/**
* do_basil_confirm - confirm an existing BASIL reservation.
* This requires the alloc_sid to equal the session ID (getsid()) of the process
* executing the aprun/mpirun commands
* Returns: SLURM_SUCCESS if ok, READY_JOB_ERROR/FATAL on transient/fatal error.
*/
extern int do_basil_confirm(struct job_record *job_ptr)
{
uint32_t resv_id;
uint64_t pagg_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) {
/* On Cray XT/XE, a reservation ID of 0 is always invalid. */
error("JobId=%u has invalid (ZERO) resId", job_ptr->job_id);
} else if (_get_select_jobinfo(job_ptr->select_jobinfo->data,
SELECT_JOBDATA_PAGG_ID, &pagg_id) != SLURM_SUCCESS) {
error("can not read pagg ID for JobId=%u", job_ptr->job_id);
} else {
int rc;
if (pagg_id == 0) {
#ifdef HAVE_REAL_CRAY
/* This fallback case is for interactive jobs only */
error("JobId %u has no pagg ID, falling back to SID",
job_ptr->job_id);
#endif
pagg_id = job_ptr->alloc_sid;
}
rc = basil_confirm(resv_id, job_ptr->job_id, pagg_id);
if (rc == 0) {
debug2("confirmed ALPS resId %u for JobId %u, pagg "
"%"PRIu64"", resv_id, job_ptr->job_id, pagg_id);
return SLURM_SUCCESS;
} else if (rc == -BE_NO_RESID) {
/*
* If ALPS can not find the reservation ID we are trying
* to confirm, it may be that the job has already been
* canceled, or that the reservation has timed out after
* waiting for the confirmation.
* It is more likely that this error occurs on a per-job
* basis, hence in this case do not drain frontend node.
*/
error("JobId %u has invalid ALPS resId %u - job "
"already canceled?", job_ptr->job_id, resv_id);
return SLURM_SUCCESS;
} else {
error("confirming ALPS resId %u of JobId %u FAILED: %s",
resv_id, job_ptr->job_id, basil_strerror(rc));
if (is_transient_error(rc))
return READY_JOB_ERROR;
}
}
return READY_JOB_FATAL;
}
/**
* do_basil_confirm - confirm an existing BASIL reservation.
* This requires the alloc_sid to equal the session ID (getsid()) of the process
* executing the aprun/mpirun commands
* Returns: SLURM_SUCCESS if ok, READY_JOB_ERROR/FATAL on transient/fatal error.
*/
extern int do_basil_confirm(struct job_record *job_ptr)
{
uint32_t resv_id;
uint64_t pagg_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) {
/* On Cray XT/XE, a reservation ID of 0 is always invalid. */
error("JobId=%u has invalid (ZERO) resId", job_ptr->job_id);
} else if (_get_select_jobinfo(job_ptr->select_jobinfo->data,
SELECT_JOBDATA_PAGG_ID, &pagg_id) != SLURM_SUCCESS) {
error("can not read pagg ID for JobId=%u", job_ptr->job_id);
} else {
int rc;
if (pagg_id == 0) {
/* This fallback case is for interactive jobs only */
error("JobId %u has no pagg ID, falling back to SID",
job_ptr->job_id);
pagg_id = job_ptr->alloc_sid;
}
rc = basil_confirm(resv_id, job_ptr->job_id, pagg_id);
if (rc == 0) {
debug2("confirmed ALPS resId %u for JobId %u, "
"pagg %"PRIu64"",
resv_id, job_ptr->job_id, pagg_id);
return SLURM_SUCCESS;
} else {
error("confirming ALPS resId %u of JobId %u FAILED: %s",
resv_id, job_ptr->job_id, basil_strerror(rc));
if (is_transient_error(rc))
return READY_JOB_ERROR;
}
}
return READY_JOB_FATAL;
}