/**
* @brief
* reclaim_cpusets()
* Given a list of cpusets, attempt to destroy each cpuset named by the list.
* If it can be destroyed, unset the bits corresponding to the cpuset's nodes
* in the mask (if supplied). This is used to reclaim cpusets that were
* supposed to be deleted, but were in fact "stuck", and placed on stucklist.
*
* @param[in] listp - pointer to cpuset list
* @param[in] maskp - pointer to mask bitfield
*
* @return int
* @retval num of cpuset reclaimed success
* @retval 0 error
*
*/
int
reclaim_cpusets(cpusetlist **listp, Bitfield *maskp)
{
cpusetlist *set, *next;
int count = 0;
/*
* Walk the list of stuck cpusets, attempting to free each one. Keep
* track of the previous and next pointers so the element can be
* unlinked and freed.
*/
for (set = *listp; set != NULL; set = next) {
next = set->next; /* Keep track of next pointer. */
/* See if this cpuset can be deleted now. If not, go on. */
if (destroy_cpuset(set->name)) {
log_err(0, __func__, "could not destroy cpuset");
continue;
}
/*
* Remove the corresponding bits from the given bitmask, if supplied,
* and return the nodes to the nodepool.
*/
if (maskp != NULL)
BITFIELD_CLRM(maskp, &(set->nodes));
BITFIELD_SETM(&nodepool, &(set->nodes));
/* Log that the cpuset was reclaimed. */
(void)sprintf(log_buffer, "stuck cpuset %s reclaimed", set->name);
log_event(PBSEVENT_ERROR, PBS_EVENTCLASS_JOB, LOG_INFO, __func__, log_buffer);
#ifdef DEBUG
(void)sprintf(log_buffer, "nodepool now %s", bitfield2hex(&nodepool));
log_event(PBSEVENT_SYSTEM, PBS_EVENTCLASS_JOB, LOG_INFO, __func__, log_buffer);
#endif /* DEBUG */
/* Now free the storage for the cpusetlist element. */
if (remove_from_cpusetlist(listp, NULL, set->name, NULL))
break;
count ++; /* Another cpuset reclaimed. */
}
/*
* Perform a quick sanity check. If there are no cpusets on the supplied
* list, then there should be no bits set in the supplied bitfield. Log
* an error message if this is not the case.
*/
if (maskp != NULL && *listp == NULL && !BITFIELD_IS_ZERO(maskp))
log_err(-1, __func__, "NULL cpusetlist but mask not empty!");
return (count);
}
static int
nodemask_overlaps(void)
{
char *id = "nodemask_overlaps";
QueueList *qlp1, *qlp2;
Bitfield overlap;
/* Check for overlapping queue nodemasks in batch queues. */
for (qlp1 = schd_BatchQueues; qlp1 != NULL; qlp1 = qlp1->next)
{
if (!(qlp1->queue->flags & QFLAGS_NODEMASK))
continue;
for (qlp2 = qlp1->next; qlp2 != NULL; qlp2 = qlp2->next)
{
/*
* If the queues are on different hosts, they may appear to
* overlap - no problem.
*/
if (strcmp(qlp1->queue->exechost, qlp2->queue->exechost))
continue;
if ((qlp2->queue->flags & QFLAGS_NODEMASK))
{
DBPRT(("comparing nodemask for queue %s with queue %s\n",
qlp1->queue->qname, qlp2->queue->qname));
BITFIELD_CPY(&overlap, &(qlp1->queue->queuemask));
BITFIELD_ANDM(&overlap, &(qlp2->queue->queuemask));
if (!BITFIELD_IS_ZERO(&overlap))
{
(void)sprintf(log_buffer,
"Queues %s and %s have overlapping nodemasks!",
qlp1->queue->qname, qlp2->queue->qname);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
return (1);
}
}
}
}
/* Make sure batch queues and external queues have differing nodemasks. */
for (qlp1 = schd_BatchQueues; qlp1 != NULL; qlp1 = qlp1->next)
{
if (!(qlp1->queue->flags & QFLAGS_NODEMASK))
continue;
for (qlp2 = schd_ExternQueues; qlp2 != NULL; qlp2 = qlp2->next)
{
/*
* If the queues are on different hosts, they may appear to
* overlap - no problem.
*/
if (strcmp(qlp1->queue->exechost, qlp2->queue->exechost))
continue;
if ((qlp2->queue->flags & QFLAGS_NODEMASK))
{
DBPRT(("comparing nodemask for queue %s with queue %s\n",
qlp1->queue->qname, qlp2->queue->qname));
BITFIELD_CPY(&overlap, &(qlp1->queue->queuemask));
BITFIELD_ANDM(&overlap, &(qlp2->queue->queuemask));
if (!BITFIELD_IS_ZERO(&overlap))
{
(void)sprintf(log_buffer,
"Queues %s and %s have overlapping nodemasks!",
qlp1->queue->qname, qlp2->queue->qname);
log_record(PBSEVENT_SYSTEM, PBS_EVENTCLASS_SERVER, id,
log_buffer);
DBPRT(("%s: %s\n", id, log_buffer));
return (1);
}
}
}
}
return (0);
}
