/*
* Note that a step has terminated.
*
* RET - slurm error code
*/
extern int post_step(slurmd_job_t *job)
{
if (slurmd_task_init())
return SLURM_ERROR;
return (*(g_task_context->ops.post_step))(job);
}
/*
* Note that a task launch is about to occur.
*
* RET - slurm error code
*/
extern int pre_launch(slurmd_job_t *job)
{
if (slurmd_task_init())
return SLURM_ERROR;
return (*(g_task_context->ops.pre_launch))(job);
}
/*
* Slurmd is releasing resources for the task.
*
* RET - slurm error code
*/
extern int slurmd_release_resources(uint32_t job_id)
{
if (slurmd_task_init())
return SLURM_ERROR;
return (*(g_task_context->ops.slurmd_release_resources))(job_id);
}
/*
* Slurmd is resuming a previously suspended job.
*
* RET - slurm error code
*/
extern int slurmd_resume_job(uint32_t job_id)
{
if (slurmd_task_init())
return SLURM_ERROR;
return (*(g_task_context->ops.slurmd_resume_job))(job_id);
}
/*
* Slurmd has received a batch job launch request.
*
* RET - slurm error code
*/
extern int slurmd_batch_request(uint32_t job_id, batch_job_launch_msg_t *req)
{
if (slurmd_task_init())
return SLURM_ERROR;
return (*(g_task_context->ops.slurmd_batch_request))(job_id, req);
}
/*
* Slurmd is reserving resources for the task.
*
* RET - slurm error code
*/
extern int slurmd_reserve_resources(uint32_t job_id,
launch_tasks_request_msg_t *req,
uint32_t node_id )
{
if (slurmd_task_init())
return SLURM_ERROR;
return (*(g_task_context->ops.slurmd_reserve_resources))(job_id, req, node_id);
}
/*
* Note that a step has terminated.
*
* RET - slurm error code
*/
extern int post_step(slurmd_job_t *job)
{
int i, rc = SLURM_SUCCESS;
if (slurmd_task_init())
return SLURM_ERROR;
slurm_mutex_lock( &g_task_context_lock );
for (i = 0; ((i < g_task_context_num) && (rc == SLURM_SUCCESS)); i++)
rc = (*(g_task_context[i]->ops.post_step))(job);
slurm_mutex_unlock( &g_task_context_lock );
return (rc);
}
/*
* Slurmd is resuming a previously suspended job.
*
* RET - slurm error code
*/
extern int slurmd_resume_job(uint32_t job_id)
{
int i, rc = SLURM_SUCCESS;
if (slurmd_task_init())
return SLURM_ERROR;
slurm_mutex_lock( &g_task_context_lock );
for (i = 0; ((i < g_task_context_num) && (rc == SLURM_SUCCESS)); i++)
rc = (*(g_task_context[i]->ops.slurmd_resume_job))(job_id);
slurm_mutex_unlock( &g_task_context_lock );
return (rc);
}
/*
* Slurmd has received a batch job launch request.
*
* RET - slurm error code
*/
extern int slurmd_batch_request(uint32_t job_id, batch_job_launch_msg_t *req)
{
int i, rc = SLURM_SUCCESS;
if (slurmd_task_init())
return SLURM_ERROR;
slurm_mutex_lock( &g_task_context_lock );
for (i = 0; ((i < g_task_context_num) && (rc == SLURM_SUCCESS)); i++) {
rc = (*(g_task_context[i]->ops.slurmd_batch_request))(job_id,
req);
}
slurm_mutex_unlock( &g_task_context_lock );
return (rc);
}
/*
* Slurmd is reserving resources for the task.
*
* RET - slurm error code
*/
extern int slurmd_reserve_resources(uint32_t job_id,
launch_tasks_request_msg_t *req,
uint32_t node_id )
{
int i, rc = SLURM_SUCCESS;
if (slurmd_task_init())
return SLURM_ERROR;
slurm_mutex_lock( &g_task_context_lock );
for (i = 0; ((i < g_task_context_num) && (rc == SLURM_SUCCESS)); i++) {
rc = (*(g_task_context[i]->ops.slurmd_reserve_resources))
(job_id, req, node_id);
}
slurm_mutex_unlock( &g_task_context_lock );
return (rc);
}
static int
_slurmd_init(void)
{
struct rlimit rlim;
slurm_ctl_conf_t *cf;
struct stat stat_buf;
uint32_t cpu_cnt;
/*
* Process commandline arguments first, since one option may be
* an alternate location for the slurm config file.
*/
_process_cmdline(*conf->argc, *conf->argv);
/*
* Build nodes table like in slurmctld
* This is required by the topology stack
* Node tables setup must preceed _read_config() so that the
* proper hostname is set.
*/
slurm_conf_init(conf->conffile);
init_node_conf();
/* slurm_select_init() must be called before
* build_all_nodeline_info() to be called with proper argument. */
if (slurm_select_init(1) != SLURM_SUCCESS )
return SLURM_FAILURE;
build_all_nodeline_info(true);
build_all_frontend_info(true);
/*
* Read global slurm config file, override necessary values from
* defaults and command line.
*/
_read_config();
cpu_cnt = MAX(conf->conf_cpus, conf->block_map_size);
if ((gres_plugin_init() != SLURM_SUCCESS) ||
(gres_plugin_node_config_load(cpu_cnt) != SLURM_SUCCESS))
return SLURM_FAILURE;
if (slurm_topo_init() != SLURM_SUCCESS)
return SLURM_FAILURE;
/*
* Get and set slurmd topology information
* Build node hash table first to speed up the topo build
*/
rehash_node();
slurm_topo_build_config();
_set_topo_info();
/*
* Check for cpu frequency set capabilities on this node
*/
cpu_freq_init(conf);
_print_conf();
if (slurm_proctrack_init() != SLURM_SUCCESS)
return SLURM_FAILURE;
if (slurmd_task_init() != SLURM_SUCCESS)
return SLURM_FAILURE;
if (slurm_auth_init(NULL) != SLURM_SUCCESS)
return SLURM_FAILURE;
if (spank_slurmd_init() < 0)
return SLURM_FAILURE;
if (getrlimit(RLIMIT_CPU, &rlim) == 0) {
rlim.rlim_cur = rlim.rlim_max;
setrlimit(RLIMIT_CPU, &rlim);
if (rlim.rlim_max != RLIM_INFINITY) {
error("Slurmd process CPU time limit is %d seconds",
(int) rlim.rlim_max);
}
}
if (getrlimit(RLIMIT_NOFILE, &rlim) == 0) {
rlim.rlim_cur = rlim.rlim_max;
setrlimit(RLIMIT_NOFILE, &rlim);
}
#ifndef NDEBUG
if (getrlimit(RLIMIT_CORE, &rlim) == 0) {
rlim.rlim_cur = rlim.rlim_max;
setrlimit(RLIMIT_CORE, &rlim);
}
#endif /* !NDEBUG */
/*
* Create a context for verifying slurm job credentials
*/
if (!(conf->vctx = slurm_cred_verifier_ctx_create(conf->pubkey)))
return SLURM_FAILURE;
if (!strcmp(conf->select_type, "select/serial")) {
/* Only cache credential for 5 seconds with select/serial
* for shorter cache searches and higher throughput */
slurm_cred_ctx_set(conf->vctx, SLURM_CRED_OPT_EXPIRY_WINDOW, 5);
}
/*
* Create slurmd spool directory if necessary.
*/
if (_set_slurmd_spooldir() < 0) {
error("Unable to initialize slurmd spooldir");
return SLURM_FAILURE;
}
if (conf->cleanstart) {
/*
* Need to kill any running slurmd's here
*/
_kill_old_slurmd();
stepd_cleanup_sockets(conf->spooldir, conf->node_name);
_stepd_cleanup_batch_dirs(conf->spooldir, conf->node_name);
}
if (conf->daemonize) {
bool success = false;
if (conf->logfile && (conf->logfile[0] == '/')) {
char *slash_ptr, *work_dir;
work_dir = xstrdup(conf->logfile);
slash_ptr = strrchr(work_dir, '/');
if (slash_ptr == work_dir)
work_dir[1] = '\0';
else
slash_ptr[0] = '\0';
if ((access(work_dir, W_OK) != 0) ||
(chdir(work_dir) < 0)) {
error("Unable to chdir to %s", work_dir);
} else
success = true;
xfree(work_dir);
}
if (!success) {
if ((access(conf->spooldir, W_OK) != 0) ||
(chdir(conf->spooldir) < 0)) {
error("Unable to chdir to %s", conf->spooldir);
} else
success = true;
}
if (!success) {
if ((access("/var/tmp", W_OK) != 0) ||
(chdir("/var/tmp") < 0)) {
error("chdir(/var/tmp): %m");
return SLURM_FAILURE;
} else
info("chdir to /var/tmp");
}
}
/*
* Cache the group access list
*/
cf = slurm_conf_lock();
if (cf->group_info & GROUP_CACHE)
init_gids_cache(1);
else
init_gids_cache(0);
slurm_conf_unlock();
if ((devnull = open_cloexec("/dev/null", O_RDWR)) < 0) {
error("Unable to open /dev/null: %m");
return SLURM_FAILURE;
}
/* make sure we have slurmstepd installed */
if (stat(conf->stepd_loc, &stat_buf))
fatal("Unable to find slurmstepd file at %s", conf->stepd_loc);
if (!S_ISREG(stat_buf.st_mode))
fatal("slurmstepd not a file at %s", conf->stepd_loc);
return SLURM_SUCCESS;
}