/* 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);
}
extern void launch_common_set_stdio_fds(srun_job_t *job,
slurm_step_io_fds_t *cio_fds)
{
bool err_shares_out = false;
int file_flags;
if (opt.open_mode == OPEN_MODE_APPEND)
file_flags = O_CREAT|O_WRONLY|O_APPEND;
else if (opt.open_mode == OPEN_MODE_TRUNCATE)
file_flags = O_CREAT|O_WRONLY|O_APPEND|O_TRUNC;
else {
slurm_ctl_conf_t *conf;
conf = slurm_conf_lock();
if (conf->job_file_append)
file_flags = O_CREAT|O_WRONLY|O_APPEND;
else
file_flags = O_CREAT|O_WRONLY|O_APPEND|O_TRUNC;
slurm_conf_unlock();
}
/*
* create stdin file descriptor
*/
if (_is_local_file(job->ifname)) {
if ((job->ifname->name == NULL) ||
(job->ifname->taskid != -1)) {
cio_fds->in.fd = STDIN_FILENO;
} else {
cio_fds->in.fd = open(job->ifname->name, O_RDONLY);
if (cio_fds->in.fd == -1) {
error("Could not open stdin file: %m");
exit(error_exit);
}
}
if (job->ifname->type == IO_ONE) {
cio_fds->in.taskid = job->ifname->taskid;
cio_fds->in.nodeid = slurm_step_layout_host_id(
launch_common_get_slurm_step_layout(job),
job->ifname->taskid);
}
}
/*
* create stdout file descriptor
*/
if (_is_local_file(job->ofname)) {
if ((job->ofname->name == NULL) ||
(job->ofname->taskid != -1)) {
cio_fds->out.fd = STDOUT_FILENO;
} else {
cio_fds->out.fd = open(job->ofname->name,
file_flags, 0644);
if (cio_fds->out.fd == -1) {
error("Could not open stdout file: %m");
exit(error_exit);
}
}
if (job->ofname->name != NULL
&& job->efname->name != NULL
&& !strcmp(job->ofname->name, job->efname->name)) {
err_shares_out = true;
}
}
/*
* create seperate stderr file descriptor only if stderr is not sharing
* the stdout file descriptor
*/
if (err_shares_out) {
debug3("stdout and stderr sharing a file");
cio_fds->err.fd = cio_fds->out.fd;
cio_fds->err.taskid = cio_fds->out.taskid;
} else if (_is_local_file(job->efname)) {
if ((job->efname->name == NULL) ||
(job->efname->taskid != -1)) {
cio_fds->err.fd = STDERR_FILENO;
} else {
cio_fds->err.fd = open(job->efname->name,
file_flags, 0644);
if (cio_fds->err.fd == -1) {
error("Could not open stderr file: %m");
exit(error_exit);
}
}
}
}
