void *_sig_basil(void *args)
{
args_sig_basil_t *args_sig_basil = (args_sig_basil_t *) args;
int rc;
sleep(args_sig_basil->delay);
rc = basil_signal_apids(args_sig_basil->resv_id,
args_sig_basil->signal, NULL);
if (rc) {
error("could not signal APIDs of resId %u: %s",
args_sig_basil->resv_id, basil_strerror(rc));
}
xfree(args);
return NULL;
}
/**
* do_basil_signal - pass job signal on to any APIDs
* IN job_ptr - job to be signalled
* IN signal - signal(7) number
* Only signal job if an ALPS reservation exists (non-0 reservation ID).
*/
extern int do_basil_signal(struct job_record *job_ptr, int signal)
{
uint32_t resv_id;
if (_get_select_jobinfo(job_ptr->select_jobinfo->data,
SELECT_JOBDATA_RESV_ID, &resv_id) != SLURM_SUCCESS) {
error("can not read resId for JobId=%u", job_ptr->job_id);
} else if (resv_id != 0) {
int rc = basil_signal_apids(resv_id, signal, NULL);
if (rc)
error("could not signal APIDs of resId %u: %s", resv_id,
basil_strerror(rc));
}
return SLURM_SUCCESS;
}
/**
* queue_basil_signal - queue job signal on to any APIDs
* IN job_ptr - job to be signalled
* IN signal - signal(7) number
* IN delay - how long to delay the signal, in seconds
* Only signal job if an ALPS reservation exists (non-0 reservation ID).
*/
extern void queue_basil_signal(struct job_record *job_ptr, int signal,
uint16_t delay)
{
args_sig_basil_t *args_sig_basil;
pthread_attr_t attr_sig_basil;
pthread_t thread_sig_basil;
uint32_t resv_id;
if (_get_select_jobinfo(job_ptr->select_jobinfo->data,
SELECT_JOBDATA_RESV_ID, &resv_id) != SLURM_SUCCESS) {
error("can not read resId for JobId=%u", job_ptr->job_id);
return;
}
if (resv_id == 0)
return;
if ((delay == 0) || (delay == (uint16_t) NO_VAL)) {
/* Send the signal now */
int rc = basil_signal_apids(resv_id, signal, NULL);
if (rc)
error("could not signal APIDs of resId %u: %s", resv_id,
basil_strerror(rc));
return;
}
/* Create a thread to send the signal later */
slurm_attr_init(&attr_sig_basil);
if (pthread_attr_setdetachstate(&attr_sig_basil,
PTHREAD_CREATE_DETACHED)) {
error("pthread_attr_setdetachstate error %m");
slurm_attr_destroy(&attr_sig_basil);
return;
}
args_sig_basil = xmalloc(sizeof(args_sig_basil_t));
args_sig_basil->resv_id = resv_id;
args_sig_basil->signal = signal;
args_sig_basil->delay = delay;
if (pthread_create(&thread_sig_basil, &attr_sig_basil,
_sig_basil, (void *) args_sig_basil)) {
error("pthread_create error %m");
slurm_attr_destroy(&attr_sig_basil);
xfree(args_sig_basil);
return;
}
slurm_attr_destroy(&attr_sig_basil);
}
/**
* basil_safe_release - release reservation after signaling job steps
* @rsvn_id: reservation to release
* @inv: recent Basil Inventory, or NULL to generate internally
* Returns 0 if ok, a negative %basil_error otherwise.
*/
int basil_safe_release(int32_t rsvn_id, struct basil_inventory *inv)
{
int rc = basil_release(rsvn_id);
/*
* If there are still any live application IDs (APIDs) associated with
* @rsvn_id, the RELEASE command will be without effect, since ALPS
* holds on to a reservation until all of its application IDs have
* disappeared.
* On normal termination, ALPS should clean up the APIDs by itself. In
* order to clean up orphaned reservations, try to terminate the APIDs
* manually using apkill(1). If this step fails, fall back to releasing
* the reservation normally and hope that ALPS resolves the situation.
* To prevent that any subsequent aprun lines get started while the
* apkill of the current one is still in progress, do the RELEASE first.
*/
basil_signal_apids(rsvn_id, SIGKILL, inv);
return rc;
}
