static void *
_task_launch_detection(void *unused)
{
spawn_resp_t *resp;
time_t start;
int rc = 0;
/*
* mpir_init() is called in plugins/launch/slurm/launch_slurm.c before
* mpi_hook_client_prelaunch() is called in api/step_launch.c
*/
start = time(NULL);
while (_tasks_launched() == 0) {
usleep(1000*50);
if (time(NULL) - start > 600) {
rc = 1;
break;
}
}
/* send a resp to spawner srun */
resp = spawn_resp_new();
resp->seq = job_info.spawn_seq;
resp->jobid = xstrdup(job_info.pmi_jobid);
resp->error_cnt = 0; /* TODO */
resp->rc = rc;
resp->pmi_port = tree_info.pmi_port;
spawn_resp_send_to_srun(resp);
spawn_resp_free(resp);
return NULL;
}
static void
_setup_exec_srun(spawn_req_t *req)
{
char **env, env_key[32];
int i, rc;
spawn_resp_t *resp;
debug3("mpi/pmi2: in _setup_exec_srun");
/* setup environments */
env = env_array_copy((const char **)job_info.job_env);
/* TODO: unset some env-vars */
env_array_overwrite_fmt(&env, "SLURM_JOB_ID", "%u", job_info.jobid);
env_array_overwrite_fmt(&env, PMI2_SPAWNER_JOBID_ENV, "%s",
job_info.pmi_jobid);
env_array_overwrite_fmt(&env, PMI2_PMI_JOBID_ENV, "%s-%u",
job_info.pmi_jobid, req->seq);
env_array_overwrite_fmt(&env, PMI2_SPAWN_SEQ_ENV, "%u", req->seq);
env_array_overwrite_fmt(&env, PMI2_SPAWNER_PORT_ENV, "%hu",
tree_info.pmi_port);
/* preput kvs */
env_array_overwrite_fmt(&env, PMI2_PREPUT_CNT_ENV, "%d",
req->preput_cnt);
for (i = 0; i < req->preput_cnt; i ++) {
snprintf(env_key, 32, PMI2_PPKEY_ENV"%d", i);
env_array_overwrite_fmt(&env, env_key, "%s", req->pp_keys[i]);
snprintf(env_key, 32, PMI2_PPVAL_ENV"%d", i);
env_array_overwrite_fmt(&env, env_key, "%s", req->pp_vals[i]);
}
if (req->subcmd_cnt == 1) {
/* no return if success */
rc = _exec_srun_single(req, env);
} else {
/* no return if success */
rc = _exec_srun_multiple(req, env);
}
resp = spawn_resp_new();
resp->seq = req->seq;
xstrfmtcat(resp->jobid, "%s-%u", job_info.pmi_jobid, req->seq);
resp->error_cnt = 0;
resp->rc = rc;
/* fake a srun address */
tree_info.srun_addr = xmalloc(sizeof(slurm_addr_t));
slurm_set_addr(tree_info.srun_addr, tree_info.pmi_port,
"127.0.0.1");
spawn_resp_send_to_srun(resp);
spawn_resp_free(resp);
exit(errno);
}
