/*-------------------------------------------------------------------*/
void sge_setup_sge_execd(sge_gdi_ctx_class_t *ctx, const char* tmp_err_file_name)
{
char err_str[MAX_STRING_SIZE];
int allowed_get_conf_errors = 5;
char* spool_dir = NULL;
const char *unqualified_hostname = ctx->get_unqualified_hostname(ctx);
const char *admin_user = ctx->get_admin_user(ctx);
DENTER(TOP_LAYER, "sge_setup_sge_execd");
/* TODO: is this the right place to switch the user ?
ports below 1024 ok */
/*
** switch to admin user
*/
if (sge_set_admin_username(admin_user, err_str)) {
CRITICAL((SGE_EVENT, SFNMAX, err_str));
/* TODO: remove */
SGE_EXIT(NULL, 1);
}
if (sge_switch2admin_user()) {
CRITICAL((SGE_EVENT, SFNMAX, MSG_ERROR_CANTSWITCHTOADMINUSER));
/* TODO: remove */
SGE_EXIT(NULL, 1);
}
while (gdi2_wait_for_conf(ctx, &Execd_Config_List)) {
if (allowed_get_conf_errors-- <= 0) {
CRITICAL((SGE_EVENT, SFNMAX, MSG_EXECD_CANT_GET_CONFIGURATION_EXIT));
/* TODO: remove */
SGE_EXIT(NULL, 1);
}
sleep(1);
ctx->get_master(ctx, true);
}
sge_show_conf();
/* get aliased hostname */
/* TODO: is this call needed ? */
ctx->reresolve_qualified_hostname(ctx);
spool_dir = mconf_get_execd_spool_dir();
DPRINTF(("chdir(\"/\")----------------------------\n"));
sge_chdir_exit("/",1);
DPRINTF(("Making directories----------------------------\n"));
sge_mkdir(spool_dir, 0755, 1, 0);
DPRINTF(("chdir(\"%s\")----------------------------\n", spool_dir));
sge_chdir_exit(spool_dir,1);
sge_mkdir(unqualified_hostname, 0755, 1, 0);
DPRINTF(("chdir(\"%s\",me.unqualified_hostname)--------------------------\n",
unqualified_hostname));
sge_chdir_exit(unqualified_hostname, 1);
/* having passed the previous statement we may
log messages into the ERR_FILE */
if ( tmp_err_file_name != NULL) {
sge_copy_append((char*)tmp_err_file_name, ERR_FILE, SGE_MODE_APPEND);
}
sge_switch2start_user();
if ( tmp_err_file_name != NULL) {
unlink(tmp_err_file_name);
}
sge_switch2admin_user();
log_state_set_log_as_admin_user(1);
sprintf(execd_messages_file, "%s/%s/%s", spool_dir,
unqualified_hostname, ERR_FILE);
log_state_set_log_file(execd_messages_file);
sprintf(execd_spool_dir, "%s/%s", spool_dir,
unqualified_hostname);
DPRINTF(("Making directories----------------------------\n"));
sge_mkdir(EXEC_DIR, 0775, 1, 0);
sge_mkdir(JOB_DIR, 0775, 1, 0);
sge_mkdir(ACTIVE_DIR, 0775, 1, 0);
#if defined(PLPA_LINUX) || defined(SOLARIS86) || defined(SOLARISAMD64)
/* initialize processor topology */
if (initialize_topology() != true) {
DPRINTF(("Couldn't initizalize topology-----------------------\n"));
}
#endif
sge_free(&spool_dir);
DRETURN_VOID;
}
/****** qmaster/sge_qmaster_main/main() ****************************************
* NAME
* main() -- qmaster entry point
*
* SYNOPSIS
* int main(int argc, char* argv[])
*
* FUNCTION
* Qmaster entry point.
*
* NOTE: The main thread must block all signals before any additional thread
* is created. Failure to do so will ruin signal handling!
*
* INPUTS
* int argc - number of commandline arguments
* char* argv[] - commandline arguments
*
* RESULT
* 0 - success
*
* NOTES
* We check whether 'SGE_ROOT' is set before we daemonize. Once qmaster is
* a daemon, we are no longer connected to a terminal and hence can not
* output an error message to stdout or stderr.
*
* We need to inovke 'prepare_enroll()' *before* the user id is switched via
* 'become_admin_user()'. This is because qmaster must be able to bind a so
* called reserved port (requires root privileges) if configured to do so.
*
*******************************************************************************/
int main(int argc, char* argv[])
{
int max_enroll_tries;
int ret_val;
int file_descriptor_settings_result = 0;
bool has_daemonized = false;
sge_gdi_ctx_class_t *ctx = NULL;
u_long32 start_time = sge_get_gmt();
monitoring_t monitor;
DENTER_MAIN(TOP_LAYER, "qmaster");
sge_monitor_init(&monitor, "MAIN", NONE_EXT, MT_WARNING, MT_ERROR);
prof_mt_init();
sge_get_root_dir(true, NULL, 0, true);
#ifdef __SGE_COMPILE_WITH_GETTEXT__
sge_init_language_func((gettext_func_type)gettext, (setlocale_func_type)setlocale, (bindtextdomain_func_type)bindtextdomain, (textdomain_func_type)textdomain);
sge_init_language(NULL,NULL);
#endif
/*
* qmaster doesn't support any commandline anymore,
* but we should show version string and -help option
*/
if (argc != 1) {
sigset_t sig_set;
sigfillset(&sig_set);
pthread_sigmask(SIG_SETMASK, &sig_set, NULL);
sge_qmaster_thread_init(&ctx, QMASTER, MAIN_THREAD, true);
sge_process_qmaster_cmdline(argv);
SGE_EXIT((void**)&ctx, 1);
}
/*
* daemonize qmaster
* set file descriptor limits
* and initialize libraries to be used in multi threaded environment
* also take care that finished child processed of this process become
* zombie jobs
*/
has_daemonized = sge_daemonize_qmaster();
file_descriptor_settings_result = set_file_descriptor_limit();
#if !defined(INTERIX) && !defined(CYGWIN)
init_sig_action_and_mask();
#endif
/* init qmaster threads without becomming admin user */
sge_qmaster_thread_init(&ctx, QMASTER, MAIN_THREAD, false);
ctx->set_daemonized(ctx, has_daemonized);
/* this must be done as root user to be able to bind ports < 1024 */
max_enroll_tries = 30;
while (cl_com_get_handle(prognames[QMASTER],1) == NULL) {
ctx->prepare_enroll(ctx);
max_enroll_tries--;
if (max_enroll_tries <= 0) {
/* exit after 30 seconds */
CRITICAL((SGE_EVENT, MSG_QMASTER_COMMUNICATION_ERRORS ));
SGE_EXIT((void**)&ctx, 1);
}
if (cl_com_get_handle(prognames[QMASTER],1) == NULL) {
/* sleep when prepare_enroll() failed */
sleep(1);
}
}
/*
* now the commlib up and running. Set qmaster application status function
* (commlib callback function for qping status information response
* messages (SIRM))
*/
ret_val = cl_com_set_status_func(sge_qmaster_application_status);
if (ret_val != CL_RETVAL_OK) {
ERROR((SGE_EVENT, cl_get_error_text(ret_val)));
}
/*
* now we become admin user change into the correct root directory set the
* the target for logging messages
*/
sge_become_admin_user(ctx->get_admin_user(ctx));
sge_chdir_exit(ctx->get_qmaster_spool_dir(ctx), 1);
log_state_set_log_file(ERR_FILE);
ctx->set_exit_func(ctx, sge_exit_func);
#if defined(SOLARIS)
/* Init shared SMF libs if necessary */
if (sge_smf_used() == 1 && sge_smf_init_libs() != 0) {
SGE_EXIT((void**)&ctx, 1);
}
#endif
/*
* We do increment the heartbeat manually here. This is the 'startup heartbeat'.
* The first time the hearbeat will be incremented through the heartbeat event
* handler is after about HEARTBEAT_INTERVAL seconds. The hardbeat event handler
* is setup during the initialisazion of the timer thread.
*/
inc_qmaster_heartbeat(QMASTER_HEARTBEAT_FILE, HEARTBEAT_INTERVAL, NULL);
/*
* Event master module has to be initialized already here because
* sge_setup_qmaster() might already access it although event delivery
* thread is not running.
*
* Corresponding shutdown is done in sge_event_master_terminate();
*
* EB: In my opinion the init function should called in
* sge_event_master_initialize(). Is it possible to move that call?
*/
sge_event_master_init();
sge_setup_qmaster(ctx, argv);
#ifndef USE_POLL
if (file_descriptor_settings_result == 1) {
WARNING((SGE_EVENT, MSG_QMASTER_FD_SETSIZE_LARGER_THAN_LIMIT_U, sge_u32c(FD_SETSIZE)));
WARNING((SGE_EVENT, MSG_QMASTER_FD_SETSIZE_COMPILE_MESSAGE1_U, sge_u32c(FD_SETSIZE - 20)));
WARNING((SGE_EVENT, MSG_QMASTER_FD_SETSIZE_COMPILE_MESSAGE2));
WARNING((SGE_EVENT, MSG_QMASTER_FD_SETSIZE_COMPILE_MESSAGE3));
}
#endif
/*
* Setup all threads and initialize corresponding modules.
* Order is important!
*/
sge_signaler_initialize(ctx);
sge_event_master_initialize(ctx);
sge_timer_initialize(ctx, &monitor);
sge_worker_initialize(ctx);
#if 0
sge_test_initialize(ctx);
#endif
sge_listener_initialize(ctx);
sge_scheduler_initialize(ctx, NULL);
#ifndef NO_JNI
sge_jvm_initialize(ctx, NULL);
#endif
INFO((SGE_EVENT, "qmaster startup took "sge_u32" seconds", sge_get_gmt() - start_time));
/*
* Block till signal from signal thread arrives us
*/
sge_thread_wait_for_signal();
/*
* Shutdown all threads and shutdown corresponding modules.
* Order is important!
*/
#ifndef NO_JNI
sge_jvm_terminate(ctx, NULL);
#endif
sge_scheduler_terminate(ctx, NULL);
sge_listener_terminate();
#if 0
sge_test_terminate(ctx);
#endif
sge_worker_terminate(ctx);
sge_timer_terminate();
sge_event_master_terminate();
sge_signaler_terminate();
/*
* Remaining shutdown operations
*/
sge_clean_lists();
sge_monitor_free(&monitor);
sge_shutdown((void**)&ctx, sge_qmaster_get_exit_state());
sge_prof_cleanup();
DEXIT;
return 0;
} /* main() */
