static int _handle_add_extern_pid_internal(stepd_step_rec_t *job, pid_t pid)
{
pthread_t thread_id;
pthread_attr_t attr;
extern_pid_t *extern_pid;
jobacct_id_t jobacct_id;
int retries = 0, rc = SLURM_SUCCESS;
if (job->stepid != SLURM_EXTERN_CONT) {
error("%s: non-extern step (%u) given for job %u.",
__func__, job->stepid, job->jobid);
return SLURM_FAILURE;
}
debug("%s: for job %u.%u, pid %d",
__func__, job->jobid, job->stepid, pid);
extern_pid = xmalloc(sizeof(extern_pid_t));
extern_pid->job = job;
extern_pid->pid = pid;
/* track pid: add outside of the below thread so that the pam module
* waits until the parent pid is added, before letting the parent spawn
* any children. */
jobacct_id.taskid = job->nodeid;
jobacct_id.nodeid = job->nodeid;
jobacct_id.job = job;
if (proctrack_g_add(job, pid) != SLURM_SUCCESS) {
error("%s: Job %u can't add pid %d to proctrack plugin in the extern_step.", __func__, job->jobid, pid);
return SLURM_FAILURE;
}
if (task_g_add_pid(pid) != SLURM_SUCCESS) {
error("%s: Job %u can't add pid %d to task plugin in the extern_step.", __func__, job->jobid, pid);
return SLURM_FAILURE;
}
if (jobacct_gather_add_task(pid, &jobacct_id, 1) != SLURM_SUCCESS) {
error("%s: Job %u can't add pid %d to jobacct_gather plugin in the extern_step.", __func__, job->jobid, pid);
return SLURM_FAILURE;
}
/* spawn a thread that will wait on the pid given */
slurm_attr_init(&attr);
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
while (pthread_create(&thread_id, &attr,
&_wait_extern_pid, (void *) extern_pid)) {
error("%s: pthread_create: %m", __func__);
if (++retries > MAX_RETRIES) {
error("%s: Can't create pthread", __func__);
rc = SLURM_FAILURE;
break;
}
usleep(10); /* sleep and again */
}
slurm_attr_destroy(&attr);
return rc;
}
static int
_handle_add_extern_pid(int fd, stepd_step_rec_t *job)
{
int rc = SLURM_SUCCESS;
pid_t pid;
jobacct_id_t jobacct_id;
safe_read(fd, &pid, sizeof(pid_t));
if (job->stepid != SLURM_EXTERN_CONT) {
error("_handle_add_extern_pid: non-extern step (%u) given for job %u.",
job->stepid, job->jobid);
rc = SLURM_FAILURE;
goto send_it;
}
debug("_handle_add_extern_pid for job %u.%u, pid %d",
job->jobid, job->stepid, pid);
jobacct_id.taskid = job->nodeid;
jobacct_id.nodeid = job->nodeid;
jobacct_id.job = job;
proctrack_g_add(job, pid);
jobacct_gather_add_task(pid, &jobacct_id, 1);
send_it:
/* Send the return code */
safe_write(fd, &rc, sizeof(int));
debug("Leaving _handle_add_extern_pid");
return SLURM_SUCCESS;
rwfail:
return SLURM_FAILURE;
}
