Simple dns resolution cache

src/ergw_node_selection.erl

@@ -238,8 +238,7 @@ lookup_host(Name, dns, NameServers) ->
 	    _ ->
 		[]
 	end,
-    %% 3GPP DNS answers are almost always large, use TCP by default....
-    case inet_res:resolve(Name, in, any, [{usevc, true} | NsOpts]) of
+    case ergw_node_selection_cache:resolve(Name, any, NsOpts) of
 	{ok, Msg} ->
 	    lists:foldl(
 	      fun(RR, R) ->
@@ -400,8 +399,7 @@ naptr(Name, NameServers) ->
 	    _ ->
 		[]
 	end,
-    %% 3GPP DNS answers are almost always large, use TCP by default....
-    inet_res:resolve(Name, in, naptr, [{usevc, true} | NsOpts]).
+    ergw_node_selection_cache:resolve(Name, naptr, NsOpts).
 
 naptr_service_filter({Order, Pref, Flag, Services, _RegExp, Host}, OrdSPSet, Acc) ->
     [Service | Protocols] = string:tokens(Services, ":"),

src/ergw_node_selection_cache.erl

@@ -0,0 +1,35 @@
+-module(ergw_node_selection_cache).
+-export([
+    init/0,
+    resolve/3
+]).
+
+init() ->
+    ?MODULE = ets:new(?MODULE, [named_table, public, ordered_set, {read_concurrency, true}]),
+    ok.
+
+resolve(Name, Type, NsOpts) ->
+    Now = erlang:system_time(second),
+    case ets:lookup(?MODULE, {Name, Type}) of
+        [{_, Update, Result}] when Update == Now orelse Update == updating ->
+            Result;
+        [{_, _, Result}] ->
+            ets:update_element(?MODULE, {Name, Type}, {2, updating}),
+            % optimistic async update, won't hold up queries
+            spawn(fun() -> 
+                do_update(Name, Type, NsOpts, Now)
+            end),
+            Result;
+        [] ->
+            do_update(Name, Type, NsOpts, Now)
+    end.
+
+do_update(Name, Type, NsOpts, Now) ->
+    %% 3GPP DNS answers are almost always large, use TCP by default....
+    case inet_res:resolve(Name, in, Type, [{usevc, true} | NsOpts]) of
+        {error, _} = Error ->
+            Error;
+        {ok, Msg} ->
+            ets:insert(?MODULE, {{Name, Type}, Now, {ok, Msg}}),
+            {ok, Msg}
+    end.

src/ergw_sup.erl

@@ -30,6 +30,7 @@ start_link() ->
 %% ===================================================================
 
 init([]) ->
+    ok = ergw_node_selection_cache:init(),
     {ok, {{one_for_one, 5, 10}, [?CHILD(gtp_path_reg, worker, []),
 				 ?CHILD(ergw_tei_mngr, worker, []),
 				 ?CHILD(gtp_path_sup, supervisor, []),

test/pgw_proxy_SUITE.erl

@@ -2464,8 +2464,26 @@ dns_node_selection(Config) ->
 	      meck:passthrough([Name, Class, Type, Opts])
       end),
 
-    {GtpC1, _, _} = create_session(Config),
-    delete_session(GtpC1),
+    lists:foreach(
+        fun(_) ->
+            {GtpC, _, _} = create_session(Config),
+            delete_session(GtpC)
+        end,
+        lists:seq(1,10)
+    ),
+
+    timer:sleep(1100),
+
+    {GtpC3, _, _} = create_session(Config),
+    delete_session(GtpC3),
+
+    % 4 DNS requests to start with (it'll remember Sx node name then)
+    % then 3 for each time it needs to resolve further
+    % since the cache time check is on exact second, it can be 
+    % total 7 or 10 if the test is done just before change
+    % of second
+    NCalls = meck:num_calls(inet_res, resolve, '_'),
+    ?equal(true, NCalls == 7 orelse NCalls == 10),
 
     ok = meck:wait(?HUT, terminate, '_', ?TIMEOUT),
     meck_validate(Config),