static int route_lost(const orte_process_name_t *route)
{
opal_list_item_t *item;
orte_routed_tree_t *child;
orte_routed_jobfam_t *jfam;
uint16_t jfamily;
int i;
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_framework.framework_output,
"%s route to %s lost",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(route)));
/* if the route is to a different job family and we are the HNP, look it up */
if ((ORTE_JOB_FAMILY(route->jobid) != ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid)) &&
ORTE_PROC_IS_HNP) {
jfamily = ORTE_JOB_FAMILY(route->jobid);
for (i=0; i < orte_routed_jobfams.size; i++) {
if (NULL == (jfam = (orte_routed_jobfam_t*)opal_pointer_array_get_item(&orte_routed_jobfams, i))) {
continue;
}
if (jfam->job_family == jfamily) {
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_framework.framework_output,
"%s routed_direct: route to %s lost",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOB_FAMILY_PRINT(route->jobid)));
opal_pointer_array_set_item(&orte_routed_jobfams, i, NULL);
OBJ_RELEASE(jfam);
break;
}
}
}
/* if we lose the connection to the lifeline and we are NOT already,
* in finalize, tell the OOB to abort.
* NOTE: we cannot call abort from here as the OOB needs to first
* release a thread-lock - otherwise, we will hang!!
*/
if (!orte_finalizing &&
NULL != lifeline &&
OPAL_EQUAL == orte_util_compare_name_fields(ORTE_NS_CMP_ALL, route, lifeline)) {
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_framework.framework_output,
"%s routed:direct: Connection to lifeline %s lost",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(lifeline)));
return ORTE_ERR_FATAL;
}
/* if we are the HNP, and the route is a daemon,
* see if it is one of our children - if so, remove it
*/
if (ORTE_PROC_IS_HNP &&
route->jobid == ORTE_PROC_MY_NAME->jobid) {
for (item = opal_list_get_first(&my_children);
item != opal_list_get_end(&my_children);
item = opal_list_get_next(item)) {
child = (orte_routed_tree_t*)item;
if (child->vpid == route->vpid) {
opal_list_remove_item(&my_children, item);
OBJ_RELEASE(item);
return ORTE_SUCCESS;
}
}
}
/* we don't care about this one, so return success */
return ORTE_SUCCESS;
}
static orte_process_name_t get_route(orte_process_name_t *target)
{
orte_process_name_t *ret, daemon;
orte_routed_jobfam_t *jfam;
int i;
uint16_t jfamily;
if (target->jobid == ORTE_JOBID_INVALID ||
target->vpid == ORTE_VPID_INVALID) {
ret = ORTE_NAME_INVALID;
goto found;
}
/* initialize */
daemon.jobid = ORTE_PROC_MY_DAEMON->jobid;
daemon.vpid = ORTE_PROC_MY_DAEMON->vpid;
if (ORTE_PROC_IS_APP) {
/* if I am an application, AND I have knowledge of
* my daemon (i.e., a daemon launched me), then I
* always route thru the daemon */
if (NULL != orte_process_info.my_daemon_uri) {
ret = ORTE_PROC_MY_DAEMON;
} else {
/* I was direct launched and do not have
* a daemon, so I have to route direct */
ret = target;
}
goto found;
}
/* if I am a tool, the route is direct if target is in
* my own job family, and to the target's HNP if not
*/
if (ORTE_PROC_IS_TOOL) {
if (ORTE_JOB_FAMILY(target->jobid) == ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid)) {
ret = target;
goto found;
} else {
ORTE_HNP_NAME_FROM_JOB(&daemon, target->jobid);
ret = &daemon;
goto found;
}
}
/****** HNP AND DAEMONS ONLY ******/
/* IF THIS IS FOR A DIFFERENT JOB FAMILY... */
if (ORTE_JOB_FAMILY(target->jobid) != ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid)) {
/* if I am a daemon, route this via the HNP */
if (ORTE_PROC_IS_DAEMON) {
ret = ORTE_PROC_MY_HNP;
goto found;
}
/* if I am the HNP, then I stored a route to
* this job family, so look it up
*/
jfamily = ORTE_JOB_FAMILY(target->jobid);
for (i=0; i < orte_routed_jobfams.size; i++) {
if (NULL == (jfam = (orte_routed_jobfam_t*)opal_pointer_array_get_item(&orte_routed_jobfams, i))) {
continue;
}
if (jfam->job_family == jfamily) {
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_framework.framework_output,
"%s routed_direct: route to %s found",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOB_FAMILY_PRINT(target->jobid)));
ret = &jfam->route;
goto found;
}
}
/* not found - so we have no route */
ret = ORTE_NAME_INVALID;
goto found;
}
/* THIS CAME FROM OUR OWN JOB FAMILY... */
if (OPAL_EQUAL == orte_util_compare_name_fields(ORTE_NS_CMP_ALL, ORTE_PROC_MY_HNP, target)) {
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_framework.framework_output,
"%s routing direct to the HNP",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
ret = ORTE_PROC_MY_HNP;
goto found;
}
daemon.jobid = ORTE_PROC_MY_NAME->jobid;
/* find out what daemon hosts this proc */
if (ORTE_VPID_INVALID == (daemon.vpid = orte_get_proc_daemon_vpid(target))) {
ret = ORTE_NAME_INVALID;
goto found;
}
/* if the daemon is me, then send direct to the target! */
if (ORTE_PROC_MY_NAME->vpid == daemon.vpid) {
ret = target;
goto found;
}
/* else route to this daemon directly */
ret = &daemon;
found:
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_framework.framework_output,
"%s routed_direct_get(%s) --> %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(target),
ORTE_NAME_PRINT(ret)));
return *ret;
}
static int update_route(orte_process_name_t *target,
orte_process_name_t *route)
{
int i;
orte_routed_jobfam_t *jfam;
uint16_t jfamily;
if (target->jobid == ORTE_JOBID_INVALID ||
target->vpid == ORTE_VPID_INVALID) {
return ORTE_ERR_BAD_PARAM;
}
/* if I am an application process, we don't update the route since
* we automatically route everything through the local daemon
*/
if (ORTE_PROC_IS_APP) {
return ORTE_SUCCESS;
}
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
"%s routed_radix_update: %s --> %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(target),
ORTE_NAME_PRINT(route)));
/* if I am a daemon and the target is my HNP, then check
* the route - if it isn't direct, then we just flag that
* we have a route to the HNP
*/
if (OPAL_EQUAL == orte_util_compare_name_fields(ORTE_NS_CMP_ALL, ORTE_PROC_MY_HNP, target) &&
OPAL_EQUAL != orte_util_compare_name_fields(ORTE_NS_CMP_ALL, ORTE_PROC_MY_HNP, route)) {
hnp_direct = false;
return ORTE_SUCCESS;
}
/* if this is from a different job family, then I need to
* track how to send messages to it
*/
if (ORTE_JOB_FAMILY(target->jobid) != ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid)) {
/* if I am a daemon, then I will automatically route
* anything to this job family via my HNP - so nothing to do
* here, just return
*/
if (ORTE_PROC_IS_DAEMON) {
return ORTE_SUCCESS;
}
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
"%s routed_radix_update: diff job family routing job %s --> %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOBID_PRINT(target->jobid),
ORTE_NAME_PRINT(route)));
/* see if this target is already present */
jfamily = ORTE_JOB_FAMILY(target->jobid);
for (i=0; i < orte_routed_jobfams.size; i++) {
if (NULL == (jfam = (orte_routed_jobfam_t*)opal_pointer_array_get_item(&orte_routed_jobfams, i))) {
continue;
}
if (jfam->job_family == jfamily) {
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
"%s routed_radix: updating route to %s via %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOB_FAMILY_PRINT(target->jobid),
ORTE_NAME_PRINT(route)));
jfam->route.jobid = route->jobid;
jfam->route.vpid = route->vpid;
return ORTE_SUCCESS;
}
}
/* not there, so add the route FOR THE JOB FAMILY*/
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
"%s routed_radix: adding route to %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOB_FAMILY_PRINT(target->jobid)));
jfam = OBJ_NEW(orte_routed_jobfam_t);
jfam->job_family = jfamily;
jfam->route.jobid = route->jobid;
jfam->route.vpid = route->vpid;
opal_pointer_array_add(&orte_routed_jobfams, jfam);
return ORTE_SUCCESS;
}
/* THIS CAME FROM OUR OWN JOB FAMILY... */
opal_output(0, "%s CALL TO UPDATE ROUTE FOR OWN JOB FAMILY", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
return ORTE_ERR_NOT_SUPPORTED;
}
static orte_process_name_t get_route(orte_process_name_t *target)
{
orte_process_name_t *ret, daemon;
opal_list_item_t *item;
orte_routed_tree_t *child;
int i;
orte_routed_jobfam_t *jfam;
uint16_t jfamily;
if (!orte_routing_is_enabled) {
ret = target;
goto found;
}
/* initialize */
daemon.jobid = ORTE_PROC_MY_DAEMON->jobid;
daemon.vpid = ORTE_PROC_MY_DAEMON->vpid;
if (target->jobid == ORTE_JOBID_INVALID ||
target->vpid == ORTE_VPID_INVALID) {
ret = ORTE_NAME_INVALID;
goto found;
}
/* if it is me, then the route is just direct */
if (OPAL_EQUAL == opal_dss.compare(ORTE_PROC_MY_NAME, target, ORTE_NAME)) {
ret = target;
goto found;
}
/* if I am an application process, always route via my local daemon */
if (ORTE_PROC_IS_APP) {
ret = ORTE_PROC_MY_DAEMON;
goto found;
}
/* if I am a tool, the route is direct if target is in
* my own job family, and to the target's HNP if not
*/
if (ORTE_PROC_IS_TOOL) {
if (ORTE_JOB_FAMILY(target->jobid) == ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid)) {
ret = target;
goto found;
} else {
ORTE_HNP_NAME_FROM_JOB(&daemon, target->jobid);
ret = &daemon;
goto found;
}
}
/****** HNP AND DAEMONS ONLY ******/
/* IF THIS IS FOR A DIFFERENT JOB FAMILY... */
if (ORTE_JOB_FAMILY(target->jobid) != ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid)) {
/* if I am a daemon, route this via the HNP */
if (ORTE_PROC_IS_DAEMON) {
ret = ORTE_PROC_MY_HNP;
goto found;
}
/* if I am the HNP or a tool, then I stored a route to
* this job family, so look it up
*/
jfamily = ORTE_JOB_FAMILY(target->jobid);
for (i=0; i < orte_routed_jobfams.size; i++) {
if (NULL == (jfam = (orte_routed_jobfam_t*)opal_pointer_array_get_item(&orte_routed_jobfams, i))) {
continue;
}
if (jfam->job_family == jfamily) {
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
"%s routed_binomial: route to %s found",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOB_FAMILY_PRINT(target->jobid)));
ret = &jfam->route;
goto found;
}
}
/* not found - so we have no route */
ret = ORTE_NAME_INVALID;
goto found;
}
/* THIS CAME FROM OUR OWN JOB FAMILY... */
/* if this is going to the HNP, then send it direct if we don't know
* how to get there - otherwise, send it via the tree
*/
if (OPAL_EQUAL == orte_util_compare_name_fields(ORTE_NS_CMP_ALL, ORTE_PROC_MY_HNP, target)) {
if (!hnp_direct || orte_static_ports || orte_use_common_port) {
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
"%s routing to the HNP through my parent %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(ORTE_PROC_MY_PARENT)));
ret = ORTE_PROC_MY_PARENT;
goto found;
} else {
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
"%s routing direct to the HNP",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
ret = ORTE_PROC_MY_HNP;
goto found;
}
}
daemon.jobid = ORTE_PROC_MY_NAME->jobid;
/* find out what daemon hosts this proc */
if (ORTE_VPID_INVALID == (daemon.vpid = orte_get_proc_daemon_vpid(target))) {
ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
ret = ORTE_NAME_INVALID;
goto found;
}
/* if the daemon is me, then send direct to the target! */
if (ORTE_PROC_MY_NAME->vpid == daemon.vpid) {
ret = target;
goto found;
} else if (orte_process_info.num_procs < mca_routed_radix_component.max_connections) {
/* if the job is small enough, send direct to the target's daemon */
ret = &daemon;
goto found;
} else {
/* search routing tree for next step to that daemon */
for (item = opal_list_get_first(&my_children);
item != opal_list_get_end(&my_children);
item = opal_list_get_next(item)) {
child = (orte_routed_tree_t*)item;
if (child->vpid == daemon.vpid) {
/* the child is hosting the proc - just send it there */
ret = &daemon;
goto found;
}
/* otherwise, see if the daemon we need is below the child */
if (opal_bitmap_is_set_bit(&child->relatives, daemon.vpid)) {
/* yep - we need to step through this child */
daemon.vpid = child->vpid;
ret = &daemon;
goto found;
}
}
}
/* if we get here, then the target daemon is not beneath
* any of our children, so we have to step up through our parent
*/
daemon.vpid = ORTE_PROC_MY_PARENT->vpid;
ret = &daemon;
found:
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
"%s routed_radix_get(%s) --> %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(target),
ORTE_NAME_PRINT(ret)));
return *ret;
}
static int delete_route(orte_process_name_t *proc)
{
int i;
orte_routed_jobfam_t *jfam;
uint16_t jfamily;
if (proc->jobid == ORTE_JOBID_INVALID ||
proc->vpid == ORTE_VPID_INVALID) {
return ORTE_ERR_BAD_PARAM;
}
/* if I am an application process, I don't have any routes
* so there is nothing for me to do
*/
if (!ORTE_PROC_IS_HNP && !ORTE_PROC_IS_DAEMON &&
!ORTE_PROC_IS_TOOL) {
return ORTE_SUCCESS;
}
OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
"%s routed_radix_delete_route for %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(proc)));
/* if this is from a different job family, then I need to
* look it up appropriately
*/
if (ORTE_JOB_FAMILY(proc->jobid) != ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid)) {
/* if I am a daemon, then I will automatically route
* anything to this job family via my HNP - so I have nothing
* in my routing table and thus have nothing to do
* here, just return
*/
if (ORTE_PROC_IS_DAEMON) {
return ORTE_SUCCESS;
}
/* see if this job family is present */
jfamily = ORTE_JOB_FAMILY(proc->jobid);
for (i=0; i < orte_routed_jobfams.size; i++) {
if (NULL == (jfam = (orte_routed_jobfam_t*)opal_pointer_array_get_item(&orte_routed_jobfams, i))) {
continue;
}
if (jfam->job_family == jfamily) {
OPAL_OUTPUT_VERBOSE((2, orte_routed_base_output,
"%s routed_binomial: deleting route to %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_JOB_FAMILY_PRINT(proc->jobid)));
opal_pointer_array_set_item(&orte_routed_jobfams, i, NULL);
OBJ_RELEASE(jfam);
return ORTE_SUCCESS;
}
}
/* not present - nothing to do */
return ORTE_SUCCESS;
}
/* THIS CAME FROM OUR OWN JOB FAMILY...there is nothing
* to do here. The routes will be redefined when we update
* the routing tree
*/
return ORTE_SUCCESS;
}
