static int do_site(void)
{
int fd;
int rv = -1;
if (!cl.debug) {
if (daemon(0, 0) < 0) {
perror("daemon error");
exit(EXIT_FAILURE);
}
}
setup_logging();
fd = lockfile();
if (fd < 0)
return fd;
log_info("BOOTH cluster site daemon started");
set_scheduler();
set_oom_adj(-16);
rv = loop(SITE);
if (rv < 0)
goto fail;
unlink_lockfile(fd);
close_logging();
return 0;
fail:
return -1;
}
int main(int argc, char *argv[])
{
char *envp = NULL;
set_oom_adj(-1000);
if (initService(argc, argv) != 0)
return -1;
envp = ::getenv("SCI_DISABLE_IPV6");
if (envp && (::strcasecmp(envp, "yes") == 0)) {
Socket::setDisableIPv6(1);
}
ExtListener *listener = new ExtListener();
listener->start();
launcherList.clear();
while (1) {
Locker::getLocker()->freeze();
log_debug("Delete unused launcher");
Locker::getLocker()->lock();
vector<ExtLauncher *>::iterator lc;
for (lc = launcherList.begin(); lc != launcherList.end(); lc++) {
while (!(*lc)->isLaunched()) {
// before join, this thread should have been launched
SysUtil::sleep(WAIT_INTERVAL);
}
(*lc)->join();
delete (*lc);
}
launcherList.clear();
Locker::getLocker()->unlock();
log_info("Begin to cleanup the jobInfo");
Locker::getLocker()->lock();
JOB_INFO::iterator it;
for (it = jobInfo.begin(); it != jobInfo.end(); ) {
if ((SysUtil::microseconds() - it->second.timestamp) > FIVE_MINUTES) {
log_crit("Erase jobInfo item %d", it->first);
jobInfo.erase(it++);
} else {
++it;
}
}
log_info("Finish cleanup the jobInfo");
Locker::getLocker()->unlock();
}
listener->join();
delete listener;
delete Log::getInstance();
return 0;
}
static int do_server(int type)
{
int fd = -1;
int rv = -1;
rv = setup(type);
if (rv < 0)
goto out;
if (!daemonize) {
if (daemon(0, 0) < 0) {
perror("daemon error");
exit(EXIT_FAILURE);
}
}
/*
The lock cannot be obtained before the call to daemon(), otherwise
the lockfile would contain the pid of the parent, not the daemon.
*/
fd = lockfile();
if (fd < 0)
return fd;
if (type == ARBITRATOR)
log_info("BOOTH arbitrator daemon started");
else if (type == SITE)
log_info("BOOTH cluster site daemon started");
set_scheduler();
set_oom_adj(-16);
rv = loop();
out:
if (fd >= 0)
unlink_lockfile(fd);
return rv;
}
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;
}
static int do_server(int type)
{
int rv = -1;
static char log_ent[128] = DAEMON_NAME "-";
rv = setup_config(type);
if (rv < 0)
return rv;
if (!local) {
log_error("Cannot find myself in the configuration.");
exit(EXIT_FAILURE);
}
if (!daemonize) {
if (daemon(0, 0) < 0) {
perror("daemon error");
exit(EXIT_FAILURE);
}
}
/* The lockfile must be written to _after_ the call to daemon(), so
* that the lockfile contains the pid of the daemon, not the parent. */
lock_fd = create_lockfile();
if (lock_fd < 0)
return lock_fd;
atexit(server_exit);
strcat(log_ent, type_to_string(local->type));
cl_log_set_entity(log_ent);
cl_log_enable_stderr(enable_stderr ? TRUE : FALSE);
cl_log_set_facility(HA_LOG_FACILITY);
cl_inherit_logging_environment(0);
log_info("BOOTH %s %s daemon is starting",
type_to_string(local->type), RELEASE_STR);
signal(SIGUSR1, (__sighandler_t)tickets_log_info);
signal(SIGTERM, (__sighandler_t)sig_exit_handler);
signal(SIGINT, (__sighandler_t)sig_exit_handler);
set_scheduler();
set_oom_adj(-16);
set_proc_title("%s %s %s for [%s]:%d",
DAEMON_NAME,
cl.configfile,
type_to_string(local->type),
local->addr_string,
booth_conf->port);
rv = limit_this_process();
if (rv)
return rv;
if (cl_enable_coredumps(TRUE) < 0){
cl_log(LOG_ERR, "enabling core dump failed");
}
cl_cdtocoredir();
prctl(PR_SET_DUMPABLE, (unsigned long)TRUE, 0UL, 0UL, 0UL);
rv = loop(lock_fd);
return rv;
}
