extern int select_p_job_signal(struct job_record *job_ptr, int signal)
{
xassert(job_ptr);
/*
* Release the ALPS reservation already here for those signals that are
* likely to terminate the job. Otherwise there is a race condition if a
* script has more than one aprun line: while the apkill of the current
* aprun line is underway, the job script proceeds to run and executes
* the next following aprun line, until reaching the end of the script.
* This not only creates large delays, it can also mess up cleaning up
* after the job. Releasing the reservation will stop any new aprun
* lines from being executed.
*/
switch (signal) {
case SIGCONT:
case SIGSTOP:
case SIGTSTP:
case SIGTTIN:
case SIGTTOU:
case SIGURG:
case SIGCHLD:
case SIGWINCH:
break;
default:
if (signal < SIGRTMIN)
do_basil_release(job_ptr);
}
if (do_basil_signal(job_ptr, signal) != SLURM_SUCCESS)
return SLURM_ERROR;
return other_job_signal(job_ptr, signal);
}
extern int select_p_job_signal(struct job_record *job_ptr, int signal)
{
xassert(job_ptr);
/*
* Release the ALPS reservation already here for those signals that are
* likely to terminate the job. Otherwise there is a race condition if a
* script has more than one aprun line: while the apkill of the current
* aprun line is underway, the job script proceeds to run and executes
* the next following aprun line, until reaching the end of the script.
* This not only creates large delays, it can also mess up cleaning up
* after the job. Releasing the reservation will stop any new aprun
* lines from being executed.
*/
if (slurmctld_primary) {
switch (signal) {
case SIGCHLD:
case SIGCONT:
case SIGSTOP:
case SIGTSTP:
case SIGTTIN:
case SIGTTOU:
case SIGURG:
case SIGWINCH:
break;
case SIGTERM:
case SIGKILL:
if (cray_conf->no_apid_signal_on_kill &&
job_ptr->batch_flag)
return other_job_signal(
job_ptr, signal);
default:
if (signal < SIGRTMIN)
do_basil_release(job_ptr);
}
}
if (slurmctld_primary && !_zero_size_job(job_ptr)) {
if (signal != SIGKILL) {
if (do_basil_signal(job_ptr, signal) != SLURM_SUCCESS)
return SLURM_ERROR;
} else {
uint16_t kill_wait = slurm_get_kill_wait();
if (do_basil_signal(job_ptr, SIGCONT) != SLURM_SUCCESS)
return SLURM_ERROR;
if (do_basil_signal(job_ptr, SIGTERM) != SLURM_SUCCESS)
return SLURM_ERROR;
queue_basil_signal(job_ptr, SIGKILL, kill_wait);
}
}
return other_job_signal(job_ptr, signal);
}
extern int select_p_job_fini(struct job_record *job_ptr)
{
if (job_ptr == NULL)
return SLURM_SUCCESS;
if (do_basil_release(job_ptr) != SLURM_SUCCESS)
return SLURM_ERROR;
/*
* Convention: like select_p_job_ready, may be called also from
* stepdmgr, where job_state == NO_VAL is used to
* distinguish the context from that of slurmctld.
*/
if (job_ptr->job_state == (uint16_t)NO_VAL)
return SLURM_SUCCESS;
return other_job_fini(job_ptr);
}
extern int select_p_job_fini(struct job_record *job_ptr)
{
if (job_ptr == NULL)
return SLURM_SUCCESS;
/* Don't run the release in the controller for batch jobs. It is
* handled on the stepd end.
*/
if (((slurmctld_primary && !job_ptr->batch_flag) ||
(job_ptr->job_state == NO_VAL))
&& !_zero_size_job(job_ptr) &&
(do_basil_release(job_ptr) != SLURM_SUCCESS))
return SLURM_ERROR;
/*
* Convention: like select_p_job_ready, may be called also from
* stepdmgr, where job_state == NO_VAL is used to
* distinguish the context from that of slurmctld.
*/
if (job_ptr->job_state == NO_VAL)
return SLURM_SUCCESS;
return other_job_fini(job_ptr);
}
