/* block_state_mutex should be unlocked before calling this */
extern void free_block_list(uint32_t job_id, List track_list,
bool destroy, bool wait)
{
bg_record_t *bg_record = NULL;
int retries;
ListIterator itr = NULL;
bg_free_block_list_t *bg_free_list;
pthread_attr_t attr_agent;
pthread_t thread_agent;
List kill_job_list = NULL;
kill_job_struct_t *freeit;
if (!track_list || !list_count(track_list))
return;
bg_free_list = xmalloc(sizeof(bg_free_block_list_t));
bg_free_list->track_list = list_create(NULL);
bg_free_list->destroy = destroy;
bg_free_list->job_id = job_id;
slurm_mutex_lock(&block_state_mutex);
list_transfer(bg_free_list->track_list, track_list);
itr = list_iterator_create(bg_free_list->track_list);
while ((bg_record = list_next(itr))) {
if (bg_record->magic != BLOCK_MAGIC) {
error("block was already destroyed %p", bg_record);
continue;
}
bg_record->free_cnt++;
/* just so we don't over write a different thread that
wants this block destroyed */
if (destroy && !bg_record->destroy)
bg_record->destroy = destroy;
if (destroy && (bg_record->state & BG_BLOCK_ERROR_FLAG))
resume_block(bg_record);
/* This means we are wanting this block free so we can
run this job on it, so it is ok to have the job
remain here. Only checking for jobs should go
below this.
*/
if (bg_record->modifying) {
debug("free_block_list: Just FYI, we are "
"freeing a block (%s) that "
"has at least one pending job.",
bg_record->bg_block_id);
continue;
}
if (bg_record->job_ptr
&& !IS_JOB_FINISHED(bg_record->job_ptr)) {
info("We are freeing a block (%s) that "
"has job %u(%u).",
bg_record->bg_block_id,
bg_record->job_ptr->job_id,
bg_record->job_running);
if (!kill_job_list)
kill_job_list =
bg_status_create_kill_job_list();
freeit = xmalloc(sizeof(kill_job_struct_t));
freeit->jobid = bg_record->job_ptr->job_id;
list_push(kill_job_list, freeit);
} else if (bg_record->job_list
&& list_count(bg_record->job_list)) {
struct job_record *job_ptr;
ListIterator itr;
if (!kill_job_list)
kill_job_list =
bg_status_create_kill_job_list();
info("We are freeing a block (%s) that has at "
"least 1 job.",
bg_record->bg_block_id);
itr = list_iterator_create(bg_record->job_list);
while ((job_ptr = list_next(itr))) {
if ((job_ptr->magic != JOB_MAGIC)
|| IS_JOB_FINISHED(job_ptr))
continue;
freeit = xmalloc(sizeof(kill_job_struct_t));
freeit->jobid = job_ptr->job_id;
list_push(kill_job_list, freeit);
}
list_iterator_destroy(itr);
}
}
list_iterator_destroy(itr);
slurm_mutex_unlock(&block_state_mutex);
if (kill_job_list) {
bg_status_process_kill_job_list(kill_job_list, JOB_FAILED, 0);
FREE_NULL_LIST(kill_job_list);
}
if (wait) {
/* Track_freeing_blocks waits until the list is done
and frees the memory of bg_free_list.
*/
_track_freeing_blocks(bg_free_list);
return;
}
/* _track_freeing_blocks handles cleanup */
slurm_attr_init(&attr_agent);
if (pthread_attr_setdetachstate(&attr_agent, PTHREAD_CREATE_DETACHED))
error("pthread_attr_setdetachstate error %m");
retries = 0;
while (pthread_create(&thread_agent, &attr_agent,
_track_freeing_blocks,
bg_free_list)) {
error("pthread_create error %m");
if (++retries > MAX_PTHREAD_RETRIES)
fatal("Can't create pthread");
/* sleep and retry */
usleep(1000);
}
slurm_attr_destroy(&attr_agent);
return;
}
/* block_state_mutex should be unlocked before calling this */
extern int free_block_list(uint32_t job_id, List track_list,
bool destroy, bool wait)
{
bg_record_t *bg_record = NULL;
int retries;
ListIterator itr = NULL;
bg_free_block_list_t *bg_free_list;
pthread_attr_t attr_agent;
pthread_t thread_agent;
if (!track_list || !list_count(track_list))
return SLURM_SUCCESS;
bg_free_list = xmalloc(sizeof(bg_free_block_list_t));
bg_free_list->track_list = list_create(NULL);
bg_free_list->destroy = destroy;
bg_free_list->job_id = job_id;
slurm_mutex_lock(&block_state_mutex);
list_transfer(bg_free_list->track_list, track_list);
itr = list_iterator_create(bg_free_list->track_list);
while ((bg_record = list_next(itr))) {
if (bg_record->magic != BLOCK_MAGIC) {
error("block was already destroyed %p", bg_record);
continue;
}
bg_record->free_cnt++;
if (bg_record->job_ptr
&& !IS_JOB_FINISHED(bg_record->job_ptr)) {
info("We are freeing a block (%s) that has job %u(%u).",
bg_record->bg_block_id,
bg_record->job_ptr->job_id,
bg_record->job_running);
/* This is not thread safe if called from
bg_job_place.c anywhere from within
submit_job() or at startup. */
slurm_mutex_unlock(&block_state_mutex);
bg_requeue_job(bg_record->job_ptr->job_id, 0);
slurm_mutex_lock(&block_state_mutex);
}
if (remove_from_bg_list(bg_lists->job_running, bg_record)
== SLURM_SUCCESS)
num_unused_cpus += bg_record->cpu_cnt;
}
list_iterator_destroy(itr);
slurm_mutex_unlock(&block_state_mutex);
if (wait) {
/* Track_freeing_blocks waits until the list is done
and frees the memory of bg_free_list.
*/
_track_freeing_blocks(bg_free_list);
return SLURM_SUCCESS;
}
/* _track_freeing_blocks handles cleanup */
slurm_attr_init(&attr_agent);
if (pthread_attr_setdetachstate(&attr_agent, PTHREAD_CREATE_DETACHED))
error("pthread_attr_setdetachstate error %m");
retries = 0;
while (pthread_create(&thread_agent, &attr_agent,
_track_freeing_blocks,
bg_free_list)) {
error("pthread_create error %m");
if (++retries > MAX_PTHREAD_RETRIES)
fatal("Can't create "
"pthread");
/* sleep and retry */
usleep(1000);
}
slurm_attr_destroy(&attr_agent);
return SLURM_SUCCESS;
}