Resolve issue with icorrectly markign slave connection as errored

server/modules/include/blr.h

@@ -400,6 +400,7 @@ static char *blrs_states[] = { "Created", "Unregistered", "Registered",
 extern void blr_start_master(ROUTER_INSTANCE *);
 extern void blr_master_response(ROUTER_INSTANCE *, GWBUF *);
 extern void blr_master_reconnect(ROUTER_INSTANCE *);
+extern int blr_master_connected(ROUTER_INSTANCE *);
 
 extern int blr_slave_request(ROUTER_INSTANCE *, ROUTER_SLAVE *, GWBUF *);
 extern void blr_slave_rotate(ROUTER_SLAVE *slave, uint8_t *ptr);
@@ -413,4 +414,5 @@ extern void blr_file_flush(ROUTER_INSTANCE *);
 extern BLFILE *blr_open_binlog(ROUTER_INSTANCE *, char *);
 extern GWBUF *blr_read_binlog(ROUTER_INSTANCE *, BLFILE *, unsigned int, REP_HEADER *);
 extern void blr_close_binlog(ROUTER_INSTANCE *, BLFILE *);
+extern unsigned long blr_file_size(BLFILE *);
 #endif

server/modules/routing/binlog/blr.c

@@ -809,10 +809,21 @@ static  void
 errorReply(ROUTER *instance, void *router_session, GWBUF *message, DCB *backend_dcb, error_action_t action, bool *succp)
 {
 ROUTER_INSTANCE	*router = (ROUTER_INSTANCE *)instance;
+int		error, len;
+char		msg[85];
+
+	len = sizeof(error);
+	if (getsockopt(router->master->fd, SOL_SOCKET, SO_ERROR, &error, &len) != 0)
+	{
+		strerror_r(error, msg, 80);
+		strcat(msg, " ");
+	}
+	else
+		strcpy(msg, "");
 
        	LOGIF(LE, (skygw_log_write_flush(
-		LOGFILE_ERROR, "Erorr Reply '%s', attempting reconnect to master",
-			message)));
+		LOGFILE_ERROR, "Erorr Reply '%s', %sattempting reconnect to master",
+			message, msg)));
 	*succp = false;
 	blr_master_reconnect(router);
 }

server/modules/routing/binlog/blr_file.c

@@ -522,6 +522,13 @@ uint32_t	rval = 0, shift = 0;
 	return rval;
 }
 
+/**
+ * Log the event header of  binlog event
+ *
+ * @param	file	The log file into which to write the entry
+ * @param 	msg	A message strign to preceed the header with
+ * @param	ptr	The event header raw data
+ */
 static void
 blr_log_header(logfile_id_t file, char *msg, uint8_t *ptr)
 {
@@ -535,3 +542,19 @@ int	i;
 	skygw_log_write_flush(file, "%s", buf);
 	
 }
+
+/**
+ * Return the size of the current binlog file
+ *
+ * @param file	The binlog file
+ * @return	The current size of the binlog file
+ */
+unsigned long
+blr_file_size(BLFILE *file)
+{
+struct	stat	statb;
+
+	if (fstat(file->fd, &statb) == 0)
+		return statb.st_size;
+	return 0;
+}

server/modules/routing/binlog/blr_master.c

@@ -138,6 +138,7 @@ blr_restart_master(ROUTER_INSTANCE *router)
 GWBUF	*ptr;
 
 	dcb_close(router->master);
+	dcb_close(router->client);
 	dcb_free(router->master);
 	dcb_free(router->client);
 
@@ -1069,3 +1070,16 @@ int	i;
 		skygw_log_write_flush(file, "%s", buf);
 	
 }
+
+/**
+ * Check if the master connection is in place and we
+ * are downlaoding binlogs
+ *
+ * @param router	The router instance
+ * @return non-zero if we are recivign binlog records
+ */
+int
+blr_master_connected(ROUTER_INSTANCE *router)
+{
+	return router->master_state == BLRM_BINLOGDUMP;
+}

server/modules/routing/binlog/blr_slave.c

@@ -810,7 +810,10 @@ uint8_t		*ptr;
 	}
 	else
 	{
-		if (router->rotating != 0 && strcmp(router->binlog_name, slave->binlogfile) != 0)
+		if (slave->binlog_pos >= blr_file_size(slave->file)
+				&& router->rotating == 0
+				&& strcmp(router->binlog_name, slave->binlogfile) != 0
+				&& blr_master_connected(router))
 		{
 			/* We may have reached the end of file of a non-current
 			 * binlog file.
@@ -821,9 +824,11 @@ uint8_t		*ptr;
 			 * we ignore these issues during the rotate processing.
 			 */
 			LOGIF(LE, (skygw_log_write(LOGFILE_ERROR,
-				"Slave reached end of file for binlong file %s "
-				"which is not the file currently being downloaded.",
-				slave->binlogfile)));
+				"Slave reached end of file for binlong file %s at %u "
+				"which is not the file currently being downloaded. "
+				"Master binlog is %s, %lu.",
+				slave->binlogfile, slave->binlog_pos,
+				router->binlog_name, router->binlog_position)));
 			slave->state = BLRS_ERRORED;
 		}
 		else