add a helper function to create all missing remote databases detected

during recovery

(This used to be ctdb commit 04758c6f7d8f61260be6d2472380cb7904984427)

diff --git a/ctdb/direct/recoverd.c b/ctdb/direct/recoverd.c
index b5c3a3784fe29a07f12c9d651ec4a7229e491dde..9cb9a9557a4284415032a10249ae73b70a4529f8 100644 (file)
--- a/ctdb/direct/recoverd.c
+++ b/ctdb/direct/recoverd.c
@@ -66,55 +66,12 @@ static int set_recovery_mode(struct ctdb_context *ctdb, struct ctdb_node_map *no
        return 0;
 }
 
-static int do_recovery(struct ctdb_context *ctdb, struct event_context *ev,
-       TALLOC_CTX *mem_ctx, uint32_t vnn, uint32_t num_active,
-       struct ctdb_node_map *nodemap, struct ctdb_vnn_map *vnnmap)
+static int create_missing_remote_databases(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap, uint32_t vnn, struct ctdb_dbid_map *dbmap, TALLOC_CTX *mem_ctx)
 {
        int i, j, db, ret;
-       uint32_t generation;
-       struct ctdb_dbid_map *dbmap;
        struct ctdb_dbid_map *remote_dbmap;
 
-       printf("we need to do recovery !!!\n");
-
-       /* pick a new generation number */
-       generation = random();
-
-       /* change the vnnmap on this node to use the new generation 
-          number but not on any other nodes.
-          this guarantees that if we abort the recovery prematurely
-          for some reason (a node stops responding?)
-          that we can just return immediately and we will reenter
-          recovery shortly again.
-          I.e. we deliberately leave the cluster with an inconsistent
-          generation id to allow us to abort recovery at any stage and
-          just restart it from scratch.
-        */
-       vnnmap->generation = generation;
-       ret = ctdb_ctrl_setvnnmap(ctdb, timeval_current_ofs(1, 0), vnn, mem_ctx, vnnmap);
-       if (ret != 0) {
-               printf("Unable to set vnnmap for node %u\n", vnn);
-               return -1;
-       }
-
-
-       /* set recovery mode to active on all nodes */
-       ret = set_recovery_mode(ctdb, nodemap, CTDB_RECOVERY_ACTIVE);
-       if (ret!=0) {
-               printf("Unable to set recovery mode to active on cluster\n");
-               return -1;
-       }
-
-
-       /* get a list of all databases */
-       ret = ctdb_ctrl_getdbmap(ctdb, timeval_current_ofs(1, 0), vnn, mem_ctx, &dbmap);
-       if (ret != 0) {
-               printf("Unable to get dbids from node %u\n", vnn);
-               return -1;
-       }
-
        /* verify that all other nodes have all our databases */
-printf("Verify all other nodes have the same databases as we have\n");
        for (j=0; j<nodemap->num; j++) {
                /* we dont need to ourself ourselves */
                if (nodemap->nodes[j].vnn == vnn) {
@@ -124,7 +81,6 @@ printf("Verify all other nodes have the same databases as we have\n");
                if (!(nodemap->nodes[j].flags&NODE_FLAGS_CONNECTED)) {
                        continue;
                }
-printf("checking node %d\n",nodemap->nodes[j].vnn);
 
                ret = ctdb_ctrl_getdbmap(ctdb, timeval_current_ofs(1, 0), nodemap->nodes[j].vnn, mem_ctx, &remote_dbmap);
                if (ret != 0) {
@@ -160,6 +116,65 @@ printf("checking node %d\n",nodemap->nodes[j].vnn);
                }
        }
 
+       return 0;
+}
+
+
+static int do_recovery(struct ctdb_context *ctdb, struct event_context *ev,
+       TALLOC_CTX *mem_ctx, uint32_t vnn, uint32_t num_active,
+       struct ctdb_node_map *nodemap, struct ctdb_vnn_map *vnnmap)
+{
+       int i, j, db, ret;
+       uint32_t generation;
+       struct ctdb_dbid_map *dbmap;
+       struct ctdb_dbid_map *remote_dbmap;
+
+       printf("we need to do recovery !!!\n");
+
+       /* pick a new generation number */
+       generation = random();
+
+       /* change the vnnmap on this node to use the new generation 
+          number but not on any other nodes.
+          this guarantees that if we abort the recovery prematurely
+          for some reason (a node stops responding?)
+          that we can just return immediately and we will reenter
+          recovery shortly again.
+          I.e. we deliberately leave the cluster with an inconsistent
+          generation id to allow us to abort recovery at any stage and
+          just restart it from scratch.
+        */
+       vnnmap->generation = generation;
+       ret = ctdb_ctrl_setvnnmap(ctdb, timeval_current_ofs(1, 0), vnn, mem_ctx, vnnmap);
+       if (ret != 0) {
+               printf("Unable to set vnnmap for node %u\n", vnn);
+               return -1;
+       }
+
+
+       /* set recovery mode to active on all nodes */
+       ret = set_recovery_mode(ctdb, nodemap, CTDB_RECOVERY_ACTIVE);
+       if (ret!=0) {
+               printf("Unable to set recovery mode to active on cluster\n");
+               return -1;
+       }
+
+
+       /* get a list of all databases */
+       ret = ctdb_ctrl_getdbmap(ctdb, timeval_current_ofs(1, 0), vnn, mem_ctx, &dbmap);
+       if (ret != 0) {
+               printf("Unable to get dbids from node :%d\n", vnn);
+               return -1;
+       }
+
+       /* verify that all other nodes have all our databases */
+       ret = create_missing_remote_databases(ctdb, nodemap, vnn, dbmap, mem_ctx);
+       if (ret != 0) {
+               printf("Unable to create missing remote databases\n");
+               return -1;
+       }
+
+
        /* verify that we have all database any other node has */
        for (j=0; j<nodemap->num; j++) {
                /* we dont need to ourself ourselves */
@@ -213,49 +228,13 @@ printf("checking node %d\n",nodemap->nodes[j].vnn);
 
 
        /* verify that all other nodes have all our databases */
-       for (j=0; j<nodemap->num; j++) {
-               /* we dont need to ourself ourselves */
-               if (nodemap->nodes[j].vnn == vnn) {
-                       continue;
-               }
-               /* dont check nodes that are unavailable */
-               if (!(nodemap->nodes[j].flags&NODE_FLAGS_CONNECTED)) {
-                       continue;
-               }
-
-               ret = ctdb_ctrl_getdbmap(ctdb, timeval_current_ofs(1, 0), nodemap->nodes[j].vnn, mem_ctx, &remote_dbmap);
-               if (ret != 0) {
-                       printf("Unable to get dbids from node %u\n", vnn);
-                       return -1;
-               }
-
-               /* step through all local databases */
-               for (db=0; db<dbmap->num;db++) {
-                       const char *name;
-
-                       for (i=0;i<remote_dbmap->num;i++) {
-                               if (dbmap->dbids[db] == remote_dbmap->dbids[i]) {
-                                       break;
-                               }
-                       }
-                       /* the remote node already have this database */
-                       if (i!=remote_dbmap->num) {
-                               continue;
-                       }
-                       /* ok so we need to create this database */
-                       ctdb_ctrl_getdbname(ctdb, timeval_current_ofs(1, 0), vnn, dbmap->dbids[db], mem_ctx, &name);
-                       if (ret != 0) {
-                               printf("Unable to get dbname from node %u\n", vnn);
-                               return -1;
-                       }
-                       ctdb_ctrl_createdb(ctdb, timeval_current_ofs(1, 0), nodemap->nodes[j].vnn, mem_ctx, name);
-                       if (ret != 0) {
-                               printf("Unable to create remote db:%s\n", name);
-                               return -1;
-                       }
-               }
+       ret = create_missing_remote_databases(ctdb, nodemap, vnn, dbmap, mem_ctx);
+       if (ret != 0) {
+               printf("Unable to create missing remote databases\n");
+               return -1;
        }
 
+
        /* pull all records from all other nodes across to this node
           (this merges based on rsn)
        */