Derived from patch provided by Artem, cleanup the "abnormal" code path for selecting TCP OOB modules to connect to a remote process. If we can't find a direct interface-to-address match, then assign all the provided addresses to the first available TCP module and let the normal failure process determine if the remote proc is truly reachable.

cmr=v1.8.2:reviewer=artpol:subject=fix abnormal code connection path in tcp oob

This commit was SVN r31536.

orte/mca/oob/tcp/oob_tcp_component.c

@@ -805,7 +805,7 @@ static int component_set_addr(orte_process_name_t *peer,
     char **addrs, *hptr;
     char *tcpuri=NULL, *host, *ports;
     int i, j, k, rc;
-    mca_oob_tcp_module_t *mod;
+    mca_oob_tcp_module_t *mod, *firstmod;
     mca_oob_tcp_component_peer_t *pr;
     uint16_t af_family = AF_UNSPEC;
     uint64_t ui64;
@@ -950,10 +950,32 @@ static int component_set_addr(orte_process_name_t *peer,
         /* if we cycled thru all their addresses without finding a match in our
          * interfaces, then it remains possible that they have a routed system
          * that can still route messages to the destination. So give it a chance
-         * to succeed by assigning our first module to try, and let the normal
-         * failure progression ultimately determine if we can reach this peer
+         * to succeed by assigning each provided address to the first module in our list,
+         * and let the normal failure progression ultimately determine if we
+         * can reach this peer. A future enhancement could be to do a better
+         * job of matching provided addresses with available interfaces, perhaps
+         * looking at the number of matching octets and assigning the address
+         * to the interface with the most matches - but that's for someone else
+         * to address :-)
          */
         if (!assigned) {
+            firstmod = NULL;
+            for (k=0; k < mca_oob_tcp_component.modules.size; k++) {
+                if (NULL == (mod = (mca_oob_tcp_module_t*)opal_pointer_array_get_item(&mca_oob_tcp_component.modules, k))) {
+                    continue;
+                }
+                if (NULL == firstmod) {
+                    firstmod = mod;
+                    break;
+                }
+            }
+            if (NULL == firstmod) {
+                ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
+                return ORTE_ERR_NOT_FOUND;
+            }
+            for (j=0; NULL != addrs[j]; j++) {
+                /* if they gave us "localhost", then just take the first conn on our list */
+                if (0 == strcasecmp(addrs[j], "localhost")) {
 #if OPAL_ENABLE_IPV6
                     if (AF_INET6 == af_family) {
                         if (NULL == mca_oob_tcp_component.ipv6conns ||
@@ -971,19 +993,8 @@ static int component_set_addr(orte_process_name_t *peer,
 #if OPAL_ENABLE_IPV6
                     }
 #endif
-            /* lookup the kernel index of this address */
-            if (0 >= (k = opal_ifaddrtokindex(host))) {
-                opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
-                                    "%s UNFOUND KERNEL INDEX %d FOR ADDRESS %s",
-                                    ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), k, host);
-                /* we don't have an interface on this subnet - ignore it */
-                continue;
-            }
-            if (NULL == (mod = (mca_oob_tcp_module_t*)opal_pointer_array_get_item(&mca_oob_tcp_component.modules, k))) {
-                opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
-                                    "%s NO MODULE AT KINDEX %d FOR ADDRESS %s",
-                                    ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), k, host);
-                continue;
+                } else {
+                    host = addrs[j];
                 }
                 /* record that this peer may be reachable via this module, but don't assign
                  * the peer to this module until later when we actually connect
@@ -991,7 +1002,7 @@ static int component_set_addr(orte_process_name_t *peer,
                 opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
                                     "%s PEER %s MAY BE REACHABLE BY ROUTING - ASSIGNING MODULE AT KINDEX %d INTERFACE %s",
                                     ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
-                                ORTE_NAME_PRINT(peer), k, mod->if_name);
+                                    ORTE_NAME_PRINT(peer), k, firstmod->if_name);
                 if (OPAL_SUCCESS != opal_hash_table_get_value_uint64(&mca_oob_tcp_component.peers,
                                                                      ui64, (void**)&pr) || NULL == pr) {
                     pr = OBJ_NEW(mca_oob_tcp_component_peer_t);
@@ -1012,11 +1023,12 @@ static int component_set_addr(orte_process_name_t *peer,
                 opal_output_verbose(OOB_TCP_DEBUG_CONNECT, orte_oob_base_framework.framework_output,
                                     "%s PASSING ADDR %s TO INTERFACE %s AT KERNEL INDEX %d",
                                     ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), host,
-                                mod->if_name, k);
-            mod->api.set_peer((struct mca_oob_tcp_module_t*)mod,
+                                    firstmod->if_name, k);
+                mod->api.set_peer((struct mca_oob_tcp_module_t*)firstmod,
                                   peer, af_family, host, ports);
                 found = true;
             }
+        }
         if (NULL != addrs) {
             opal_argv_free(addrs);
         }