Fix/vdf client incomplete prev session

apps/arweave/src/ar_http_iface_client.erl

@@ -522,7 +522,7 @@ get_vdf_update(Peer) ->
 		{ok, {{<<"200">>, _}, _, Bin, _, _}} ->
 			ar_serialize:binary_to_nonce_limiter_update(Bin);
 		{ok, {{<<"404">>, _}, _, _, _, _}} ->
-			not_found;
+			{error, not_found};
 		{ok, {{Status, _}, _, ResponseBody, _, _}} ->
 			{error, {Status, ResponseBody}};
 		Reply ->
@@ -535,7 +535,7 @@ get_vdf_session(Peer) ->
 		{ok, {{<<"200">>, _}, _, Bin, _, _}} ->
 			ar_serialize:binary_to_nonce_limiter_update(Bin);
 		{ok, {{<<"404">>, _}, _, _, _, _}} ->
-			not_found;
+			{error, not_found};
 		{ok, {{Status, _}, _, ResponseBody, _, _}} ->
 			{error, {Status, ResponseBody}};
 		Reply ->
@@ -548,7 +548,7 @@ get_previous_vdf_session(Peer) ->
 		{ok, {{<<"200">>, _}, _, Bin, _, _}} ->
 			ar_serialize:binary_to_nonce_limiter_update(Bin);
 		{ok, {{<<"404">>, _}, _, _, _, _}} ->
-			not_found;
+			{error, not_found};
 		{ok, {{Status, _}, _, ResponseBody, _, _}} ->
 			{error, {Status, ResponseBody}};
 		Reply ->

apps/arweave/src/ar_nonce_limiter.erl

@@ -307,6 +307,7 @@ request_validation(H, #nonce_limiter_info{ output = Output,
 				true ->
 					%% Wait for our VDF server(s) to validate the remaining steps.
 					%% Alternatively, the network may abandon this block.
+					ar_nonce_limiter_client:request_sessions(),
 					spawn(fun() -> ar_events:send(nonce_limiter, {refuse_validation, H}) end);
 				false ->
 					%% Validate the remaining steps.

apps/arweave/src/ar_nonce_limiter_client.erl

@@ -2,7 +2,7 @@
 
 -behaviour(gen_server).
 
--export([start_link/0]).
+-export([start_link/0, request_sessions/0]).
 
 -export([init/1, handle_cast/2, handle_call/3, handle_info/2, terminate/2]).
 
@@ -21,6 +21,11 @@
 start_link() ->
 	gen_server:start_link({local, ?MODULE}, ?MODULE, [], []).
 
+%% @doc Pick a server from the list of configured VDF servers and request complete
+%% VDF sessions from them. Failures to resolve a domain of make a request are not retried.
+request_sessions() ->
+	gen_server:cast(?MODULE, request_sessions).
+
 %%%===================================================================
 %%% Generic server callbacks.
 %%%===================================================================
@@ -80,7 +85,7 @@ handle_cast(pull, State) ->
 									{noreply, State2}
 							end
 					end;
-				not_found ->
+				{error, not_found} ->
 					%% The server might be restarting. Try another one, if there are any.
 					{noreply, State#state{ remote_servers = queue:in(RawPeer, Q2) }};
 				{error, Reason} ->
@@ -92,6 +97,23 @@ handle_cast(pull, State) ->
 			end
 	end;
 
+handle_cast(request_sessions, State) ->
+	#state{ remote_servers = Q } = State,
+	{{value, RawPeer}, Q2} = queue:out(Q),
+	State2 = State#state{ remote_servers = queue:in(RawPeer, Q2) },
+	case resolve_peer(RawPeer) of
+		error ->
+			%% Push the peer to the back of the queue.
+			{noreply, State2};
+		{ok, Peer} ->
+			case fetch_and_apply_session_and_previous_session(Peer) of
+				ok ->
+					{noreply, State};
+				_ ->
+					{noreply, State2}
+			end
+	end;
+
 handle_cast(Cast, State) ->
 	?LOG_WARNING([{event, unhandled_cast}, {module, ?MODULE}, {cast, Cast}]),
 	{noreply, State}.
@@ -127,8 +149,7 @@ fetch_and_apply_session_and_previous_session(Peer) ->
 			case ar_http_iface_client:get_previous_vdf_session(Peer) of
 				{ok, #nonce_limiter_update{ session_key = PrevSessionKey } = Update2} ->
 					ar_nonce_limiter:apply_external_update(Update2, Peer),
-					ar_nonce_limiter:apply_external_update(Update, Peer),
-					ok;
+					ar_nonce_limiter:apply_external_update(Update, Peer);
 				{ok, _} ->
 					%% The session should have just changed, retry.
 					fetch_and_apply_session_and_previous_session(Peer);
@@ -148,10 +169,7 @@ fetch_and_apply_session_and_previous_session(Peer) ->
 fetch_and_apply_session(Peer) ->
 	case ar_http_iface_client:get_vdf_session(Peer) of
 		{ok, Update} ->
-			ar_nonce_limiter:apply_external_update(Update, Peer),
-			ok;
-		not_found ->
-			not_found;
+			ar_nonce_limiter:apply_external_update(Update, Peer);
 		{error, Reason} = Error ->
 			?LOG_WARNING([{event, failed_to_fetch_vdf_session},
 					{peer, ar_util:format_peer(Peer)},

apps/arweave/src/ar_nonce_limiter_server_worker.erl

@@ -129,8 +129,15 @@ push_update(SessionKey, StepNumber, Output, Peer, Format, State) ->
 							%% Client requested the full session
 							PrevSessionKey = Session#vdf_session.prev_session_key,
 							PrevSession = ar_nonce_limiter:get_session(PrevSessionKey),
-							push_session(PrevSessionKey, PrevSession, Peer, Format),
-							push_session(SessionKey, Session, Peer, Format),
+							case push_session(PrevSessionKey, PrevSession, Peer, Format) of
+								ok ->
+									%% Do not push the new session until the previous
+									%% session is in line with our view (i.e., has steps
+									%% at least up to StepNumber where the new session begins).
+									push_session(SessionKey, Session, Peer, Format);
+								fail ->
+									ok
+							end,
 							State;
 						{true, true, true, false} ->
 							%% Client requested missing steps
@@ -158,9 +165,11 @@ push_session(SessionKey, Session, Peer, Format) ->
 						session_found = ReportedSessionFound }} ->
 					log_failure(Peer, SessionKey, Update, behind_client,
 						[{client_step_number, ClientStepNumber},
-						{session_found, ReportedSessionFound}]);
+						{session_found, ReportedSessionFound}]),
+					fail;
 				{error, Error} ->
-					log_failure(Peer, SessionKey, Update, Error, [])
+					log_failure(Peer, SessionKey, Update, Error, []),
+					fail
 			end
 	end.
 

apps/arweave/src/ar_storage_module.erl

@@ -84,8 +84,15 @@ packing_label(Packing) ->
 label_by_id("default") ->
 	"default";
 label_by_id(StoreID) ->
-	M = get_by_id(StoreID),
-	label(M).
+	case get_by_id(StoreID) of
+		not_found ->
+			%% Occurs in tests on application shutdown.
+			?LOG_WARNING([{event, store_id_for_label_not_found},
+					{store_id, StoreID}]),
+			"error";
+		M ->
+			label(M)
+	end.
 
 %% @doc Return the storage module with the given identifier or not_found.
 %% Search across both attached modules and repacked in-place modules.