Improve socketchannel, try the next host in backup list when auth failed. See issue #1145

lualib/skynet/db/mongo.lua

@@ -112,38 +112,15 @@ local function mongo_auth(mongoc)
 				mongoc.__sock:changebackup(backup)
 			end
 			if rs_data.ismaster	then
-				if rawget(mongoc, "__pickserver") then
-					rawset(mongoc, "__pickserver", nil)
-				end
 				return
-			else
+			elseif rs_data.primary then
-				if rs_data.primary then
 				local host,	port = __parse_addr(rs_data.primary)
 				mongoc.host	= host
 				mongoc.port	= port
 				mongoc.__sock:changehost(host, port)
 			else
-					skynet.error("WARNING: NO PRIMARY RETURN " .. rs_data.me)
+				-- socketchannel would try the next host in backup list
-					-- determine the primary db using hosts
+				error ("No primary return : " .. tostring(rs_data.me))
-					local pickserver = {}
-					if rawget(mongoc, "__pickserver") == nil then
-						for _, v in ipairs(rs_data.hosts) do
-							if v ~= rs_data.me then
-								table.insert(pickserver, v)
-							end
-							rawset(mongoc, "__pickserver", pickserver)
-						end
-					end
-					if #mongoc.__pickserver <= 0 then
-						error("CAN NOT DETERMINE THE PRIMARY DB")
-					end
-					skynet.error("INFO: TRY TO CONNECT " .. mongoc.__pickserver[1])
-					local host, port = __parse_addr(mongoc.__pickserver[1])
-					table.remove(mongoc.__pickserver, 1)
-					mongoc.host	= host
-					mongoc.port	= port
-					mongoc.__sock:changehost(host, port)
-				end
 			end
 		end
 	end

lualib/skynet/socketchannel.lua

@@ -201,27 +201,6 @@ local function dispatch_function(self)
 	end
 end
 
-local function connect_backup(self)
-	if self.__backup then
-		for _, addr in ipairs(self.__backup) do
-			local host, port
-			if type(addr) == "table" then
-				host, port = addr.host, addr.port
-			else
-				host = addr
-				port = self.__port
-			end
-			skynet.error("socket: connect to backup host", host, port)
-			local fd = socket.open(host, port)
-			if fd then
-				self.__host = host
-				self.__port = port
-				return fd
-			end
-		end
-	end
-end
-
 local function term_dispatch_thread(self)
 	if not self.__response and self.__dispatch_thread then
 		-- dispatch by order, send close signal to dispatch thread
@@ -233,17 +212,57 @@ local function connect_once(self)
 	if self.__closed then
 		return false
 	end
+
+	local addr_list = {}
+	local addr_set = {}
+
+	local function _add_backup()
+		if self.__backup then
+			for _, addr in ipairs(self.__backup) do
+				local host, port
+				if type(addr) == "table" then
+					host,port = addr.host, addr.port
+				else
+					host = addr
+					port = self.__port
+				end
+
+				-- don't add the same host
+				local hostkey = host..":"..port
+				if not addr_set[hostkey] then
+					addr_set[hostkey] = true
+					table.insert(addr_list, { host = host, port = port })
+				end
+			end
+		end
+	end
+
+	local function _next_addr()
+		local addr =  table.remove(addr_list,1)
+		if addr then
+			skynet.error("socket: connect to backup host", addr.host, addr.port)
+		end
+		return addr
+	end
+
+	local function _connect_once(self, addr)
+		local fd,err = socket.open(addr.host, addr.port)
+		if not fd then
+			-- try next one
+			addr = _next_addr()
+			if addr == nil then
+				return false, err
+			end
+			return _connect_once(self, addr)
+		end
+
+		self.__host = addr.host
+		self.__port = addr.port
+
 		assert(not self.__sock and not self.__authcoroutine)
 		-- term current dispatch thread (send a signal)
 		term_dispatch_thread(self)
 
-	local fd,err = socket.open(self.__host, self.__port)
-	if not fd then
-		fd = connect_backup(self)
-		if not fd then
-			return false, err
-		end
-	end
 		if self.__nodelay then
 			socketdriver.nodelay(fd)
 		end
@@ -297,14 +316,28 @@ local function connect_once(self)
 				end
 			end
 			self.__authcoroutine = false
-		if ok and not self.__sock then
+			if ok then
+				if not self.__sock then
 					-- auth may change host, so connect again
 					return connect_once(self)
 				end
-		return ok
+				-- auth succ, go through
+			else
+				-- auth failed, try next addr
+				_add_backup()	-- auth may add new backup hosts
+				addr = _next_addr()
+				if addr == nil then
+					return false, "no more backup host"
+				end
+				return _connect_once(self, addr)
+			end
 		end
 
 		return true
+	end
+
+	_add_backup()
+	return _connect_once(self, { host = self.__host, port = self.__port })
 end
 
 local function try_connect(self , once)