno);
	log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
	badreply = 1;
	goto bail;
    } else {
	if (schd_bits2mask(response, &new_rsrcs->availmask) != 0) {
	    if (schd_str2mask(response, &new_rsrcs->availmask) != 0) {
		(void)sprintf(log_buffer, "can't parse availmask '%s'", response);
		log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
		badreply = 1;
		goto bail;
	    }
	}
	(void)free(response);
    }
#endif	/* NODEMASK */

    if (schd_MANAGE_HPM) {
	/* Receive HPM_CTL response from resource monitor. */
	response = getreq(rm);
	if (response != NULL) {
	    if (strcmp(response, HPM_CTL_USERMODE_STR) == 0)
		new_rsrcs->flags |= RSRCS_FLAGS_HPM_USER;
	    else if (strcmp(response, HPM_CTL_GLOBALMODE_STR) == 0)
		new_rsrcs->flags &= ~RSRCS_FLAGS_HPM_USER;
	    else {
		(void)sprintf(log_buffer, "bad response '%s' for '%s@%s'", 
		    response, hpm_ctl, exechost);
		log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, 
		    log_buffer);
		badreply = 1;
		goto bail;
	    }
	} else {
	    (void)sprintf(log_buffer, "bad return from getreq(%s), %d, %d", 
		hpm_ctl, pbs_errno, errno);
	    log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
	    badreply = 1;
	    goto bail;
	}
    }
    /* 
     * NOTE: response will be free()'d in bail.  Be sure to explicitly free()
     * response if more getreq() calls are added before the code below.
     */

bail:
    if (response != NULL)
	(void)free(response);

    /* Disconnect from the resource monitor. */
    if (rm >= 0)		/* resmom handle "0" is valid in RPP. */
	closerm(rm);

    /* And unset the alarm and handler. */
    alarm(0);
    sigaction(SIGALRM, &oact, &act);

    /* Reset the old alarm, taking into account how much time has passed. */
    if (remain) {
	DBPRT(("%s: old alarm had %d secs remaining, %d elapsed, ", id,
		remain, (time(NULL) - then)));
	/* How much time remains even after the time spent above? */
	remain -= (time(NULL) - then);

	/* 
	 * Would the previous time have already expired?  If so, schedule
	 * an alarm call in 1 second (close enough, hopefully).
	 */
	if (remain < 1)
	    remain = 1;

	DBPRT(("reset to %d secs\n", remain));
	alarm (remain);
    }

    /*
     * Verify all the data came back as expected; if not, abort this 
     * iteration of the scheduler.
     */
    if (badreply) {
	(void)sprintf(log_buffer, 
	    "Got bad info from mom@%s - aborting sched run", exechost);
	log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
	DBPRT(("%s: %s\n", id, log_buffer));

	free(new_rsrcs);
	return (NULL);
    }

    /* Make a copy of the hostname for the resources struct. */
    new_rsrcs->exechost = schd_strdup(exechost);
    if (new_rsrcs->exechost == NULL) {
	(void)sprintf(log_buffer, "Unable to copy exechost %s to rsrcs",
	    exechost);
	log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
	DBPRT(("%s: %s\n", id, log_buffer));

	free(new_rsrcs);
	return (NULL);
    }

    new_rsrcs->nodes_total = NODES_REQD(cpus_avail, pmem_avail);

#ifdef NODEMASK
    /* Copy the availmask schd_FAKE_MACH_MULT times to match avail cpus. */
    BITFIELD_CPY(&cpy, &(new_rsrcs->availmask));
    for (i = 2; i <= schd_FAKE_MACH_MULT; i++) {
	for (j = 0; j < (cpus_avail / schd_FAKE_MACH_MULT / 2); j++)
	    BITFIELD_SHIFTL(&cpy);
	BITFIELD_SETM(&(new_rsrcs->availmask), &cpy);
    }
#endif	/* NODEMASK */

    if (schd_RsrcsList == NULL) {
	schd_RsrcsList  = new_rsrcs;	/* Start the list. */
    } else {
	for (rptr = schd_RsrcsList; rptr->next != NULL; rptr = rptr->next)
	    /* Find the last element in the list. */ ;
	rptr->next = new_rsrcs;
    }
    /* Next pointer for the tail of the list points to nothing. */
    new_rsrcs->next = NULL;

    return (new_rsrcs);
}

void
schd_dump_rsrclist(void)
{
    Resources *rsrcs;

    for (rsrcs = schd_RsrcsList; rsrcs != NULL; rsrcs = rsrcs->next)
	dump_resources(rsrcs);
}

static void
dump_resources(Resources *rsrcs)
{
    char   *id = "dump_resources";

    /* Log the system's status */

#if 0
    (void)sprintf(log_buffer, " %d%% usr, %d%% sys, %d%% idl",
	rsrcs->usrtime,rsrcs->systime,rsrcs->idltime);
    log_record(PBSEVENT_SYSTEM,PBS_EVENTCLASS_SERVER,id,log_buffer);
#endif /* 0 */

    (void)sprintf(log_buffer,
		  "Resources for host %s", rsrcs->exechost);
    log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);

    (void)sprintf(log_buffer, " :: %-24s = %s", "Memory (free):",
	schd_byte2val(rsrcs->freemem));
    log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);

    (void)sprintf(log_buffer, " :: %-24s = %d / %d (%.2f%% utilization)", 
	"Nodes allocated:", rsrcs->nodes_alloc, rsrcs->nodes_total,
	(rsrcs->nodes_alloc * 100.0) / rsrcs->nodes_total);
    log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);

#ifdef NODEMASK
    (void)sprintf(log_buffer, " :: %-24s = %s", "Nodes configured:",
	schd_format_nodemask(&(rsrcs->availmask), &(rsrcs->availmask)));
    log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);

    (void)sprintf(log_buffer, " :: %-24s = %s", "Nodes in use:",
	schd_format_nodemask(&(rsrcs->availmask), &(rsrcs->nodes_used)));
    log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);
#endif	/* NODEMASK */

    (void)sprintf(log_buffer, " :: %-24s = %0.2f", "CPU Load average:", 
	rsrcs->loadave);
    log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id, log_buffer);

    (void)sprintf(log_buffer, " :: %-24s = %d", "Running jobs:",
	rsrcs->njobs);
    log_record(PBSEVENT_SYSTEM,PBS_EVENTCLASS_SERVER,id,log_buffer);

    if (schd_MANAGE_HPM) {
	(void)sprintf(log_buffer, " :: %-24s = %s mode, %sin use", 
	    "HPM Counters:",
	    rsrcs->flags & RSRCS_FLAGS_HPM_USER ? "user" : "global",
	    rsrcs->flags & RSRCS_FLAGS_HPM_IN_USE ? "" : "not ");
	log_record(PBSEVENT_SYSTEM,PBS_EVENTCLASS_SERVER,id,log_buffer);
    }
}