extern int jobacct_gather_endpoll(void)
{
int retval = SLURM_SUCCESS;
if (jobacct_gather_init() < 0)
return SLURM_ERROR;
jobacct_shutdown = true;
slurm_mutex_lock(&task_list_lock);
FREE_NULL_LIST(task_list);
retval = (*(ops.endpoll))();
slurm_mutex_unlock(&task_list_lock);
return retval;
}
extern int jobacct_gather_add_task(pid_t pid, jobacct_id_t *jobacct_id,
int poll)
{
struct jobacctinfo *jobacct;
if (jobacct_gather_init() < 0)
return SLURM_ERROR;
if (!plugin_polling)
return SLURM_SUCCESS;
if (_jobacct_shutdown_test())
return SLURM_ERROR;
jobacct = jobacctinfo_create(jobacct_id);
slurm_mutex_lock(&task_list_lock);
if (pid <= 0) {
error("invalid pid given (%d) for task acct", pid);
goto error;
} else if (!task_list) {
error("no task list created!");
goto error;
}
jobacct->pid = pid;
memcpy(&jobacct->id, jobacct_id, sizeof(jobacct_id_t));
jobacct->min_cpu = 0;
debug2("adding task %u pid %d on node %u to jobacct",
jobacct_id->taskid, pid, jobacct_id->nodeid);
list_push(task_list, jobacct);
slurm_mutex_unlock(&task_list_lock);
(*(ops.add_task))(pid, jobacct_id);
if (poll == 1)
_poll_data(1);
return SLURM_SUCCESS;
error:
slurm_mutex_unlock(&task_list_lock);
jobacctinfo_destroy(jobacct);
return SLURM_ERROR;
}
extern int jobacct_gather_startpoll(uint16_t frequency)
{
int retval = SLURM_SUCCESS;
pthread_attr_t attr;
pthread_t _watch_tasks_thread_id;
if (!plugin_polling)
return SLURM_SUCCESS;
if (jobacct_gather_init() < 0)
return SLURM_ERROR;
if (!jobacct_shutdown) {
error("jobacct_gather_startpoll: poll already started!");
return retval;
}
jobacct_shutdown = false;
freq = frequency;
task_list = list_create(jobacctinfo_destroy);
if (frequency == 0) { /* don't want dynamic monitoring? */
debug2("jobacct_gather dynamic logging disabled");
return retval;
}
/* create polling thread */
slurm_attr_init(&attr);
if (pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED))
error("pthread_attr_setdetachstate error %m");
if (pthread_create(&_watch_tasks_thread_id, &attr,
&_watch_tasks, NULL)) {
debug("jobacct_gather failed to create _watch_tasks "
"thread: %m");
frequency = 0;
} else
debug3("jobacct_gather dynamic logging enabled");
slurm_attr_destroy(&attr);
return retval;
}
extern int jobacct_gather_startpoll(uint16_t frequency)
{
int retval = SLURM_SUCCESS;
pthread_attr_t attr;
if (!plugin_polling)
return SLURM_SUCCESS;
if (jobacct_gather_init() < 0)
return SLURM_ERROR;
if (!_jobacct_shutdown_test()) {
error("jobacct_gather_startpoll: poll already started!");
return retval;
}
slurm_mutex_lock(&jobacct_shutdown_mutex);
jobacct_shutdown = false;
slurm_mutex_unlock(&jobacct_shutdown_mutex);
freq = frequency;
task_list = list_create(jobacctinfo_destroy);
if (frequency == 0) { /* don't want dynamic monitoring? */
debug2("jobacct_gather dynamic logging disabled");
return retval;
}
/* create polling thread */
slurm_attr_init(&attr);
if (pthread_create(&watch_tasks_thread_id, &attr,
&_watch_tasks, NULL)) {
debug("jobacct_gather failed to create _watch_tasks "
"thread: %m");
} else
debug3("jobacct_gather dynamic logging enabled");
slurm_attr_destroy(&attr);
return retval;
}
extern int jobacctinfo_unpack(jobacctinfo_t **jobacct,
uint16_t rpc_version, uint16_t protocol_type,
Buf buffer, bool alloc)
{
uint32_t uint32_tmp;
uint8_t uint8_tmp;
if (jobacct_gather_init() < 0)
return SLURM_ERROR;
if (rpc_version >= SLURM_MIN_PROTOCOL_VERSION) {
safe_unpack8(&uint8_tmp, buffer);
if (uint8_tmp == (uint8_t) 0)
return SLURM_SUCCESS;
if (alloc)
*jobacct = xmalloc(sizeof(struct jobacctinfo));
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->user_cpu_sec = uint32_tmp;
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->user_cpu_usec = uint32_tmp;
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->sys_cpu_sec = uint32_tmp;
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->sys_cpu_usec = uint32_tmp;
safe_unpack64(&(*jobacct)->max_vsize, buffer);
safe_unpack64(&(*jobacct)->tot_vsize, buffer);
safe_unpack64(&(*jobacct)->max_rss, buffer);
safe_unpack64(&(*jobacct)->tot_rss, buffer);
safe_unpack64(&(*jobacct)->max_pages, buffer);
safe_unpack64(&(*jobacct)->tot_pages, buffer);
safe_unpack32(&(*jobacct)->min_cpu, buffer);
safe_unpackdouble(&(*jobacct)->tot_cpu, buffer);
safe_unpack32(&(*jobacct)->act_cpufreq, buffer);
safe_unpack64(&(*jobacct)->energy.consumed_energy, buffer);
safe_unpackdouble(&(*jobacct)->max_disk_read, buffer);
safe_unpackdouble(&(*jobacct)->tot_disk_read, buffer);
safe_unpackdouble(&(*jobacct)->max_disk_write, buffer);
safe_unpackdouble(&(*jobacct)->tot_disk_write, buffer);
if (_unpack_jobacct_id(&(*jobacct)->max_vsize_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_rss_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_pages_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->min_cpu_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_disk_read_id,
rpc_version, buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_disk_write_id,
rpc_version, buffer) != SLURM_SUCCESS)
goto unpack_error;
} else {
info("jobacctinfo_unpack version %u not supported",
rpc_version);
return SLURM_ERROR;
}
return SLURM_SUCCESS;
unpack_error:
debug2("jobacctinfo_unpack: unpack_error: size_buf(buffer) %u",
size_buf(buffer));
if (alloc)
xfree(*jobacct);
return SLURM_ERROR;
}
int
main (int argc, char *argv[])
{
int i, pidfd;
int blocked_signals[] = {SIGPIPE, 0};
int cc;
char *oom_value;
uint32_t slurmd_uid = 0;
uint32_t curr_uid = 0;
char time_stamp[256];
log_options_t lopts = LOG_OPTS_INITIALIZER;
/* NOTE: logfile is NULL at this point */
log_init(argv[0], lopts, LOG_DAEMON, NULL);
/*
* Make sure we have no extra open files which
* would be propagated to spawned tasks.
*/
cc = sysconf(_SC_OPEN_MAX);
for (i = 3; i < cc; i++)
close(i);
/*
* Drop supplementary groups.
*/
if (geteuid() == 0) {
if (setgroups(0, NULL) != 0) {
fatal("Failed to drop supplementary groups, "
"setgroups: %m");
}
} else {
debug("Not running as root. Can't drop supplementary groups");
}
/*
* Create and set default values for the slurmd global
* config variable "conf"
*/
conf = xmalloc(sizeof(slurmd_conf_t));
_init_conf();
conf->argv = &argv;
conf->argc = &argc;
if (_slurmd_init() < 0) {
error( "slurmd initialization failed" );
fflush( NULL );
exit(1);
}
slurmd_uid = slurm_get_slurmd_user_id();
curr_uid = getuid();
if (curr_uid != slurmd_uid) {
struct passwd *pw = NULL;
char *slurmd_user = NULL;
char *curr_user = NULL;
/* since when you do a getpwuid you get a pointer to a
* structure you have to do a xstrdup on the first
* call or your information will just get over
* written. This is a memory leak, but a fatal is
* called right after so it isn't that big of a deal.
*/
if ((pw=getpwuid(slurmd_uid)))
slurmd_user = xstrdup(pw->pw_name);
if ((pw=getpwuid(curr_uid)))
curr_user = pw->pw_name;
fatal("You are running slurmd as something "
"other than user %s(%d). If you want to "
"run as this user add SlurmdUser=%s "
"to the slurm.conf file.",
slurmd_user, slurmd_uid, curr_user);
}
init_setproctitle(argc, argv);
xsignal(SIGTERM, &_term_handler);
xsignal(SIGINT, &_term_handler);
xsignal(SIGHUP, &_hup_handler );
xsignal_block(blocked_signals);
debug3("slurmd initialization successful");
/*
* Become a daemon if desired.
* Do not chdir("/") or close all fd's
*/
if (conf->daemonize) {
if (daemon(1,1) == -1)
error("Couldn't daemonize slurmd: %m");
}
test_core_limit();
info("slurmd version %s started", SLURM_VERSION_STRING);
debug3("finished daemonize");
if ((oom_value = getenv("SLURMD_OOM_ADJ"))) {
i = atoi(oom_value);
debug("Setting slurmd oom_adj to %d", i);
set_oom_adj(i);
}
_kill_old_slurmd();
if (conf->mlock_pages) {
/*
* Call mlockall() if available to ensure slurmd
* doesn't get swapped out
*/
#ifdef _POSIX_MEMLOCK
if (mlockall (MCL_FUTURE | MCL_CURRENT) < 0)
error ("failed to mlock() slurmd pages: %m");
#else
error ("mlockall() system call does not appear to be available");
#endif /* _POSIX_MEMLOCK */
}
/*
* Restore any saved revoked credential information
*/
if (!conf->cleanstart && (_restore_cred_state(conf->vctx) < 0))
return SLURM_FAILURE;
if (jobacct_gather_init() != SLURM_SUCCESS)
fatal("Unable to initialize jobacct_gather");
if (job_container_init() < 0)
fatal("Unable to initialize job_container plugin.");
if (container_g_restore(conf->spooldir, !conf->cleanstart))
error("Unable to restore job_container state.");
if (switch_g_node_init() < 0)
fatal("Unable to initialize interconnect.");
if (conf->cleanstart && switch_g_clear_node_state())
fatal("Unable to clear interconnect state.");
switch_g_slurmd_init();
_create_msg_socket();
conf->pid = getpid();
/* This has to happen after daemon(), which closes all fd's,
so we keep the write lock of the pidfile.
*/
pidfd = create_pidfile(conf->pidfile, 0);
rfc2822_timestamp(time_stamp, sizeof(time_stamp));
info("%s started on %s", slurm_prog_name, time_stamp);
_install_fork_handlers();
list_install_fork_handlers();
slurm_conf_install_fork_handlers();
/*
* Initialize any plugins
*/
if (slurmd_plugstack_init())
fatal("failed to initialize slurmd_plugstack");
_spawn_registration_engine();
_msg_engine();
/*
* Close fd here, otherwise we'll deadlock since create_pidfile()
* flocks the pidfile.
*/
if (pidfd >= 0) /* valid pidfd, non-error */
(void) close(pidfd); /* Ignore errors */
if (unlink(conf->pidfile) < 0)
error("Unable to remove pidfile `%s': %m", conf->pidfile);
_wait_for_all_threads(120);
_slurmd_fini();
_destroy_conf();
slurm_crypto_fini(); /* must be after _destroy_conf() */
info("Slurmd shutdown completing");
log_fini();
return 0;
}
extern int jobacctinfo_unpack(jobacctinfo_t **jobacct,
uint16_t rpc_version, uint16_t protocol_type,
Buf buffer, bool alloc)
{
uint32_t uint32_tmp;
uint8_t uint8_tmp;
bool no_pack;
jobacct_gather_init();
no_pack = (!plugin_polling && (protocol_type != PROTOCOL_TYPE_DBD));
/* The function can take calls from both DBD and from regular
* SLURM functions. We choose to standardize on using the
* SLURM_PROTOCOL_VERSION here so if PROTOCOL_TYPE_DBD comes
* in we need to translate the DBD rpc_version to use the
* SLURM protocol_version.
*
* If this function ever changes make sure the
* slurmdbd_translate_rpc function has been updated with the
* new protocol version.
*/
if (protocol_type == PROTOCOL_TYPE_DBD)
rpc_version = slurmdbd_translate_rpc(rpc_version);
if (rpc_version >= SLURM_14_03_PROTOCOL_VERSION) {
safe_unpack8(&uint8_tmp, buffer);
if (uint8_tmp == (uint8_t) 0)
return SLURM_SUCCESS;
if (alloc)
*jobacct = xmalloc(sizeof(struct jobacctinfo));
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->user_cpu_sec = uint32_tmp;
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->user_cpu_usec = uint32_tmp;
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->sys_cpu_sec = uint32_tmp;
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->sys_cpu_usec = uint32_tmp;
safe_unpack64(&(*jobacct)->max_vsize, buffer);
safe_unpack64(&(*jobacct)->tot_vsize, buffer);
safe_unpack64(&(*jobacct)->max_rss, buffer);
safe_unpack64(&(*jobacct)->tot_rss, buffer);
safe_unpack64(&(*jobacct)->max_pages, buffer);
safe_unpack64(&(*jobacct)->tot_pages, buffer);
safe_unpack32(&(*jobacct)->min_cpu, buffer);
safe_unpack32(&(*jobacct)->tot_cpu, buffer);
safe_unpack32(&(*jobacct)->act_cpufreq, buffer);
safe_unpack32(&(*jobacct)->energy.consumed_energy, buffer);
safe_unpackdouble(&(*jobacct)->max_disk_read, buffer);
safe_unpackdouble(&(*jobacct)->tot_disk_read, buffer);
safe_unpackdouble(&(*jobacct)->max_disk_write, buffer);
safe_unpackdouble(&(*jobacct)->tot_disk_write, buffer);
if (_unpack_jobacct_id(&(*jobacct)->max_vsize_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_rss_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_pages_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->min_cpu_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_disk_read_id,
rpc_version, buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_disk_write_id,
rpc_version, buffer) != SLURM_SUCCESS)
goto unpack_error;
} else if (rpc_version >= SLURM_2_6_PROTOCOL_VERSION) {
if (no_pack)
return SLURM_SUCCESS;
if (alloc)
*jobacct = xmalloc(sizeof(struct jobacctinfo));
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->user_cpu_sec = uint32_tmp;
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->user_cpu_usec = uint32_tmp;
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->sys_cpu_sec = uint32_tmp;
safe_unpack32(&uint32_tmp, buffer);
(*jobacct)->sys_cpu_usec = uint32_tmp;
safe_unpack32((uint32_t *)&(*jobacct)->max_vsize, buffer);
safe_unpack32((uint32_t *)&(*jobacct)->tot_vsize, buffer);
safe_unpack32((uint32_t *)&(*jobacct)->max_rss, buffer);
safe_unpack32((uint32_t *)&(*jobacct)->tot_rss, buffer);
safe_unpack32((uint32_t *)&(*jobacct)->max_pages, buffer);
safe_unpack32((uint32_t *)&(*jobacct)->tot_pages, buffer);
safe_unpack32(&(*jobacct)->min_cpu, buffer);
safe_unpack32(&(*jobacct)->tot_cpu, buffer);
safe_unpack32(&(*jobacct)->act_cpufreq, buffer);
safe_unpack32(&(*jobacct)->energy.consumed_energy, buffer);
safe_unpackdouble(&(*jobacct)->max_disk_read, buffer);
safe_unpackdouble(&(*jobacct)->tot_disk_read, buffer);
safe_unpackdouble(&(*jobacct)->max_disk_write, buffer);
safe_unpackdouble(&(*jobacct)->tot_disk_write, buffer);
if (_unpack_jobacct_id(&(*jobacct)->max_vsize_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_rss_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_pages_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->min_cpu_id, rpc_version,
buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_disk_read_id,
rpc_version, buffer) != SLURM_SUCCESS)
goto unpack_error;
if (_unpack_jobacct_id(&(*jobacct)->max_disk_write_id,
rpc_version, buffer) != SLURM_SUCCESS)
goto unpack_error;
} else {
info("jobacctinfo_unpack version %u not supported",
rpc_version);
return SLURM_ERROR;
}
return SLURM_SUCCESS;
unpack_error:
debug2("jobacctinfo_unpack: unpack_error: size_buf(buffer) %u",
size_buf(buffer));
if (alloc)
xfree(*jobacct);
return SLURM_ERROR;
}