Update the routed framework to:

1. add a new API delete_route(orte_process_name_t*) to delete the specified proc from the routing table

2. modify update_route so that it actually updates pre-existing routes instead of only adding routing info the end of the hash table

This fixes ticket #1403

This commit was SVN r18970.

orte/mca/routed/binomial/routed_binomial.c

@@ -34,6 +34,7 @@
 
 static int init(void);
 static int finalize(void);
+static int delete_route(orte_process_name_t *proc);
 static int update_route(orte_process_name_t *target,
                         orte_process_name_t *route);
 static orte_process_name_t get_route(orte_process_name_t *target);
@@ -52,6 +53,7 @@ static int binomial_ft_event(int state);
 orte_routed_module_t orte_routed_binomial_module = {
     init,
     finalize,
+    delete_route,
     update_route,
     get_route,
     init_routes,
@@ -162,6 +164,116 @@ static int finalize(void)
     return ORTE_SUCCESS;
 }
 
+static int delete_route(orte_process_name_t *proc)
+{
+    int rc;
+    orte_process_name_t *route_copy;
+    
+    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_process_info.hnp && !orte_process_info.daemon &&
+        !orte_process_info.tool) {
+        return ORTE_SUCCESS;
+    }
+    
+    OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
+                         "%s routed_binomial_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_process_info.daemon) {
+            return ORTE_SUCCESS;
+        }
+        
+        /* see if this proc is present - it will have a wildcard vpid,
+         * so we have to look for it with that condition
+         */
+        rc = opal_hash_table_get_value_uint32(&vpid_wildcard_list,
+                                              ORTE_JOB_FAMILY(proc->jobid),
+                                              (void**)&route_copy);
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* proc is present - remove the data */
+            free(route_copy);
+            rc = opal_hash_table_remove_value_uint32(&vpid_wildcard_list,
+                                                     ORTE_JOB_FAMILY(proc->jobid));
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
+        }
+        
+        /* not present - nothing to do */
+        return ORTE_SUCCESS;
+    }
+    
+    /* THIS CAME FROM OUR OWN JOB FAMILY... */
+    
+    /* treat vpid wildcards separately so they go onto the correct list */
+    if (proc->jobid != ORTE_JOBID_WILDCARD &&
+        proc->vpid == ORTE_VPID_WILDCARD) {
+        /* see if this target is already present - it will have a wildcard vpid,
+         * so we have to look for it on that list
+         */
+        rc = opal_hash_table_get_value_uint32(&vpid_wildcard_list,
+                                              proc->jobid,
+                                              (void**)&route_copy);
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* proc is present - remove the data */
+            free(route_copy);
+            rc = opal_hash_table_remove_value_uint32(&vpid_wildcard_list, proc->jobid);
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
+        }
+        /* not already present - nothing to do */
+        return ORTE_SUCCESS;
+    }
+    
+    /* check for an exact match */
+    if (proc->jobid != ORTE_JOBID_WILDCARD &&
+        proc->vpid != ORTE_VPID_WILDCARD) {
+        /* see if this route already exists in our table */
+        rc = opal_hash_table_get_value_uint64(&peer_list,
+                                              orte_util_hash_name(proc),
+                                              (void**)&route_copy);
+        
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* proc is present - remove the data */
+            free(route_copy);
+            rc = opal_hash_table_remove_value_uint64(&peer_list, orte_util_hash_name(proc));
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
+        }
+        /* not already present - nothing to do */
+        return ORTE_SUCCESS;
+    }
+    
+    /* this must be a process that doesn't match any of the
+     * prior conditions - sorry!
+     */
+    return ORTE_ERR_NOT_SUPPORTED;
+}
+
 static int update_route(orte_process_name_t *target,
                         orte_process_name_t *route)
 { 
@@ -214,8 +326,16 @@ static int update_route(orte_process_name_t *target,
                                               ORTE_JOB_FAMILY(target->jobid),
                                               (void**)&route_copy);
         if (ORTE_SUCCESS == rc && NULL != route_copy) {
-            /* target already present - no need for duplicate entry */
+            /* target already present - update the route info
-            return ORTE_SUCCESS;
+             * in case it has changed
+             */
+            *route_copy = *route;
+            rc = opal_hash_table_set_value_uint32(&vpid_wildcard_list,
+                                                  ORTE_JOB_FAMILY(target->jobid), route_copy);
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
         }
         
         /* not there, so add the route FOR THE JOB FAMILY*/
@@ -231,11 +351,52 @@ static int update_route(orte_process_name_t *target,
     
     /* THIS CAME FROM OUR OWN JOB FAMILY... */
     
+    /* treat vpid wildcards separately so they go onto the correct list */
+    if (target->jobid != ORTE_JOBID_WILDCARD &&
+        target->vpid == ORTE_VPID_WILDCARD) {
+        /* see if this target is already present - it will have a wildcard vpid,
+         * so we have to look for it on that list
+         */
+        rc = opal_hash_table_get_value_uint32(&vpid_wildcard_list,
+                                              target->jobid,
+                                              (void**)&route_copy);
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* target already present - update the route info
+             * in case it has changed
+             */
+            *route_copy = *route;
+            rc = opal_hash_table_set_value_uint32(&vpid_wildcard_list,
+                                                  target->jobid, route_copy);
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
+        }
+        
+        /* not already present, so let's add it */
         route_copy = malloc(sizeof(orte_process_name_t));
         *route_copy = *route;
-    /* exact match */
+        rc = opal_hash_table_set_value_uint32(&vpid_wildcard_list,
+                                              target->jobid, route_copy);
+        if (ORTE_SUCCESS != rc) {
+            ORTE_ERROR_LOG(rc);
+        }
+        return rc;
+    }
+    
+    /* check for an exact match */
     if (target->jobid != ORTE_JOBID_WILDCARD &&
         target->vpid != ORTE_VPID_WILDCARD) {
+        /* see if this route already exists in our table */
+        rc = opal_hash_table_get_value_uint64(&peer_list,
+                                              orte_util_hash_name(target),
+                                              (void**)&route_copy);
+        
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* target already present - update the route info
+             * in case it has changed
+             */
+            *route_copy = *route;
             rc = opal_hash_table_set_value_uint64(&peer_list,
                                                   orte_util_hash_name(target), route_copy);
             if (ORTE_SUCCESS != rc) {
@@ -244,18 +405,20 @@ static int update_route(orte_process_name_t *target,
             return rc;
         }
         
-    /* vpid wildcard */
+        /* not present - add it to the table */
-    if (target->jobid != ORTE_JOBID_WILDCARD &&
+        route_copy = malloc(sizeof(orte_process_name_t));
-        target->vpid == ORTE_VPID_WILDCARD) {
+        *route_copy = *route;
-        rc = opal_hash_table_set_value_uint32(&vpid_wildcard_list,
+        rc = opal_hash_table_set_value_uint64(&peer_list,
-                                              target->jobid, route_copy);
+                                              orte_util_hash_name(target), route_copy);
         if (ORTE_SUCCESS != rc) {
             ORTE_ERROR_LOG(rc);
         }
         return rc;
     }
-    free(route_copy);
 
+    /* this must be a process that doesn't match any of the
+     * prior conditions - sorry!
+     */
     return ORTE_ERR_NOT_SUPPORTED;
 }
 

orte/mca/routed/direct/routed_direct.c

@@ -45,6 +45,7 @@ static orte_process_name_t  *lifeline=NULL;
 /* API functions */
 static int init(void);
 static int finalize(void);
+static int delete_route(orte_process_name_t *proc);
 static int update_route(orte_process_name_t *target,
                         orte_process_name_t *route);
 static orte_process_name_t get_route(orte_process_name_t *target);
@@ -63,6 +64,7 @@ static int direct_ft_event(int state);
 orte_routed_module_t orte_routed_direct_module = {
     init,
     finalize,
+    delete_route,
     update_route,
     get_route,
     init_routes,
@@ -141,6 +143,51 @@ static int finalize(void)
 }
 
 
+static int delete_route(orte_process_name_t *proc)
+{
+    orte_process_name_t *route_copy;
+    int rc;
+    
+    if (proc->jobid == ORTE_JOBID_INVALID ||
+        proc->vpid == ORTE_VPID_INVALID) {
+        return ORTE_ERR_BAD_PARAM;
+    }
+    
+    /* if this isn't from a different job family, then there is
+     * nothing for us to do as all routes are direct - nothing
+     * is in the routing table
+     */
+    if (ORTE_JOB_FAMILY(proc->jobid) == ORTE_JOB_FAMILY(ORTE_PROC_MY_NAME->jobid)) {
+        return ORTE_SUCCESS;
+    }
+    
+    /* if I am -not- the HNP or a tool, then I will automatically route
+     * anything to this job family via my HNP - so nothing to do
+     * here since nothing is in my routing table
+     */
+    if (!orte_process_info.hnp && !orte_process_info.tool) {
+        return ORTE_SUCCESS;
+    }
+
+    /* must need to look it up */
+    rc = opal_hash_table_get_value_uint32(&peer_list,
+                                          ORTE_JOB_FAMILY(proc->jobid),
+                                          (void**)&route_copy);
+    if (ORTE_SUCCESS == rc && NULL != route_copy) {
+        /* proc is present - remove the data */
+        free(route_copy);
+        rc = opal_hash_table_remove_value_uint32(&peer_list,
+                                                 ORTE_JOB_FAMILY(proc->jobid));
+        if (ORTE_SUCCESS != rc) {
+            ORTE_ERROR_LOG(rc);
+        }            
+        return rc;
+    }
+    
+    /* wasn't here - nothing to do */
+    return ORTE_SUCCESS;
+}
+
 static int update_route(orte_process_name_t *target,
                         orte_process_name_t *route)
 {
@@ -176,8 +223,16 @@ static int update_route(orte_process_name_t *target,
                                               ORTE_JOB_FAMILY(target->jobid),
                                               (void**)&route_copy);
         if (ORTE_SUCCESS == rc && NULL != route_copy) {
-            /* target already present - no need for duplicate entry */
+            /* target already present - update the route info
-            return ORTE_SUCCESS;
+             * in case it has changed
+             */
+            *route_copy = *route;
+            rc = opal_hash_table_set_value_uint32(&peer_list,
+                                                  ORTE_JOB_FAMILY(target->jobid), route_copy);
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
         }
         
         /* not there, so add the route FOR THE JOB FAMILY*/

orte/mca/routed/linear/routed_linear.c

@@ -33,6 +33,7 @@
 
 static int init(void);
 static int finalize(void);
+static int delete_route(orte_process_name_t *proc);
 static int update_route(orte_process_name_t *target,
                         orte_process_name_t *route);
 static orte_process_name_t get_route(orte_process_name_t *target);
@@ -51,6 +52,7 @@ static int linear_ft_event(int state);
 orte_routed_module_t orte_routed_linear_module = {
     init,
     finalize,
+    delete_route,
     update_route,
     get_route,
     init_routes,
@@ -146,6 +148,116 @@ static int finalize(void)
     return ORTE_SUCCESS;
 }
 
+static int delete_route(orte_process_name_t *proc)
+{
+    int rc;
+    orte_process_name_t *route_copy;
+    
+    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_process_info.hnp && !orte_process_info.daemon &&
+        !orte_process_info.tool) {
+        return ORTE_SUCCESS;
+    }
+    
+    OPAL_OUTPUT_VERBOSE((1, orte_routed_base_output,
+                         "%s routed_binomial_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_process_info.daemon) {
+            return ORTE_SUCCESS;
+        }
+        
+        /* see if this proc is present - it will have a wildcard vpid,
+         * so we have to look for it with that condition
+         */
+        rc = opal_hash_table_get_value_uint32(&vpid_wildcard_list,
+                                              ORTE_JOB_FAMILY(proc->jobid),
+                                              (void**)&route_copy);
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* proc is present - remove the data */
+            free(route_copy);
+            rc = opal_hash_table_remove_value_uint32(&vpid_wildcard_list,
+                                                     ORTE_JOB_FAMILY(proc->jobid));
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
+        }
+        
+        /* not present - nothing to do */
+        return ORTE_SUCCESS;
+    }
+    
+    /* THIS CAME FROM OUR OWN JOB FAMILY... */
+    
+    /* treat vpid wildcards separately so they go onto the correct list */
+    if (proc->jobid != ORTE_JOBID_WILDCARD &&
+        proc->vpid == ORTE_VPID_WILDCARD) {
+        /* see if this target is already present - it will have a wildcard vpid,
+         * so we have to look for it on that list
+         */
+        rc = opal_hash_table_get_value_uint32(&vpid_wildcard_list,
+                                              proc->jobid,
+                                              (void**)&route_copy);
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* proc is present - remove the data */
+            free(route_copy);
+            rc = opal_hash_table_remove_value_uint32(&vpid_wildcard_list, proc->jobid);
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
+        }
+        /* not already present - nothing to do */
+        return ORTE_SUCCESS;
+    }
+    
+    /* check for an exact match */
+    if (proc->jobid != ORTE_JOBID_WILDCARD &&
+        proc->vpid != ORTE_VPID_WILDCARD) {
+        /* see if this route already exists in our table */
+        rc = opal_hash_table_get_value_uint64(&peer_list,
+                                              orte_util_hash_name(proc),
+                                              (void**)&route_copy);
+        
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* proc is present - remove the data */
+            free(route_copy);
+            rc = opal_hash_table_remove_value_uint64(&peer_list, orte_util_hash_name(proc));
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
+        }
+        /* not already present - nothing to do */
+        return ORTE_SUCCESS;
+    }
+    
+    /* this must be a process that doesn't match any of the
+     * prior conditions - sorry!
+     */
+    return ORTE_ERR_NOT_SUPPORTED;
+}
+
 static int update_route(orte_process_name_t *target,
                         orte_process_name_t *route)
 { 
@@ -198,8 +310,16 @@ static int update_route(orte_process_name_t *target,
                                               ORTE_JOB_FAMILY(target->jobid),
                                               (void**)&route_copy);
         if (ORTE_SUCCESS == rc && NULL != route_copy) {
-            /* target already present - no need for duplicate entry */
+            /* target already present - update the route info
-            return ORTE_SUCCESS;
+             * in case it has changed
+             */
+            *route_copy = *route;
+            rc = opal_hash_table_set_value_uint32(&vpid_wildcard_list,
+                                                  ORTE_JOB_FAMILY(target->jobid), route_copy);
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
         }
         
         /* not there, so add the route FOR THE JOB FAMILY*/
@@ -215,11 +335,52 @@ static int update_route(orte_process_name_t *target,
     
     /* THIS CAME FROM OUR OWN JOB FAMILY... */
     
+    /* treat vpid wildcards separately so they go onto the correct list */
+    if (target->jobid != ORTE_JOBID_WILDCARD &&
+        target->vpid == ORTE_VPID_WILDCARD) {
+        /* see if this target is already present - it will have a wildcard vpid,
+         * so we have to look for it on that list
+         */
+        rc = opal_hash_table_get_value_uint32(&vpid_wildcard_list,
+                                              target->jobid,
+                                              (void**)&route_copy);
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* target already present - update the route info
+             * in case it has changed
+             */
+            *route_copy = *route;
+            rc = opal_hash_table_set_value_uint32(&vpid_wildcard_list,
+                                                  target->jobid, route_copy);
+            if (ORTE_SUCCESS != rc) {
+                ORTE_ERROR_LOG(rc);
+            }            
+            return rc;
+        }
+        
+        /* not already present, so let's add it */
         route_copy = malloc(sizeof(orte_process_name_t));
         *route_copy = *route;
-    /* exact match */
+        rc = opal_hash_table_set_value_uint32(&vpid_wildcard_list,
+                                              target->jobid, route_copy);
+        if (ORTE_SUCCESS != rc) {
+            ORTE_ERROR_LOG(rc);
+        }
+        return rc;
+    }
+    
+    /* check for an exact match */
     if (target->jobid != ORTE_JOBID_WILDCARD &&
         target->vpid != ORTE_VPID_WILDCARD) {
+        /* see if this route already exists in our table */
+        rc = opal_hash_table_get_value_uint64(&peer_list,
+                                              orte_util_hash_name(target),
+                                              (void**)&route_copy);
+        
+        if (ORTE_SUCCESS == rc && NULL != route_copy) {
+            /* target already present - update the route info
+             * in case it has changed
+             */
+            *route_copy = *route;
             rc = opal_hash_table_set_value_uint64(&peer_list,
                                                   orte_util_hash_name(target), route_copy);
             if (ORTE_SUCCESS != rc) {
@@ -228,18 +389,20 @@ static int update_route(orte_process_name_t *target,
             return rc;
         }
         
-    /* vpid wildcard */
+        /* not present - add it to the table */
-    if (target->jobid != ORTE_JOBID_WILDCARD &&
+        route_copy = malloc(sizeof(orte_process_name_t));
-        target->vpid == ORTE_VPID_WILDCARD) {
+        *route_copy = *route;
-        rc = opal_hash_table_set_value_uint32(&vpid_wildcard_list,
+        rc = opal_hash_table_set_value_uint64(&peer_list,
-                                              target->jobid, route_copy);
+                                              orte_util_hash_name(target), route_copy);
         if (ORTE_SUCCESS != rc) {
             ORTE_ERROR_LOG(rc);
         }
         return rc;
     }
-    free(route_copy);
     
+    /* this must be a process that doesn't match any of the
+     * prior conditions - sorry!
+     */
     return ORTE_ERR_NOT_SUPPORTED;
 }
 

orte/mca/routed/routed.h

@@ -94,6 +94,15 @@ typedef int (*orte_routed_module_init_fn_t)(void);
 typedef int (*orte_routed_module_finalize_fn_t)(void);
 
 
+/*
+ * Delete route
+ *
+ * Delete the route to the specified proc from the routing table. Note
+ * that wildcards are supported to remove routes from, for example, all
+ * procs in a given job
+ */
+typedef int (*orte_routed_module_delete_route_fn_t)(orte_process_name_t *proc);
+
 /**
  * Update route table with new information
  *
@@ -229,6 +238,7 @@ struct orte_routed_module_t {
     orte_routed_module_init_fn_t                    initialize;
     orte_routed_module_finalize_fn_t                finalize;
     /* API functions */
+    orte_routed_module_delete_route_fn_t            delete_route;
     orte_routed_module_update_route_fn_t            update_route;
     orte_routed_module_get_route_fn_t               get_route;
     orte_routed_module_init_routes_fn_t             init_routes;