fixes for direct routed component under mpirun
Этот коммит содержится в:
родитель
8cae899a42
Коммит
b17ea23ce0
@ -391,6 +391,8 @@ static void xcast_recv(int status, orte_process_name_t* sender,
|
|||||||
goto relay;
|
goto relay;
|
||||||
}
|
}
|
||||||
if (0 == flag) {
|
if (0 == flag) {
|
||||||
|
/* copy the remainder of the payload */
|
||||||
|
opal_dss.copy_payload(relay, buffer);
|
||||||
/* no - just return */
|
/* no - just return */
|
||||||
goto relay;
|
goto relay;
|
||||||
}
|
}
|
||||||
|
@ -117,8 +117,12 @@ static orte_process_name_t get_route(orte_process_name_t *target)
|
|||||||
goto found;
|
goto found;
|
||||||
}
|
}
|
||||||
|
|
||||||
/* all routes go direct */
|
if (ORTE_PROC_IS_APP) {
|
||||||
ret = target;
|
ret = ORTE_PROC_MY_HNP;
|
||||||
|
} else {
|
||||||
|
/* all routes go direct */
|
||||||
|
ret = target;
|
||||||
|
}
|
||||||
|
|
||||||
found:
|
found:
|
||||||
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_framework.framework_output,
|
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_framework.framework_output,
|
||||||
@ -217,25 +221,16 @@ static int init_routes(orte_jobid_t job, opal_buffer_t *ndat)
|
|||||||
|
|
||||||
/*** MUST BE A PROC ***/
|
/*** MUST BE A PROC ***/
|
||||||
|
|
||||||
/* if ndat=NULL, then we are being called during orte_init */
|
if (NULL != orte_process_info.my_hnp_uri) {
|
||||||
if (NULL == ndat) {
|
/* set the contact info into the hash table */
|
||||||
if (NULL != orte_process_info.my_daemon_uri) {
|
orte_rml.set_contact_info(orte_process_info.my_hnp_uri);
|
||||||
/* we are being launched by a daemon, so we need to
|
|
||||||
* register a sync with it to get our nidmap back
|
/* extract the hnp name and store it */
|
||||||
*/
|
if (ORTE_SUCCESS != (rc = orte_rml_base_parse_uris(orte_process_info.my_hnp_uri,
|
||||||
/* Set the contact info in the RML - this won't actually establish
|
ORTE_PROC_MY_HNP, NULL))) {
|
||||||
* the connection, but just tells the RML how to reach the daemon
|
ORTE_ERROR_LOG(rc);
|
||||||
* if/when we attempt to send to it
|
return rc;
|
||||||
*/
|
|
||||||
orte_rml.set_contact_info(orte_process_info.my_daemon_uri);
|
|
||||||
/* extract the daemon's name so we can update the routing table */
|
|
||||||
if (ORTE_SUCCESS != (rc = orte_rml_base_parse_uris(orte_process_info.my_daemon_uri,
|
|
||||||
ORTE_PROC_MY_DAEMON, NULL))) {
|
|
||||||
ORTE_ERROR_LOG(rc);
|
|
||||||
return rc;
|
|
||||||
}
|
|
||||||
}
|
}
|
||||||
return ORTE_SUCCESS;
|
|
||||||
}
|
}
|
||||||
|
|
||||||
/* if ndat != NULL, then this is being invoked by the proc to
|
/* if ndat != NULL, then this is being invoked by the proc to
|
||||||
@ -243,13 +238,15 @@ static int init_routes(orte_jobid_t job, opal_buffer_t *ndat)
|
|||||||
* job family. It really doesn't matter as everything must
|
* job family. It really doesn't matter as everything must
|
||||||
* go direct
|
* go direct
|
||||||
*/
|
*/
|
||||||
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_framework.framework_output,
|
if (NULL != ndat) {
|
||||||
"%s routed_direct: init routes w/non-NULL data",
|
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_framework.framework_output,
|
||||||
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
|
"%s routed_direct: init routes w/non-NULL data",
|
||||||
|
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
|
||||||
|
|
||||||
if (ORTE_SUCCESS != (rc = orte_rml_base_update_contact_info(ndat))) {
|
if (ORTE_SUCCESS != (rc = orte_rml_base_update_contact_info(ndat))) {
|
||||||
ORTE_ERROR_LOG(rc);
|
ORTE_ERROR_LOG(rc);
|
||||||
return rc;
|
return rc;
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
return ORTE_SUCCESS;
|
return ORTE_SUCCESS;
|
||||||
@ -290,7 +287,7 @@ static void get_routing_list(opal_list_t *coll)
|
|||||||
/* if I am anything other than daemons and the HNP, this
|
/* if I am anything other than daemons and the HNP, this
|
||||||
* is a meaningless command as I am not allowed to route
|
* is a meaningless command as I am not allowed to route
|
||||||
*/
|
*/
|
||||||
if (!ORTE_PROC_IS_DAEMON || !ORTE_PROC_IS_HNP) {
|
if (!ORTE_PROC_IS_DAEMON && !ORTE_PROC_IS_HNP) {
|
||||||
return;
|
return;
|
||||||
}
|
}
|
||||||
|
|
||||||
@ -331,14 +328,14 @@ static void get_routing_list(opal_list_t *coll)
|
|||||||
|
|
||||||
static int get_wireup_info(opal_buffer_t *buf)
|
static int get_wireup_info(opal_buffer_t *buf)
|
||||||
{
|
{
|
||||||
opal_byte_object_t bo, *boptr;
|
int rc;
|
||||||
|
|
||||||
/* this is a meaningless command for a direct as I am not allowed to route */
|
if (ORTE_PROC_IS_HNP) {
|
||||||
bo.bytes = NULL;
|
if (ORTE_SUCCESS != (rc = orte_rml_base_get_contact_info(ORTE_PROC_MY_NAME->jobid, buf))) {
|
||||||
bo.size = 0;
|
ORTE_ERROR_LOG(rc);
|
||||||
boptr = &bo;
|
}
|
||||||
|
return rc;
|
||||||
opal_dss.pack(buf, &boptr, 1, OPAL_BYTE_OBJECT);
|
}
|
||||||
return ORTE_SUCCESS;
|
return ORTE_SUCCESS;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
Загрузка…
Ссылка в новой задаче
Block a user