/* Convert an array of task IDs into a list of host names
* RET: the string, caller must xfree() this value */
static char *_task_ids_to_host_list(int ntasks, uint32_t taskids[])
{
int i;
hostset_t hs;
char *hosts;
slurm_step_layout_t *sl;
if ((sl = launch_common_get_slurm_step_layout(local_srun_job)) == NULL)
return (xstrdup("Unknown"));
hs = hostset_create(NULL);
for (i = 0; i < ntasks; i++) {
char *host = slurm_step_layout_host_name(sl, taskids[i]);
if (host) {
hostset_insert(hs, host);
free(host);
} else {
error("Could not identify host name for task %u",
taskids[i]);
}
}
hosts = _hostset_to_string(hs);
hostset_destroy(hs);
return (hosts);
}
/* Convert an array of task IDs into a list of host names
* RET: the string, caller must xfree() this value */
static char *_task_ids_to_host_list(int ntasks, uint32_t *taskids)
{
int i, task_cnt = 0;
hostset_t hs;
char *hosts;
slurm_step_layout_t *sl;
if ((sl = launch_common_get_slurm_step_layout(local_srun_job)) == NULL)
return (xstrdup("Unknown"));
/* If overhead of determining the hostlist is too high then srun
* communications will timeout and fail, so return "Unknown" instead.
*
* See slurm_step_layout_host_id() in src/common/slurm_step_layout.c
* for details. */
for (i = 0; i < sl->node_cnt; i++) {
task_cnt += sl->tasks[i];
}
if (task_cnt > 100000)
return (xstrdup("Unknown"));
hs = hostset_create(NULL);
for (i = 0; i < ntasks; i++) {
char *host = slurm_step_layout_host_name(sl, taskids[i]);
if (host) {
hostset_insert(hs, host);
free(host);
} else {
error("Could not identify host name for task %u",
taskids[i]);
}
}
hosts = _hostset_to_string(hs);
hostset_destroy(hs);
return (hosts);
}
