Add an MCA param to retain all aliases based on IP addrs for node names so that procs can look them up by interface, if desired. If the param is set, pass aliases around to all daemons and procs for local use

This commit was SVN r27619.

orte/mca/dfs/app/dfs_app.c

@@ -504,10 +504,9 @@ static void process_opens(int fd, short args, void *cbdata)
     orte_dfs_request_t *dfs = (orte_dfs_request_t*)cbdata;
     int rc;
     opal_buffer_t *buffer;
-    char *scheme, *host, *filename, *hostname;
+    char *scheme, *host, *filename;
     orte_process_name_t daemon;
-    bool found;
+    orte_vpid_t *v;
-    orte_vpid_t v;
 
     /* get the scheme to determine if we can process locally or not */
     if (NULL == (scheme = opal_uri_get_scheme(dfs->uri))) {
@@ -538,8 +537,7 @@ static void process_opens(int fd, short args, void *cbdata)
     }
 
     /* if the host is our own, then treat it as a local file */
-    if (NULL == host ||
+    if (0 == strcmp(host, orte_process_info.nodename) ||
-        0 == strcmp(host, orte_process_info.nodename) ||
         0 == strcmp(host, "localhost") ||
         opal_ifislocal(host)) {
         opal_output_verbose(1, orte_dfs_base.output,
@@ -554,24 +552,16 @@ static void process_opens(int fd, short args, void *cbdata)
 
     /* ident the daemon on that host */
     daemon.jobid = ORTE_PROC_MY_DAEMON->jobid;
-    found = false;
+    /* fetch the daemon for this hostname */
-    for (v=0; v < orte_process_info.num_daemons; v++) {
+    opal_output_verbose(1, orte_dfs_base.output,
-        daemon.vpid = v;
+                        "%s looking for daemon on host %s",
-        /* fetch the hostname where this daemon is located */
+                        ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), host);
-        if (ORTE_SUCCESS != (rc = orte_db.fetch_pointer(&daemon, ORTE_DB_HOSTNAME, (void**)&hostname, OPAL_STRING))) {
+    v = &daemon.vpid;
+    if (ORTE_SUCCESS != (rc = orte_db.fetch(ORTE_NAME_WILDCARD, host, (void**)&v, ORTE_VPID))) {
         ORTE_ERROR_LOG(rc);
         goto complete;
     }
-        opal_output(0, "%s GOT HOST %s HOSTNAME %s", ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), host, hostname);
+
-        if (0 == strcmp(host, hostname)) {
-            found = true;
-            break;
-        }
-    }
-    if (!found) {
-        ORTE_ERROR_LOG(ORTE_ERR_NOT_FOUND);
-        goto complete;
-    }
     opal_output_verbose(1, orte_dfs_base.output,
                         "%s file %s on host %s daemon %s",
                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),

orte/mca/ess/hnp/ess_hnp_module.c

@@ -550,6 +550,11 @@ static int rte_init(void)
     node->daemon_launched = true;
     node->state = ORTE_NODE_STATE_UP;
     
+    /* if we are to retain aliases, get ours */
+    if (orte_retain_aliases) {
+        opal_ifgetaliases(&node->alias);
+    }
+
     /* record that the daemon job is running */
     jdata->num_procs = 1;
     jdata->state = ORTE_JOB_STATE_RUNNING;

orte/mca/plm/base/plm_base_launch_support.c

@@ -674,14 +674,6 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
             orted_failed_launch = true;
             goto CLEANUP;
         }
-        
-        OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
-                             "%s plm:base:orted_report_launch from daemon %s on node %s",
-                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
-                             ORTE_NAME_PRINT(&dname), nodename));
-        
-        /* look this node up, if necessary */
-        if (!orte_plm_globals.daemon_nodes_assigned_at_launch) {
         if (!orte_have_fqdn_allocation) {
             /* remove any domain info */
             if (NULL != (ptr = strchr(nodename, '.'))) {
@@ -691,6 +683,14 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
                 nodename = ptr;
             }
         }
+        
+        OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
+                             "%s plm:base:orted_report_launch from daemon %s on node %s",
+                             ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
+                             ORTE_NAME_PRINT(&dname), nodename));
+        
+        /* look this node up, if necessary */
+        if (!orte_plm_globals.daemon_nodes_assigned_at_launch) {
             OPAL_OUTPUT_VERBOSE((5, orte_plm_globals.output,
                                  "%s plm:base:orted_report_launch attempting to assign daemon %s to node %s",
                                  ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
@@ -733,14 +733,6 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
             }
         }
 
-#if OPAL_HAVE_HWLOC
-        /* store the local resources for that node */
-        if (1 == dname.vpid || orte_hetero_nodes) {
-            hwloc_topology_t topo, t;
-            int i;
-            bool found;
-            
-            idx=1;
         node = daemon->node;
         if (NULL == node) {
             /* this shouldn't happen - it indicates an error in the
@@ -751,6 +743,45 @@ void orte_plm_base_daemon_callback(int status, orte_process_name_t* sender,
             orted_failed_launch = true;
             goto CLEANUP;
         }
+        
+        if (orte_retain_aliases) {
+            char *alias;
+            uint8_t naliases, ni;
+            /* first, store the nodename itself as an alias. We do
+             * this in case the nodename isn't the same as what we
+             * were given by the allocation. For example, a hostfile
+             * might contain an IP address instead of the value returned
+             * by gethostname, yet the daemon will have returned the latter
+             * and apps may refer to the host by that name
+             */
+            opal_argv_append_nosize(&node->alias, nodename);
+            /* unpack and store the provided aliases */
+            idx = 1;
+            if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &naliases, &idx, OPAL_UINT8))) {
+                ORTE_ERROR_LOG(rc);
+                orted_failed_launch = true;
+                goto CLEANUP;
+            }
+            for (ni=0; ni < naliases; ni++) {
+                idx = 1;
+                if (ORTE_SUCCESS != (rc = opal_dss.unpack(buffer, &alias, &idx, OPAL_STRING))) {
+                    ORTE_ERROR_LOG(rc);
+                    orted_failed_launch = true;
+                    goto CLEANUP;
+                }
+                opal_argv_append_nosize(&node->alias, alias);
+                free(alias);
+            }
+        }
+
+#if OPAL_HAVE_HWLOC
+        /* store the local resources for that node */
+        if (1 == dname.vpid || orte_hetero_nodes) {
+            hwloc_topology_t topo, t;
+            int i;
+            bool found;
+            
+            idx=1;
             if (OPAL_SUCCESS != (rc = opal_dss.unpack(buffer, &topo, &idx, OPAL_HWLOC_TOPO))) {
                 ORTE_ERROR_LOG(rc);
                 orted_failed_launch = true;

orte/mca/routed/base/routed_base_fns.c

@@ -303,21 +303,7 @@ int orte_routed_base_register_sync(bool setup)
         return rc;
     }
 
-    /* send the sync command to our daemon */
+    /* setup to receive the response */
-    if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_DAEMON, buffer,
-                                          ORTE_RML_TAG_DAEMON, 0,
-                                          orte_rml_send_callback, NULL))) {
-        ORTE_ERROR_LOG(rc);
-        return rc;
-    }
-    
-    /* get the ack - need this to ensure that the sync communication
-     * gets serviced by the event library on the orted prior to the
-     * process exiting
-     */
-    OPAL_OUTPUT_VERBOSE((5, orte_routed_base_output,
-                         "%s registering sync waiting for ack",
-                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
     sync_waiting = true;
     rc = orte_rml.recv_buffer_nb(ORTE_NAME_WILDCARD, ORTE_RML_TAG_SYNC,
                                  ORTE_RML_NON_PERSISTENT, report_sync, NULL);
@@ -326,9 +312,24 @@ int orte_routed_base_register_sync(bool setup)
         return rc;
     }
 
-    /* it is okay to block here as we are -not- in an event */
+    /* send the sync command to our daemon */
-    ORTE_WAIT_FOR_COMPLETION(sync_waiting);
+    if (0 > (rc = orte_rml.send_buffer_nb(ORTE_PROC_MY_DAEMON, buffer,
+                                          ORTE_RML_TAG_DAEMON, 0,
+                                          orte_rml_send_callback, NULL))) {
+        ORTE_ERROR_LOG(rc);
+        return rc;
+    }
     
+    OPAL_OUTPUT_VERBOSE((5, orte_routed_base_output,
+                         "%s registering sync waiting for ack",
+                         ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));
+
+    
+    /* get the ack - need this to ensure that the sync communication
+     * gets serviced by the event library on the orted prior to the
+     * process exiting
+     */
+    ORTE_WAIT_FOR_COMPLETION(sync_waiting);
     OPAL_OUTPUT_VERBOSE((5, orte_routed_base_output,
                          "%s registering sync ack recvd",
                          ORTE_NAME_PRINT(ORTE_PROC_MY_NAME)));

orte/orted/orted_main.c

@@ -52,6 +52,7 @@
 #include "opal/mca/base/base.h"
 #include "opal/util/output.h"
 #include "opal/util/cmd_line.h"
+#include "opal/util/if.h"
 #include "opal/util/opal_environ.h"
 #include "opal/util/os_path.h"
 #include "opal/util/printf.h"
@@ -704,6 +705,27 @@ int orte_daemon(int argc, char *argv[])
         /* include our node name */
         opal_dss.pack(buffer, &orte_process_info.nodename, 1, OPAL_STRING);
         
+        /* if requested, include any non-loopback aliases for this node */
+        if (orte_retain_aliases) {
+            char **aliases=NULL;
+            uint8_t naliases, ni;
+            opal_ifgetaliases(&aliases);
+            naliases = opal_argv_count(aliases);
+            if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &naliases, 1, OPAL_UINT8))) {
+                ORTE_ERROR_LOG(ret);
+                OBJ_RELEASE(buffer);
+                goto DONE;
+            }
+            for (ni=0; ni < naliases; ni++) {
+                if (ORTE_SUCCESS != (ret = opal_dss.pack(buffer, &aliases[ni], 1, OPAL_STRING))) {
+                    ORTE_ERROR_LOG(ret);
+                    OBJ_RELEASE(buffer);
+                    goto DONE;
+                }
+            }
+            opal_argv_free(aliases);
+        }
+
 #if OPAL_HAVE_HWLOC
         /* add the local topology */
         if (NULL != opal_hwloc_topology &&

orte/runtime/orte_globals.c

@@ -80,6 +80,8 @@ bool orte_use_common_port = false;
 bool orte_keep_fqdn_hostnames = false;
 bool orte_have_fqdn_allocation = false;
 bool orte_show_resolved_nodenames;
+bool orte_retain_aliases;
+
 int orted_debug_failure;
 int orted_debug_failure_delay;
 bool orte_homogeneous_nodes = false;

orte/runtime/orte_globals.h

@@ -603,14 +603,21 @@ ORTE_DECLSPEC extern char *orte_oob_static_ports;
 ORTE_DECLSPEC extern bool orte_standalone_operation;
 ORTE_DECLSPEC extern bool orte_use_common_port;
 
+/* nodename flags */
 ORTE_DECLSPEC extern bool orte_keep_fqdn_hostnames;
 ORTE_DECLSPEC extern bool orte_have_fqdn_allocation;
 ORTE_DECLSPEC extern bool orte_show_resolved_nodenames;
+ORTE_DECLSPEC extern bool orte_retain_aliases;
+
+/* debug flags */
 ORTE_DECLSPEC extern int orted_debug_failure;
 ORTE_DECLSPEC extern int orted_debug_failure_delay;
+
+/* homegeneity flags */
 ORTE_DECLSPEC extern bool orte_homogeneous_nodes;
 ORTE_DECLSPEC extern bool orte_hetero_apps;
 ORTE_DECLSPEC extern bool orte_hetero_nodes;
+
 ORTE_DECLSPEC extern bool orte_never_launched;
 ORTE_DECLSPEC extern bool orte_devel_level_output;
 ORTE_DECLSPEC extern bool orte_display_topo_with_map;

orte/runtime/orte_mca_params.c

@@ -345,6 +345,12 @@ int orte_register_params(void)
                                 false, false, (int)false, &value);
     orte_keep_fqdn_hostnames = OPAL_INT_TO_BOOL(value);
     
+    /* whether or not to retain aliases of hostnames */
+    mca_base_param_reg_int_name("orte", "retain_aliases",
+                                "Whether or not to keep aliases for host names [default: no]",
+                                false, false, (int)false, &value);
+    orte_retain_aliases = OPAL_INT_TO_BOOL(value);
+
     /* whether to tag output */
     mca_base_param_reg_int_name("orte", "tag_output",
                                 "Tag all output with [job,rank] (default: false)",

orte/test/system/opal_interface.c

@@ -16,12 +16,14 @@ int main(int argc, char* argv[])
     int rc, idx;
     uint32_t addr, netmask, netaddr;
     struct sockaddr_in inaddr;
+    char **aliases=NULL;
 
     if (0 > (rc = opal_init(&argc, &argv))) {
         fprintf(stderr, "orte_interface: couldn't init opal - error code %d\n", rc);
         return rc;
     }
     
+    if (2 == argc) {
         rc = opal_iftupletoaddr(argv[1], &netaddr, &netmask);
     
         fprintf(stderr, "netaddr %03d.%03d.%03d.%03d netmask %03d.%03d.%03d.%03d rc %d\n",
@@ -47,6 +49,15 @@ int main(int argc, char* argv[])
             }
             idx = opal_ifnext(idx);
         }
+    }
+
+    /* check the aliases */
+    opal_ifgetaliases(&aliases);
+    idx = 0;
+    while (NULL != aliases[idx]) {
+        fprintf(stderr, "alias: %s\n", aliases[idx]);
+        idx++;
+    }
 
     opal_finalize();
     return 0;

orte/util/nidmap.c

@@ -311,6 +311,22 @@ int orte_util_encode_nodemap(opal_byte_object_t *boptr, bool update)
                 return rc;
             }
         }
+        /* if requested, pack any aliases */
+        if (orte_retain_aliases) {
+            uint8_t naliases, ni;
+            naliases = opal_argv_count(node->alias);
+            if (ORTE_SUCCESS != (rc = opal_dss.pack(&buf, &naliases, 1, OPAL_UINT8))) {
+                ORTE_ERROR_LOG(rc);
+                return rc;
+            }
+            for (ni=0; ni < naliases; ni++) {
+                if (ORTE_SUCCESS != (rc = opal_dss.pack(&buf, &node->alias[ni], 1, OPAL_STRING))) {
+                    ORTE_ERROR_LOG(rc);
+                    return rc;
+                }
+            }
+        }
+
         /* pack the oversubscribed flag */
         if (ORTE_SUCCESS != (rc = opal_dss.pack(&buf, &node->oversubscribed, 1, OPAL_UINT8))) {
             ORTE_ERROR_LOG(rc);
@@ -366,10 +382,17 @@ int orte_util_decode_nodemap(opal_byte_object_t *bo)
             ORTE_ERROR_LOG(rc);
             return rc;
         }
+        /* now store a direct reference so we can quickly lookup the daemon from a hostname */
+        if (ORTE_SUCCESS != (rc = orte_db.store(ORTE_NAME_WILDCARD, nodename, &daemon.vpid, ORTE_VPID))) {
+            ORTE_ERROR_LOG(rc);
+            return rc;
+        }
+
         OPAL_OUTPUT_VERBOSE((2, orte_nidmap_output,
                              "%s orte:util:decode:nidmap daemon %s node %s",
                              ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
                              ORTE_VPID_PRINT(daemon.vpid), nodename));
+
         /* if this is my daemon, then store the data for me too */
         if (daemon.vpid == ORTE_PROC_MY_DAEMON->vpid) {
             if (ORTE_SUCCESS != (rc = orte_db.store(ORTE_PROC_MY_NAME, ORTE_DB_HOSTNAME, nodename, OPAL_STRING))) {
@@ -381,6 +404,31 @@ int orte_util_decode_nodemap(opal_byte_object_t *bo)
                 return rc;
             }
         }
+
+        /* if requested, unpack any aliases */
+        if (orte_retain_aliases) {
+            char *alias;
+            uint8_t naliases, ni;
+            n=1;
+            if (ORTE_SUCCESS != (rc = opal_dss.unpack(&buf, &naliases, &n, OPAL_UINT8))) {
+                ORTE_ERROR_LOG(rc);
+                return rc;
+            }
+            for (ni=0; ni < naliases; ni++) {
+                n=1;
+                if (ORTE_SUCCESS != (rc = opal_dss.unpack(&buf, &alias, &n, OPAL_STRING))) {
+                    ORTE_ERROR_LOG(rc);
+                    return rc;
+                }
+                /* store a cross-reference to the daemon for this nodename */
+                if (ORTE_SUCCESS != (rc = orte_db.store(ORTE_NAME_WILDCARD, alias, &daemon.vpid, ORTE_VPID))) {
+                    ORTE_ERROR_LOG(rc);
+                    return rc;
+                }
+                free(alias);
+            }
+        }
+
         /* unpack and discard the oversubscribed flag - procs don't need it */
         n=1;
         if (ORTE_SUCCESS != (rc = opal_dss.unpack(&buf, &oversub, &n, OPAL_UINT8))) {
@@ -447,6 +495,25 @@ int orte_util_decode_daemon_nodemap(opal_byte_object_t *bo)
         } else {
             free(name);
         }
+        /* if requested, unpack any aliases */
+        if (orte_retain_aliases) {
+            char *alias;
+            uint8_t naliases, ni;
+            n=1;
+            if (ORTE_SUCCESS != (rc = opal_dss.unpack(&buf, &naliases, &n, OPAL_UINT8))) {
+                ORTE_ERROR_LOG(rc);
+                return rc;
+            }
+            for (ni=0; ni < naliases; ni++) {
+                n=1;
+                if (ORTE_SUCCESS != (rc = opal_dss.unpack(&buf, &alias, &n, OPAL_STRING))) {
+                    ORTE_ERROR_LOG(rc);
+                    return rc;
+                }
+                opal_argv_append_nosize(&node->alias, alias);
+                free(alias);
+            }
+        }
         /* unpack the oversubscribed flag */
         n=1;
         if (ORTE_SUCCESS != (rc = opal_dss.unpack(&buf, &oversub, &n, OPAL_UINT8))) {