repmgr: have "standby follow" delete old replication slot, if possible

Addresses GitHub #272

HISTORY

@@ -7,6 +7,8 @@
           monitoring table (Ian)
         repmgrd: fix error in XLogRecPtr conversion when calculating
           monitoring statistics (Ian)
+        repmgr: if replication slots in use, where possible delete slot on old
+          upstream node after following new upstream (Ian)
 
 3.3     2016-12-27
         repmgr: always log to STDERR even if log facility defined (Ian)

repmgr.c

@@ -166,6 +166,8 @@ static void config_file_list_add(t_configfile_list *list, const char *file, cons
 static void matrix_set_node_status(t_node_matrix_rec **matrix_rec_list, int n, int node_id, int connection_node_id, int connection_status);
 static void cube_set_node_status(t_node_status_cube **cube, int n, int node_id, int matrix_node_id, int connection_node_id, int connection_status);
 
+static void drop_replication_slot_if_exists(PGconn *conn, int node_id, char *slot_name);
+
 /* Global variables */
 static PQconninfoOption *opts = NULL;
 
@@ -4527,7 +4529,8 @@ do_standby_follow(void)
 	char		script[MAXLEN];
 	char		master_conninfo[MAXLEN];
 	PGconn	   *master_conn;
-	int			master_id = 0;
+	int			master_id = UNKNOWN_NODE_ID,
+				original_upstream_node_id = UNKNOWN_NODE_ID;
 
 	int			r,
 				retval;
@@ -4674,7 +4677,11 @@ do_standby_follow(void)
 	}
 
 
-	/* Fetch our node record so we can write application_name, if set */
+	/*
+	 * Fetch our node record so we can write application_name, if set,
+	 * and to get the upstream node ID, which we'll need to know if
+	 * replication slots are in use and we want to delete the old slot.
+	 */
 	query_result = get_node_record(master_conn,
 								   options.cluster_name,
 								   options.node,
@@ -4707,6 +4714,15 @@ do_standby_follow(void)
 			if (application_name != NULL && strlen(application_name))
 				param_set(&recovery_conninfo, "application_name", application_name);
 		}
+
+		if (local_node_record.upstream_node_id != UNKNOWN_NODE_ID)
+		{
+			original_upstream_node_id = local_node_record.upstream_node_id;
+		}
+		else
+		{
+			original_upstream_node_id = master_id;
+		}
 	}
 
 	log_info(_("changing standby's master to node %i\n"), master_id);
@@ -4736,6 +4752,43 @@ do_standby_follow(void)
 		exit(ERR_NO_RESTART);
 	}
 
+
+	/*
+	 * If replication slots are in use, and an inactive one for this node
+	 * (a former standby) exists on the former upstream, drop it.
+	 */
+
+	if (options.use_replication_slots)
+	{
+		t_node_info upstream_node_record  = T_NODE_INFO_INITIALIZER;
+		int			upstream_query_result;
+
+		/* XXX should we poll for server restart? */
+		conn = establish_db_connection(options.conninfo, true);
+
+		upstream_query_result = get_node_record(conn,
+												options.cluster_name,
+												original_upstream_node_id,
+												&upstream_node_record);
+
+		PQfinish(conn);
+
+		conn = establish_db_connection(upstream_node_record.conninfo_str, false);
+		if (PQstatus(conn) != CONNECTION_OK)
+		{
+			log_warning("unable to connect to old upstream node %i to remove replication slot\n",
+						original_upstream_node_id);
+		}
+		else
+		{
+			log_debug("attempting to remove replication slot from old upstream node %i\n",
+					  original_upstream_node_id);
+			drop_replication_slot_if_exists(conn,
+											original_upstream_node_id,
+											local_node_record.slot_name);
+		}
+	}
+
 	/*
 	 * It's possible this node was an inactive primary - update the
 	 * relevant fields to ensure it's marked as an active standby
@@ -5629,30 +5682,9 @@ do_standby_switchover(void)
 		}
 		else
 		{
-			t_replication_slot  slot_info;
+			drop_replication_slot_if_exists(remote_conn,
-			int					query_res;
+											remote_node_id,
-
+											local_node_record.slot_name);
-			query_res = get_slot_record(remote_conn, local_node_record.slot_name, &slot_info);
-
-			if (query_res)
-			{
-				if (slot_info.active == false)
-				{
-					if (drop_replication_slot(remote_conn, local_node_record.slot_name) == true)
-					{
-						log_notice(_("replication slot \"%s\" deleted on former master\n"), local_node_record.slot_name);
-					}
-					else
-					{
-						log_err(_("unable to delete replication slot \"%s\" on former master\n"), local_node_record.slot_name);
-					}
-				}
-				/* if active replication slot exists, call Houston as we have a problem */
-				else
-				{
-					log_err(_("replication slot \"%s\" is still active on former master\n"), local_node_record.slot_name);
-				}
-			}
 		}
 
 		PQfinish(remote_conn);
@@ -8770,3 +8802,33 @@ config_file_list_add(t_configfile_list *list, const char *file, const char *file
 
 	list->entries ++;
 }
+
+
+static void
+drop_replication_slot_if_exists(PGconn *conn, int node_id, char *slot_name)
+{
+	t_replication_slot  slot_info;
+	int					query_res;
+
+	query_res = get_slot_record(conn,slot_name, &slot_info);
+
+	if (query_res)
+	{
+		if (slot_info.active == false)
+		{
+			if (drop_replication_slot(conn, slot_name) == true)
+			{
+				log_notice(_("replication slot \"%s\" deleted on node %i\n"), slot_name, node_id);
+			}
+			else
+			{
+				log_err(_("unable to delete replication slot \"%s\" on node %i\n"), slot_name, node_id);
+			}
+		}
+		/* if active replication slot exists, call Houston as we have a problem */
+		else
+		{
+			log_err(_("replication slot \"%s\" is still active on node %i\n"), slot_name, node_id);
+		}
+	}
+}