/*
* task_p_slurmd_suspend_job()
*/
extern int task_p_slurmd_suspend_job (uint32_t job_id)
{
debug("task_p_slurmd_suspend_job: %u", job_id);
#ifdef HAVE_NATIVE_CRAY
_step_epilogue();
#endif
return SLURM_SUCCESS;
}
/*
* task_p_slurmd_suspend_job()
*/
extern int task_p_slurmd_suspend_job (uint32_t job_id)
{
DEF_TIMERS;
START_TIMER;
debug("task_p_slurmd_suspend_job: %u", job_id);
#ifdef HAVE_NATIVE_CRAY
_step_epilogue();
#endif
END_TIMER;
if (debug_flags & DEBUG_FLAG_TIME_CRAY)
INFO_LINE("call took: %s", TIME_STR);
return SLURM_SUCCESS;
}
/*
* task_p_post_step() is called after termination of the step
* (all the tasks)
*/
extern int task_p_post_step (stepd_step_rec_t *job)
{
#ifdef HAVE_NATIVE_CRAY
char llifile[LLI_STATUS_FILE_BUF_SIZE];
int rc, cnt;
char *err_msg = NULL, path[PATH_MAX];
int32_t *numa_nodes;
cpu_set_t *cpuMasks;
if (track_status) {
// Get the lli file name
snprintf(llifile, sizeof(llifile), LLI_STATUS_FILE,
SLURM_ID_HASH(job->jobid, job->stepid));
// Unlink the file
errno = 0;
rc = unlink(llifile);
if (rc == -1 && errno != ENOENT) {
CRAY_ERR("unlink(%s) failed: %m", llifile);
} else if (rc == 0) {
info("Unlinked %s", llifile);
}
}
/*
* Compact Memory
*
* Determine which NUMA nodes and CPUS an application is using. It will
* be used to compact the memory.
*
* You'll find the information in the following location.
* For a normal job step:
* /dev/cpuset/slurm/uid_<uid>/job_<jobID>/step_<stepID>/
*
* For a batch job step (only on the head node and only for batch jobs):
* /dev/cpuset/slurm/uid_<uid>/job_<jobID>/step_batch/
*
* NUMA node: mems
* CPU Masks: cpus
*/
if (job->batch) {
// Batch Job Step
rc = snprintf(path, sizeof(path),
"/dev/cpuset/slurm/uid_%d/job_%"
PRIu32 "/step_batch", job->uid, job->jobid);
if (rc < 0) {
CRAY_ERR("snprintf failed. Return code: %d", rc);
return SLURM_ERROR;
}
} else {
// Normal Job Step
/* Only run epilogue on non-batch steps */
_step_epilogue();
rc = snprintf(path, sizeof(path),
"/dev/cpuset/slurm/uid_%d/job_%"
PRIu32 "/step_%" PRIu32,
job->uid, job->jobid, job->stepid);
if (rc < 0) {
CRAY_ERR("snprintf failed. Return code: %d", rc);
return SLURM_ERROR;
}
}
rc = _get_numa_nodes(path, &cnt, &numa_nodes);
if (rc < 0) {
CRAY_ERR("get_numa_nodes failed. Return code: %d", rc);
return SLURM_ERROR;
}
rc = _get_cpu_masks(cnt, numa_nodes, &cpuMasks);
if (rc < 0) {
CRAY_ERR("get_cpu_masks failed. Return code: %d", rc);
return SLURM_ERROR;
}
/*
* Compact Memory
* The last argument which is a path to the cpuset directory has to be
* NULL because the CPUSET directory has already been cleaned up.
*/
rc = alpsc_compact_mem(&err_msg, cnt, numa_nodes, cpuMasks, NULL);
_ALPSC_DEBUG("alpsc_compact_mem");
xfree(numa_nodes);
xfree(cpuMasks);
if (rc != 1) {
return SLURM_ERROR;
}
#endif
return SLURM_SUCCESS;
}
