diff --git a/lualib-src/lua-cluster.c b/lualib-src/lua-cluster.c
index e4e2f57f..8ea8ad08 100644
--- a/lualib-src/lua-cluster.c
+++ b/lualib-src/lua-cluster.c
@@ -208,8 +208,10 @@ lpackpush(lua_State *L) {
 	return 	
 		uint32_t or string addr
 		int session
-		string msg
+		lightuserdata msg
+		int sz
 		boolean padding
+		boolean is_push
  */
 
 static inline uint32_t
@@ -217,6 +219,14 @@ unpack_uint32(const uint8_t * buf) {
 	return buf[0] | buf[1]<<8 | buf[2]<<16 | buf[3]<<24;
 }
 
+static void
+return_buffer(lua_State *L, const char * buffer, int sz) {
+	void * ptr = skynet_malloc(sz);
+	memcpy(ptr, buffer, sz);
+	lua_pushlightuserdata(L, ptr);
+	lua_pushinteger(L, sz);
+}
+
 static int
 unpackreq_number(lua_State *L, const uint8_t * buf, int sz) {
 	if (sz < 9) {
@@ -226,14 +236,15 @@ unpackreq_number(lua_State *L, const uint8_t * buf, int sz) {
 	uint32_t session = unpack_uint32(buf+5);
 	lua_pushinteger(L, address);
 	lua_pushinteger(L, session);
-	lua_pushlstring(L, (const char *)buf+9, sz-9);
+
+	return_buffer(L, (const char *)buf+9, sz-9);
 	if (session == 0) {
 		lua_pushnil(L);
 		lua_pushboolean(L,1);	// is_push, no reponse
-		return 5;
+		return 6;
 	}
 
-	return 3;
+	return 4;
 }
 
 static int
@@ -246,11 +257,12 @@ unpackmreq_number(lua_State *L, const uint8_t * buf, int sz, int is_push) {
 	uint32_t size = unpack_uint32(buf+9);
 	lua_pushinteger(L, address);
 	lua_pushinteger(L, session);
+	lua_pushnil(L);
 	lua_pushinteger(L, size);
 	lua_pushboolean(L, 1);	// padding multi part
 	lua_pushboolean(L, is_push);
 
-	return 5;
+	return 6;
 }
 
 static int
@@ -262,10 +274,10 @@ unpackmreq_part(lua_State *L, const uint8_t * buf, int sz) {
 	uint32_t session = unpack_uint32(buf+1);
 	lua_pushboolean(L, 0);	// no address
 	lua_pushinteger(L, session);
-	lua_pushlstring(L, (const char *)buf+5, sz-5);
+	return_buffer(L, (const char *)buf+5, sz-5);
 	lua_pushboolean(L, padding);
 
-	return 4;
+	return 5;
 }
 
 static int
@@ -280,14 +292,14 @@ unpackreq_string(lua_State *L, const uint8_t * buf, int sz) {
 	lua_pushlstring(L, (const char *)buf+2, namesz);
 	uint32_t session = unpack_uint32(buf + namesz + 2);
 	lua_pushinteger(L, (uint32_t)session);
-	lua_pushlstring(L, (const char *)buf+2+namesz+4, sz - namesz - 6);
+	return_buffer(L, (const char *)buf+2+namesz+4, sz - namesz - 6);
 	if (session == 0) {
 		lua_pushnil(L);
 		lua_pushboolean(L,1);	// is_push, no reponse
-		return 5;
+		return 6;
 	}
 
-	return 3;
+	return 4;
 }
 
 static int
@@ -303,11 +315,12 @@ unpackmreq_string(lua_State *L, const uint8_t * buf, int sz, int is_push) {
 	uint32_t session = unpack_uint32(buf + namesz + 2);
 	uint32_t size = unpack_uint32(buf + namesz + 6);
 	lua_pushinteger(L, session);
+	lua_pushnil(L);
 	lua_pushinteger(L, size);
 	lua_pushboolean(L, 1);	// padding multipart
 	lua_pushboolean(L, is_push);
 
-	return 5;
+	return 6;
 }
 
 static int
@@ -481,6 +494,32 @@ lunpackresponse(lua_State *L) {
 	}
 }
 
+/*
+	table
+	pointer
+	sz
+
+	push (pointer/sz) as string into table, and free pointer
+ */
+static int
+lappend(lua_State *L) {
+	luaL_checktype(L, 1, LUA_TTABLE);
+	int n = lua_rawlen(L, 1);
+	if (lua_isnil(L, 2)) {
+		lua_settop(L, 3);
+		lua_seti(L, 1, n + 1);
+		return 0;
+	}
+	void * buffer = lua_touserdata(L, 2);
+	if (buffer == NULL)
+		return luaL_error(L, "Need lightuserdata");
+	int sz = luaL_checkinteger(L, 3);
+	lua_pushlstring(L, (const char *)buffer, sz);
+	skynet_free((void *)buffer);
+	lua_seti(L, 1, n+1);
+	return 0;
+}
+
 static int
 lconcat(lua_State *L) {
 	if (!lua_istable(L,1))
@@ -522,6 +561,7 @@ luaopen_skynet_cluster_core(lua_State *L) {
 		{ "unpackrequest", lunpackrequest },
 		{ "packresponse", lpackresponse },
 		{ "unpackresponse", lunpackresponse },
+		{ "append", lappend },
 		{ "concat", lconcat },
 		{ NULL, NULL },
 	};
diff --git a/service/clusteragent.lua b/service/clusteragent.lua
index d297dbb7..be201552 100644
--- a/service/clusteragent.lua
+++ b/service/clusteragent.lua
@@ -11,18 +11,17 @@ fd = tonumber(fd)
 local large_request = {}
 local register_name = {}
 
-local function dispatch_request(_,_,addr, session, msg, padding, is_push)
-	local sz
+local function dispatch_request(_,_,addr, session, msg, sz, padding, is_push)
 	if padding then
 		local req = large_request[session] or { addr = addr , is_push = is_push }
 		large_request[session] = req
-		table.insert(req, msg)
+		cluster.append(req, msg, sz)
 		return
 	else
 		local req = large_request[session]
 		if req then
 			large_request[session] = nil
-			table.insert(req, msg)
+			cluster.append(req, msg, sz)
 			msg,sz = cluster.concat(req)
 			addr = req.addr
 			is_push = req.is_push
@@ -36,6 +35,7 @@ local function dispatch_request(_,_,addr, session, msg, padding, is_push)
 	local ok, response
 	if addr == 0 then
 		local name = skynet.unpack(msg, sz)
+		skynet.trash(msg, sz)
 		local addr = register_name[name]
 		if addr == nil then
 			addr = skynet.call(clusterd, "lua", "queryname", name)
diff --git a/skynet-src/malloc_hook.c b/skynet-src/malloc_hook.c
index 25c3491b..7174f0a9 100644
--- a/skynet-src/malloc_hook.c
+++ b/skynet-src/malloc_hook.c
@@ -246,7 +246,7 @@ dump_c_mem() {
 		struct mem_data* data = &mem_stats[i];
 		if(data->handle != 0 && data->allocated != 0) {
 			total += data->allocated;
-			skynet_error(NULL, "0x%x -> %zdkb", data->handle, data->allocated >> 10);
+			skynet_error(NULL, ":%08x -> %zdkb %db", data->handle, data->allocated >> 10, (int)(data->allocated % 1024));
 		}
 	}
 	skynet_error(NULL, "+total: %zdkb",total >> 10);