%%%------------------------------------------------------------------- %%% @doc Relay-routed Erlang distribution. %%% %%% When MACULA_DIST_MODE=relay, Erlang distribution connections %%% are tunneled through the Macula relay mesh instead of direct QUIC. %%% This enables distribution across firewalls and NATs — nodes only %%% need outbound connectivity to a relay. %%% %%% The tunnel works by: %%% 1. Node A publishes a DIST_CONNECT request to the mesh %%% 2. The relay forwards it to Node B's handler %%% 3. Node B opens a distribution stream back through the relay %%% 4. Raw ETF bytes flow through the relay tunnel %%% %%% This is EXPERIMENTAL. Supports Pid ! Msg, gen_server:call, %%% pg groups, process monitoring. Does NOT guarantee Mnesia or %%% global module compatibility over WAN latency. %%% @end %%%------------------------------------------------------------------- -module(macula_dist_relay). -include_lib("kernel/include/logger.hrl"). -export([connect/2, accept_dist/2]). -export([is_relay_mode/0, get_mesh_client/0]). -export([register_mesh_client/1]). -export([advertise_dist_accept/0]). -define(DIST_TOPIC, <<"_dist.connect">>). -define(DIST_TIMEOUT, 10000). %% @doc Register a mesh relay client for distribution tunneling. %% Called by the application that owns the relay client (e.g., hecate_mesh). -spec register_mesh_client(pid()) -> ok. register_mesh_client(Pid) when is_pid(Pid) -> persistent_term:put(macula_dist_mesh_client, Pid), ?LOG_INFO("[dist_relay] Mesh client registered: ~p", [Pid]), ok. %% @doc Check if relay distribution mode is enabled. -spec is_relay_mode() -> boolean(). is_relay_mode() -> case os:getenv("MACULA_DIST_MODE") of "relay" -> true; _ -> false end. %% @doc Connect to a remote node via relay mesh. %% Returns {ok, Conn, Stream} where the stream carries ETF bytes %% through the relay tunnel. -spec connect(string(), integer()) -> {ok, reference(), reference()} | {error, term()}. connect(Host, Port) -> NodeName = lists:flatten(io_lib:format("~b@~s", [Port, Host])), ?LOG_INFO("[dist_relay] Connecting to ~s via mesh", [NodeName]), %% Step 1: Find which relay the target node is connected to %% For now, use the same relay we're connected to (they share the mesh) case get_mesh_client() of undefined -> {error, no_mesh_connection}; MeshClient -> %% Step 2: Request a distribution tunnel via mesh RPC Procedure = <<"_dist.tunnel.", (list_to_binary(NodeName))/binary>>, case macula_relay_client:call(MeshClient, Procedure, #{ <<"from_node">> => atom_to_binary(node()), <<"target_node">> => list_to_binary(NodeName) }, ?DIST_TIMEOUT) of {ok, #{<<"tunnel_id">> := TunnelId}} -> ?LOG_INFO("[dist_relay] Tunnel established: ~s", [TunnelId]), %% Step 3: The tunnel_id identifies a relay-side bridge. %% We use the mesh QUIC connection's stream for ETF bytes. %% For the experimental version, we reuse the existing %% relay_client connection and multiplex dist traffic %% as pub/sub messages on a private topic. create_dist_socket(MeshClient, TunnelId); {error, Reason} -> ?LOG_WARNING("[dist_relay] Tunnel request failed: ~p", [Reason]), {error, {tunnel_failed, Reason}} end end. %% @doc Accept an incoming distribution connection via relay mesh. %% Called when a remote node requests a tunnel to us. -spec accept_dist(binary(), map()) -> {ok, pid()} | {error, term()}. accept_dist(TunnelId, Opts) -> ?LOG_INFO("[dist_relay] Accepting dist tunnel: ~s", [TunnelId]), %% The accepting side creates a process that bridges %% mesh pub/sub messages to the dist_util handshake FromNode = maps:get(<<"from_node">>, Opts, <<>>), spawn_link(fun() -> dist_tunnel_process(TunnelId, FromNode) end), {ok, TunnelId}. %% @doc Advertise this node as accepting distribution connections via relay. %% Registers a `_dist.tunnel.{nodename}` RPC procedure on the mesh. -spec advertise_dist_accept() -> ok. advertise_dist_accept() -> case get_mesh_client() of undefined -> ?LOG_WARNING("[dist_relay] Cannot advertise — no mesh client registered"), ok; Client -> NodeName = atom_to_binary(node()), Procedure = <<"_dist.tunnel.", NodeName/binary>>, Handler = fun(Args) -> handle_tunnel_request(Args) end, macula_relay_client:advertise(Client, Procedure, Handler), ?LOG_INFO("[dist_relay] Advertised distribution accept: ~s", [Procedure]), ok end. %%==================================================================== %% Internal — tunnel negotiation %%==================================================================== handle_tunnel_request(Args) -> FromNode = maps:get(<<"from_node">>, Args, <<>>), TunnelId = base64:encode(crypto:strong_rand_bytes(16)), ?LOG_INFO("[dist_relay] Tunnel request from ~s → id: ~s", [FromNode, TunnelId]), %% Set up the receiving side of the tunnel: %% Subscribe to the remote node's outgoing data topic %% Publish to the remote node's incoming data topic SendTopic = <<"_dist.data.", TunnelId/binary, ".in">>, RecvTopic = <<"_dist.data.", TunnelId/binary, ".out">>, case get_mesh_client() of undefined -> {error, <<"no_mesh_client">>}; Client -> %% Subscribe to data from the connecting node Self = self(), {ok, _SubRef} = macula_relay_client:subscribe(Client, RecvTopic, fun(#{payload := Data}) -> Self ! {dist_data, Data} end), %% Notify net_kernel about the incoming connection %% For now, return the tunnel_id so the connecting side can proceed {ok, #{<<"tunnel_id">> => TunnelId, <<"send_topic">> => SendTopic, <<"recv_topic">> => RecvTopic}} end. %%==================================================================== %% Internal — mesh client lookup %%==================================================================== %% Find the mesh relay client. Uses a registered name convention — %% any application that provides a relay client registers it as %% macula_dist_mesh_client. No dependency on hecate or any specific app. get_mesh_client() -> case persistent_term:get(macula_dist_mesh_client, undefined) of undefined -> undefined; Pid when is_pid(Pid) -> case is_process_alive(Pid) of true -> Pid; false -> undefined end end. create_dist_socket(MeshClient, TunnelId) -> %% Create a pair of linked processes that act as a socket: %% - Send: publish ETF bytes to tunnel topic %% - Recv: subscribe to tunnel topic, deliver as {quic, Data, ...} SendTopic = <<"_dist.data.", TunnelId/binary, ".out">>, RecvTopic = <<"_dist.data.", TunnelId/binary, ".in">>, Self = self(), %% Subscribe to incoming data {ok, _SubRef} = macula_relay_client:subscribe(MeshClient, RecvTopic, fun(#{payload := Data}) -> Self ! {dist_data, Data} end), {ok, MeshClient, {tunnel, TunnelId, SendTopic, RecvTopic}}. dist_tunnel_process(_TunnelId, _FromNode) -> %% This process lives for the duration of the distribution connection receive stop -> ok end.