Skip to main content

self_hosted_node/
host.rs

1#![allow(clippy::too_many_arguments)]
2use std::collections::{HashMap, HashSet};
3use std::sync::atomic::{AtomicBool, Ordering};
4use std::sync::{mpsc as std_mpsc, Arc, Mutex};
5use std::thread;
6use std::time::{Duration, Instant};
7
8use crate::config::{Config, DeckSelection, SelfPlayConfig};
9use crate::engine_backend::{java_backend, rust_backend, EngineBackendKind, HostedGameOver};
10use crate::shell_bridge::{ShellBridge, ShellCommand};
11use crate::updater::{run_stale_monitor, StaleConfig};
12use futures_util::stream::{SplitSink, SplitStream};
13use futures_util::{SinkExt, StreamExt};
14use manabot::{run_bot, AgentKind, BotConfig};
15use manabrew_agent_interface::ids_codec::{parse_player_slot, player_slot};
16use manabrew_agent_interface::prompt::{AgentMessage, ClientToServerMessage, PromptOutput};
17use manabrew_agent_interface::protocol::{
18    identity_token, ClientMessage, ClientPlatform, EngineKind, GameFormat, GameOutcomeReport,
19    IdentityProof, PlayerDeckInfo, ResumeRoomRequest, RoomInfo, RoomStatus, ServerMessage,
20    StateEnvelope, PROTOCOL_VERSION,
21};
22use manabrew_protocol::deck_dto::Deck;
23use manabrew_protocol::game::{GameViewDto, PlayerStatus};
24use manabrew_protocol::transport::{DirectiveInput, StateUpdate};
25use serde::Deserialize;
26use serde_json::{json, Value};
27use tokio::net::TcpStream;
28use tokio::sync::mpsc as tokio_mpsc;
29use tokio::task::JoinHandle;
30use tokio::time;
31use tokio_tungstenite::tungstenite::Message;
32use tokio_tungstenite::{connect_async_with_config, MaybeTlsStream, WebSocketStream};
33use tracing::{debug, error, info, warn};
34use uuid::Uuid;
35
36type WsStream = WebSocketStream<MaybeTlsStream<TcpStream>>;
37type WsWrite = SplitSink<WsStream, Message>;
38type WsRead = SplitStream<WsStream>;
39
40const SELF_HOSTED_NODE_PROTOCOL: &str = "self-hosted-node";
41const PRIVATE_MESSAGE_MISSING_PLAYER: &str =
42    "cannot forward private engine message without a mapped player";
43
44struct RelayClient {
45    username: String,
46    outbound: tokio_mpsc::UnboundedSender<Message>,
47    read: WsRead,
48    writer: JoinHandle<()>,
49}
50
51enum EngineSession {
52    Manabrew {
53        game_id: String,
54        remote_response_txs: HashMap<usize, std_mpsc::Sender<ClientToServerMessage>>,
55        engine_clock: EngineClock,
56    },
57    Forge {
58        game_id: String,
59        remote_response_txs: HashMap<usize, std_mpsc::Sender<ClientToServerMessage>>,
60        cancel: Arc<AtomicBool>,
61        engine_clock: EngineClock,
62    },
63}
64
65impl EngineSession {
66    fn game_id(&self) -> &str {
67        match self {
68            EngineSession::Manabrew { game_id, .. } | EngineSession::Forge { game_id, .. } => {
69                game_id
70            }
71        }
72    }
73
74    fn engine_clock(&self) -> &EngineClock {
75        match self {
76            EngineSession::Manabrew { engine_clock, .. }
77            | EngineSession::Forge { engine_clock, .. } => engine_clock,
78        }
79    }
80}
81
82struct BotState {
83    handle: JoinHandle<()>,
84    shutdown: Arc<tokio::sync::Notify>,
85}
86
87#[derive(Debug, Deserialize)]
88#[serde(rename_all = "camelCase")]
89struct SpawnBotPayload {
90    #[serde(default)]
91    deck: Option<SpawnBotDeckPayload>,
92    #[serde(default)]
93    decks: Option<Vec<SpawnBotDeckPayload>>,
94}
95
96#[derive(Debug, Deserialize)]
97#[serde(rename_all = "camelCase")]
98struct SpawnBotDeckPayload {
99    deck_name: String,
100    deck: Deck,
101    #[serde(default)]
102    commander_name: Option<String>,
103}
104
105type SharedEngineSession = Arc<Mutex<Option<EngineSession>>>;
106type SessionRegistry = Arc<Mutex<Vec<SharedEngineSession>>>;
107
108fn registry_idle(registry: &SessionRegistry) -> bool {
109    registry
110        .lock()
111        .map(|sessions| {
112            sessions
113                .iter()
114                .all(|session| session.lock().map(|guard| guard.is_none()).unwrap_or(false))
115        })
116        .unwrap_or(false)
117}
118
119static DRAINING: AtomicBool = AtomicBool::new(false);
120
121fn draining() -> bool {
122    DRAINING.load(Ordering::Relaxed)
123}
124
125type SharedBotState = Arc<Mutex<Vec<BotState>>>;
126
127#[derive(Clone)]
128struct GameStart {
129    game_id: String,
130    player_order: Vec<String>,
131    player_decks: Vec<PlayerDeckInfo>,
132    starting_life: i32,
133}
134
135#[derive(Default)]
136struct HostSnapshot {
137    room_info: Option<RoomInfo>,
138    resume_token: Option<String>,
139    game: Option<GameStart>,
140    last_state: Option<Value>,
141    last_state_by_slot: HashMap<String, Value>,
142    pending_prompts: HashMap<String, Value>,
143    pending_end_game: Option<String>,
144}
145
146type SharedHostSnapshot = Arc<Mutex<HostSnapshot>>;
147
148const RECONNECT_BACKOFF_SECS: [u64; 6] = [1, 2, 4, 8, 15, 30];
149
150/// Pseudo-seat used by per-recipient backends for the public (spectator) view.
151pub(crate) const OBSERVER_SEAT: usize = usize::MAX;
152const CLOSE_DRAIN_TIMEOUT: Duration = Duration::from_secs(3);
153const BOT_STOP_TIMEOUT: Duration = Duration::from_secs(5);
154
155enum LoopExit {
156    Cancelled,
157    Disconnected,
158}
159
160pub async fn cli_entry() {
161    {
162        use tracing_subscriber::layer::SubscriberExt;
163        use tracing_subscriber::util::SubscriberInitExt;
164        tracing_subscriber::registry()
165            .with(
166                tracing_subscriber::EnvFilter::try_from_default_env()
167                    .unwrap_or_else(|_| "self_hosted_node=info".into()),
168            )
169            .with(tracing_subscriber::fmt::layer())
170            .with(crate::logs::layer_from_env())
171            .init();
172    }
173    crate::metrics::init_from_env();
174
175    if std::env::var("SELF_HOSTED_NODE_JAVA_SMOKE").is_ok() {
176        let max_prompts = std::env::var("SELF_HOSTED_NODE_JAVA_SMOKE_PROMPTS")
177            .ok()
178            .and_then(|value| value.parse().ok())
179            .unwrap_or(4);
180        if let Err(error) = java_backend::run_smoke_game(max_prompts) {
181            error!(%error, "java-forge smoke failed");
182            std::process::exit(1);
183        }
184        info!(max_prompts, "java-forge smoke completed");
185        return;
186    }
187    if std::env::var("SELF_HOSTED_NODE_GRAAL_SMOKE").is_ok() {
188        if let Err(error) = java_backend::run_graal_smoke() {
189            error!(%error, "graal-forge smoke failed");
190            std::process::exit(1);
191        }
192        info!("graal-forge smoke completed");
193        return;
194    }
195    if std::env::var("SELF_HOSTED_NODE_JAVA_CONCEDE_SMOKE").is_ok() {
196        if let Err(error) = java_backend::run_concede_smoke() {
197            error!(%error, "java-forge concede smoke failed");
198            std::process::exit(1);
199        }
200        info!("java-forge concede smoke completed");
201        return;
202    }
203    if let Ok(scenario_name) = std::env::var("SELF_HOSTED_NODE_JAVA_SCENARIO") {
204        let max_prompts = std::env::var("SELF_HOSTED_NODE_JAVA_SCENARIO_PROMPTS")
205            .ok()
206            .and_then(|value| value.parse().ok())
207            .unwrap_or(20);
208        if let Err(error) = java_backend::run_scenario(&scenario_name, max_prompts) {
209            error!(scenario_name, %error, "java-forge scenario failed");
210            std::process::exit(1);
211        }
212        info!(scenario_name, max_prompts, "java-forge scenario completed");
213        return;
214    }
215    if std::env::var("SELF_HOSTED_NODE_JAVA_SELF_PLAY").is_ok() {
216        let cfg = SelfPlayConfig::from_env();
217        let max_prompts = std::env::var("SELF_HOSTED_NODE_JAVA_SELF_PLAY_PROMPTS")
218            .ok()
219            .and_then(|value| value.parse().ok())
220            .unwrap_or(2_000);
221        let games = std::env::var("SELF_HOSTED_NODE_JAVA_SELF_PLAY_GAMES")
222            .ok()
223            .and_then(|value| value.parse().ok())
224            .unwrap_or(1);
225        if let Err(error) =
226            java_backend::run_self_play(&cfg.seats, cfg.starting_life, cfg.seed, max_prompts, games)
227        {
228            error!(%error, "java-forge self-play failed");
229            std::process::exit(1);
230        }
231        info!(
232            players = cfg.seats.len(),
233            max_prompts, games, "java-forge self-play completed"
234        );
235        return;
236    }
237    if let Ok(value) = std::env::var("SELF_HOSTED_NODE_JAVA_CONCURRENT_GAMES") {
238        let cfg = SelfPlayConfig::from_env();
239        let max_prompts = std::env::var("SELF_HOSTED_NODE_JAVA_SELF_PLAY_PROMPTS")
240            .ok()
241            .and_then(|value| value.parse().ok())
242            .unwrap_or(2_000);
243        let concurrency = value.parse().unwrap_or(2);
244        if let Err(error) = java_backend::run_concurrent_self_play(
245            &cfg.seats,
246            cfg.starting_life,
247            cfg.seed,
248            max_prompts,
249            concurrency,
250        ) {
251            error!(%error, "java-forge concurrent self-play failed");
252            std::process::exit(1);
253        }
254        info!(
255            players = cfg.seats.len(),
256            concurrency, "java-forge concurrent self-play completed"
257        );
258        return;
259    }
260    if std::env::var("SELF_HOSTED_NODE_RUST_SELF_PLAY").is_ok() {
261        let cfg = SelfPlayConfig::from_env();
262        let max_turns = std::env::var("SELF_HOSTED_NODE_RUST_SELF_PLAY_TURNS")
263            .ok()
264            .and_then(|value| value.parse().ok())
265            .unwrap_or(200);
266        if let Err(error) =
267            rust_backend::run_self_play(&cfg.seats, cfg.starting_life, cfg.seed, max_turns)
268        {
269            error!(%error, "rust self-play failed");
270            std::process::exit(1);
271        }
272        info!(
273            players = cfg.seats.len(),
274            max_turns, "rust self-play completed"
275        );
276        return;
277    }
278
279    let config = Config::from_env();
280    if let Err(error) = run(config).await {
281        error!(%error, "room node exited");
282        std::process::exit(1);
283    }
284}
285
286/// Signal a hosted room to shut down: call `notify_one()` to stop it.
287pub type RoomCancel = Arc<tokio::sync::Notify>;
288
289fn ensure_engine_ready(config: &Config) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
290    if config.engine_enabled
291        && config.backend.is_supported()
292        && matches!(config.backend, EngineBackendKind::Forge)
293    {
294        info!("initializing forge engine backend");
295        java_backend::init_engine()?;
296    }
297    Ok(())
298}
299
300/// Host a single relay room until it ends or `cancel` is signalled. The crate's
301/// public entry point — env-free; the caller supplies a fully-built `Config`.
302pub async fn host_room(
303    config: Config,
304    cancel: RoomCancel,
305    ready: tokio::sync::oneshot::Sender<String>,
306) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
307    ensure_engine_ready(&config)?;
308    host_one_room(config, None, cancel, Some(ready), None, None).await
309}
310
311/// The webview's half of the data plane, handed in by a desktop shell.
312pub struct ShellBridgeHandle {
313    pub bridge: Arc<ShellBridge>,
314    pub commands: tokio_mpsc::UnboundedReceiver<ShellCommand>,
315}
316
317pub async fn host_room_bridged(
318    config: Config,
319    cancel: RoomCancel,
320    ready: tokio::sync::oneshot::Sender<String>,
321    bridge: ShellBridgeHandle,
322) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
323    ensure_engine_ready(&config)?;
324    host_one_room(config, None, cancel, Some(ready), None, Some(bridge)).await
325}
326
327async fn run(mut config: Config) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
328    ensure_engine_ready(&config)?;
329
330    let registry: SessionRegistry = Arc::new(Mutex::new(Vec::new()));
331
332    let slots = config.max_games.max(1);
333    let single = config.room_id.is_some() || slots <= 1;
334    let room_count = if single { 1 } else { slots };
335    let cancels: Vec<RoomCancel> = (0..room_count)
336        .map(|_| Arc::new(tokio::sync::Notify::new()))
337        .collect();
338
339    let monitor_registry = registry.clone();
340    let stale_cancels = cancels.clone();
341    tokio::spawn(run_stale_monitor(
342        StaleConfig::from_env_and_args(),
343        move || registry_idle(&monitor_registry),
344        move || notify_all(&stale_cancels),
345    ));
346
347    let signal_cancels = cancels.clone();
348    tokio::spawn(async move {
349        wait_for_shutdown_signal().await;
350        info!("shutdown signal received; closing rooms");
351        notify_all(&signal_cancels);
352    });
353
354    #[cfg(unix)]
355    tokio::spawn(async move {
356        let mut usr1 =
357            match tokio::signal::unix::signal(tokio::signal::unix::SignalKind::user_defined1()) {
358                Ok(usr1) => usr1,
359                Err(error) => {
360                    warn!(%error, "failed to install SIGUSR1 handler");
361                    return;
362                }
363            };
364        while usr1.recv().await.is_some() {
365            DRAINING.store(true, Ordering::Relaxed);
366            info!("drain signal received; refusing new games and closing each room once idle");
367        }
368    });
369
370    #[cfg(all(unix, feature = "graal-forge"))]
371    tokio::spawn(async move {
372        let mut usr2 =
373            match tokio::signal::unix::signal(tokio::signal::unix::SignalKind::user_defined2()) {
374                Ok(usr2) => usr2,
375                Err(error) => {
376                    warn!(%error, "failed to install SIGUSR2 handler");
377                    return;
378                }
379            };
380        while usr2.recv().await.is_some() {
381            let dir = std::env::var("SELF_HOSTED_NODE_HEAP_DUMP_DIR")
382                .unwrap_or_else(|_| "/tmp".to_string());
383            let stamp = std::time::SystemTime::now()
384                .duration_since(std::time::UNIX_EPOCH)
385                .map(|since| since.as_secs())
386                .unwrap_or(0);
387            let path = format!("{dir}/self-hosted-node-{stamp}.hprof");
388            info!(path, "heap dump requested");
389            match tokio::task::spawn_blocking({
390                let path = path.clone();
391                move || java_backend::dump_shared_heap(&path)
392            })
393            .await
394            {
395                Ok(Ok(())) => info!(path, "heap dump written"),
396                Ok(Err(error)) => warn!(%error, "heap dump failed"),
397                Err(error) => warn!(%error, "heap dump task failed"),
398            }
399        }
400    });
401
402    if single {
403        return host_one_room(config, None, cancels[0].clone(), None, Some(registry), None).await;
404    }
405
406    config.format = GameFormat::Any;
407    let hosts: Vec<(Config, String)> = (0..slots)
408        .map(|slot| (config.clone(), (slot + 1).to_string()))
409        .collect();
410
411    info!(rooms = hosts.len(), "hosting multiple rooms on one node");
412    let mut handles = Vec::with_capacity(hosts.len());
413    for ((cfg, label), cancel) in hosts.into_iter().zip(cancels) {
414        let registry = registry.clone();
415        handles.push(tokio::spawn(async move {
416            if let Err(error) =
417                host_one_room(cfg, Some(label.clone()), cancel, None, Some(registry), None).await
418            {
419                error!(%error, label, "room host exited");
420            }
421        }));
422    }
423    for handle in handles {
424        let _ = handle.await;
425    }
426    Ok(())
427}
428
429fn notify_all(cancels: &[RoomCancel]) {
430    for cancel in cancels {
431        cancel.notify_one();
432    }
433}
434
435async fn wait_for_shutdown_signal() {
436    #[cfg(unix)]
437    {
438        let mut term =
439            match tokio::signal::unix::signal(tokio::signal::unix::SignalKind::terminate()) {
440                Ok(term) => term,
441                Err(error) => {
442                    warn!(%error, "failed to install SIGTERM handler");
443                    let _ = tokio::signal::ctrl_c().await;
444                    return;
445                }
446            };
447        tokio::select! {
448            _ = tokio::signal::ctrl_c() => {}
449            _ = term.recv() => {}
450        }
451    }
452    #[cfg(not(unix))]
453    {
454        let _ = tokio::signal::ctrl_c().await;
455    }
456}
457
458async fn host_one_room(
459    mut config: Config,
460    label: Option<String>,
461    cancel: RoomCancel,
462    ready: Option<tokio::sync::oneshot::Sender<String>>,
463    sessions: Option<SessionRegistry>,
464    shell: Option<ShellBridgeHandle>,
465) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
466    if let Some(label) = &label {
467        config.username = format!("{}-{label}", config.username);
468        config.bot_username = format!("{}-{label}", config.bot_username);
469        config.room_name = format!("{} ({label})", config.room_name);
470    }
471
472    let _rooms_hosted = crate::metrics::RoomHostedGuard::new(if label.is_some() {
473        crate::metrics::PoolKind::Pod
474    } else {
475        crate::metrics::PoolKind::Solo
476    });
477
478    info!(
479        relay_url = %config.relay_url,
480        username = %config.username,
481        room_name = %config.room_name,
482        auto_start = config.auto_start,
483        engine_enabled = config.engine_enabled,
484        host_plays = config.host_plays,
485        bot_enabled = config.bot_enabled,
486        "starting room host"
487    );
488
489    let snapshot: SharedHostSnapshot = Arc::new(Mutex::new(HostSnapshot::default()));
490    let engine_session: SharedEngineSession = Arc::new(Mutex::new(None));
491    if let Some(sessions) = &sessions {
492        if let Ok(mut registered) = sessions.lock() {
493            registered.push(engine_session.clone());
494        }
495    }
496    let bot_state: SharedBotState = Arc::new(Mutex::new(Vec::new()));
497    let (outbound_tx, mut outbound_rx) = tokio_mpsc::unbounded_channel::<ClientMessage>();
498
499    let mut host =
500        RelayClient::connect(&config.relay_url, &config.username, &config.password).await?;
501    let mut room_id = establish_room(&mut host, &config, &snapshot).await?;
502
503    // Keep the sender alive without a shell, or `recv()` spins the select.
504    let (bridge, mut bridge_rx, _idle_bridge_tx) = match shell {
505        Some(handle) => (Some(handle.bridge), handle.commands, None),
506        None => {
507            let (tx, rx) = tokio_mpsc::unbounded_channel::<ShellCommand>();
508            (None, rx, Some(tx))
509        }
510    };
511    if let Some(bridge) = &bridge {
512        let reprime_snapshot = snapshot.clone();
513        let reprime_tx = outbound_tx.clone();
514        bridge.set_on_fallback(move |username| {
515            reprime_relay_cache(&reprime_snapshot, &reprime_tx, username);
516        });
517    }
518
519    if bridge.is_some() {
520        announce_webrtc_only(&config, &outbound_tx);
521    }
522
523    if let Some(ready) = ready {
524        let _ = ready.send(room_id.clone());
525    }
526
527    if config.bot_enabled {
528        spawn_bots(
529            &config,
530            std::slice::from_ref(&config.bot_deck),
531            &room_id,
532            &bot_state,
533        );
534    }
535
536    loop {
537        let exit = run_client_loop(
538            &mut host,
539            &config,
540            &room_id,
541            &engine_session,
542            &snapshot,
543            &bot_state,
544            &outbound_tx,
545            &mut outbound_rx,
546            &cancel,
547            bridge.as_ref(),
548            &mut bridge_rx,
549        )
550        .await;
551        if matches!(exit, LoopExit::Cancelled) {
552            stop_bots(&bot_state);
553            host.close().await;
554            return Ok(());
555        }
556
557        crate::metrics::record_relay_reconnect();
558        let mut attempt: usize = 0;
559        host = loop {
560            let delay = RECONNECT_BACKOFF_SECS[attempt.min(RECONNECT_BACKOFF_SECS.len() - 1)];
561            tokio::select! {
562                _ = cancel.notified() => {
563                    info!(username = %config.username, "room host cancelled while reconnecting");
564                    cancel_engine(&engine_session);
565                    stop_bots(&bot_state);
566                    return Ok(());
567                }
568                _ = time::sleep(Duration::from_secs(delay)) => {}
569            }
570            attempt += 1;
571            let mut client =
572                match RelayClient::connect(&config.relay_url, &config.username, &config.password)
573                    .await
574                {
575                    Ok(client) => client,
576                    Err(error) => {
577                        warn!(%error, attempt, "relay reconnect failed");
578                        continue;
579                    }
580                };
581            match reestablish_room(&mut client, &config, &engine_session, &snapshot).await {
582                Ok(new_room_id) => {
583                    if new_room_id != room_id {
584                        stop_bots(&bot_state);
585                        if config.bot_enabled {
586                            spawn_bots(
587                                &config,
588                                std::slice::from_ref(&config.bot_deck),
589                                &new_room_id,
590                                &bot_state,
591                            );
592                        }
593                        room_id = new_room_id;
594                    }
595                    break client;
596                }
597                Err(error) => {
598                    warn!(%error, attempt, "failed to re-establish room; retrying");
599                }
600            }
601        };
602        if let Some(bridge) = &bridge {
603            bridge.clear_game();
604            announce_webrtc_only(&config, &outbound_tx);
605        }
606        info!(username = %config.username, room_id, "relay connection re-established");
607    }
608}
609
610/// Re-primes the relay's replay cache for a seat leaving the plane.
611fn reprime_relay_cache(
612    snapshot: &SharedHostSnapshot,
613    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
614    username: &str,
615) {
616    let Some(index) = seat_index_of(snapshot, username) else {
617        return;
618    };
619    let slot = player_slot(index);
620    let Ok(snap) = snapshot.lock() else {
621        return;
622    };
623    let state = snap.last_state_by_slot.get(&slot).cloned();
624    let prompt = snap.pending_prompts.get(&slot).cloned();
625    drop(snap);
626
627    // Stored before `patch_against_last` runs, so these are full states.
628    for state in [state, prompt].into_iter().flatten() {
629        let _ = outbound_tx.send(ClientMessage::BroadcastState {
630            state,
631            target_player: Some(username.to_string()),
632        });
633    }
634    info!(
635        username,
636        "re-primed the relay cache for a seat that fell back"
637    );
638}
639
640fn announce_webrtc_only(config: &Config, outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>) {
641    let endpoint = manabrew_relay_protocol::TransportEndpoint {
642        endpoint_id: format!("webrtc:{}", config.username),
643        relay_url: None,
644        direct_addrs: Vec::new(),
645        kinds: vec![manabrew_relay_protocol::TRANSPORT_KIND_WEBRTC.to_string()],
646    };
647    info!(endpoint_id = %endpoint.endpoint_id, "announcing a webrtc-only endpoint");
648    let _ = outbound_tx.send(ClientMessage::AnnounceTransport {
649        endpoint: Some(endpoint),
650    });
651}
652
653async fn establish_room(
654    client: &mut RelayClient,
655    config: &Config,
656    snapshot: &SharedHostSnapshot,
657) -> Result<String, Box<dyn std::error::Error + Send + Sync>> {
658    let room_id = if let Some(room_id) = &config.room_id {
659        client
660            .send(&ClientMessage::JoinRoom {
661                room_id: room_id.clone(),
662                observe: !config.host_plays,
663                as_bot: false,
664                password: config.room_password.clone(),
665            })
666            .await?;
667        info!(room_id, "joining configured room");
668        room_id.clone()
669    } else {
670        client
671            .send(&ClientMessage::CreateRoom {
672                room_name: config.room_name.clone(),
673                max_players: config.max_players,
674                format: config.format.clone(),
675                protocol_version: PROTOCOL_VERSION,
676                hosted: !config.host_plays,
677                engine: engine_kind(config),
678                draft_config: None,
679                sealed_config: None,
680                official_key: config.official_key.clone(),
681                password: config.room_password.clone(),
682                reconnect_timeout_s: config.reconnect_timeout_s,
683            })
684            .await?;
685        info!(room_name = %config.room_name, "creating room");
686        wait_for_host_room(client, config, snapshot).await?
687    };
688
689    if config.host_plays {
690        seat_client(client, &config.host_deck).await?;
691    } else {
692        info!(
693            username = %config.username,
694            "hosting room without occupying a player seat"
695        );
696    }
697    Ok(room_id)
698}
699
700async fn reestablish_room(
701    client: &mut RelayClient,
702    config: &Config,
703    engine_session: &SharedEngineSession,
704    snapshot: &SharedHostSnapshot,
705) -> Result<String, Box<dyn std::error::Error + Send + Sync>> {
706    let resume = {
707        let snap = snapshot.lock().map_err(|error| error.to_string())?;
708        match (&snap.resume_token, &snap.game, &snap.room_info) {
709            (Some(token), Some(game), Some(room_info)) => {
710                Some(resume_room_request(config, token, game, room_info))
711            }
712            _ => None,
713        }
714    };
715    let game_active = engine_session
716        .lock()
717        .map(|guard| guard.is_some())
718        .unwrap_or(false);
719
720    if let (Some(request), true) = (resume, game_active) {
721        let room_id = request.room_id.clone();
722        client.send(&ClientMessage::ResumeRoom(request)).await?;
723        match wait_for_room_resumed(client, engine_session, snapshot).await? {
724            Some(room) => {
725                let (last_state, seat_states, prompts, player_order, pending_end_game) = {
726                    let mut snap = snapshot.lock().map_err(|error| error.to_string())?;
727                    snap.room_info = Some(room);
728                    (
729                        snap.last_state.clone(),
730                        snap.last_state_by_slot.clone(),
731                        snap.pending_prompts.clone(),
732                        snap.game
733                            .as_ref()
734                            .map(|game| game.player_order.clone())
735                            .unwrap_or_default(),
736                        snap.pending_end_game.clone(),
737                    )
738                };
739                if let Some(state) = last_state {
740                    client
741                        .send(&ClientMessage::BroadcastState {
742                            state,
743                            target_player: None,
744                        })
745                        .await?;
746                }
747                for (slot, state) in seat_states {
748                    let Some(target_player) =
749                        parse_player_slot(&slot).and_then(|index| player_order.get(index).cloned())
750                    else {
751                        warn!(slot, "cannot replay private state without a mapped player");
752                        continue;
753                    };
754                    client
755                        .send(&ClientMessage::BroadcastState {
756                            state,
757                            target_player: Some(target_player),
758                        })
759                        .await?;
760                }
761                for (slot, prompt) in prompts {
762                    let Some(target_player) =
763                        parse_player_slot(&slot).and_then(|index| player_order.get(index).cloned())
764                    else {
765                        warn!(slot, "cannot replay prompt without a mapped player");
766                        continue;
767                    };
768                    client
769                        .send(&ClientMessage::BroadcastState {
770                            state: prompt,
771                            target_player: Some(target_player),
772                        })
773                        .await?;
774                }
775                if let Some(game_id) = pending_end_game {
776                    client.send(&ClientMessage::EndGame { game_id }).await?;
777                }
778                info!(room_id, "room resumed; snapshot re-broadcast");
779                return Ok(room_id);
780            }
781            None => {
782                warn!(room_id, "room resume rejected; abandoning hosted game");
783                abort_engine_session(engine_session);
784            }
785        }
786    }
787
788    establish_room(client, config, snapshot).await
789}
790
791fn engine_kind(config: &Config) -> EngineKind {
792    if matches!(config.backend, EngineBackendKind::Forge) {
793        EngineKind::Forge
794    } else {
795        EngineKind::Manabrew
796    }
797}
798
799fn resume_room_request(
800    config: &Config,
801    token: &str,
802    game: &GameStart,
803    room_info: &RoomInfo,
804) -> ResumeRoomRequest {
805    ResumeRoomRequest {
806        room_id: room_info.room_id.clone(),
807        resume_token: token.to_string(),
808        room_name: room_info.room_name.clone(),
809        max_players: room_info.max_players,
810        format: room_info.format.clone(),
811        hosted: !config.host_plays,
812        engine: engine_kind(config),
813        official_key: config.official_key.clone(),
814        password: config.room_password.clone(),
815        reconnect_timeout_s: Some(room_info.reconnect_timeout_s),
816        draft_config: room_info.draft_config.clone(),
817        sealed_config: room_info.sealed_config.clone(),
818        player_order: game.player_order.clone(),
819        player_decks: game.player_decks.clone(),
820        starting_life: game.starting_life,
821        bot_players: room_info
822            .players
823            .iter()
824            .filter(|player| player.is_bot)
825            .map(|player| player.username.clone())
826            .collect(),
827        game_id: game.game_id.clone(),
828    }
829}
830
831async fn wait_for_room_resumed(
832    client: &mut RelayClient,
833    engine_session: &SharedEngineSession,
834    snapshot: &SharedHostSnapshot,
835) -> Result<Option<RoomInfo>, Box<dyn std::error::Error + Send + Sync>> {
836    loop {
837        match client.recv().await? {
838            Some(ServerMessage::RoomResumed { room }) => {
839                if room.status == RoomStatus::Lobby {
840                    abort_stale_engine_session(
841                        engine_session,
842                        snapshot,
843                        "resumed room is no longer in game",
844                    );
845                } else {
846                    concede_missing_active_seats(engine_session, snapshot, &room);
847                }
848                return Ok(Some(room));
849            }
850            Some(ServerMessage::RoomUpdate { room }) => {
851                if room.status == RoomStatus::Lobby {
852                    abort_stale_engine_session(
853                        engine_session,
854                        snapshot,
855                        "room reset while reconnecting",
856                    );
857                } else {
858                    concede_missing_active_seats(engine_session, snapshot, &room);
859                }
860                if let Ok(mut snap) = snapshot.lock() {
861                    snap.room_info = Some(room);
862                }
863            }
864            Some(ServerMessage::StateUpdate { from_player, state }) => {
865                let Ok(envelope) = serde_json::from_value::<StateEnvelope>(state.clone()) else {
866                    continue;
867                };
868                match envelope {
869                    StateEnvelope::Response { .. } => {
870                        route_remote_response(engine_session, snapshot, &from_player, &state);
871                    }
872                    StateEnvelope::Directive {
873                        from_player: claimed_slot,
874                        directive,
875                    } => route_remote_directive(
876                        engine_session,
877                        snapshot,
878                        &from_player,
879                        &claimed_slot,
880                        &directive,
881                    ),
882                    _ => {}
883                }
884            }
885            Some(ServerMessage::Error { code, message }) => {
886                warn!(code, message, "resume rejected by relay");
887                return Ok(None);
888            }
889            Some(other) => debug!(?other, "ignored message while waiting for room resume"),
890            None => return Err("relay closed while waiting for room resume".into()),
891        }
892    }
893}
894
895fn cancel_engine(engine_session: &SharedEngineSession) {
896    if let Ok(guard) = engine_session.lock() {
897        if let Some(EngineSession::Forge { cancel, .. }) = guard.as_ref() {
898            cancel.store(true, Ordering::Relaxed);
899        }
900    }
901}
902
903fn abort_engine_session(engine_session: &SharedEngineSession) {
904    cancel_engine(engine_session);
905    // Dropping the response channels makes the Manabrew backend's transports
906    // observe a disconnect and concede, which ends the engine thread.
907    clear_engine_session(engine_session);
908}
909
910fn abort_stale_engine_session(
911    engine_session: &SharedEngineSession,
912    snapshot: &SharedHostSnapshot,
913    reason: &str,
914) {
915    let stale_game_id = engine_session
916        .lock()
917        .ok()
918        .and_then(|guard| guard.as_ref().map(|s| s.game_id().to_string()));
919    let Some(stale_game_id) = stale_game_id else {
920        return;
921    };
922    warn!(game_id = %stale_game_id, reason, "aborting stale engine session");
923    abort_engine_session(engine_session);
924    clear_game_snapshot(snapshot, &stale_game_id);
925}
926
927fn clear_game_snapshot(snapshot: &SharedHostSnapshot, game_id: &str) {
928    if let Ok(mut snap) = snapshot.lock() {
929        if snap
930            .game
931            .as_ref()
932            .is_some_and(|game| game.game_id == game_id)
933        {
934            snap.game = None;
935            snap.last_state = None;
936            snap.last_state_by_slot.clear();
937            snap.pending_prompts.clear();
938            snap.pending_end_game = None;
939        }
940    }
941}
942
943/// A game must have at least one human seat. Seats survive disconnects and
944/// concessions (reconnect grace / spectating), but an explicit leave removes
945/// the seat for good — once every human seat is gone the game has no
946/// stakeholders and ends now, instead of bots playing it out until the
947/// relay's humanless sweep reclaims the room minutes later.
948fn end_game_without_humans(
949    engine_session: &SharedEngineSession,
950    snapshot: &SharedHostSnapshot,
951    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
952    room: &RoomInfo,
953) {
954    if room.players.iter().any(|seat| !seat.is_bot) {
955        return;
956    }
957    let game_id = engine_session
958        .lock()
959        .ok()
960        .and_then(|guard| guard.as_ref().map(|s| s.game_id().to_string()));
961    let Some(game_id) = game_id else {
962        return;
963    };
964    info!(game_id, "no human seats remain; ending hosted game");
965    abort_engine_session(engine_session);
966    clear_game_snapshot(snapshot, &game_id);
967    let _ = outbound_tx.send(ClientMessage::EndGame { game_id });
968}
969
970fn spawn_bots(config: &Config, decks: &[DeckSelection], room_id: &str, bot_state: &SharedBotState) {
971    stop_bots(bot_state);
972    let mut guard = match bot_state.lock() {
973        Ok(guard) => guard,
974        Err(error) => {
975            warn!(%error, "bot state lock poisoned");
976            return;
977        }
978    };
979    let open_seats = config.max_players.saturating_sub(1) as usize;
980    for (index, deck) in decks.iter().take(open_seats).enumerate() {
981        let username = if index == 0 {
982            config.bot_username.clone()
983        } else {
984            format!("{} {}", config.bot_username, index + 1)
985        };
986        let relay_url = config.relay_url.clone();
987        let bot_config = BotConfig {
988            username,
989            password: config.password.clone(),
990            room_id: room_id.to_string(),
991            room_password: config.room_password.clone(),
992            deck_name: deck.name.clone(),
993            deck: deck.deck.clone(),
994            commander_name: deck.commander_name.clone(),
995            agent: AgentKind::Simple,
996            answer_delay_ms: None,
997        };
998        let shutdown = Arc::new(tokio::sync::Notify::new());
999        let bot_shutdown = shutdown.clone();
1000        let handle = tokio::spawn(async move {
1001            if let Err(error) = run_bot(relay_url, bot_config, bot_shutdown).await {
1002                error!(%error, "bot task exited");
1003            }
1004        });
1005        guard.push(BotState { handle, shutdown });
1006    }
1007}
1008
1009fn stop_bots(bot_state: &SharedBotState) {
1010    match bot_state.lock() {
1011        Ok(mut state) => {
1012            for bot in state.drain(..) {
1013                bot.shutdown.notify_one();
1014                let mut handle = bot.handle;
1015                tokio::spawn(async move {
1016                    if time::timeout(BOT_STOP_TIMEOUT, &mut handle).await.is_err() {
1017                        handle.abort();
1018                    }
1019                });
1020            }
1021        }
1022        Err(error) => warn!(%error, "bot state lock poisoned"),
1023    }
1024}
1025
1026async fn wait_for_host_room(
1027    client: &mut RelayClient,
1028    config: &Config,
1029    snapshot: &SharedHostSnapshot,
1030) -> Result<String, Box<dyn std::error::Error + Send + Sync>> {
1031    loop {
1032        match client.recv().await? {
1033            Some(ServerMessage::RoomCreated {
1034                room_id,
1035                room_name,
1036                room,
1037                resume_token,
1038            }) => {
1039                info!(room_id, room_name, "room created");
1040                if let Ok(mut snap) = snapshot.lock() {
1041                    snap.resume_token = resume_token;
1042                    snap.room_info = Some(room);
1043                }
1044                return Ok(room_id);
1045            }
1046            Some(ServerMessage::RoomUpdate { room }) if room.host == config.username => {
1047                info!(room_id = %room.room_id, room_name = %room.room_name, "host room update");
1048                return Ok(room.room_id);
1049            }
1050            Some(ServerMessage::AuthResult { success, error, .. }) => {
1051                if !success {
1052                    return Err(format!("authentication failed: {error:?}").into());
1053                }
1054            }
1055            Some(ServerMessage::Error { code, message }) => {
1056                return Err(format!("server error while creating room: {code}: {message}").into());
1057            }
1058            Some(other) => debug!(?other, "ignored message while waiting for room creation"),
1059            None => return Err("relay closed while waiting for room creation".into()),
1060        }
1061    }
1062}
1063
1064async fn seat_client(
1065    client: &mut RelayClient,
1066    deck: &DeckSelection,
1067) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
1068    client
1069        .send(&ClientMessage::SetDeckSelection {
1070            deck_name: deck.name.clone(),
1071            deck: deck.deck.clone(),
1072            published_deck_id: None,
1073            commander_name: deck.commander_name.clone(),
1074            avatar_url: None,
1075        })
1076        .await?;
1077    client
1078        .send(&ClientMessage::SetReady { ready: true })
1079        .await?;
1080    info!(
1081        username = %client.username,
1082        deck = %deck.name,
1083        cards = deck.deck.cards.len(),
1084        "selected deck and marked ready"
1085    );
1086    Ok(())
1087}
1088
1089async fn run_client_loop(
1090    client: &mut RelayClient,
1091    config: &Config,
1092    room_id: &str,
1093    engine_session: &SharedEngineSession,
1094    snapshot: &SharedHostSnapshot,
1095    bot_state: &SharedBotState,
1096    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
1097    outbound_rx: &mut tokio_mpsc::UnboundedReceiver<ClientMessage>,
1098    cancel: &RoomCancel,
1099    bridge: Option<&Arc<ShellBridge>>,
1100    bridge_rx: &mut tokio_mpsc::UnboundedReceiver<ShellCommand>,
1101) -> LoopExit {
1102    let mut heartbeat = time::interval(Duration::from_secs(30));
1103    heartbeat.set_missed_tick_behavior(time::MissedTickBehavior::Delay);
1104    let mut drain_tick = time::interval(Duration::from_secs(5));
1105    drain_tick.set_missed_tick_behavior(time::MissedTickBehavior::Delay);
1106    let mut bot_usernames: HashSet<String> = HashSet::new();
1107
1108    loop {
1109        tokio::select! {
1110            _ = cancel.notified() => {
1111                info!(username = %client.username, "room host cancelled; shutting down");
1112                cancel_engine(engine_session);
1113                return LoopExit::Cancelled;
1114            }
1115            _ = drain_tick.tick() => {
1116                let idle = engine_session
1117                    .lock()
1118                    .map(|guard| guard.is_none())
1119                    .unwrap_or(false);
1120                if draining() && idle {
1121                    info!(username = %client.username, "draining and no active game; closing room");
1122                    return LoopExit::Cancelled;
1123                }
1124            }
1125            _ = heartbeat.tick() => {
1126                if let Err(error) = client.broadcast_room_message(SELF_HOSTED_NODE_PROTOCOL, json!({
1127                    "type": "heartbeat",
1128                    "node": client.username,
1129                    "capacity": config.max_games,
1130                })).await {
1131                    warn!(%error, username = %client.username, "relay send failed");
1132                    return LoopExit::Disconnected;
1133                }
1134            }
1135            outbound = outbound_rx.recv() => {
1136                let Some(outbound) = outbound else {
1137                    warn!(username = %client.username, "outbound channel closed");
1138                    return LoopExit::Cancelled;
1139                };
1140                if let ClientMessage::BroadcastState { state, target_player: Some(target) } = &outbound {
1141                    if bridge.is_some_and(|shell| shell.try_send(target, state)) {
1142                        continue;
1143                    }
1144                }
1145                if let Err(error) = client.send(&outbound).await {
1146                    warn!(%error, username = %client.username, "relay send failed");
1147                    return LoopExit::Disconnected;
1148                }
1149            }
1150            command = bridge_rx.recv() => {
1151                let Some(command) = command else {
1152                    warn!(username = %client.username, "shell bridge closed");
1153                    return LoopExit::Cancelled;
1154                };
1155                let Some(shell) = bridge else { continue };
1156                match command {
1157                    ShellCommand::Serving { seats } => shell.set_serving(seats),
1158                    ShellCommand::Signal { to, payload } => {
1159                        if let Err(error) = client
1160                            .send(&ClientMessage::SignalPeer { to, payload })
1161                            .await
1162                        {
1163                            warn!(%error, username = %client.username, "relay send failed");
1164                            return LoopExit::Disconnected;
1165                        }
1166                    }
1167                    ShellCommand::SeatEnvelope { from, envelope } => {
1168                        // `from` came from a relay-attested roster.
1169                        route_seat_envelope(engine_session, snapshot, &from, &envelope);
1170                    }
1171                }
1172            }
1173            message = client.recv() => {
1174                let message = match message {
1175                    Ok(Some(message)) => message,
1176                    Ok(None) => {
1177                        warn!(username = %client.username, "relay websocket closed");
1178                        return LoopExit::Disconnected;
1179                    }
1180                    Err(error) => {
1181                        warn!(%error, username = %client.username, "relay receive failed");
1182                        return LoopExit::Disconnected;
1183                    }
1184                };
1185                if let Err(error) = handle_server_message(
1186                    client,
1187                    config,
1188                    room_id,
1189                    engine_session,
1190                    snapshot,
1191                    bot_state,
1192                    outbound_tx,
1193                    &mut bot_usernames,
1194                    bridge,
1195                    message,
1196                ).await {
1197                    warn!(%error, username = %client.username, "relay send failed");
1198                    return LoopExit::Disconnected;
1199                }
1200            }
1201        }
1202    }
1203}
1204
1205async fn handle_server_message(
1206    client: &mut RelayClient,
1207    config: &Config,
1208    room_id: &str,
1209    engine_session: &SharedEngineSession,
1210    snapshot: &SharedHostSnapshot,
1211    bot_state: &SharedBotState,
1212    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
1213    bot_usernames: &mut HashSet<String>,
1214    bridge: Option<&Arc<ShellBridge>>,
1215    message: ServerMessage,
1216) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
1217    match message {
1218        ServerMessage::RoomUpdate { room } => {
1219            if let Ok(mut snap) = snapshot.lock() {
1220                snap.room_info = Some(room.clone());
1221            }
1222            *bot_usernames = room
1223                .players
1224                .iter()
1225                .filter(|p| p.is_bot)
1226                .map(|p| p.username.clone())
1227                .collect();
1228            log_room_update(&client.username, &room);
1229            if room.status == RoomStatus::Lobby {
1230                abort_stale_engine_session(engine_session, snapshot, "room reset to lobby");
1231            } else {
1232                end_game_without_humans(engine_session, snapshot, outbound_tx, &room);
1233            }
1234            maybe_auto_start_room(client, config, &room).await?;
1235        }
1236        ServerMessage::StateUpdate { from_player, state } => {
1237            if let Some(shell) = bridge {
1238                shell.note_relay_message(&from_player);
1239            }
1240            handle_state_update(
1241                client,
1242                config,
1243                room_id,
1244                engine_session,
1245                snapshot,
1246                bot_state,
1247                from_player,
1248                state,
1249            )
1250            .await?;
1251        }
1252        ServerMessage::PeerSignal { from, payload } => match bridge {
1253            Some(shell) => shell.forward_signal(&from, payload),
1254            None => debug!(from, "signalling arrived with no shell to hand it to"),
1255        },
1256        ServerMessage::ReadyStateChanged { username, ready } => {
1257            info!(username, ready, observer = %client.username, "ready changed");
1258        }
1259        ServerMessage::PlayerJoined { username, room_id } => {
1260            info!(username, room_id, observer = %client.username, "player joined");
1261        }
1262        ServerMessage::PlayerLeft { username, room_id } => {
1263            info!(username, room_id, observer = %client.username, "player left");
1264            let leaver_still_needed =
1265                active_player_usernames(snapshot).is_some_and(|active| active.contains(&username));
1266            if leaver_still_needed {
1267                if let Some(index) = seat_index_of(snapshot, &username) {
1268                    info!(
1269                        username,
1270                        index, "player abandoned mid-game; conceding their seat"
1271                    );
1272                    concede_seat(engine_session, index);
1273                }
1274            }
1275        }
1276        ServerMessage::GameStarted {
1277            room_id,
1278            game_id,
1279            player_order,
1280            player_decks,
1281            starting_life,
1282        } => {
1283            info!(room_id, game_id, ?player_order, observer = %client.username, "game started");
1284            abort_stale_engine_session(engine_session, snapshot, "relay started a new game");
1285            if let Ok(mut snap) = snapshot.lock() {
1286                snap.game = Some(GameStart {
1287                    game_id: game_id.clone(),
1288                    player_order: player_order.clone(),
1289                    player_decks: player_decks.clone(),
1290                    starting_life,
1291                });
1292                snap.last_state = None;
1293                snap.last_state_by_slot.clear();
1294                snap.pending_prompts.clear();
1295                snap.pending_end_game = None;
1296            }
1297            let mut left_the_relay = Vec::new();
1298            if let Some(shell) = bridge {
1299                let seats = shell.freeze_for_game(&player_order, &[]);
1300                if !seats.is_empty() {
1301                    info!(
1302                        game_id,
1303                        ?seats,
1304                        "seats playing this game through the webview"
1305                    );
1306                    left_the_relay.extend(shell.transport_report(&seats));
1307                }
1308            }
1309            if !left_the_relay.is_empty() {
1310                let _ = outbound_tx.send(ClientMessage::ReportTransport {
1311                    game_id: game_id.clone(),
1312                    seats: left_the_relay,
1313                });
1314            }
1315            maybe_start_hosted_engine(
1316                config,
1317                engine_session,
1318                snapshot,
1319                outbound_tx,
1320                game_id,
1321                player_order,
1322                player_decks,
1323                starting_life,
1324                bot_usernames,
1325            );
1326        }
1327        ServerMessage::RoomTransport { members, .. } => {
1328            if let Some(shell) = bridge {
1329                // Empty until every seat has opted in.
1330                shell.set_roster(members.iter().map(|member| member.username.clone()));
1331            }
1332        }
1333        ServerMessage::ServerShuttingDown { reconnect_in_s } => {
1334            info!(reconnect_in_s, observer = %client.username, "relay is restarting");
1335        }
1336        ServerMessage::Error { code, message } => {
1337            warn!(code, message, observer = %client.username, "server error");
1338        }
1339        other => {
1340            debug!(?other, observer = %client.username, "server message");
1341        }
1342    }
1343
1344    Ok(())
1345}
1346
1347async fn handle_state_update(
1348    client: &mut RelayClient,
1349    config: &Config,
1350    room_id: &str,
1351    engine_session: &SharedEngineSession,
1352    snapshot: &SharedHostSnapshot,
1353    bot_state: &SharedBotState,
1354    from_player: String,
1355    state: Value,
1356) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
1357    let Ok(envelope) = serde_json::from_value::<StateEnvelope>(state.clone()) else {
1358        debug!(from_player, state = %state, observer = %client.username, "state update");
1359        return Ok(());
1360    };
1361
1362    match envelope {
1363        StateEnvelope::Response { .. } | StateEnvelope::Directive { .. } => {
1364            route_seat_envelope(engine_session, snapshot, &from_player, &state);
1365            Ok(())
1366        }
1367        StateEnvelope::RoomRelay {
1368            protocol,
1369            from_player: _,
1370            room_id: requested_room_id,
1371            payload,
1372            ..
1373        } => {
1374            info!(
1375                from_player,
1376                protocol,
1377                payload = %payload,
1378                observer = %client.username,
1379                "room relay message"
1380            );
1381            if protocol != SELF_HOSTED_NODE_PROTOCOL {
1382                return Ok(());
1383            }
1384            let payload_type = payload.get("type").and_then(Value::as_str);
1385            match payload_type {
1386                Some("removeBot") => {
1387                    info!(observer = %client.username, "received removeBot request");
1388                    stop_bots(bot_state);
1389                }
1390                Some("spawnBot") => {
1391                    let effective_room_id = requested_room_id.as_deref().unwrap_or(room_id);
1392                    if effective_room_id == room_id {
1393                        info!(observer = %client.username, room_id, "received spawnBot request");
1394                        let bot_decks = bot_decks_from_payload(config, &payload);
1395                        spawn_bots(config, &bot_decks, room_id, bot_state);
1396                    } else {
1397                        debug!(
1398                            observer = %client.username,
1399                            current_room_id = room_id,
1400                            requested_room_id = effective_room_id,
1401                            "ignoring spawnBot request for another room"
1402                        );
1403                    }
1404                }
1405                _ => {}
1406            }
1407            Ok(())
1408        }
1409        _ => {
1410            debug!(from_player, state = %state, observer = %client.username, "state update");
1411            Ok(())
1412        }
1413    }
1414}
1415
1416fn bot_decks_from_payload(config: &Config, payload: &Value) -> Vec<DeckSelection> {
1417    let parsed = serde_json::from_value::<SpawnBotPayload>(payload.clone()).ok();
1418    let requested = match parsed {
1419        Some(SpawnBotPayload {
1420            decks: Some(decks), ..
1421        }) if !decks.is_empty() => decks,
1422        Some(SpawnBotPayload {
1423            deck: Some(deck), ..
1424        }) => vec![deck],
1425        _ => return vec![config.bot_deck.clone()],
1426    };
1427    requested
1428        .into_iter()
1429        .map(deck_selection_from_payload)
1430        .collect()
1431}
1432
1433fn deck_selection_from_payload(deck: SpawnBotDeckPayload) -> DeckSelection {
1434    info!(
1435        deck = %deck.deck_name,
1436        cards = deck.deck.cards.len(),
1437        commander = ?deck.commander_name,
1438        "using requested bot deck"
1439    );
1440    let commander_name = deck.commander_name.or_else(|| {
1441        deck.deck
1442            .commanders
1443            .as_ref()
1444            .and_then(|commanders| commanders.first())
1445            .map(|commander| commander.identity.name.clone())
1446    });
1447    DeckSelection {
1448        name: deck.deck_name,
1449        deck: deck.deck,
1450        commander_name,
1451    }
1452}
1453
1454async fn maybe_auto_start_room(
1455    client: &mut RelayClient,
1456    config: &Config,
1457    room: &RoomInfo,
1458) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
1459    if !config.auto_start || draining() {
1460        return Ok(());
1461    }
1462    if config.format == GameFormat::Any {
1463        return Ok(());
1464    }
1465    if room.host != config.username
1466        || room.status != RoomStatus::Lobby
1467        || room.players.len() < config.max_players as usize
1468    {
1469        return Ok(());
1470    }
1471    let all_ready = room
1472        .players
1473        .iter()
1474        .all(|player| player.connected && player.ready && player.selected_deck_name.is_some());
1475    if all_ready {
1476        info!(
1477            room_id = %room.room_id,
1478            players = room.players.len(),
1479            "all players ready; auto-starting hosted game"
1480        );
1481        client
1482            .send(&ClientMessage::StartGame { format: None })
1483            .await?;
1484    }
1485    Ok(())
1486}
1487
1488const ENGINE_STACK_BYTES: usize = 64 * 1024 * 1024;
1489
1490fn spawn_engine_thread<F: FnOnce() + Send + 'static>(body: F) {
1491    if let Err(error) = thread::Builder::new()
1492        .name("hosted-engine".to_string())
1493        .stack_size(ENGINE_STACK_BYTES)
1494        .spawn(body)
1495    {
1496        error!(%error, "failed to spawn hosted engine thread");
1497    }
1498}
1499
1500// Any stays commander-capable: hosted "Any" rooms resolve their real format
1501// at StartGame, and dropping commanders there would break commander games.
1502fn is_commander_variant(format: GameFormat) -> bool {
1503    match format {
1504        GameFormat::Any | GameFormat::Commander | GameFormat::Brawl | GameFormat::Oathbreaker => {
1505            true
1506        }
1507        GameFormat::Standard
1508        | GameFormat::Pioneer
1509        | GameFormat::Modern
1510        | GameFormat::Legacy
1511        | GameFormat::Vintage
1512        | GameFormat::Pauper
1513        | GameFormat::Premodern
1514        | GameFormat::Draft
1515        | GameFormat::Sealed => false,
1516    }
1517}
1518
1519// Forge GameType name for the java backend; empty = let the adapter infer
1520// commander-ness the pre-variant way (Any rooms carry no concrete format here).
1521fn java_game_variant(format: GameFormat) -> &'static str {
1522    match format {
1523        GameFormat::Any => "",
1524        GameFormat::Commander => "Commander",
1525        GameFormat::Brawl => "Brawl",
1526        GameFormat::Oathbreaker => "Oathbreaker",
1527        GameFormat::Standard
1528        | GameFormat::Pioneer
1529        | GameFormat::Modern
1530        | GameFormat::Legacy
1531        | GameFormat::Vintage
1532        | GameFormat::Pauper
1533        | GameFormat::Premodern
1534        | GameFormat::Draft
1535        | GameFormat::Sealed => "Constructed",
1536    }
1537}
1538
1539fn maybe_start_hosted_engine(
1540    config: &Config,
1541    engine_session: &SharedEngineSession,
1542    snapshot: &SharedHostSnapshot,
1543    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
1544    game_id: String,
1545    player_order: Vec<String>,
1546    player_decks: Vec<PlayerDeckInfo>,
1547    starting_life: i32,
1548    bot_usernames: &HashSet<String>,
1549) {
1550    if !config.engine_enabled {
1551        debug!("hosted engine disabled for this node");
1552        return;
1553    }
1554    let backend = config.backend;
1555    if !backend.is_supported() {
1556        if let Err(error) = java_backend::JavaRuntimeConfig::from_env().validate() {
1557            warn!(
1558                backend = backend.label(),
1559                %error,
1560                "hosted engine backend runtime is not ready"
1561            );
1562            return;
1563        }
1564        warn!(
1565            backend = backend.label(),
1566            message = java_backend::unsupported_message(),
1567            "hosted engine backend is not implemented yet"
1568        );
1569        return;
1570    }
1571
1572    let local_player_index = if config.host_plays {
1573        match player_order
1574            .iter()
1575            .position(|name| name == &config.username)
1576        {
1577            Some(index) => Some(index),
1578            None => {
1579                warn!(
1580                    username = %config.username,
1581                    ?player_order,
1582                    "room node is configured as a player but is not in player order; not starting engine"
1583                );
1584                return;
1585            }
1586        }
1587    } else {
1588        None
1589    };
1590
1591    let mut guard = match engine_session.lock() {
1592        Ok(guard) => guard,
1593        Err(error) => {
1594            warn!(%error, "engine session lock poisoned");
1595            return;
1596        }
1597    };
1598    if let Some(session) = guard.as_ref() {
1599        warn!(
1600            game_id,
1601            stale_game_id = session.game_id(),
1602            "engine session still present at game start; not starting engine"
1603        );
1604        return;
1605    }
1606
1607    let session_handle = engine_session.clone();
1608    let num_players = player_order.len();
1609    if num_players < 2 {
1610        warn!(num_players, "not enough players to start hosted engine");
1611        return;
1612    }
1613
1614    let mut deck_map: HashMap<String, PlayerDeckInfo> = player_decks
1615        .into_iter()
1616        .map(|deck| (deck.username.clone(), deck))
1617        .collect();
1618    let mut ordered_decks = Vec::with_capacity(num_players);
1619    let mut commander_names = Vec::with_capacity(num_players);
1620    let mut ai_player_indices = Vec::new();
1621    for (index, username) in player_order.iter().enumerate() {
1622        let Some(deck) = deck_map.remove(username) else {
1623            warn!(username, "missing deck for player; not starting engine");
1624            return;
1625        };
1626        if config.forge_ai && bot_usernames.contains(username) {
1627            ai_player_indices.push(index);
1628        }
1629        ordered_decks.push(deck.deck);
1630        commander_names.push(deck.commander_name);
1631    }
1632
1633    let player_names = player_order;
1634    let room_format = snapshot
1635        .lock()
1636        .ok()
1637        .and_then(|snap| snap.room_info.as_ref().map(|room| room.format.clone()))
1638        .unwrap_or(GameFormat::Any);
1639    let commander_variant = is_commander_variant(room_format.clone());
1640    let game_variant = java_game_variant(room_format).to_string();
1641
1642    match backend {
1643        EngineBackendKind::Manabrew => {
1644            let (remote_prompt_tx, remote_prompt_rx) = std_mpsc::channel::<(usize, AgentMessage)>();
1645            let mut remote_response_txs = HashMap::new();
1646            let mut remote_response_rxs = Vec::new();
1647            for i in 0..num_players {
1648                if Some(i) == local_player_index {
1649                    continue;
1650                }
1651                let (response_tx, response_rx) = std_mpsc::channel::<ClientToServerMessage>();
1652                remote_response_txs.insert(i, response_tx);
1653                remote_response_rxs.push((i, response_rx));
1654            }
1655            let engine_clock = EngineClock::default();
1656            *guard = Some(EngineSession::Manabrew {
1657                game_id: game_id.clone(),
1658                remote_response_txs,
1659                engine_clock: engine_clock.clone(),
1660            });
1661            drop(guard);
1662
1663            spawn_remote_prompt_forwarder(
1664                outbound_tx.clone(),
1665                snapshot.clone(),
1666                engine_session.clone(),
1667                game_id.clone(),
1668                remote_prompt_rx,
1669                Some(player_names.clone()),
1670                config.state_delta,
1671                engine_clock,
1672            );
1673            let (game_over_tx, game_over_rx) = std_mpsc::channel::<HostedGameOver>();
1674            spawn_game_over_forwarder(
1675                outbound_tx.clone(),
1676                game_over_rx,
1677                engine_session.clone(),
1678                snapshot.clone(),
1679                game_id.clone(),
1680                Some(player_names.clone()),
1681            );
1682            let outbound_tx = outbound_tx.clone();
1683            let snapshot = snapshot.clone();
1684            spawn_engine_thread(move || {
1685                info!(
1686                    game_id,
1687                    backend = backend.label(),
1688                    players = num_players,
1689                    local_player_index,
1690                    "starting hosted engine thread"
1691                );
1692                crate::metrics::record_engine_session_started();
1693                let started = Instant::now();
1694                let result = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
1695                    rust_backend::run_hosted_engine_game(
1696                        game_id.clone(),
1697                        player_names,
1698                        ordered_decks,
1699                        commander_names,
1700                        local_player_index,
1701                        starting_life,
1702                        remote_prompt_tx,
1703                        remote_response_rxs,
1704                        game_over_tx,
1705                    )
1706                }));
1707                finish_hosted_engine(
1708                    result,
1709                    &game_id,
1710                    num_players,
1711                    started,
1712                    &outbound_tx,
1713                    &session_handle,
1714                    &snapshot,
1715                );
1716            });
1717        }
1718        EngineBackendKind::Forge => {
1719            let (remote_prompt_tx, remote_prompt_rx) = std_mpsc::channel::<(usize, AgentMessage)>();
1720            let mut remote_response_txs = HashMap::new();
1721            let mut remote_response_rxs = Vec::new();
1722            for i in 0..num_players {
1723                if Some(i) == local_player_index {
1724                    continue;
1725                }
1726                let (response_tx, response_rx) = std_mpsc::channel::<ClientToServerMessage>();
1727                remote_response_txs.insert(i, response_tx);
1728                remote_response_rxs.push((i, response_rx));
1729            }
1730            let cancel = Arc::new(AtomicBool::new(false));
1731            let engine_clock = EngineClock::default();
1732            *guard = Some(EngineSession::Forge {
1733                game_id: game_id.clone(),
1734                remote_response_txs,
1735                cancel: cancel.clone(),
1736                engine_clock: engine_clock.clone(),
1737            });
1738            drop(guard);
1739
1740            spawn_remote_prompt_forwarder(
1741                outbound_tx.clone(),
1742                snapshot.clone(),
1743                engine_session.clone(),
1744                game_id.clone(),
1745                remote_prompt_rx,
1746                Some(player_names.clone()),
1747                config.state_delta,
1748                engine_clock,
1749            );
1750            let (game_over_tx, game_over_rx) = std_mpsc::channel::<HostedGameOver>();
1751            spawn_game_over_forwarder(
1752                outbound_tx.clone(),
1753                game_over_rx,
1754                engine_session.clone(),
1755                snapshot.clone(),
1756                game_id.clone(),
1757                Some(player_names.clone()),
1758            );
1759            let outbound_tx = outbound_tx.clone();
1760            let snapshot = snapshot.clone();
1761            spawn_engine_thread(move || {
1762                info!(
1763                    game_id,
1764                    backend = backend.label(),
1765                    players = num_players,
1766                    local_player_index,
1767                    "starting hosted engine thread"
1768                );
1769                crate::metrics::record_engine_session_started();
1770                let started = Instant::now();
1771                let result = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
1772                    java_backend::run_hosted_engine_game(
1773                        game_id.clone(),
1774                        player_names,
1775                        ordered_decks,
1776                        commander_names,
1777                        commander_variant,
1778                        game_variant,
1779                        local_player_index,
1780                        ai_player_indices,
1781                        starting_life,
1782                        remote_prompt_tx,
1783                        remote_response_rxs,
1784                        game_over_tx,
1785                        cancel,
1786                    )
1787                }));
1788                finish_hosted_engine(
1789                    result,
1790                    &game_id,
1791                    num_players,
1792                    started,
1793                    &outbound_tx,
1794                    &session_handle,
1795                    &snapshot,
1796                );
1797            });
1798        }
1799    }
1800}
1801
1802fn clear_engine_session(engine_session: &SharedEngineSession) {
1803    match engine_session.lock() {
1804        Ok(mut guard) => *guard = None,
1805        Err(error) => warn!(%error, "engine session lock poisoned on reset"),
1806    }
1807}
1808
1809/// Usernames of engine players the game still needs — seated in the current
1810/// game and not yet eliminated per the last state broadcast. Everyone else
1811/// (eliminated players, non-playing members) is effectively a spectator whose
1812/// absence never matters. `None` = no hosted game.
1813fn active_player_usernames(snapshot: &SharedHostSnapshot) -> Option<HashSet<String>> {
1814    let snap = snapshot.lock().ok()?;
1815    let game = snap.game.as_ref()?;
1816    let mut active: HashSet<String> = game.player_order.iter().cloned().collect();
1817    let players = snap
1818        .last_state
1819        .as_ref()
1820        .and_then(|state| state.pointer("/state/gameView/players"))
1821        .and_then(Value::as_array);
1822    if let Some(players) = players {
1823        for player in players {
1824            let playing = player
1825                .get("status")
1826                .and_then(Value::as_str)
1827                .is_none_or(|status| status == "playing");
1828            if playing {
1829                continue;
1830            }
1831            let index = player
1832                .get("id")
1833                .and_then(Value::as_str)
1834                .and_then(parse_player_slot);
1835            if let Some(username) = index.and_then(|i| game.player_order.get(i)) {
1836                active.remove(username);
1837            }
1838        }
1839    }
1840    Some(active)
1841}
1842
1843fn seat_index_of(snapshot: &SharedHostSnapshot, username: &str) -> Option<usize> {
1844    let snap = snapshot.lock().ok()?;
1845    let game = snap.game.as_ref()?;
1846    game.player_order.iter().position(|name| name == username)
1847}
1848
1849fn concede_missing_active_seats(
1850    engine_session: &SharedEngineSession,
1851    snapshot: &SharedHostSnapshot,
1852    room: &RoomInfo,
1853) {
1854    let Some(active) = active_player_usernames(snapshot) else {
1855        return;
1856    };
1857    let present: HashSet<&str> = room
1858        .players
1859        .iter()
1860        .map(|player| player.username.as_str())
1861        .collect();
1862    for username in active {
1863        if !present.contains(username.as_str()) {
1864            if let Some(index) = seat_index_of(snapshot, &username) {
1865                concede_seat(engine_session, index);
1866            }
1867        }
1868    }
1869}
1870
1871fn concede_seat(engine_session: &SharedEngineSession, player_index: usize) {
1872    send_seat_message(
1873        engine_session,
1874        player_index,
1875        ClientToServerMessage::Directive {
1876            directive: manabrew_protocol::transport::DirectiveInput::Concede,
1877        },
1878    );
1879}
1880
1881fn send_seat_message(
1882    engine_session: &SharedEngineSession,
1883    player_index: usize,
1884    message: ClientToServerMessage,
1885) {
1886    let guard = match engine_session.lock() {
1887        Ok(guard) => guard,
1888        Err(error) => {
1889            warn!(%error, "engine session lock poisoned");
1890            return;
1891        }
1892    };
1893    let Some(session) = guard.as_ref() else {
1894        debug!(player_index, "no engine session for seat action");
1895        return;
1896    };
1897    let txs = match session {
1898        EngineSession::Manabrew {
1899            remote_response_txs,
1900            ..
1901        }
1902        | EngineSession::Forge {
1903            remote_response_txs,
1904            ..
1905        } => remote_response_txs,
1906    };
1907    let Some(tx) = txs.get(&player_index) else {
1908        debug!(player_index, "no response channel for player");
1909        return;
1910    };
1911    if let Err(error) = tx.send(message) {
1912        warn!(player_index, %error, "failed to route seat message");
1913    }
1914}
1915
1916fn finish_hosted_engine(
1917    result: std::thread::Result<Result<(), String>>,
1918    game_id: &str,
1919    players: usize,
1920    started: Instant,
1921    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
1922    session_handle: &SharedEngineSession,
1923    snapshot: &SharedHostSnapshot,
1924) {
1925    let fatal = match result {
1926        Ok(Ok(())) => {
1927            info!(game_id, "hosted engine thread finished");
1928            None
1929        }
1930        Ok(Err(message)) => {
1931            error!(game_id, message, "hosted engine exited with a fatal error");
1932            Some(message)
1933        }
1934        Err(panic) => {
1935            let message = if let Some(message) = panic.downcast_ref::<String>() {
1936                message.clone()
1937            } else if let Some(message) = panic.downcast_ref::<&str>() {
1938                message.to_string()
1939            } else {
1940                "the host engine panicked".to_string()
1941            };
1942            error!(game_id, message, "hosted engine thread panicked");
1943            Some(message)
1944        }
1945    };
1946    crate::metrics::record_engine_session_finished(players, started, fatal.as_deref());
1947    let still_owner = session_handle
1948        .lock()
1949        .map(|guard| guard.as_ref().is_some_and(|s| s.game_id() == game_id))
1950        .unwrap_or(false);
1951    if let Some(message) = fatal {
1952        if still_owner {
1953            if let Ok(mut snap) = snapshot.lock() {
1954                if snap
1955                    .game
1956                    .as_ref()
1957                    .is_some_and(|game| game.game_id == game_id)
1958                {
1959                    snap.pending_end_game = Some(game_id.to_string());
1960                }
1961            }
1962            let _ = outbound_tx.send(ClientMessage::ReportGameOutcome {
1963                game_id: game_id.to_string(),
1964                outcome: GameOutcomeReport {
1965                    fatal_message: Some(message.clone()),
1966                    ..GameOutcomeReport::default()
1967                },
1968            });
1969            if let Ok(state) = serde_json::to_value(StateEnvelope::Fatal { message }) {
1970                let _ = outbound_tx.send(ClientMessage::BroadcastState {
1971                    state,
1972                    target_player: None,
1973                });
1974            }
1975            let _ = outbound_tx.send(ClientMessage::EndGame {
1976                game_id: game_id.to_string(),
1977            });
1978        } else {
1979            warn!(game_id, message, "stale engine session finished with error");
1980        }
1981    }
1982}
1983
1984/// `from_player` is relay-attested; never client supplied.
1985fn route_seat_envelope(
1986    engine_session: &SharedEngineSession,
1987    snapshot: &SharedHostSnapshot,
1988    from_player: &str,
1989    state: &Value,
1990) {
1991    match serde_json::from_value::<StateEnvelope>(state.clone()) {
1992        Ok(StateEnvelope::Response { .. }) => {
1993            route_remote_response(engine_session, snapshot, from_player, state)
1994        }
1995        Ok(StateEnvelope::Directive {
1996            from_player: claimed_slot,
1997            directive,
1998        }) => route_remote_directive(
1999            engine_session,
2000            snapshot,
2001            from_player,
2002            &claimed_slot,
2003            &directive,
2004        ),
2005        _ => debug!(from_player, state = %state, "unroutable seat envelope"),
2006    }
2007}
2008
2009fn route_remote_response(
2010    engine_session: &SharedEngineSession,
2011    snapshot: &SharedHostSnapshot,
2012    authenticated_username: &str,
2013    state: &Value,
2014) {
2015    let envelope: StateEnvelope = match serde_json::from_value(state.clone()) {
2016        Ok(envelope) => envelope,
2017        Err(error) => {
2018            warn!(%error, state = %state, "relay response invalid envelope");
2019            return;
2020        }
2021    };
2022    let StateEnvelope::Response {
2023        from_player,
2024        prompt_id,
2025        action: action_value,
2026    } = envelope
2027    else {
2028        warn!(state = %state, "expected response envelope");
2029        return;
2030    };
2031    let Some(player_index) =
2032        authenticated_player_index(snapshot, authenticated_username, &from_player, "response")
2033    else {
2034        return;
2035    };
2036    let action: PromptOutput = match serde_json::from_value(action_value) {
2037        Ok(action) => action,
2038        Err(error) => {
2039            warn!(from_player, %error, "relay response has invalid action");
2040            return;
2041        }
2042    };
2043
2044    let guard = match engine_session.lock() {
2045        Ok(guard) => guard,
2046        Err(error) => {
2047            warn!(%error, "engine session lock poisoned");
2048            return;
2049        }
2050    };
2051    let Some(session) = guard.as_ref() else {
2052        debug!(from_player, "no engine session for relay response");
2053        return;
2054    };
2055    let tx = match session {
2056        EngineSession::Manabrew {
2057            remote_response_txs,
2058            ..
2059        } => remote_response_txs.get(&player_index),
2060        EngineSession::Forge {
2061            remote_response_txs,
2062            ..
2063        } => {
2064            debug!(
2065                from_player,
2066                player_index, "routing relay response to java engine"
2067            );
2068            remote_response_txs.get(&player_index)
2069        }
2070    };
2071    let Some(tx) = tx else {
2072        debug!(from_player, player_index, "no response channel for player");
2073        return;
2074    };
2075    session.engine_clock().mark_response();
2076    if let Err(error) = tx.send(ClientToServerMessage::Response { prompt_id, action }) {
2077        warn!(from_player, %error, "failed to route relay response");
2078        return;
2079    }
2080    drop(guard);
2081    if let Ok(mut snap) = snapshot.lock() {
2082        snap.pending_prompts.remove(&from_player);
2083    }
2084}
2085
2086fn route_remote_directive(
2087    engine_session: &SharedEngineSession,
2088    snapshot: &SharedHostSnapshot,
2089    authenticated_username: &str,
2090    claimed_slot: &str,
2091    directive: &Value,
2092) {
2093    let directive: DirectiveInput = match serde_json::from_value(directive.clone()) {
2094        Ok(directive) => directive,
2095        Err(error) => {
2096            warn!(claimed_slot, %error, "relay directive is invalid");
2097            return;
2098        }
2099    };
2100    let Some(player_index) =
2101        authenticated_player_index(snapshot, authenticated_username, claimed_slot, "directive")
2102    else {
2103        return;
2104    };
2105    info!(claimed_slot, player_index, ?directive, "routing directive");
2106    match directive {
2107        DirectiveInput::Concede => concede_seat(engine_session, player_index),
2108    }
2109}
2110
2111fn authenticated_player_index(
2112    snapshot: &SharedHostSnapshot,
2113    authenticated_username: &str,
2114    claimed_slot: &str,
2115    message_kind: &str,
2116) -> Option<usize> {
2117    let Some(player_index) = seat_index_of(snapshot, authenticated_username) else {
2118        warn!(
2119            authenticated_username,
2120            message_kind, "relay sender has no engine seat"
2121        );
2122        return None;
2123    };
2124    let authenticated_slot = player_slot(player_index);
2125    if claimed_slot != authenticated_slot {
2126        warn!(
2127            authenticated_username,
2128            claimed_slot,
2129            authenticated_slot,
2130            message_kind,
2131            "relay sender claimed another engine seat"
2132        );
2133        return None;
2134    }
2135    Some(player_index)
2136}
2137
2138/// Add the fingerprint of the carried state to a full state envelope, so a
2139/// receiver can tell whether a later patch applies to what it holds.
2140fn stamp_fingerprint(mut envelope: Value) -> Value {
2141    let Some(state) = envelope.get("state") else {
2142        return envelope;
2143    };
2144    let fingerprint = manabrew_relay_protocol::state_delta::fingerprint(state);
2145    if let Some(object) = envelope.as_object_mut() {
2146        object.insert("fingerprint".to_string(), Value::String(fingerprint));
2147    }
2148    envelope
2149}
2150
2151/// Time spent building this envelope after `engine_ms` was sampled: serialising
2152/// the state, fingerprinting it, diffing it. Must be stamped last, or it reaches
2153/// the dedup comparisons and the diff base, neither of which may see it.
2154fn stamp_emit_ms(mut envelope: Value, elapsed: Duration) -> Value {
2155    if let Some(object) = envelope.as_object_mut() {
2156        object.insert(
2157            "emitMs".to_string(),
2158            Value::from(elapsed.as_millis().min(u128::from(u32::MAX)) as u32),
2159        );
2160    }
2161    envelope
2162}
2163
2164/// Patch form of a fingerprinted state envelope, against the last one sent to
2165/// this seat. `None` the first time a seat is served, or if anything is missing,
2166/// and the caller falls back to the full state, which is always correct.
2167/// When the engine host last received a player's response, as milliseconds on
2168/// its own monotonic clock. The response path writes it and the emit path
2169/// reads it, so an outgoing envelope can say how long the engine took without
2170/// anyone comparing clocks across hosts.
2171#[derive(Clone, Default)]
2172pub struct EngineClock(std::sync::Arc<std::sync::atomic::AtomicU64>);
2173
2174impl EngineClock {
2175    fn now() -> u64 {
2176        use std::sync::OnceLock;
2177        static START: OnceLock<Instant> = OnceLock::new();
2178        START.get_or_init(Instant::now).elapsed().as_millis() as u64
2179    }
2180
2181    pub fn mark_response(&self) {
2182        self.0
2183            .store(Self::now(), std::sync::atomic::Ordering::Relaxed);
2184    }
2185
2186    /// Cleared once the decision it was timing completes, so an envelope emitted
2187    /// later, while a player is thinking, cannot report that wait as engine time.
2188    pub fn mark_decision_complete(&self) {
2189        self.0.store(0, std::sync::atomic::Ordering::Relaxed);
2190    }
2191
2192    /// Milliseconds since the last response, or `None` if none has arrived yet
2193    /// or the decision it was timing has already completed.
2194    ///
2195    /// There is deliberately no upper bound. Idle time used to be excluded by
2196    /// discarding anything past 120s, which also discarded the slow decisions
2197    /// the metric exists to expose: a real 175s decision reported no `engineMs`
2198    /// at all, so the field was absent exactly when it mattered and every
2199    /// quantile built from it was understated. `mark_decision_complete` draws
2200    /// that line by state instead of by duration.
2201    pub fn elapsed_ms(&self) -> Option<u32> {
2202        let at = self.0.load(std::sync::atomic::Ordering::Relaxed);
2203        if at == 0 {
2204            return None;
2205        }
2206        let delta = Self::now().saturating_sub(at);
2207        Some(delta.min(u32::MAX as u64) as u32)
2208    }
2209}
2210
2211fn patch_against_last(
2212    bases: &mut HashMap<usize, (Value, String)>,
2213    player_index: usize,
2214    per_seat: bool,
2215    slot: &str,
2216    envelope: &Value,
2217    engine_ms: Option<u32>,
2218) -> Option<Value> {
2219    let next = envelope.get("state")?.clone();
2220    let fingerprint = envelope.get("fingerprint")?.as_str()?.to_string();
2221    let previous = bases.insert(player_index, (next.clone(), fingerprint.clone()));
2222    let (base_state, base) = previous?;
2223    let patch = manabrew_relay_protocol::state_delta::diff(&base_state, &next)?;
2224    serde_json::to_value(StateEnvelope::StateDelta {
2225        for_player: per_seat.then(|| slot.to_string()),
2226        base,
2227        fingerprint,
2228        patch,
2229        engine_ms,
2230        emit_ms: None,
2231    })
2232    .ok()
2233}
2234
2235fn spawn_remote_prompt_forwarder(
2236    outbound_tx: tokio_mpsc::UnboundedSender<ClientMessage>,
2237    snapshot: SharedHostSnapshot,
2238    engine_session: SharedEngineSession,
2239    game_id: String,
2240    remote_prompt_rx: std_mpsc::Receiver<(usize, AgentMessage)>,
2241    seat_usernames: Option<Vec<String>>,
2242    state_delta: bool,
2243    engine_clock: EngineClock,
2244) {
2245    thread::spawn(move || {
2246        let mut last_state_by_seat: HashMap<usize, Value> = HashMap::new();
2247        let mut delta_bases: HashMap<usize, (Value, String)> = HashMap::new();
2248        let mut last_state: Option<Value> = None;
2249        let mut last_display: Option<Value> = None;
2250        while let Ok((player_index, message)) = remote_prompt_rx.recv() {
2251            let Ok(session) = engine_session.lock() else {
2252                break;
2253            };
2254            if session
2255                .as_ref()
2256                .is_none_or(|session| session.game_id() != game_id)
2257            {
2258                break;
2259            }
2260            let per_seat = seat_usernames.is_some() && player_index != OBSERVER_SEAT;
2261            let slot = player_slot(player_index);
2262            let engine_ms = engine_clock.elapsed_ms();
2263            if matches!(message, AgentMessage::Prompt(_)) {
2264                engine_clock.mark_decision_complete();
2265            }
2266            let emit_started = Instant::now();
2267            let envelope = match &message {
2268                AgentMessage::State(state_update) if !per_seat => StateEnvelope::State {
2269                    for_player: None,
2270                    state: serde_json::to_value(state_update).unwrap_or(Value::Null),
2271                    fingerprint: None,
2272                    engine_ms,
2273                    emit_ms: None,
2274                },
2275                _ => StateEnvelope::for_agent_message_timed(slot.clone(), &message, engine_ms),
2276            };
2277            let Ok(state) = serde_json::to_value(envelope) else {
2278                continue;
2279            };
2280            let state = match &message {
2281                AgentMessage::State(_) if state_delta => stamp_fingerprint(state),
2282                _ => state,
2283            };
2284            match &message {
2285                AgentMessage::State(_) if per_seat => {
2286                    if last_state_by_seat.get(&player_index) == Some(&state) {
2287                        continue;
2288                    }
2289                    last_state_by_seat.insert(player_index, state.clone());
2290                }
2291                AgentMessage::State(_) if last_state.as_ref() == Some(&state) => continue,
2292                AgentMessage::State(_) => last_state = Some(state.clone()),
2293                AgentMessage::Display(_) if last_display.as_ref() == Some(&state) => continue,
2294                AgentMessage::Display(_) => last_display = Some(state.clone()),
2295                AgentMessage::Prompt(_) | AgentMessage::Error(_) => {}
2296            }
2297            if let Ok(mut snap) = snapshot.lock() {
2298                match &message {
2299                    AgentMessage::State(_) if per_seat => {
2300                        snap.last_state_by_slot.insert(slot.clone(), state.clone());
2301                    }
2302                    AgentMessage::State(_) => snap.last_state = Some(state.clone()),
2303                    AgentMessage::Prompt(_) => {
2304                        snap.pending_prompts.insert(slot.clone(), state.clone());
2305                    }
2306                    AgentMessage::Display(_) | AgentMessage::Error(_) => {}
2307                }
2308            }
2309            let state = match &message {
2310                AgentMessage::State(_) if state_delta => patch_against_last(
2311                    &mut delta_bases,
2312                    player_index,
2313                    per_seat,
2314                    &slot,
2315                    &state,
2316                    engine_ms,
2317                )
2318                .unwrap_or(state),
2319                _ => state,
2320            };
2321            let target_player = if per_seat
2322                && matches!(
2323                    message,
2324                    AgentMessage::State(_) | AgentMessage::Prompt(_) | AgentMessage::Error(_)
2325                ) {
2326                let Some(target_player) = seat_usernames
2327                    .as_ref()
2328                    .and_then(|names| names.get(player_index).cloned())
2329                else {
2330                    warn!(player_index, PRIVATE_MESSAGE_MISSING_PLAYER);
2331                    continue;
2332                };
2333                Some(target_player)
2334            } else {
2335                None
2336            };
2337            if outbound_tx
2338                .send(ClientMessage::BroadcastState {
2339                    state: stamp_emit_ms(state, emit_started.elapsed()),
2340                    target_player,
2341                })
2342                .is_err()
2343            {
2344                break;
2345            }
2346        }
2347    });
2348}
2349
2350fn outcome_report(view: &GameViewDto) -> GameOutcomeReport {
2351    GameOutcomeReport {
2352        game_over: view.game_over,
2353        winner_slot: view.winner_id.clone(),
2354        conceded_slots: view
2355            .players
2356            .iter()
2357            .filter(|player| player.status == PlayerStatus::Conceded)
2358            .map(|player| player.id.clone())
2359            .collect(),
2360        fatal_message: None,
2361        turns: Some(view.turn),
2362    }
2363}
2364
2365fn cached_outcome_report(envelope: &Value) -> Option<GameOutcomeReport> {
2366    let update: StateUpdate = serde_json::from_value(envelope.get("state")?.clone()).ok()?;
2367    Some(outcome_report(&update.game_view))
2368}
2369
2370fn spawn_game_over_forwarder(
2371    outbound_tx: tokio_mpsc::UnboundedSender<ClientMessage>,
2372    game_over_rx: std_mpsc::Receiver<HostedGameOver>,
2373    session_handle: SharedEngineSession,
2374    snapshot: SharedHostSnapshot,
2375    game_id: String,
2376    seat_usernames: Option<Vec<String>>,
2377) {
2378    thread::spawn(move || {
2379        while let Ok(game_over) = game_over_rx.recv() {
2380            let Ok(session) = session_handle.lock() else {
2381                return;
2382            };
2383            if session
2384                .as_ref()
2385                .is_none_or(|session| session.game_id() != game_id)
2386            {
2387                warn!(
2388                    game_id,
2389                    "stale engine session reached game over; not ending the relay game"
2390                );
2391                continue;
2392            }
2393            if let Ok(mut snap) = snapshot.lock() {
2394                if snap
2395                    .game
2396                    .as_ref()
2397                    .is_some_and(|game| game.game_id == game_id)
2398                {
2399                    snap.pending_end_game = Some(game_id.clone());
2400                }
2401            }
2402            let outcome = game_over
2403                .messages
2404                .iter()
2405                .rev()
2406                .find_map(|(_, message)| match message {
2407                    AgentMessage::State(update) => Some(outcome_report(&update.game_view)),
2408                    _ => None,
2409                })
2410                .or_else(|| {
2411                    let snap = snapshot.lock().ok()?;
2412                    snap.last_state
2413                        .as_ref()
2414                        .or_else(|| snap.last_state_by_slot.values().next())
2415                        .and_then(cached_outcome_report)
2416                        .filter(|outcome| outcome.game_over)
2417                });
2418            let mut last_state_by_seat: HashMap<usize, Value> = HashMap::new();
2419            let mut last_state: Option<Value> = None;
2420            for (player_index, message) in game_over.messages {
2421                let per_seat = seat_usernames.is_some() && player_index != OBSERVER_SEAT;
2422                let envelope = match &message {
2423                    AgentMessage::State(state_update) if !per_seat => StateEnvelope::State {
2424                        for_player: None,
2425                        state: serde_json::to_value(state_update).unwrap_or(Value::Null),
2426                        fingerprint: None,
2427                        engine_ms: None,
2428                        emit_ms: None,
2429                    },
2430                    _ => StateEnvelope::for_agent_message(player_slot(player_index), &message),
2431                };
2432                let Ok(state) = serde_json::to_value(envelope) else {
2433                    continue;
2434                };
2435                match &message {
2436                    AgentMessage::State(_) if per_seat => {
2437                        if last_state_by_seat.get(&player_index) == Some(&state) {
2438                            continue;
2439                        }
2440                        last_state_by_seat.insert(player_index, state.clone());
2441                    }
2442                    AgentMessage::State(_) if last_state.as_ref() == Some(&state) => continue,
2443                    AgentMessage::State(_) => last_state = Some(state.clone()),
2444                    AgentMessage::Display(_) | AgentMessage::Prompt(_) | AgentMessage::Error(_) => {
2445                    }
2446                }
2447                let target_player = if per_seat
2448                    && matches!(
2449                        message,
2450                        AgentMessage::State(_) | AgentMessage::Prompt(_) | AgentMessage::Error(_)
2451                    ) {
2452                    let Some(target_player) = seat_usernames
2453                        .as_ref()
2454                        .and_then(|names| names.get(player_index).cloned())
2455                    else {
2456                        warn!(player_index, PRIVATE_MESSAGE_MISSING_PLAYER);
2457                        continue;
2458                    };
2459                    Some(target_player)
2460                } else {
2461                    None
2462                };
2463                if outbound_tx
2464                    .send(ClientMessage::BroadcastState {
2465                        state,
2466                        target_player,
2467                    })
2468                    .is_err()
2469                {
2470                    return;
2471                }
2472            }
2473            let Ok(state) = serde_json::to_value(StateEnvelope::RoomRelay {
2474                protocol: SELF_HOSTED_NODE_PROTOCOL.to_string(),
2475                version: 1,
2476                message_id: Uuid::new_v4().to_string(),
2477                from_player: None,
2478                target_player: None,
2479                room_id: None,
2480                payload: json!({ "type": "gameOver", "gameId": game_over.game_id }),
2481            }) else {
2482                continue;
2483            };
2484            if outbound_tx
2485                .send(ClientMessage::BroadcastState {
2486                    state,
2487                    target_player: None,
2488                })
2489                .is_err()
2490            {
2491                return;
2492            }
2493            if let Some(outcome) = outcome {
2494                let _ = outbound_tx.send(ClientMessage::ReportGameOutcome {
2495                    game_id: game_id.clone(),
2496                    outcome,
2497                });
2498            }
2499            if outbound_tx
2500                .send(ClientMessage::EndGame {
2501                    game_id: game_id.clone(),
2502                })
2503                .is_err()
2504            {
2505                return;
2506            }
2507        }
2508    });
2509}
2510
2511async fn relay_writer(mut write: WsWrite, mut outbound: tokio_mpsc::UnboundedReceiver<Message>) {
2512    while let Some(message) = outbound.recv().await {
2513        let started = Instant::now();
2514        let closing = matches!(message, Message::Close(_));
2515        if let Err(error) = write.send(message).await {
2516            warn!(%error, "relay write failed");
2517            break;
2518        }
2519        crate::metrics::record_relay_send(started.elapsed());
2520        if closing {
2521            break;
2522        }
2523    }
2524    let _ = write.close().await;
2525}
2526
2527fn self_minted_identity_token(kind: &str, handle: &str) -> String {
2528    let iat = std::time::SystemTime::now()
2529        .duration_since(std::time::UNIX_EPOCH)
2530        .map(|elapsed| elapsed.as_secs() as i64)
2531        .unwrap_or_default();
2532    identity_token::mint_unsigned(&format!("{kind}:{handle}"), handle, iat, 24 * 60 * 60)
2533}
2534
2535impl RelayClient {
2536    async fn connect(
2537        relay_url: &str,
2538        username: &str,
2539        password: &str,
2540    ) -> Result<Self, Box<dyn std::error::Error + Send + Sync>> {
2541        info!(relay_url, username, "connecting relay client");
2542        let (socket, _) = connect_async_with_config(relay_url, None, true).await?;
2543        let (write, read) = socket.split();
2544        let (outbound, outbound_rx) = tokio_mpsc::unbounded_channel();
2545        let writer = tokio::spawn(relay_writer(write, outbound_rx));
2546        let mut client = Self {
2547            username: username.to_string(),
2548            outbound,
2549            read,
2550            writer,
2551        };
2552        client
2553            .send(&ClientMessage::Authenticate {
2554                username: username.to_string(),
2555                password: password.to_string(),
2556                service: true,
2557                identity: Some(IdentityProof {
2558                    token: Some(self_minted_identity_token("node", username)),
2559                    device: None,
2560                }),
2561                client_platform: ClientPlatform::Unknown,
2562                client_version: None,
2563            })
2564            .await?;
2565        client.wait_for_auth().await?;
2566        Ok(client)
2567    }
2568
2569    async fn wait_for_auth(&mut self) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
2570        loop {
2571            match self.recv().await? {
2572                Some(ServerMessage::AuthResult { success, error, .. }) => {
2573                    if success {
2574                        info!(username = %self.username, "authenticated");
2575                        return Ok(());
2576                    }
2577                    return Err(format!("authentication failed: {error:?}").into());
2578                }
2579                Some(other) => {
2580                    debug!(?other, username = %self.username, "ignored pre-auth message")
2581                }
2582                None => return Err("relay closed before authentication".into()),
2583            }
2584        }
2585    }
2586
2587    async fn send(
2588        &mut self,
2589        message: &ClientMessage,
2590    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
2591        self.enqueue(Message::Text(serde_json::to_string(message)?))
2592    }
2593
2594    fn enqueue(&self, message: Message) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
2595        self.outbound
2596            .send(message)
2597            .map_err(|_| "relay writer stopped".into())
2598    }
2599
2600    async fn recv(
2601        &mut self,
2602    ) -> Result<Option<ServerMessage>, Box<dyn std::error::Error + Send + Sync>> {
2603        loop {
2604            let Some(message) = self.read.next().await else {
2605                return Ok(None);
2606            };
2607            let message = message?;
2608            match message {
2609                Message::Text(text) => {
2610                    return Ok(Some(serde_json::from_str(&text)?));
2611                }
2612                Message::Ping(payload) => {
2613                    self.enqueue(Message::Pong(payload))?;
2614                }
2615                Message::Close(_) => return Ok(None),
2616                _ => {}
2617            }
2618        }
2619    }
2620
2621    async fn close(&mut self) {
2622        if self.enqueue(Message::Close(None)).is_err() {
2623            return;
2624        }
2625        let _ = time::timeout(CLOSE_DRAIN_TIMEOUT, async {
2626            while let Some(Ok(_)) = self.read.next().await {}
2627        })
2628        .await;
2629        self.writer.abort();
2630    }
2631
2632    async fn broadcast_room_message(
2633        &mut self,
2634        protocol: &str,
2635        payload: Value,
2636    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
2637        let envelope = StateEnvelope::RoomRelay {
2638            protocol: protocol.to_string(),
2639            version: 1,
2640            message_id: Uuid::new_v4().to_string(),
2641            from_player: Some(self.username.clone()),
2642            target_player: None,
2643            room_id: None,
2644            payload,
2645        };
2646        self.send(&ClientMessage::BroadcastState {
2647            state: serde_json::to_value(envelope)?,
2648            target_player: None,
2649        })
2650        .await
2651    }
2652}
2653
2654fn log_room_update(observer: &str, room: &RoomInfo) {
2655    let players = room
2656        .players
2657        .iter()
2658        .map(|player| {
2659            format!(
2660                "{}{}{}",
2661                player.username,
2662                if player.ready { ":ready" } else { "" },
2663                if player.connected { "" } else { ":offline" },
2664            )
2665        })
2666        .collect::<Vec<_>>()
2667        .join(",");
2668    info!(
2669        observer,
2670        room_id = %room.room_id,
2671        room_name = %room.room_name,
2672        host = %room.host,
2673        players,
2674        "room update"
2675    );
2676}