repmgrd: optionally exclude/include witness server from child node checks

configfile.c

@@ -364,11 +364,13 @@ _parse_config(t_configuration_options *options, ItemList *error_list, ItemList *
 	options->primary_visibility_consensus = false;
 	memset(options->failover_validation_command, 0, sizeof(options->failover_validation_command));
 	options->election_rerun_interval = DEFAULT_ELECTION_RERUN_INTERVAL;
+
 	options->child_nodes_check_interval = DEFAULT_CHILD_NODES_CHECK_INTERVAL;
-	memset(options->child_nodes_disconnect_command, 0, sizeof(options->child_nodes_disconnect_command));
 	options->child_nodes_disconnect_min_count = DEFAULT_CHILD_NODES_DISCONNECT_MIN_COUNT;
 	options->child_nodes_connected_min_count = DEFAULT_CHILD_NODES_CONNECTED_MIN_COUNT;
+	options->child_nodes_connected_include_witness = DEFAULT_CHILD_NODES_CONNECTED_INCLUDE_WITNESS;
 	options->child_nodes_disconnect_timeout = DEFAULT_CHILD_NODES_DISCONNECT_TIMEOUT;
+	memset(options->child_nodes_disconnect_command, 0, sizeof(options->child_nodes_disconnect_command));
 
 	/*-------------
 	 * witness settings
@@ -675,6 +677,8 @@ _parse_config(t_configuration_options *options, ItemList *error_list, ItemList *
 			options->child_nodes_disconnect_min_count = repmgr_atoi(value, name, error_list, -1);
 		else if (strcmp(name, "child_nodes_connected_min_count") == 0)
 			options->child_nodes_connected_min_count = repmgr_atoi(value, name, error_list, -1);
+		else if (strcmp(name, "child_nodes_connected_include_witness") == 0)
+			options->child_nodes_connected_include_witness = parse_bool(value, name, error_list);
 		else if (strcmp(name, "child_nodes_disconnect_timeout") == 0)
 			options->child_nodes_disconnect_timeout = repmgr_atoi(value, name, error_list, 0);
 
@@ -1117,6 +1121,7 @@ parse_time_unit_parameter(const char *name, const char *value, char *dest, ItemL
  * - bdr_recovery_timeout
  * - child_nodes_check_interval
  * - child_nodes_connected_min_count
+ * - child_nodes_connected_include_witness
  * - child_nodes_disconnect_command
  * - child_nodes_disconnect_min_count
  * - child_nodes_disconnect_timeout
@@ -1328,6 +1333,15 @@ reload_config(t_configuration_options *orig_options, t_server_type server_type)
 		}
 	}
 
+	/* child_nodes_connected_include_witness */
+	if (orig_options->child_nodes_connected_include_witness != new_options.child_nodes_connected_include_witness)
+	{
+		orig_options->child_nodes_connected_include_witness = new_options.child_nodes_connected_include_witness;
+		log_info(_("\"child_nodes_connected_include_witness\" is now \"%i\""), new_options.child_nodes_connected_include_witness);
+
+		config_changed = true;
+	}
+
 	/* child_nodes_disconnect_timeout */
 	if (orig_options->child_nodes_disconnect_timeout != new_options.child_nodes_disconnect_timeout)
 	{

configfile.h

@@ -151,6 +151,7 @@ typedef struct
 	int			child_nodes_check_interval;
 	int			child_nodes_disconnect_min_count;
 	int			child_nodes_connected_min_count;
+	bool		child_nodes_connected_include_witness;
 	int			child_nodes_disconnect_timeout;
 	char		child_nodes_disconnect_command[MAXPGPATH];
 
@@ -229,6 +230,7 @@ typedef struct
 		DEFAULT_CHILD_NODES_CHECK_INTERVAL, \
 		DEFAULT_CHILD_NODES_DISCONNECT_MIN_COUNT, \
 		DEFAULT_CHILD_NODES_CONNECTED_MIN_COUNT, \
+		DEFAULT_CHILD_NODES_CONNECTED_INCLUDE_WITNESS, \
 		DEFAULT_CHILD_NODES_DISCONNECT_TIMEOUT, "", \
 		/* BDR settings */ \
 		false, DEFAULT_BDR_RECOVERY_TIMEOUT, \

repmgr.h

@@ -100,7 +100,7 @@
 #define DEFAULT_CHILD_NODES_DISCONNECT_MIN_COUNT -1
 #define DEFAULT_CHILD_NODES_CONNECTED_MIN_COUNT -1
 #define DEFAULT_CHILD_NODES_DISCONNECT_TIMEOUT 30 /* seconds */
-
+#define DEFAULT_CHILD_NODES_CONNECTED_INCLUDE_WITNESS false
 
 #define WALRECEIVER_DISABLE_TIMEOUT_VALUE    86400000 /* milliseconds */
 

repmgrd-physical.c

@@ -1026,7 +1026,10 @@ check_primary_child_nodes(t_child_node_info_list *local_child_nodes)
 		bool repmgrd_paused = repmgrd_is_paused(local_conn);
 
 		if (repmgrd_paused == false)
+		{
+			/* check criteria for execution, and execute if criteria met */
 			execute_child_nodes_disconnect_command(&db_child_node_records, local_child_nodes);
+		}
 	}
 
 	clear_child_node_info_list(&disconnected_child_nodes);
@@ -1060,14 +1063,34 @@ execute_child_nodes_disconnect_command(NodeInfoList *db_child_node_records, t_ch
 	}
 	else if (config_file_options.child_nodes_disconnect_min_count > 0)
 	{
+		int child_node_count = db_child_node_records->node_count;
+
+		if (config_file_options.child_nodes_connected_include_witness == false)
+		{
+			/* reduce total, if witness server in child node list */
+			for (cell = db_child_node_records->head; cell; cell = cell->next)
+			{
+				if (cell->node_info->type == WITNESS)
+				{
+					child_node_count--;
+					break;
+				}
+			}
+		}
+
 		min_required_connected_count =
-			(db_child_node_records->node_count - config_file_options.child_nodes_disconnect_min_count)
+			(child_node_count - config_file_options.child_nodes_disconnect_min_count)
 			+ 1;
 	}
 
 	/* calculate number of connected child nodes */
 	for (cell = db_child_node_records->head; cell; cell = cell->next)
 	{
+		/* exclude witness server from total, if necessay */
+		if (config_file_options.child_nodes_connected_include_witness == false &&
+			cell->node_info->type == WITNESS)
+			continue;
+
 		if (cell->node_info->attached == NODE_ATTACHED)
 			connected_count ++;
 	}
@@ -1101,6 +1124,11 @@ execute_child_nodes_disconnect_command(NodeInfoList *db_child_node_records, t_ch
 				instr_time  current_time = current_time_base;
 				int seconds_since_detached;
 
+				/* exclude witness server from calculatin if neccessary */
+				if (config_file_options.child_nodes_connected_include_witness == false &&
+					child_node_rec->type == WITNESS)
+					continue;
+
 				if (child_node_rec->attached != NODE_DETACHED)
 					continue;