/****** qmaster/setup_qmaster/sge_setup_job_resend() ***************************
* NAME
* sge_setup_job_resend() -- Setup job resend events.
*
* SYNOPSIS
* void sge_setup_job_resend(void)
*
* FUNCTION
* Register a job resend event for each job or array task which does have a
* 'JTRANSFERING' status.
*
* INPUTS
* void - none
*
* RESULT
* void - none
*
* NOTES
* MT-NOTE: sge_setup_job_resend() is not MT safe
*
*******************************************************************************/
void sge_setup_job_resend(void)
{
lListElem *job = NULL;
object_description *object_base = object_type_get_object_description();
DENTER(TOP_LAYER, "sge_setup_job_resend");
job = lFirst(*object_base[SGE_TYPE_JOB].list);
while (NULL != job)
{
lListElem *task;
u_long32 job_num;
job_num = lGetUlong(job, JB_job_number);
task = lFirst(lGetList(job, JB_ja_tasks));
while (NULL != task)
{
if (lGetUlong(task, JAT_status) == JTRANSFERING)
{
lListElem *granted_queue, *qinstance, *host;
const char *qname;
u_long32 task_num, when;
te_event_t ev;
task_num = lGetUlong(task, JAT_task_number);
granted_queue = lFirst(lGetList(task, JAT_granted_destin_identifier_list));
qname = lGetString(granted_queue, JG_qname);
qinstance = cqueue_list_locate_qinstance(*object_base[SGE_TYPE_CQUEUE].list, qname);
host = host_list_locate(*object_base[SGE_TYPE_EXECHOST].list, lGetHost(qinstance, QU_qhostname));
when = lGetUlong(task, JAT_start_time);
when += MAX(load_report_interval(host), MAX_JOB_DELIVER_TIME);
ev = te_new_event((time_t)when, TYPE_JOB_RESEND_EVENT, ONE_TIME_EVENT, job_num, task_num, "job-resend_event");
te_add_event(ev);
te_free_event(&ev);
DPRINTF(("Did add job resend for "sge_u32"/"sge_u32" at %d\n", job_num, task_num, when));
}
task = lNext(task);
}
job = lNext(job);
}
DEXIT;
return;
} /* sge_setup_job_resend() */
/****** qmaster/sge_qmaster_main/sge_start_heartbeat() *****************************
* NAME
* sge_start_heartbeat() -- Start qmaster heartbeat.
*
* SYNOPSIS
* static void sge_start_heartbeat(void)
*
* FUNCTION
* Add heartbeat event and register according event handler.
*
* INPUTS
* void - none
*
* RESULT
* void - none
*
* NOTES
* MT-NOTE: sge_start_heartbeat() is MT safe
*
*******************************************************************************/
void heartbeat_initialize(void)
{
te_event_t ev = NULL;
DENTER(TOP_LAYER, "heartbeat_initialize");
te_register_event_handler(increment_heartbeat, TYPE_HEARTBEAT_EVENT);
ev = te_new_event(HEARTBEAT_INTERVAL, TYPE_HEARTBEAT_EVENT, RECURRING_EVENT,
0, 0, "heartbeat-event");
te_add_event(ev);
te_free_event(&ev);
/* this is for testsuite shadowd test */
if (getenv("SGE_TEST_HEARTBEAT_TIMEOUT") != NULL) {
int test_timeout = atoi(getenv("SGE_TEST_HEARTBEAT_TIMEOUT"));
set_inc_qmaster_heartbeat_test_mode(test_timeout);
DPRINTF(("heartbeat timeout test enabled (timeout="sge_U32CFormat")\n", sge_u32c(test_timeout)));
}
DEXIT;
return;
} /* sge_start_heartbeat(void) */
/****** qmaster/sge_mod_configuration() ****************************************
* NAME
* sge_mod_configuration() -- modify cluster configuration
*
* SYNOPSIS
* int sge_mod_configuration(lListElem *aConf, lList **anAnswer, char *aUser,
* char *aHost)
*
* FUNCTION
* Modify cluster configuration. 'confp' is a pointer to a 'CONF_Type' list
* element and does contain the modified configuration entry. Adding a new
* configuration entry is also viewed as a modification.
*
* INPUTS
* lListElem *aConf - CONF_Type element containing the modified conf
* lList **anAnswer - answer list
* char *aUser - target user
* char *aHost - target host
*
* RESULT
* int - 0 success
* -1 error
*
* NOTES
* MT-NOTE: sge_mod_configuration() is MT safe
*
*******************************************************************************/
int sge_mod_configuration(sge_gdi_ctx_class_t *ctx, lListElem *aConf, lList **anAnswer, char *aUser, char *aHost)
{
lListElem *old_conf;
const char *tmp_name = NULL;
char unique_name[CL_MAXHOSTLEN];
int ret = -1;
const char *cell_root = ctx->get_cell_root(ctx);
const char *qualified_hostname = ctx->get_qualified_hostname(ctx);
u_long32 progid = ctx->get_who(ctx);
DENTER(TOP_LAYER, "sge_mod_configuration");
if (!aConf || !aUser || !aHost) {
CRITICAL((SGE_EVENT, MSG_SGETEXT_NULLPTRPASSED_S, SGE_FUNC));
answer_list_add(anAnswer, SGE_EVENT, STATUS_EUNKNOWN, ANSWER_QUALITY_ERROR);
DRETURN(STATUS_EUNKNOWN);
}
if ((tmp_name = lGetHost(aConf, CONF_name)) == NULL) {
CRITICAL((SGE_EVENT, MSG_SGETEXT_MISSINGCULLFIELD_SS, lNm2Str(CONF_name), SGE_FUNC));
answer_list_add(anAnswer, SGE_EVENT, STATUS_EUNKNOWN, ANSWER_QUALITY_ERROR);
DRETURN(STATUS_EUNKNOWN);
}
if ((ret = sge_resolve_hostname(tmp_name, unique_name, EH_name, sizeof(unique_name)))
!= CL_RETVAL_OK) {
DPRINTF(("%s: error %s resolving host %s\n", SGE_FUNC, cl_get_error_text(ret), tmp_name));
ERROR((SGE_EVENT, MSG_SGETEXT_CANTRESOLVEHOST_S, tmp_name));
answer_list_add(anAnswer, SGE_EVENT, STATUS_EUNKNOWN, ANSWER_QUALITY_ERROR);
DRETURN(STATUS_EUNKNOWN);
}
if ((ret = check_config(anAnswer, aConf))) {
DRETURN(ret);
}
if ((old_conf = sge_get_configuration_for_host(unique_name)) != NULL) {
int ret = -1;
ret = do_mod_config(ctx, unique_name, old_conf, aConf, anAnswer);
lFreeElem(&old_conf);
if (ret == 0) {
INFO((SGE_EVENT, MSG_SGETEXT_MODIFIEDINLIST_SSSS, aUser, aHost, unique_name, MSG_OBJ_CONF));
answer_list_add(anAnswer, SGE_EVENT, STATUS_OK, ANSWER_QUALITY_INFO);
} else {
DRETURN(STATUS_EUNKNOWN);
}
} else {
do_add_config(ctx, unique_name, aConf, anAnswer);
INFO((SGE_EVENT, MSG_SGETEXT_ADDEDTOLIST_SSSS, aUser, aHost, unique_name, MSG_OBJ_CONF));
answer_list_add(anAnswer, SGE_EVENT, STATUS_OK, ANSWER_QUALITY_INFO);
}
if (strcmp(SGE_GLOBAL_NAME, unique_name) == 0) {
sge_add_event(0, sgeE_GLOBAL_CONFIG, 0, 0, NULL, NULL, NULL, NULL);
}
/*
** is the configuration change relevant for the qmaster itsself?
** if so, initialise conf struct anew
*/
if (strcmp(unique_name, SGE_GLOBAL_NAME) == 0 || sge_hostcmp(unique_name, qualified_hostname) == 0) {
lListElem *local = NULL;
lListElem *global = NULL;
lList *answer_list = NULL;
char* qmaster_params = NULL;
int accounting_flush_time = mconf_get_accounting_flush_time();
if ((local = sge_get_configuration_for_host(qualified_hostname)) == NULL) {
WARNING((SGE_EVENT, MSG_CONF_NOLOCAL_S, qualified_hostname));
}
if ((global = sge_get_configuration_for_host(SGE_GLOBAL_NAME)) == NULL) {
ERROR((SGE_EVENT, SFNMAX, MSG_CONF_NOGLOBAL));
}
if (merge_configuration(&answer_list, progid, cell_root, global, local, NULL) != 0) {
ERROR((SGE_EVENT, MSG_CONF_CANTMERGECONFIGURATIONFORHOST_S, qualified_hostname));
}
answer_list_output(&answer_list);
/* Restart the accounting flush event if needed. */
if ((accounting_flush_time == 0) &&
(mconf_get_accounting_flush_time() != 0)) {
te_event_t ev = te_new_event(time(NULL), TYPE_ACCOUNTING_TRIGGER, ONE_TIME_EVENT, 1, 0, NULL);
te_add_event(ev);
te_free_event(&ev);
}
lFreeElem(&local);
lFreeElem(&global);
sge_show_conf();
/* 'max_unheard' may have changed */
cl_commlib_set_connection_param(cl_com_get_handle("qmaster", 1), HEARD_FROM_TIMEOUT, mconf_get_max_unheard());
/* fetching qmaster_params and begin to parse */
qmaster_params = mconf_get_qmaster_params();
/* updating the commlib paramterlist and gdi_timeout with new or changed parameters */
cl_com_update_parameter_list(qmaster_params);
sge_free(&qmaster_params);
}
/* invalidate configuration cache */
mconf_set_new_config(true);
DRETURN(STATUS_OK);
}