@@ -832,121 +832,162 @@ static void _toolconn(int sd, short args, void *cbdata)
832832 orte_job_t * jdata ;
833833 orte_app_context_t * app ;
834834 orte_proc_t * proc ;
835- orte_node_t * node ;
836- orte_process_name_t tool ;
837- int rc ;
835+ orte_node_t * node , * nptr ;
836+ char * hostname = NULL ;
837+ orte_process_name_t tool = {ORTE_JOBID_INVALID , ORTE_VPID_INVALID };
838+ int rc , i ;
838839 opal_value_t * val ;
839- bool flag ;
840+ bool flag = false, flag_given = false; ;
840841
841842 ORTE_ACQUIRE_OBJECT (cd );
842843
843844 opal_output_verbose (2 , orte_pmix_server_globals .output ,
844845 "%s TOOL CONNECTION PROCESSING" ,
845846 ORTE_NAME_PRINT (ORTE_PROC_MY_NAME ));
846847
847- /* if we are the HNP, we can directly assign the jobid */
848- if (ORTE_PROC_IS_HNP || ORTE_PROC_IS_MASTER ) {
849- jdata = OBJ_NEW (orte_job_t );
850- rc = orte_plm_base_create_jobid (jdata );
851- if (ORTE_SUCCESS != rc ) {
852- tool .jobid = ORTE_JOBID_INVALID ;
848+ /* check for directives */
849+ if (NULL != cd -> info ) {
850+ OPAL_LIST_FOREACH (val , cd -> info , opal_value_t ) {
851+ if (0 == strcmp (val -> key , OPAL_PMIX_EVENT_SILENT_TERMINATION )) {
852+ if (OPAL_UNDEF == val -> type || val -> data .flag ) {
853+ flag = true;
854+ flag_given = true;
855+ }
856+ } else if (0 == strcmp (val -> key , OPAL_PMIX_NSPACE )) {
857+ tool .jobid = val -> data .name .jobid ;
858+ } else if (0 == strcmp (val -> key , OPAL_PMIX_RANK )) {
859+ tool .vpid = val -> data .name .vpid ;
860+ } else if (0 == strcmp (val -> key , OPAL_PMIX_HOSTNAME )) {
861+ hostname = strdup (val -> data .string );
862+ }
863+ }
864+ }
865+
866+ /* if we are not the HNP or master, and the tool doesn't
867+ * already have a name (i.e., we didn't spawn it), then
868+ * there is nothing we can currently do.
869+ * Eventually, when we switch to nspace instead of an
870+ * integer jobid, we'll just locally assign this value */
871+ if (ORTE_JOBID_INVALID == tool .jobid ||
872+ ORTE_VPID_INVALID == tool .vpid ) {
873+ /* if we are the HNP, we can directly assign the jobid */
874+ if (ORTE_PROC_IS_HNP || ORTE_PROC_IS_MASTER ) {
875+ jdata = OBJ_NEW (orte_job_t );
876+ rc = orte_plm_base_create_jobid (jdata );
877+ if (ORTE_SUCCESS != rc ) {
878+ OBJ_RELEASE (jdata );
879+ if (NULL != cd -> toolcbfunc ) {
880+ cd -> toolcbfunc (ORTE_ERROR , tool , cd -> cbdata );
881+ }
882+ OBJ_RELEASE (cd );
883+ return ;
884+ }
885+ tool .jobid = jdata -> jobid ;
853886 tool .vpid = 0 ;
887+ } else {
888+ /* we currently do not support connections to non-HNP/master
889+ * daemons from tools that were not spawned by a daemon */
854890 if (NULL != cd -> toolcbfunc ) {
855- cd -> toolcbfunc (rc , tool , cd -> cbdata );
891+ cd -> toolcbfunc (ORTE_ERR_NOT_SUPPORTED , tool , cd -> cbdata );
856892 }
857893 OBJ_RELEASE (cd );
858894 return ;
859895 }
860- opal_hash_table_set_value_uint32 (orte_job_data , jdata -> jobid , jdata );
861- /* setup some required job-level fields in case this
862- * tool calls spawn, or uses some other functions that
863- * need them */
864- /* must create a map for it (even though it has no
865- * info in it) so that the job info will be picked
866- * up in subsequent pidmaps or other daemons won't
867- * know how to route
868- */
869- jdata -> map = OBJ_NEW (orte_job_map_t );
870-
871- /* setup an app_context for the singleton */
872- app = OBJ_NEW (orte_app_context_t );
873- app -> app = strdup ("tool" );
874- app -> num_procs = 1 ;
875- opal_pointer_array_add (jdata -> apps , app );
876- jdata -> num_apps = 1 ;
877-
878- /* setup a proc object for the singleton - since we
879- * -must- be the HNP, and therefore we stored our
880- * node on the global node pool, and since the singleton
881- * -must- be on the same node as us, indicate that
882- */
883- proc = OBJ_NEW (orte_proc_t );
884- proc -> name .jobid = jdata -> jobid ;
885- proc -> name .vpid = 0 ;
886- proc -> parent = ORTE_PROC_MY_NAME -> vpid ;
887- ORTE_FLAG_SET (proc , ORTE_PROC_FLAG_ALIVE );
888- proc -> state = ORTE_PROC_STATE_RUNNING ;
889- proc -> app_idx = 0 ;
890- /* obviously, it is on my node */
896+ } else {
897+ jdata = OBJ_NEW (orte_job_t );
898+ jdata -> jobid = tool .jobid ;
899+ }
900+
901+ opal_hash_table_set_value_uint32 (orte_job_data , jdata -> jobid , jdata );
902+ /* setup some required job-level fields in case this
903+ * tool calls spawn, or uses some other functions that
904+ * need them */
905+ /* must create a map for it (even though it has no
906+ * info in it) so that the job info will be picked
907+ * up in subsequent pidmaps or other daemons won't
908+ * know how to route
909+ */
910+ jdata -> map = OBJ_NEW (orte_job_map_t );
911+
912+ /* setup an app_context for the singleton */
913+ app = OBJ_NEW (orte_app_context_t );
914+ app -> app = strdup ("tool" );
915+ app -> num_procs = 1 ;
916+ opal_pointer_array_add (jdata -> apps , app );
917+ jdata -> num_apps = 1 ;
918+
919+ /* setup a proc object for the singleton - since we
920+ * -must- be the HNP, and therefore we stored our
921+ * node on the global node pool, and since the singleton
922+ * -must- be on the same node as us, indicate that
923+ */
924+ proc = OBJ_NEW (orte_proc_t );
925+ proc -> name .jobid = jdata -> jobid ;
926+ proc -> name .vpid = tool .vpid ;
927+ proc -> parent = ORTE_PROC_MY_NAME -> vpid ;
928+ ORTE_FLAG_SET (proc , ORTE_PROC_FLAG_ALIVE );
929+ ORTE_FLAG_SET (proc , ORTE_PROC_FLAG_TOOL );
930+ proc -> state = ORTE_PROC_STATE_RUNNING ;
931+ /* set the trivial */
932+ proc -> local_rank = 0 ;
933+ proc -> node_rank = 0 ;
934+ proc -> app_rank = 0 ;
935+ proc -> app_idx = 0 ;
936+ if (NULL == hostname ) {
937+ /* it is on my node */
891938 node = (orte_node_t * )opal_pointer_array_get_item (orte_node_pool , 0 );
892- proc -> node = node ;
893- OBJ_RETAIN (node ); /* keep accounting straight */
894- opal_pointer_array_add (jdata -> procs , proc );
895- jdata -> num_procs = 1 ;
896- /* add the node to the job map */
897- OBJ_RETAIN (node );
898- opal_pointer_array_add (jdata -> map -> nodes , node );
899- jdata -> map -> num_nodes ++ ;
900- /* and it obviously is on the node - note that
901- * we do _not_ increment the #procs on the node
902- * as the tool doesn't count against the slot
903- * allocation */
904- OBJ_RETAIN (proc );
905- opal_pointer_array_add (node -> procs , proc );
906- /* set the trivial */
907- proc -> local_rank = 0 ;
908- proc -> node_rank = 0 ;
909- proc -> app_rank = 0 ;
910- proc -> state = ORTE_PROC_STATE_RUNNING ;
911- proc -> app_idx = 0 ;
912939 ORTE_FLAG_SET (proc , ORTE_PROC_FLAG_LOCAL );
913-
914- /* check for directives */
915- if (NULL != cd -> info ) {
916- OPAL_LIST_FOREACH (val , cd -> info , opal_value_t ) {
917- if (0 == strcmp (val -> key , OPAL_PMIX_EVENT_SILENT_TERMINATION )) {
918- if (OPAL_UNDEF == val -> type || val -> data .flag ) {
919- flag = true;
920- orte_set_attribute (& jdata -> attributes , ORTE_JOB_SILENT_TERMINATION ,
921- ORTE_ATTR_GLOBAL , & flag , OPAL_BOOL );
922- }
923- }
940+ } else {
941+ /* we need to locate it */
942+ node = NULL ;
943+ for (i = 0 ; i < orte_node_pool -> size ; i ++ ) {
944+ if (NULL == (nptr = (orte_node_t * )opal_pointer_array_get_item (orte_node_pool , i ))) {
945+ continue ;
924946 }
947+ if (0 == strcmp (hostname , nptr -> name )) {
948+ node = nptr ;
949+ break ;
950+ }
951+ }
952+ if (NULL == node ) {
953+ /* not in our allocation - which is still okay */
954+ node = OBJ_NEW (orte_node_t );
955+ node -> name = strdup (hostname );
956+ ORTE_FLAG_SET (node , ORTE_NODE_NON_USABLE );
957+ opal_pointer_array_add (orte_node_pool , node );
925958 }
959+ }
960+ proc -> node = node ;
961+ OBJ_RETAIN (node ); /* keep accounting straight */
962+ opal_pointer_array_add (jdata -> procs , proc );
963+ jdata -> num_procs = 1 ;
964+ /* add the node to the job map */
965+ OBJ_RETAIN (node );
966+ opal_pointer_array_add (jdata -> map -> nodes , node );
967+ jdata -> map -> num_nodes ++ ;
968+ /* and it obviously is on the node - note that
969+ * we do _not_ increment the #procs on the node
970+ * as the tool doesn't count against the slot
971+ * allocation */
972+ OBJ_RETAIN (proc );
973+ opal_pointer_array_add (node -> procs , proc );
974+ /* if they indicated a preference for termination, set it */
975+ if (flag_given ) {
976+ orte_set_attribute (& jdata -> attributes , ORTE_JOB_SILENT_TERMINATION ,
977+ ORTE_ATTR_GLOBAL , & flag , OPAL_BOOL );
978+ } else {
979+ /* we default to silence */
926980 flag = true;
927981 orte_set_attribute (& jdata -> attributes , ORTE_JOB_SILENT_TERMINATION ,
928982 ORTE_ATTR_GLOBAL , & flag , OPAL_BOOL );
929-
930- /* pass back the assigned jobid */
931- tool .jobid = jdata -> jobid ;
932- tool .vpid = 0 ;
933- if (NULL != cd -> toolcbfunc ) {
934- cd -> toolcbfunc (rc , tool , cd -> cbdata );
935- }
936- OBJ_RELEASE (cd );
937- return ;
938983 }
939984
940- /* otherwise, we have to send the request to the HNP.
941- * Eventually, when we switch to nspace instead of an
942- * integer jobid, we'll just locally assign this value */
943- tool .jobid = ORTE_JOBID_INVALID ;
944- tool .vpid = ORTE_VPID_INVALID ;
945985 if (NULL != cd -> toolcbfunc ) {
946986 cd -> toolcbfunc (ORTE_ERR_NOT_SUPPORTED , tool , cd -> cbdata );
947987 }
948988 OBJ_RELEASE (cd );
949989}
990+
950991void pmix_tool_connected_fn (opal_list_t * info ,
951992 opal_pmix_tool_connection_cbfunc_t cbfunc ,
952993 void * cbdata )
0 commit comments