/* Remove a specific bb_alloc_t from global records.
* RET true if found, false otherwise */
extern bool bb_free_alloc_rec(bb_state_t *state_ptr, bb_alloc_t *bb_alloc)
{
bb_alloc_t *bb_link, **bb_plink;
int i;
xassert(state_ptr);
xassert(state_ptr->bb_ahash);
xassert(bb_alloc);
i = bb_alloc->user_id % BB_HASH_SIZE;
bb_plink = &state_ptr->bb_ahash[i];
bb_link = state_ptr->bb_ahash[i];
while (bb_link) {
if (bb_link == bb_alloc) {
xassert(bb_link->magic == BB_ALLOC_MAGIC);
*bb_plink = bb_alloc->next;
bb_free_alloc_buf(bb_alloc);
state_ptr->last_update_time = time(NULL);
return true;
}
bb_plink = &bb_link->next;
bb_link = bb_link->next;
}
return false;
}
/* Clear all cached burst buffer records, freeing all memory. */
extern void bb_clear_cache(bb_state_t *state_ptr)
{
bb_alloc_t *bb_current, *bb_next;
bb_job_t *job_current, *job_next;
bb_user_t *user_current, *user_next;
int i;
if (state_ptr->bb_ahash) {
for (i = 0; i < BB_HASH_SIZE; i++) {
bb_current = state_ptr->bb_ahash[i];
while (bb_current) {
xassert(bb_current->magic == BB_ALLOC_MAGIC);
bb_next = bb_current->next;
bb_free_alloc_buf(bb_current);
bb_current = bb_next;
}
}
xfree(state_ptr->bb_ahash);
}
if (state_ptr->bb_jhash) {
for (i = 0; i < BB_HASH_SIZE; i++) {
job_current = state_ptr->bb_jhash[i];
while (job_current) {
xassert(job_current->magic == BB_JOB_MAGIC);
job_next = job_current->next;
_bb_job_del2(job_current);
job_current = job_next;
}
}
xfree(state_ptr->bb_jhash);
}
if (state_ptr->bb_uhash) {
for (i = 0; i < BB_HASH_SIZE; i++) {
user_current = state_ptr->bb_uhash[i];
while (user_current) {
xassert(user_current->magic == BB_USER_MAGIC);
user_next = user_current->next;
xfree(user_current);
user_current = user_next;
}
}
xfree(state_ptr->bb_uhash);
}
xfree(state_ptr->name);
FREE_NULL_LIST(state_ptr->persist_resv_rec);
}
/* Handle timeout of burst buffer events:
* 1. Purge per-job burst buffer records when the stage-out has completed and
* the job has been purged from Slurm
* 2. Test for StageInTimeout events
* 3. Test for StageOutTimeout events
*/
static void _timeout_bb_rec(void)
{
struct job_record *job_ptr;
bb_alloc_t **bb_pptr, *bb_ptr = NULL;
uint32_t age;
time_t now = time(NULL);
int i;
for (i = 0; i < BB_HASH_SIZE; i++) {
bb_pptr = &bb_state.bb_ahash[i];
bb_ptr = bb_state.bb_ahash[i];
while (bb_ptr) {
if (bb_ptr->seen_time < bb_state.last_load_time) {
if (bb_ptr->job_id == 0) {
info("%s: Persistent burst buffer %s "
"purged",
__func__, bb_ptr->name);
} else if (bb_state.bb_config.debug_flag) {
info("%s: burst buffer for job %u "
"purged",
__func__, bb_ptr->job_id);
}
//FIXME: VESTIGIAL: Use bb_limit_rem
// bb_remove_user_load(bb_ptr, &bb_state);
*bb_pptr = bb_ptr->next;
bb_free_alloc_buf(bb_ptr);
break;
}
if ((bb_ptr->job_id != 0) &&
(bb_ptr->state >= BB_STATE_STAGED_OUT) &&
!find_job_record(bb_ptr->job_id)) {
_stop_stage_out(bb_ptr->job_id);
bb_ptr->cancelled = true;
bb_ptr->end_time = 0;
*bb_pptr = bb_ptr->next;
bb_free_alloc_buf(bb_ptr);
break;
}
age = difftime(now, bb_ptr->state_time);
if ((bb_ptr->job_id != 0) &&
bb_state.bb_config.stop_stage_in &&
(bb_ptr->state == BB_STATE_STAGING_IN) &&
(bb_state.bb_config.stage_in_timeout != 0) &&
(!bb_ptr->cancelled) &&
(age >= bb_state.bb_config.stage_in_timeout)) {
_stop_stage_in(bb_ptr->job_id);
bb_ptr->cancelled = true;
bb_ptr->end_time = 0;
job_ptr = find_job_record(bb_ptr->job_id);
if (job_ptr) {
error("%s: StageIn timed out, holding "
"job %u",
__func__, bb_ptr->job_id);
job_ptr->priority = 0;
job_ptr->direct_set_prio = 1;
job_ptr->state_reason = WAIT_HELD;
xfree(job_ptr->state_desc);
job_ptr->state_desc = xstrdup(
"Burst buffer stage-in timeout");
last_job_update = now;
} else {
error("%s: StageIn timed out for "
"vestigial job %u ",
__func__, bb_ptr->job_id);
}
}
if ((bb_ptr->job_id != 0) &&
bb_state.bb_config.stop_stage_out &&
(bb_ptr->state == BB_STATE_STAGING_OUT) &&
(bb_state.bb_config.stage_out_timeout != 0) &&
(!bb_ptr->cancelled) &&
(age >= bb_state.bb_config.stage_out_timeout)) {
error("%s: StageOut for job %u timed out",
__func__, bb_ptr->job_id);
_stop_stage_out(bb_ptr->job_id);
bb_ptr->cancelled = true;
bb_ptr->end_time = 0;
}
bb_pptr = &bb_ptr->next;
bb_ptr = bb_ptr->next;
}
}
}
