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, BotConfig, BotResponder};
15use manabrew_agent_interface::agent_impl::Responder;
16use manabrew_agent_interface::ids_codec::{parse_player_slot, player_slot};
17use manabrew_agent_interface::prompt::{AgentMessage, ClientToServerMessage, PromptOutput};
18use manabrew_agent_interface::protocol::{
19    identity_token, ClientMessage, ClientPlatform, EngineGate, EngineKind, GameFormat,
20    GameOutcomeReport, IdentityProof, PlayerDeckInfo, ResumeRoomRequest, RoomInfo, RoomStatus,
21    ServerMessage, StateEnvelope, PROTOCOL_VERSION,
22};
23use manabrew_protocol::deck_dto::Deck;
24use manabrew_protocol::game::{GameViewDto, PlayerStatus};
25use manabrew_protocol::transport::{DirectiveInput, EngineDirective, StateUpdate};
26use rand::Rng;
27use serde::Deserialize;
28use serde_json::{json, Value};
29use tokio::net::TcpStream;
30use tokio::sync::mpsc as tokio_mpsc;
31use tokio::task::JoinHandle;
32use tokio::time;
33use tokio_tungstenite::tungstenite::Message;
34use tokio_tungstenite::{connect_async_with_config, MaybeTlsStream, WebSocketStream};
35use tracing::{debug, error, info, warn};
36use uuid::Uuid;
37
38type WsStream = WebSocketStream<MaybeTlsStream<TcpStream>>;
39type WsWrite = SplitSink<WsStream, Message>;
40type WsRead = SplitStream<WsStream>;
41
42const SELF_HOSTED_NODE_PROTOCOL: &str = "self-hosted-node";
43const PRIVATE_MESSAGE_MISSING_PLAYER: &str =
44    "cannot forward private engine message without a mapped player";
45
46struct RelayClient {
47    username: String,
48    outbound: tokio_mpsc::UnboundedSender<Message>,
49    read: WsRead,
50    writer: JoinHandle<()>,
51}
52
53/// Highest prompt id each local bot seat answered; the relay's later copy
54/// of that answer is dropped against it.
55type LocalBotAnswers = Arc<Mutex<HashMap<usize, u32>>>;
56
57enum EngineSession {
58    Manabrew {
59        game_id: String,
60        remote_response_txs: HashMap<usize, std_mpsc::Sender<ClientToServerMessage>>,
61        engine_clock: EngineClock,
62        local_bot_answers: LocalBotAnswers,
63    },
64    Forge {
65        game_id: String,
66        remote_response_txs: HashMap<usize, std_mpsc::Sender<ClientToServerMessage>>,
67        cancel: Arc<AtomicBool>,
68        engine_clock: EngineClock,
69        local_bot_answers: LocalBotAnswers,
70    },
71}
72
73impl EngineSession {
74    fn game_id(&self) -> &str {
75        match self {
76            EngineSession::Manabrew { game_id, .. } | EngineSession::Forge { game_id, .. } => {
77                game_id
78            }
79        }
80    }
81
82    fn engine_clock(&self) -> &EngineClock {
83        match self {
84            EngineSession::Manabrew { engine_clock, .. }
85            | EngineSession::Forge { engine_clock, .. } => engine_clock,
86        }
87    }
88
89    fn response_tx(&self, player_index: usize) -> Option<&std_mpsc::Sender<ClientToServerMessage>> {
90        match self {
91            EngineSession::Manabrew {
92                remote_response_txs,
93                ..
94            }
95            | EngineSession::Forge {
96                remote_response_txs,
97                ..
98            } => remote_response_txs.get(&player_index),
99        }
100    }
101
102    fn local_bot_answers(&self) -> &LocalBotAnswers {
103        match self {
104            EngineSession::Manabrew {
105                local_bot_answers, ..
106            }
107            | EngineSession::Forge {
108                local_bot_answers, ..
109            } => local_bot_answers,
110        }
111    }
112
113    /// The relay's copy of an answer already given in-process.
114    fn is_local_bot_duplicate(&self, player_index: usize, prompt_id: u32) -> bool {
115        prompt_id != 0
116            && self
117                .local_bot_answers()
118                .lock()
119                .ok()
120                .and_then(|answers| answers.get(&player_index).copied())
121                .is_some_and(|answered| answered >= prompt_id)
122    }
123}
124
125struct BotState {
126    handle: JoinHandle<()>,
127    shutdown: Arc<tokio::sync::Notify>,
128}
129
130#[derive(Debug, Deserialize)]
131#[serde(rename_all = "camelCase")]
132struct SpawnBotPayload {
133    #[serde(default)]
134    deck: Option<SpawnBotDeckPayload>,
135    #[serde(default)]
136    decks: Option<Vec<SpawnBotDeckPayload>>,
137}
138
139#[derive(Debug, Deserialize)]
140#[serde(rename_all = "camelCase")]
141struct SpawnBotDeckPayload {
142    deck_name: String,
143    deck: Deck,
144    #[serde(default)]
145    commander_name: Option<String>,
146}
147
148type SharedEngineSession = Arc<Mutex<Option<EngineSession>>>;
149type SessionRegistry = Arc<Mutex<Vec<SharedEngineSession>>>;
150
151fn registry_idle(registry: &SessionRegistry) -> bool {
152    registry
153        .lock()
154        .map(|sessions| {
155            sessions
156                .iter()
157                .all(|session| session.lock().map(|guard| guard.is_none()).unwrap_or(false))
158        })
159        .unwrap_or(false)
160}
161
162static DRAINING: AtomicBool = AtomicBool::new(false);
163
164fn draining() -> bool {
165    DRAINING.load(Ordering::Relaxed)
166}
167
168type SharedBotState = Arc<Mutex<Vec<BotState>>>;
169
170#[derive(Clone)]
171struct GameStart {
172    game_id: String,
173    player_order: Vec<String>,
174    player_decks: Vec<PlayerDeckInfo>,
175    starting_life: i32,
176}
177
178#[derive(Default)]
179struct HostSnapshot {
180    room_info: Option<RoomInfo>,
181    resume_token: Option<String>,
182    game: Option<GameStart>,
183    last_state: Option<Value>,
184    last_state_by_slot: HashMap<String, Value>,
185    pending_prompts: HashMap<String, Value>,
186    pending_directives: HashMap<(String, &'static str), Value>,
187    pending_end_game: Option<String>,
188}
189
190type SharedHostSnapshot = Arc<Mutex<HostSnapshot>>;
191
192const RECONNECT_BACKOFF_SECS: [u64; 6] = [1, 2, 4, 8, 15, 30];
193
194/// Pseudo-seat used by per-recipient backends for the public (spectator) view.
195pub(crate) const OBSERVER_SEAT: usize = usize::MAX;
196const CLOSE_DRAIN_TIMEOUT: Duration = Duration::from_secs(3);
197const BOT_STOP_TIMEOUT: Duration = Duration::from_secs(5);
198
199enum LoopExit {
200    Cancelled,
201    Disconnected,
202}
203
204pub async fn cli_entry() {
205    {
206        use tracing_subscriber::layer::SubscriberExt;
207        use tracing_subscriber::util::SubscriberInitExt;
208        tracing_subscriber::registry()
209            .with(
210                tracing_subscriber::EnvFilter::try_from_default_env()
211                    .unwrap_or_else(|_| "self_hosted_node=info".into()),
212            )
213            .with(tracing_subscriber::fmt::layer())
214            .with(crate::logs::layer_from_env())
215            .init();
216    }
217    crate::metrics::init_from_env();
218
219    if std::env::var("SELF_HOSTED_NODE_JAVA_SMOKE").is_ok() {
220        let max_prompts = std::env::var("SELF_HOSTED_NODE_JAVA_SMOKE_PROMPTS")
221            .ok()
222            .and_then(|value| value.parse().ok())
223            .unwrap_or(4);
224        if let Err(error) = java_backend::run_smoke_game(max_prompts) {
225            error!(%error, "java-forge smoke failed");
226            std::process::exit(1);
227        }
228        info!(max_prompts, "java-forge smoke completed");
229        return;
230    }
231    if std::env::var("SELF_HOSTED_NODE_GRAAL_SMOKE").is_ok() {
232        if let Err(error) = java_backend::run_graal_smoke() {
233            error!(%error, "graal-forge smoke failed");
234            std::process::exit(1);
235        }
236        info!("graal-forge smoke completed");
237        return;
238    }
239    if std::env::var("SELF_HOSTED_NODE_JAVA_CONCEDE_SMOKE").is_ok() {
240        if let Err(error) = java_backend::run_concede_smoke() {
241            error!(%error, "java-forge concede smoke failed");
242            std::process::exit(1);
243        }
244        info!("java-forge concede smoke completed");
245        return;
246    }
247    if let Ok(scenario_name) = std::env::var("SELF_HOSTED_NODE_JAVA_SCENARIO") {
248        let max_prompts = std::env::var("SELF_HOSTED_NODE_JAVA_SCENARIO_PROMPTS")
249            .ok()
250            .and_then(|value| value.parse().ok())
251            .unwrap_or(20);
252        if let Err(error) = java_backend::run_scenario(&scenario_name, max_prompts) {
253            error!(scenario_name, %error, "java-forge scenario failed");
254            std::process::exit(1);
255        }
256        info!(scenario_name, max_prompts, "java-forge scenario completed");
257        return;
258    }
259    if std::env::var("SELF_HOSTED_NODE_JAVA_SELF_PLAY").is_ok() {
260        let cfg = SelfPlayConfig::from_env();
261        let max_prompts = std::env::var("SELF_HOSTED_NODE_JAVA_SELF_PLAY_PROMPTS")
262            .ok()
263            .and_then(|value| value.parse().ok())
264            .unwrap_or(2_000);
265        let games = std::env::var("SELF_HOSTED_NODE_JAVA_SELF_PLAY_GAMES")
266            .ok()
267            .and_then(|value| value.parse().ok())
268            .unwrap_or(1);
269        if let Err(error) =
270            java_backend::run_self_play(&cfg.seats, cfg.starting_life, cfg.seed, max_prompts, games)
271        {
272            error!(%error, "java-forge self-play failed");
273            std::process::exit(1);
274        }
275        info!(
276            players = cfg.seats.len(),
277            max_prompts, games, "java-forge self-play completed"
278        );
279        return;
280    }
281    if let Ok(value) = std::env::var("SELF_HOSTED_NODE_JAVA_CONCURRENT_GAMES") {
282        let cfg = SelfPlayConfig::from_env();
283        let max_prompts = std::env::var("SELF_HOSTED_NODE_JAVA_SELF_PLAY_PROMPTS")
284            .ok()
285            .and_then(|value| value.parse().ok())
286            .unwrap_or(2_000);
287        let concurrency = value.parse().unwrap_or(2);
288        if let Err(error) = java_backend::run_concurrent_self_play(
289            &cfg.seats,
290            cfg.starting_life,
291            cfg.seed,
292            max_prompts,
293            concurrency,
294        ) {
295            error!(%error, "java-forge concurrent self-play failed");
296            std::process::exit(1);
297        }
298        info!(
299            players = cfg.seats.len(),
300            concurrency, "java-forge concurrent self-play completed"
301        );
302        return;
303    }
304    if std::env::var("SELF_HOSTED_NODE_RUST_SELF_PLAY").is_ok() {
305        let cfg = SelfPlayConfig::from_env();
306        let max_turns = std::env::var("SELF_HOSTED_NODE_RUST_SELF_PLAY_TURNS")
307            .ok()
308            .and_then(|value| value.parse().ok())
309            .unwrap_or(200);
310        if let Err(error) =
311            rust_backend::run_self_play(&cfg.seats, cfg.starting_life, cfg.seed, max_turns)
312        {
313            error!(%error, "rust self-play failed");
314            std::process::exit(1);
315        }
316        info!(
317            players = cfg.seats.len(),
318            max_turns, "rust self-play completed"
319        );
320        return;
321    }
322
323    let config = Config::from_env();
324    if let Err(error) = run(config).await {
325        error!(%error, "room node exited");
326        std::process::exit(1);
327    }
328}
329
330/// Signal a hosted room to shut down: call `notify_one()` to stop it.
331pub type RoomCancel = Arc<tokio::sync::Notify>;
332
333fn ensure_engine_ready(config: &Config) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
334    if config.engine_enabled
335        && config.backend.is_supported()
336        && matches!(config.backend, EngineBackendKind::Forge)
337    {
338        info!("initializing forge engine backend");
339        java_backend::init_engine()?;
340    }
341    Ok(())
342}
343
344/// Host a single relay room until it ends or `cancel` is signalled. The crate's
345/// public entry point — env-free; the caller supplies a fully-built `Config`.
346pub async fn host_room(
347    config: Config,
348    cancel: RoomCancel,
349    ready: tokio::sync::oneshot::Sender<String>,
350) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
351    ensure_engine_ready(&config)?;
352    host_one_room(config, None, cancel, Some(ready), None, None).await
353}
354
355/// The webview's half of the data plane, handed in by a desktop shell.
356pub struct ShellBridgeHandle {
357    pub bridge: Arc<ShellBridge>,
358    pub commands: tokio_mpsc::UnboundedReceiver<ShellCommand>,
359}
360
361pub async fn host_room_bridged(
362    config: Config,
363    cancel: RoomCancel,
364    ready: tokio::sync::oneshot::Sender<String>,
365    bridge: ShellBridgeHandle,
366) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
367    ensure_engine_ready(&config)?;
368    host_one_room(config, None, cancel, Some(ready), None, Some(bridge)).await
369}
370
371async fn run(mut config: Config) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
372    ensure_engine_ready(&config)?;
373
374    let registry: SessionRegistry = Arc::new(Mutex::new(Vec::new()));
375
376    let slots = config.max_games.max(1);
377    let single = config.room_id.is_some() || slots <= 1;
378    let room_count = if single { 1 } else { slots };
379    let cancels: Vec<RoomCancel> = (0..room_count)
380        .map(|_| Arc::new(tokio::sync::Notify::new()))
381        .collect();
382
383    let monitor_registry = registry.clone();
384    let stale_cancels = cancels.clone();
385    tokio::spawn(run_stale_monitor(
386        StaleConfig::from_env_and_args(),
387        move || registry_idle(&monitor_registry),
388        move || notify_all(&stale_cancels),
389    ));
390
391    let signal_cancels = cancels.clone();
392    tokio::spawn(async move {
393        wait_for_shutdown_signal().await;
394        info!("shutdown signal received; closing rooms");
395        notify_all(&signal_cancels);
396    });
397
398    #[cfg(unix)]
399    tokio::spawn(async move {
400        let mut usr1 =
401            match tokio::signal::unix::signal(tokio::signal::unix::SignalKind::user_defined1()) {
402                Ok(usr1) => usr1,
403                Err(error) => {
404                    warn!(%error, "failed to install SIGUSR1 handler");
405                    return;
406                }
407            };
408        while usr1.recv().await.is_some() {
409            DRAINING.store(true, Ordering::Relaxed);
410            info!("drain signal received; refusing new games and closing each room once idle");
411        }
412    });
413
414    #[cfg(all(unix, feature = "graal-forge"))]
415    tokio::spawn(async move {
416        let mut usr2 =
417            match tokio::signal::unix::signal(tokio::signal::unix::SignalKind::user_defined2()) {
418                Ok(usr2) => usr2,
419                Err(error) => {
420                    warn!(%error, "failed to install SIGUSR2 handler");
421                    return;
422                }
423            };
424        while usr2.recv().await.is_some() {
425            let dir = std::env::var("SELF_HOSTED_NODE_HEAP_DUMP_DIR")
426                .unwrap_or_else(|_| "/tmp".to_string());
427            let stamp = std::time::SystemTime::now()
428                .duration_since(std::time::UNIX_EPOCH)
429                .map(|since| since.as_secs())
430                .unwrap_or(0);
431            let path = format!("{dir}/self-hosted-node-{stamp}.hprof");
432            info!(path, "heap dump requested");
433            match tokio::task::spawn_blocking({
434                let path = path.clone();
435                move || java_backend::dump_shared_heap(&path)
436            })
437            .await
438            {
439                Ok(Ok(())) => info!(path, "heap dump written"),
440                Ok(Err(error)) => warn!(%error, "heap dump failed"),
441                Err(error) => warn!(%error, "heap dump task failed"),
442            }
443        }
444    });
445
446    if single {
447        return host_one_room(config, None, cancels[0].clone(), None, Some(registry), None).await;
448    }
449
450    config.format = GameFormat::Any;
451    let hosts: Vec<(Config, String)> = (0..slots)
452        .map(|slot| (config.clone(), (slot + 1).to_string()))
453        .collect();
454
455    info!(rooms = hosts.len(), "hosting multiple rooms on one node");
456    let mut handles = Vec::with_capacity(hosts.len());
457    for ((cfg, label), cancel) in hosts.into_iter().zip(cancels) {
458        let registry = registry.clone();
459        handles.push(tokio::spawn(async move {
460            if let Err(error) =
461                host_one_room(cfg, Some(label.clone()), cancel, None, Some(registry), None).await
462            {
463                error!(%error, label, "room host exited");
464            }
465        }));
466    }
467    for handle in handles {
468        let _ = handle.await;
469    }
470    Ok(())
471}
472
473fn notify_all(cancels: &[RoomCancel]) {
474    for cancel in cancels {
475        cancel.notify_one();
476    }
477}
478
479async fn wait_for_shutdown_signal() {
480    #[cfg(unix)]
481    {
482        let mut term =
483            match tokio::signal::unix::signal(tokio::signal::unix::SignalKind::terminate()) {
484                Ok(term) => term,
485                Err(error) => {
486                    warn!(%error, "failed to install SIGTERM handler");
487                    let _ = tokio::signal::ctrl_c().await;
488                    return;
489                }
490            };
491        tokio::select! {
492            _ = tokio::signal::ctrl_c() => {}
493            _ = term.recv() => {}
494        }
495    }
496    #[cfg(not(unix))]
497    {
498        let _ = tokio::signal::ctrl_c().await;
499    }
500}
501
502async fn host_one_room(
503    mut config: Config,
504    label: Option<String>,
505    cancel: RoomCancel,
506    ready: Option<tokio::sync::oneshot::Sender<String>>,
507    sessions: Option<SessionRegistry>,
508    shell: Option<ShellBridgeHandle>,
509) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
510    if let Some(label) = &label {
511        config.username = format!("{}-{label}", config.username);
512        config.room_name = format!("{} ({label})", config.room_name);
513    }
514
515    let _rooms_hosted = crate::metrics::RoomHostedGuard::new(if label.is_some() {
516        crate::metrics::PoolKind::Pod
517    } else {
518        crate::metrics::PoolKind::Solo
519    });
520
521    info!(
522        relay_url = %config.relay_url,
523        username = %config.username,
524        room_name = %config.room_name,
525        auto_start = config.auto_start,
526        engine_enabled = config.engine_enabled,
527        host_plays = config.host_plays,
528        bot_enabled = config.bot_enabled,
529        "starting room host"
530    );
531
532    let snapshot: SharedHostSnapshot = Arc::new(Mutex::new(HostSnapshot::default()));
533    let engine_session: SharedEngineSession = Arc::new(Mutex::new(None));
534    if let Some(sessions) = &sessions {
535        if let Ok(mut registered) = sessions.lock() {
536            registered.push(engine_session.clone());
537        }
538    }
539    let bot_state: SharedBotState = Arc::new(Mutex::new(Vec::new()));
540    let (outbound_tx, mut outbound_rx) = tokio_mpsc::unbounded_channel::<ClientMessage>();
541
542    let mut host =
543        RelayClient::connect(&config.relay_url, &config.username, &config.password).await?;
544    let mut room_id = establish_room(&mut host, &config, &snapshot).await?;
545
546    // Keep the sender alive without a shell, or `recv()` spins the select.
547    let (bridge, mut bridge_rx, _idle_bridge_tx) = match shell {
548        Some(handle) => (Some(handle.bridge), handle.commands, None),
549        None => {
550            let (tx, rx) = tokio_mpsc::unbounded_channel::<ShellCommand>();
551            (None, rx, Some(tx))
552        }
553    };
554    if let Some(bridge) = &bridge {
555        let reprime_snapshot = snapshot.clone();
556        let reprime_tx = outbound_tx.clone();
557        bridge.set_on_fallback(move |username| {
558            reprime_relay_cache(&reprime_snapshot, &reprime_tx, username);
559        });
560    }
561
562    if bridge.is_some() {
563        announce_webrtc_only(&config, &outbound_tx);
564    }
565
566    if let Some(ready) = ready {
567        let _ = ready.send(room_id.clone());
568    }
569
570    if config.bot_enabled {
571        spawn_bots(
572            &config,
573            std::slice::from_ref(&config.bot_deck),
574            &room_id,
575            &bot_state,
576        );
577    }
578
579    loop {
580        let exit = run_client_loop(
581            &mut host,
582            &config,
583            &room_id,
584            &engine_session,
585            &snapshot,
586            &bot_state,
587            &outbound_tx,
588            &mut outbound_rx,
589            &cancel,
590            bridge.as_ref(),
591            &mut bridge_rx,
592        )
593        .await;
594        if matches!(exit, LoopExit::Cancelled) {
595            stop_bots(&bot_state);
596            host.close().await;
597            return Ok(());
598        }
599
600        crate::metrics::record_relay_reconnect();
601        let mut attempt: usize = 0;
602        host = loop {
603            let delay = RECONNECT_BACKOFF_SECS[attempt.min(RECONNECT_BACKOFF_SECS.len() - 1)];
604            tokio::select! {
605                _ = cancel.notified() => {
606                    info!(username = %config.username, "room host cancelled while reconnecting");
607                    cancel_engine(&engine_session);
608                    stop_bots(&bot_state);
609                    return Ok(());
610                }
611                _ = time::sleep(Duration::from_secs(delay)) => {}
612            }
613            attempt += 1;
614            let mut client =
615                match RelayClient::connect(&config.relay_url, &config.username, &config.password)
616                    .await
617                {
618                    Ok(client) => client,
619                    Err(error) => {
620                        warn!(%error, attempt, "relay reconnect failed");
621                        continue;
622                    }
623                };
624            match reestablish_room(&mut client, &config, &engine_session, &snapshot).await {
625                Ok(new_room_id) => {
626                    if new_room_id != room_id {
627                        stop_bots(&bot_state);
628                        if config.bot_enabled {
629                            spawn_bots(
630                                &config,
631                                std::slice::from_ref(&config.bot_deck),
632                                &new_room_id,
633                                &bot_state,
634                            );
635                        }
636                        room_id = new_room_id;
637                    }
638                    break client;
639                }
640                Err(error) => {
641                    warn!(%error, attempt, "failed to re-establish room; retrying");
642                }
643            }
644        };
645        if let Some(bridge) = &bridge {
646            bridge.clear_game();
647            announce_webrtc_only(&config, &outbound_tx);
648        }
649        info!(username = %config.username, room_id, "relay connection re-established");
650    }
651}
652
653/// Re-primes the relay's replay cache for a seat leaving the plane.
654fn reprime_relay_cache(
655    snapshot: &SharedHostSnapshot,
656    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
657    username: &str,
658) {
659    let Some(index) = seat_index_of(snapshot, username) else {
660        return;
661    };
662    let slot = player_slot(index);
663    let Ok(snap) = snapshot.lock() else {
664        return;
665    };
666    let state = snap.last_state_by_slot.get(&slot).cloned();
667    let prompt = snap.pending_prompts.get(&slot).cloned();
668    let directives: Vec<Value> = snap
669        .pending_directives
670        .iter()
671        .filter(|((directive_slot, _), _)| *directive_slot == slot)
672        .map(|(_, directive)| directive.clone())
673        .collect();
674    drop(snap);
675
676    // Stored before `patch_against_last` runs, so these are full states.
677    for state in [state, prompt].into_iter().flatten().chain(directives) {
678        let _ = outbound_tx.send(ClientMessage::BroadcastState {
679            state,
680            target_player: Some(username.to_string()),
681        });
682    }
683    info!(
684        username,
685        "re-primed the relay cache for a seat that fell back"
686    );
687}
688
689fn announce_webrtc_only(config: &Config, outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>) {
690    let endpoint = manabrew_relay_protocol::TransportEndpoint {
691        endpoint_id: format!("webrtc:{}", config.username),
692        relay_url: None,
693        direct_addrs: Vec::new(),
694        kinds: vec![manabrew_relay_protocol::TRANSPORT_KIND_WEBRTC.to_string()],
695    };
696    info!(endpoint_id = %endpoint.endpoint_id, "announcing a webrtc-only endpoint");
697    let _ = outbound_tx.send(ClientMessage::AnnounceTransport {
698        endpoint: Some(endpoint),
699    });
700}
701
702async fn establish_room(
703    client: &mut RelayClient,
704    config: &Config,
705    snapshot: &SharedHostSnapshot,
706) -> Result<String, Box<dyn std::error::Error + Send + Sync>> {
707    let room_id = if let Some(room_id) = &config.room_id {
708        client
709            .send(&ClientMessage::JoinRoom {
710                room_id: room_id.clone(),
711                observe: !config.host_plays,
712                as_bot: false,
713                password: config.room_password.clone(),
714            })
715            .await?;
716        info!(room_id, "joining configured room");
717        room_id.clone()
718    } else {
719        client
720            .send(&ClientMessage::CreateRoom {
721                room_name: config.room_name.clone(),
722                max_players: config.max_players,
723                format: config.format.clone(),
724                protocol_version: PROTOCOL_VERSION,
725                hosted: !config.host_plays,
726                engine: engine_kind(config),
727                draft_config: None,
728                sealed_config: None,
729                official_key: config.official_key.clone(),
730                password: config.room_password.clone(),
731                reconnect_timeout_s: config.reconnect_timeout_s,
732                table_style: config.table_style.clone(),
733            })
734            .await?;
735        info!(room_name = %config.room_name, "creating room");
736        wait_for_host_room(client, config, snapshot).await?
737    };
738
739    if config.host_plays {
740        seat_client(client, &config.host_deck).await?;
741    } else {
742        info!(
743            username = %config.username,
744            "hosting room without occupying a player seat"
745        );
746    }
747    Ok(room_id)
748}
749
750async fn reestablish_room(
751    client: &mut RelayClient,
752    config: &Config,
753    engine_session: &SharedEngineSession,
754    snapshot: &SharedHostSnapshot,
755) -> Result<String, Box<dyn std::error::Error + Send + Sync>> {
756    let resume = {
757        let snap = snapshot.lock().map_err(|error| error.to_string())?;
758        match (&snap.resume_token, &snap.game, &snap.room_info) {
759            (Some(token), Some(game), Some(room_info)) => {
760                Some(resume_room_request(config, token, game, room_info))
761            }
762            _ => None,
763        }
764    };
765    let game_active = engine_session
766        .lock()
767        .map(|guard| guard.is_some())
768        .unwrap_or(false);
769
770    if let (Some(request), true) = (resume, game_active) {
771        let room_id = request.room_id.clone();
772        client.send(&ClientMessage::ResumeRoom(request)).await?;
773        match wait_for_room_resumed(client, engine_session, snapshot).await? {
774            Some(room) => {
775                let (last_state, seat_states, prompts, directives, player_order, pending_end_game) = {
776                    let mut snap = snapshot.lock().map_err(|error| error.to_string())?;
777                    snap.room_info = Some(room);
778                    (
779                        snap.last_state.clone(),
780                        snap.last_state_by_slot.clone(),
781                        snap.pending_prompts.clone(),
782                        snap.pending_directives.clone(),
783                        snap.game
784                            .as_ref()
785                            .map(|game| game.player_order.clone())
786                            .unwrap_or_default(),
787                        snap.pending_end_game.clone(),
788                    )
789                };
790                if let Some(state) = last_state {
791                    client
792                        .send(&ClientMessage::BroadcastState {
793                            state,
794                            target_player: None,
795                        })
796                        .await?;
797                }
798                for (slot, state) in seat_states {
799                    let Some(target_player) =
800                        parse_player_slot(&slot).and_then(|index| player_order.get(index).cloned())
801                    else {
802                        warn!(slot, "cannot replay private state without a mapped player");
803                        continue;
804                    };
805                    client
806                        .send(&ClientMessage::BroadcastState {
807                            state,
808                            target_player: Some(target_player),
809                        })
810                        .await?;
811                }
812                for (slot, prompt) in prompts {
813                    let Some(target_player) =
814                        parse_player_slot(&slot).and_then(|index| player_order.get(index).cloned())
815                    else {
816                        warn!(slot, "cannot replay prompt without a mapped player");
817                        continue;
818                    };
819                    client
820                        .send(&ClientMessage::BroadcastState {
821                            state: prompt,
822                            target_player: Some(target_player),
823                        })
824                        .await?;
825                }
826                for ((slot, _), directive) in directives {
827                    let Some(target_player) =
828                        parse_player_slot(&slot).and_then(|index| player_order.get(index).cloned())
829                    else {
830                        warn!(slot, "cannot replay directive without a mapped player");
831                        continue;
832                    };
833                    client
834                        .send(&ClientMessage::BroadcastState {
835                            state: directive,
836                            target_player: Some(target_player),
837                        })
838                        .await?;
839                }
840                if let Some(game_id) = pending_end_game {
841                    client.send(&ClientMessage::EndGame { game_id }).await?;
842                }
843                info!(room_id, "room resumed; snapshot re-broadcast");
844                return Ok(room_id);
845            }
846            None => {
847                warn!(room_id, "room resume rejected; abandoning hosted game");
848                abort_engine_session(engine_session);
849            }
850        }
851    }
852
853    establish_room(client, config, snapshot).await
854}
855
856fn engine_kind(config: &Config) -> EngineKind {
857    if matches!(config.backend, EngineBackendKind::Forge) {
858        EngineKind::Forge
859    } else {
860        EngineKind::Manabrew
861    }
862}
863
864fn resume_room_request(
865    config: &Config,
866    token: &str,
867    game: &GameStart,
868    room_info: &RoomInfo,
869) -> ResumeRoomRequest {
870    ResumeRoomRequest {
871        room_id: room_info.room_id.clone(),
872        resume_token: token.to_string(),
873        room_name: room_info.room_name.clone(),
874        max_players: room_info.max_players,
875        format: room_info.format.clone(),
876        hosted: !config.host_plays,
877        engine: engine_kind(config),
878        official_key: config.official_key.clone(),
879        password: config.room_password.clone(),
880        reconnect_timeout_s: Some(room_info.reconnect_timeout_s),
881        table_style: room_info.table_style.clone(),
882        draft_config: room_info.draft_config.clone(),
883        sealed_config: room_info.sealed_config.clone(),
884        player_order: game.player_order.clone(),
885        player_decks: game.player_decks.clone(),
886        starting_life: game.starting_life,
887        bot_players: room_info
888            .players
889            .iter()
890            .filter(|player| player.is_bot)
891            .map(|player| player.username.clone())
892            .collect(),
893        game_id: game.game_id.clone(),
894    }
895}
896
897async fn wait_for_room_resumed(
898    client: &mut RelayClient,
899    engine_session: &SharedEngineSession,
900    snapshot: &SharedHostSnapshot,
901) -> Result<Option<RoomInfo>, Box<dyn std::error::Error + Send + Sync>> {
902    loop {
903        match client.recv().await? {
904            Some(ServerMessage::RoomResumed { room }) => {
905                if room.status == RoomStatus::Lobby {
906                    abort_stale_engine_session(
907                        engine_session,
908                        snapshot,
909                        "resumed room is no longer in game",
910                    );
911                } else {
912                    concede_missing_active_seats(engine_session, snapshot, &room);
913                }
914                return Ok(Some(room));
915            }
916            Some(ServerMessage::RoomUpdate { room }) => {
917                if room.status == RoomStatus::Lobby {
918                    abort_stale_engine_session(
919                        engine_session,
920                        snapshot,
921                        "room reset while reconnecting",
922                    );
923                } else {
924                    concede_missing_active_seats(engine_session, snapshot, &room);
925                }
926                if let Ok(mut snap) = snapshot.lock() {
927                    snap.room_info = Some(room);
928                }
929            }
930            Some(ServerMessage::StateUpdate { from_player, state }) => {
931                let Ok(envelope) = serde_json::from_value::<StateEnvelope>(state.clone()) else {
932                    continue;
933                };
934                match envelope {
935                    StateEnvelope::Response { .. } => {
936                        route_remote_response(engine_session, snapshot, &from_player, &state);
937                    }
938                    StateEnvelope::Directive {
939                        from_player: claimed_slot,
940                        directive,
941                    } => route_remote_directive(
942                        engine_session,
943                        snapshot,
944                        &from_player,
945                        &claimed_slot,
946                        &directive,
947                    ),
948                    _ => {}
949                }
950            }
951            Some(ServerMessage::Error { code, message }) => {
952                warn!(code, message, "resume rejected by relay");
953                return Ok(None);
954            }
955            Some(other) => debug!(?other, "ignored message while waiting for room resume"),
956            None => return Err("relay closed while waiting for room resume".into()),
957        }
958    }
959}
960
961fn cancel_engine(engine_session: &SharedEngineSession) {
962    if let Ok(guard) = engine_session.lock() {
963        if let Some(EngineSession::Forge { cancel, .. }) = guard.as_ref() {
964            cancel.store(true, Ordering::Relaxed);
965        }
966    }
967}
968
969fn abort_engine_session(engine_session: &SharedEngineSession) {
970    cancel_engine(engine_session);
971    // Dropping the response channels makes the Manabrew backend's transports
972    // observe a disconnect and concede, which ends the engine thread.
973    clear_engine_session(engine_session);
974}
975
976fn abort_stale_engine_session(
977    engine_session: &SharedEngineSession,
978    snapshot: &SharedHostSnapshot,
979    reason: &str,
980) {
981    let stale_game_id = engine_session
982        .lock()
983        .ok()
984        .and_then(|guard| guard.as_ref().map(|s| s.game_id().to_string()));
985    let Some(stale_game_id) = stale_game_id else {
986        return;
987    };
988    warn!(game_id = %stale_game_id, reason, "aborting stale engine session");
989    abort_engine_session(engine_session);
990    clear_game_snapshot(snapshot, &stale_game_id);
991}
992
993fn clear_game_snapshot(snapshot: &SharedHostSnapshot, game_id: &str) {
994    if let Ok(mut snap) = snapshot.lock() {
995        if snap
996            .game
997            .as_ref()
998            .is_some_and(|game| game.game_id == game_id)
999        {
1000            snap.game = None;
1001            snap.last_state = None;
1002            snap.last_state_by_slot.clear();
1003            snap.pending_prompts.clear();
1004            snap.pending_directives.clear();
1005            snap.pending_end_game = None;
1006        }
1007    }
1008}
1009
1010/// A game must have at least one human seat. Seats survive disconnects and
1011/// concessions (reconnect grace / spectating), but an explicit leave removes
1012/// the seat for good — once every human seat is gone the game has no
1013/// stakeholders and ends now, instead of bots playing it out until the
1014/// relay's humanless sweep reclaims the room minutes later.
1015fn end_game_without_humans(
1016    engine_session: &SharedEngineSession,
1017    snapshot: &SharedHostSnapshot,
1018    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
1019    room: &RoomInfo,
1020) {
1021    if room.players.iter().any(|seat| !seat.is_bot) {
1022        return;
1023    }
1024    let game_id = engine_session
1025        .lock()
1026        .ok()
1027        .and_then(|guard| guard.as_ref().map(|s| s.game_id().to_string()));
1028    let Some(game_id) = game_id else {
1029        return;
1030    };
1031    info!(game_id, "no human seats remain; ending hosted game");
1032    abort_engine_session(engine_session);
1033    clear_game_snapshot(snapshot, &game_id);
1034    let _ = outbound_tx.send(ClientMessage::EndGame { game_id });
1035}
1036
1037fn spawn_bots(config: &Config, decks: &[DeckSelection], room_id: &str, bot_state: &SharedBotState) {
1038    stop_bots(bot_state);
1039    let mut guard = match bot_state.lock() {
1040        Ok(guard) => guard,
1041        Err(error) => {
1042            warn!(%error, "bot state lock poisoned");
1043            return;
1044        }
1045    };
1046    let open_seats = config.max_players.saturating_sub(1) as usize;
1047    for (index, deck) in decks.iter().take(open_seats).enumerate() {
1048        let tag = rand::thread_rng().gen_range(1000..10000);
1049        let username = format!("{}{}@{tag}", config.bot_username, index + 1);
1050        let relay_url = config.relay_url.clone();
1051        let bot_config = BotConfig {
1052            username,
1053            password: config.password.clone(),
1054            room_id: room_id.to_string(),
1055            room_password: config.room_password.clone(),
1056            deck_name: deck.name.clone(),
1057            deck: deck.deck.clone(),
1058            commander_name: deck.commander_name.clone(),
1059            agent: config.bot_agent,
1060            answer_delay_ms: None,
1061        };
1062        let shutdown = Arc::new(tokio::sync::Notify::new());
1063        let bot_shutdown = shutdown.clone();
1064        let handle = tokio::spawn(async move {
1065            if let Err(error) = run_bot(relay_url, bot_config, bot_shutdown).await {
1066                error!(%error, "bot task exited");
1067            }
1068        });
1069        guard.push(BotState { handle, shutdown });
1070    }
1071}
1072
1073fn stop_bots(bot_state: &SharedBotState) {
1074    match bot_state.lock() {
1075        Ok(mut state) => {
1076            for bot in state.drain(..) {
1077                bot.shutdown.notify_one();
1078                let mut handle = bot.handle;
1079                tokio::spawn(async move {
1080                    if time::timeout(BOT_STOP_TIMEOUT, &mut handle).await.is_err() {
1081                        handle.abort();
1082                    }
1083                });
1084            }
1085        }
1086        Err(error) => warn!(%error, "bot state lock poisoned"),
1087    }
1088}
1089
1090async fn wait_for_host_room(
1091    client: &mut RelayClient,
1092    config: &Config,
1093    snapshot: &SharedHostSnapshot,
1094) -> Result<String, Box<dyn std::error::Error + Send + Sync>> {
1095    loop {
1096        match client.recv().await? {
1097            Some(ServerMessage::RoomCreated {
1098                room_id,
1099                room_name,
1100                room,
1101                resume_token,
1102            }) => {
1103                info!(room_id, room_name, "room created");
1104                if let Ok(mut snap) = snapshot.lock() {
1105                    snap.resume_token = resume_token;
1106                    snap.room_info = Some(room);
1107                }
1108                return Ok(room_id);
1109            }
1110            Some(ServerMessage::RoomUpdate { room }) if room.host == config.username => {
1111                info!(room_id = %room.room_id, room_name = %room.room_name, "host room update");
1112                return Ok(room.room_id);
1113            }
1114            Some(ServerMessage::AuthResult { success, error, .. }) => {
1115                if !success {
1116                    return Err(format!("authentication failed: {error:?}").into());
1117                }
1118            }
1119            Some(ServerMessage::Error { code, message }) => {
1120                return Err(format!("server error while creating room: {code}: {message}").into());
1121            }
1122            Some(other) => debug!(?other, "ignored message while waiting for room creation"),
1123            None => return Err("relay closed while waiting for room creation".into()),
1124        }
1125    }
1126}
1127
1128async fn seat_client(
1129    client: &mut RelayClient,
1130    deck: &DeckSelection,
1131) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
1132    client
1133        .send(&ClientMessage::SetDeckSelection {
1134            deck_name: deck.name.clone(),
1135            deck: deck.deck.clone(),
1136            published_deck_id: None,
1137            commander_name: deck.commander_name.clone(),
1138            avatar_url: None,
1139        })
1140        .await?;
1141    client
1142        .send(&ClientMessage::SetReady { ready: true })
1143        .await?;
1144    info!(
1145        username = %client.username,
1146        deck = %deck.name,
1147        cards = deck.deck.cards.len(),
1148        "selected deck and marked ready"
1149    );
1150    Ok(())
1151}
1152
1153async fn run_client_loop(
1154    client: &mut RelayClient,
1155    config: &Config,
1156    room_id: &str,
1157    engine_session: &SharedEngineSession,
1158    snapshot: &SharedHostSnapshot,
1159    bot_state: &SharedBotState,
1160    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
1161    outbound_rx: &mut tokio_mpsc::UnboundedReceiver<ClientMessage>,
1162    cancel: &RoomCancel,
1163    bridge: Option<&Arc<ShellBridge>>,
1164    bridge_rx: &mut tokio_mpsc::UnboundedReceiver<ShellCommand>,
1165) -> LoopExit {
1166    let mut heartbeat = time::interval(Duration::from_secs(30));
1167    heartbeat.set_missed_tick_behavior(time::MissedTickBehavior::Delay);
1168    let mut drain_tick = time::interval(Duration::from_secs(5));
1169    drain_tick.set_missed_tick_behavior(time::MissedTickBehavior::Delay);
1170    let mut bot_usernames: HashSet<String> = HashSet::new();
1171    let mut forge_ai = config.forge_ai;
1172
1173    loop {
1174        tokio::select! {
1175            _ = cancel.notified() => {
1176                info!(username = %client.username, "room host cancelled; shutting down");
1177                cancel_engine(engine_session);
1178                return LoopExit::Cancelled;
1179            }
1180            _ = drain_tick.tick() => {
1181                let idle = engine_session
1182                    .lock()
1183                    .map(|guard| guard.is_none())
1184                    .unwrap_or(false);
1185                if draining() && idle {
1186                    info!(username = %client.username, "draining and no active game; closing room");
1187                    return LoopExit::Cancelled;
1188                }
1189            }
1190            _ = heartbeat.tick() => {
1191                if let Err(error) = client.broadcast_room_message(SELF_HOSTED_NODE_PROTOCOL, json!({
1192                    "type": "heartbeat",
1193                    "node": client.username,
1194                    "capacity": config.max_games,
1195                })).await {
1196                    warn!(%error, username = %client.username, "relay send failed");
1197                    return LoopExit::Disconnected;
1198                }
1199            }
1200            outbound = outbound_rx.recv() => {
1201                let Some(outbound) = outbound else {
1202                    warn!(username = %client.username, "outbound channel closed");
1203                    return LoopExit::Cancelled;
1204                };
1205                if let ClientMessage::BroadcastState { state, target_player: Some(target) } = &outbound {
1206                    if bridge.is_some_and(|shell| shell.try_send(target, state)) {
1207                        continue;
1208                    }
1209                }
1210                if let Err(error) = client.send(&outbound).await {
1211                    warn!(%error, username = %client.username, "relay send failed");
1212                    return LoopExit::Disconnected;
1213                }
1214            }
1215            command = bridge_rx.recv() => {
1216                let Some(command) = command else {
1217                    warn!(username = %client.username, "shell bridge closed");
1218                    return LoopExit::Cancelled;
1219                };
1220                let Some(shell) = bridge else { continue };
1221                match command {
1222                    ShellCommand::Serving { seats } => shell.set_serving(seats),
1223                    ShellCommand::Signal { to, payload } => {
1224                        if let Err(error) = client
1225                            .send(&ClientMessage::SignalPeer { to, payload })
1226                            .await
1227                        {
1228                            warn!(%error, username = %client.username, "relay send failed");
1229                            return LoopExit::Disconnected;
1230                        }
1231                    }
1232                    ShellCommand::SeatEnvelope { from, envelope } => {
1233                        // `from` came from a relay-attested roster.
1234                        route_seat_envelope(engine_session, snapshot, &from, &envelope);
1235                    }
1236                }
1237            }
1238            message = client.recv() => {
1239                let message = match message {
1240                    Ok(Some(message)) => message,
1241                    Ok(None) => {
1242                        warn!(username = %client.username, "relay websocket closed");
1243                        return LoopExit::Disconnected;
1244                    }
1245                    Err(error) => {
1246                        warn!(%error, username = %client.username, "relay receive failed");
1247                        return LoopExit::Disconnected;
1248                    }
1249                };
1250                if let Err(error) = handle_server_message(
1251                    client,
1252                    config,
1253                    room_id,
1254                    engine_session,
1255                    snapshot,
1256                    bot_state,
1257                    outbound_tx,
1258                    &mut bot_usernames,
1259                    &mut forge_ai,
1260                    bridge,
1261                    message,
1262                ).await {
1263                    warn!(%error, username = %client.username, "relay send failed");
1264                    return LoopExit::Disconnected;
1265                }
1266            }
1267        }
1268    }
1269}
1270
1271async fn handle_server_message(
1272    client: &mut RelayClient,
1273    config: &Config,
1274    room_id: &str,
1275    engine_session: &SharedEngineSession,
1276    snapshot: &SharedHostSnapshot,
1277    bot_state: &SharedBotState,
1278    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
1279    bot_usernames: &mut HashSet<String>,
1280    forge_ai: &mut bool,
1281    bridge: Option<&Arc<ShellBridge>>,
1282    message: ServerMessage,
1283) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
1284    match message {
1285        ServerMessage::RoomUpdate { room } => {
1286            if let Ok(mut snap) = snapshot.lock() {
1287                snap.room_info = Some(room.clone());
1288            }
1289            *bot_usernames = room
1290                .players
1291                .iter()
1292                .filter(|p| p.is_bot)
1293                .map(|p| p.username.clone())
1294                .collect();
1295            log_room_update(&client.username, &room);
1296            if room.status == RoomStatus::Lobby {
1297                abort_stale_engine_session(engine_session, snapshot, "room reset to lobby");
1298            } else {
1299                end_game_without_humans(engine_session, snapshot, outbound_tx, &room);
1300            }
1301            maybe_auto_start_room(client, config, &room).await?;
1302        }
1303        ServerMessage::StateUpdate { from_player, state } => {
1304            if let Some(shell) = bridge {
1305                shell.note_relay_message(&from_player);
1306            }
1307            handle_state_update(
1308                client,
1309                config,
1310                room_id,
1311                engine_session,
1312                snapshot,
1313                bot_state,
1314                forge_ai,
1315                from_player,
1316                state,
1317            )
1318            .await?;
1319        }
1320        ServerMessage::PeerSignal { from, payload } => match bridge {
1321            Some(shell) => shell.forward_signal(&from, payload),
1322            None => debug!(from, "signalling arrived with no shell to hand it to"),
1323        },
1324        ServerMessage::ReadyStateChanged { username, ready } => {
1325            info!(username, ready, observer = %client.username, "ready changed");
1326        }
1327        ServerMessage::PlayerJoined { username, room_id } => {
1328            info!(username, room_id, observer = %client.username, "player joined");
1329        }
1330        ServerMessage::PlayerLeft { username, room_id } => {
1331            info!(username, room_id, observer = %client.username, "player left");
1332            let leaver_still_needed =
1333                active_player_usernames(snapshot).is_some_and(|active| active.contains(&username));
1334            if leaver_still_needed {
1335                if let Some(index) = seat_index_of(snapshot, &username) {
1336                    info!(
1337                        username,
1338                        index, "player abandoned mid-game; conceding their seat"
1339                    );
1340                    concede_seat(engine_session, index);
1341                }
1342            }
1343        }
1344        ServerMessage::GameStarted {
1345            room_id,
1346            game_id,
1347            player_order,
1348            player_decks,
1349            starting_life,
1350        } => {
1351            info!(room_id, game_id, ?player_order, observer = %client.username, "game started");
1352            abort_stale_engine_session(engine_session, snapshot, "relay started a new game");
1353            if let Ok(mut snap) = snapshot.lock() {
1354                snap.game = Some(GameStart {
1355                    game_id: game_id.clone(),
1356                    player_order: player_order.clone(),
1357                    player_decks: player_decks.clone(),
1358                    starting_life,
1359                });
1360                snap.last_state = None;
1361                snap.last_state_by_slot.clear();
1362                snap.pending_prompts.clear();
1363                snap.pending_directives.clear();
1364                snap.pending_end_game = None;
1365            }
1366            let mut left_the_relay = Vec::new();
1367            if let Some(shell) = bridge {
1368                let seats = shell.freeze_for_game(&player_order, &[]);
1369                if !seats.is_empty() {
1370                    info!(
1371                        game_id,
1372                        ?seats,
1373                        "seats playing this game through the webview"
1374                    );
1375                    left_the_relay.extend(shell.transport_report(&seats));
1376                }
1377            }
1378            if !left_the_relay.is_empty() {
1379                let _ = outbound_tx.send(ClientMessage::ReportTransport {
1380                    game_id: game_id.clone(),
1381                    seats: left_the_relay,
1382                });
1383            }
1384            maybe_start_hosted_engine(
1385                config,
1386                engine_session,
1387                snapshot,
1388                outbound_tx,
1389                game_id,
1390                player_order,
1391                player_decks,
1392                starting_life,
1393                bot_usernames,
1394                *forge_ai,
1395            );
1396        }
1397        ServerMessage::RoomTransport { members, .. } => {
1398            if let Some(shell) = bridge {
1399                // Empty until every seat has opted in.
1400                shell.set_roster(members.iter().map(|member| member.username.clone()));
1401            }
1402        }
1403        ServerMessage::ServerShuttingDown { reconnect_in_s } => {
1404            info!(reconnect_in_s, observer = %client.username, "relay is restarting");
1405        }
1406        ServerMessage::Error { code, message } => {
1407            warn!(code, message, observer = %client.username, "server error");
1408        }
1409        other => {
1410            debug!(?other, observer = %client.username, "server message");
1411        }
1412    }
1413
1414    Ok(())
1415}
1416
1417async fn handle_state_update(
1418    client: &mut RelayClient,
1419    config: &Config,
1420    room_id: &str,
1421    engine_session: &SharedEngineSession,
1422    snapshot: &SharedHostSnapshot,
1423    bot_state: &SharedBotState,
1424    forge_ai: &mut bool,
1425    from_player: String,
1426    state: Value,
1427) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
1428    let Ok(envelope) = serde_json::from_value::<StateEnvelope>(state.clone()) else {
1429        debug!(from_player, state = %state, observer = %client.username, "state update");
1430        return Ok(());
1431    };
1432
1433    match envelope {
1434        StateEnvelope::Response { .. } | StateEnvelope::Directive { .. } => {
1435            route_seat_envelope(engine_session, snapshot, &from_player, &state);
1436            Ok(())
1437        }
1438        StateEnvelope::RoomRelay {
1439            protocol,
1440            from_player: _,
1441            room_id: requested_room_id,
1442            payload,
1443            ..
1444        } => {
1445            info!(
1446                from_player,
1447                protocol,
1448                payload = %payload,
1449                observer = %client.username,
1450                "room relay message"
1451            );
1452            if protocol != SELF_HOSTED_NODE_PROTOCOL {
1453                return Ok(());
1454            }
1455            let payload_type = payload.get("type").and_then(Value::as_str);
1456            match payload_type {
1457                Some("removeBot") => {
1458                    info!(observer = %client.username, "received removeBot request");
1459                    stop_bots(bot_state);
1460                }
1461                Some("spawnBot") => {
1462                    let effective_room_id = requested_room_id.as_deref().unwrap_or(room_id);
1463                    if effective_room_id == room_id {
1464                        *forge_ai = payload
1465                            .get("forgeAi")
1466                            .and_then(Value::as_bool)
1467                            .unwrap_or(config.forge_ai);
1468                        info!(observer = %client.username, room_id, forge_ai = *forge_ai, "received spawnBot request");
1469                        let bot_decks = bot_decks_from_payload(config, &payload);
1470                        spawn_bots(config, &bot_decks, room_id, bot_state);
1471                    } else {
1472                        debug!(
1473                            observer = %client.username,
1474                            current_room_id = room_id,
1475                            requested_room_id = effective_room_id,
1476                            "ignoring spawnBot request for another room"
1477                        );
1478                    }
1479                }
1480                _ => {}
1481            }
1482            Ok(())
1483        }
1484        _ => {
1485            debug!(from_player, state = %state, observer = %client.username, "state update");
1486            Ok(())
1487        }
1488    }
1489}
1490
1491fn bot_decks_from_payload(config: &Config, payload: &Value) -> Vec<DeckSelection> {
1492    let parsed = serde_json::from_value::<SpawnBotPayload>(payload.clone()).ok();
1493    let requested = match parsed {
1494        Some(SpawnBotPayload {
1495            decks: Some(decks), ..
1496        }) if !decks.is_empty() => decks,
1497        Some(SpawnBotPayload {
1498            deck: Some(deck), ..
1499        }) => vec![deck],
1500        _ => return vec![config.bot_deck.clone()],
1501    };
1502    requested
1503        .into_iter()
1504        .map(deck_selection_from_payload)
1505        .collect()
1506}
1507
1508fn deck_selection_from_payload(deck: SpawnBotDeckPayload) -> DeckSelection {
1509    info!(
1510        deck = %deck.deck_name,
1511        cards = deck.deck.cards.len(),
1512        commander = ?deck.commander_name,
1513        "using requested bot deck"
1514    );
1515    let commander_name = deck.commander_name.or_else(|| {
1516        deck.deck
1517            .commanders
1518            .as_ref()
1519            .and_then(|commanders| commanders.first())
1520            .map(|commander| commander.identity.name.clone())
1521    });
1522    DeckSelection {
1523        name: deck.deck_name,
1524        deck: deck.deck,
1525        commander_name,
1526    }
1527}
1528
1529async fn maybe_auto_start_room(
1530    client: &mut RelayClient,
1531    config: &Config,
1532    room: &RoomInfo,
1533) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
1534    if !config.auto_start || draining() {
1535        return Ok(());
1536    }
1537    if config.format == GameFormat::Any {
1538        return Ok(());
1539    }
1540    if room.host != config.username
1541        || room.status != RoomStatus::Lobby
1542        || room.players.len() < config.max_players as usize
1543    {
1544        return Ok(());
1545    }
1546    let all_ready = room
1547        .players
1548        .iter()
1549        .all(|player| player.connected && player.ready && player.selected_deck_name.is_some());
1550    if all_ready {
1551        info!(
1552            room_id = %room.room_id,
1553            players = room.players.len(),
1554            "all players ready; auto-starting hosted game"
1555        );
1556        client
1557            .send(&ClientMessage::StartGame { format: None })
1558            .await?;
1559    }
1560    Ok(())
1561}
1562
1563const ENGINE_STACK_BYTES: usize = 64 * 1024 * 1024;
1564
1565fn spawn_engine_thread<F: FnOnce() + Send + 'static>(body: F) {
1566    if let Err(error) = thread::Builder::new()
1567        .name("hosted-engine".to_string())
1568        .stack_size(ENGINE_STACK_BYTES)
1569        .spawn(body)
1570    {
1571        error!(%error, "failed to spawn hosted engine thread");
1572    }
1573}
1574
1575// Any stays commander-capable: hosted "Any" rooms resolve their real format
1576// at StartGame, and dropping commanders there would break commander games.
1577fn is_commander_variant(format: GameFormat) -> bool {
1578    match format {
1579        GameFormat::Any
1580        | GameFormat::Commander
1581        | GameFormat::PauperCommander
1582        | GameFormat::Brawl
1583        | GameFormat::Oathbreaker => true,
1584        GameFormat::Standard
1585        | GameFormat::Pioneer
1586        | GameFormat::Modern
1587        | GameFormat::Legacy
1588        | GameFormat::Vintage
1589        | GameFormat::Pauper
1590        | GameFormat::Premodern
1591        | GameFormat::Draft
1592        | GameFormat::Sealed => false,
1593    }
1594}
1595
1596// Forge GameType name for the java backend; empty = let the adapter infer
1597// commander-ness the pre-variant way (Any rooms carry no concrete format here).
1598fn java_game_variant(format: GameFormat) -> &'static str {
1599    match format {
1600        GameFormat::Any => "",
1601        GameFormat::Commander => "Commander",
1602        GameFormat::PauperCommander => "PauperCommander",
1603        GameFormat::Brawl => "Brawl",
1604        GameFormat::Oathbreaker => "Oathbreaker",
1605        GameFormat::Standard
1606        | GameFormat::Pioneer
1607        | GameFormat::Modern
1608        | GameFormat::Legacy
1609        | GameFormat::Vintage
1610        | GameFormat::Pauper
1611        | GameFormat::Premodern
1612        | GameFormat::Draft
1613        | GameFormat::Sealed => "Constructed",
1614    }
1615}
1616
1617fn maybe_start_hosted_engine(
1618    config: &Config,
1619    engine_session: &SharedEngineSession,
1620    snapshot: &SharedHostSnapshot,
1621    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
1622    game_id: String,
1623    player_order: Vec<String>,
1624    player_decks: Vec<PlayerDeckInfo>,
1625    starting_life: i32,
1626    bot_usernames: &HashSet<String>,
1627    forge_ai: bool,
1628) {
1629    if !config.engine_enabled {
1630        debug!("hosted engine disabled for this node");
1631        return;
1632    }
1633    let backend = config.backend;
1634    if !backend.is_supported() {
1635        if let Err(error) = java_backend::JavaRuntimeConfig::from_env().validate() {
1636            warn!(
1637                backend = backend.label(),
1638                %error,
1639                "hosted engine backend runtime is not ready"
1640            );
1641            return;
1642        }
1643        warn!(
1644            backend = backend.label(),
1645            message = java_backend::unsupported_message(),
1646            "hosted engine backend is not implemented yet"
1647        );
1648        return;
1649    }
1650
1651    let local_player_index = if config.host_plays {
1652        match player_order
1653            .iter()
1654            .position(|name| name == &config.username)
1655        {
1656            Some(index) => Some(index),
1657            None => {
1658                warn!(
1659                    username = %config.username,
1660                    ?player_order,
1661                    "room node is configured as a player but is not in player order; not starting engine"
1662                );
1663                return;
1664            }
1665        }
1666    } else {
1667        None
1668    };
1669
1670    let mut guard = match engine_session.lock() {
1671        Ok(guard) => guard,
1672        Err(error) => {
1673            warn!(%error, "engine session lock poisoned");
1674            return;
1675        }
1676    };
1677    if let Some(session) = guard.as_ref() {
1678        warn!(
1679            game_id,
1680            stale_game_id = session.game_id(),
1681            "engine session still present at game start; not starting engine"
1682        );
1683        return;
1684    }
1685
1686    let session_handle = engine_session.clone();
1687    let num_players = player_order.len();
1688    if num_players < 2 {
1689        warn!(num_players, "not enough players to start hosted engine");
1690        return;
1691    }
1692
1693    let mut deck_map: HashMap<String, PlayerDeckInfo> = player_decks
1694        .into_iter()
1695        .map(|deck| (deck.username.clone(), deck))
1696        .collect();
1697    let mut ordered_decks = Vec::with_capacity(num_players);
1698    let mut commander_names = Vec::with_capacity(num_players);
1699    let mut ai_player_indices = Vec::new();
1700    let mut bot_player_indices = Vec::new();
1701    for (index, username) in player_order.iter().enumerate() {
1702        let Some(deck) = deck_map.remove(username) else {
1703            warn!(username, "missing deck for player; not starting engine");
1704            return;
1705        };
1706        if bot_usernames.contains(username) {
1707            if forge_ai {
1708                ai_player_indices.push(index);
1709            } else {
1710                bot_player_indices.push(index);
1711            }
1712        }
1713        ordered_decks.push(deck.deck);
1714        commander_names.push(deck.commander_name);
1715    }
1716
1717    let player_names = player_order;
1718    let room_format = snapshot
1719        .lock()
1720        .ok()
1721        .and_then(|snap| snap.room_info.as_ref().map(|room| room.format.clone()))
1722        .unwrap_or(GameFormat::Any);
1723    let commander_variant = is_commander_variant(room_format.clone());
1724    let game_variant = java_game_variant(room_format).to_string();
1725
1726    match backend {
1727        EngineBackendKind::Manabrew => {
1728            let (remote_prompt_tx, remote_prompt_rx) = std_mpsc::channel::<(usize, AgentMessage)>();
1729            let mut remote_response_txs = HashMap::new();
1730            let mut remote_response_rxs = Vec::new();
1731            for i in 0..num_players {
1732                if Some(i) == local_player_index {
1733                    continue;
1734                }
1735                let (response_tx, response_rx) = std_mpsc::channel::<ClientToServerMessage>();
1736                remote_response_txs.insert(i, response_tx);
1737                remote_response_rxs.push((i, response_rx));
1738            }
1739            let engine_clock = EngineClock::default();
1740            *guard = Some(EngineSession::Manabrew {
1741                game_id: game_id.clone(),
1742                remote_response_txs,
1743                engine_clock: engine_clock.clone(),
1744                local_bot_answers: LocalBotAnswers::default(),
1745            });
1746            drop(guard);
1747
1748            spawn_remote_prompt_forwarder(
1749                outbound_tx.clone(),
1750                snapshot.clone(),
1751                engine_session.clone(),
1752                game_id.clone(),
1753                remote_prompt_rx,
1754                Some(player_names.clone()),
1755                config.state_delta,
1756                engine_clock,
1757                local_bot_seats(config, false, &player_names, bot_usernames),
1758            );
1759            let (game_over_tx, game_over_rx) = std_mpsc::channel::<HostedGameOver>();
1760            spawn_game_over_forwarder(
1761                outbound_tx.clone(),
1762                game_over_rx,
1763                engine_session.clone(),
1764                snapshot.clone(),
1765                game_id.clone(),
1766                Some(player_names.clone()),
1767            );
1768            let outbound_tx = outbound_tx.clone();
1769            let snapshot = snapshot.clone();
1770            spawn_engine_thread(move || {
1771                info!(
1772                    game_id,
1773                    backend = backend.label(),
1774                    players = num_players,
1775                    local_player_index,
1776                    "starting hosted engine thread"
1777                );
1778                crate::metrics::record_engine_session_started();
1779                let started = Instant::now();
1780                let result = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
1781                    rust_backend::run_hosted_engine_game(
1782                        game_id.clone(),
1783                        player_names,
1784                        ordered_decks,
1785                        commander_names,
1786                        local_player_index,
1787                        starting_life,
1788                        remote_prompt_tx,
1789                        remote_response_rxs,
1790                        game_over_tx,
1791                    )
1792                }));
1793                finish_hosted_engine(
1794                    result,
1795                    &game_id,
1796                    num_players,
1797                    started,
1798                    &outbound_tx,
1799                    &session_handle,
1800                    &snapshot,
1801                );
1802            });
1803        }
1804        EngineBackendKind::Forge => {
1805            let (remote_prompt_tx, remote_prompt_rx) = std_mpsc::channel::<(usize, AgentMessage)>();
1806            let mut remote_response_txs = HashMap::new();
1807            let mut remote_response_rxs = Vec::new();
1808            for i in 0..num_players {
1809                if Some(i) == local_player_index {
1810                    continue;
1811                }
1812                let (response_tx, response_rx) = std_mpsc::channel::<ClientToServerMessage>();
1813                remote_response_txs.insert(i, response_tx);
1814                remote_response_rxs.push((i, response_rx));
1815            }
1816            let cancel = Arc::new(AtomicBool::new(false));
1817            let engine_clock = EngineClock::default();
1818            *guard = Some(EngineSession::Forge {
1819                game_id: game_id.clone(),
1820                remote_response_txs,
1821                cancel: cancel.clone(),
1822                engine_clock: engine_clock.clone(),
1823                local_bot_answers: LocalBotAnswers::default(),
1824            });
1825            drop(guard);
1826
1827            spawn_remote_prompt_forwarder(
1828                outbound_tx.clone(),
1829                snapshot.clone(),
1830                engine_session.clone(),
1831                game_id.clone(),
1832                remote_prompt_rx,
1833                Some(player_names.clone()),
1834                config.state_delta,
1835                engine_clock,
1836                local_bot_seats(config, forge_ai, &player_names, bot_usernames),
1837            );
1838            let (game_over_tx, game_over_rx) = std_mpsc::channel::<HostedGameOver>();
1839            spawn_game_over_forwarder(
1840                outbound_tx.clone(),
1841                game_over_rx,
1842                engine_session.clone(),
1843                snapshot.clone(),
1844                game_id.clone(),
1845                Some(player_names.clone()),
1846            );
1847            let outbound_tx = outbound_tx.clone();
1848            let snapshot = snapshot.clone();
1849            spawn_engine_thread(move || {
1850                info!(
1851                    game_id,
1852                    backend = backend.label(),
1853                    players = num_players,
1854                    local_player_index,
1855                    "starting hosted engine thread"
1856                );
1857                crate::metrics::record_engine_session_started();
1858                let started = Instant::now();
1859                let result = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
1860                    java_backend::run_hosted_engine_game(
1861                        game_id.clone(),
1862                        player_names,
1863                        ordered_decks,
1864                        commander_names,
1865                        commander_variant,
1866                        game_variant,
1867                        local_player_index,
1868                        ai_player_indices,
1869                        bot_player_indices,
1870                        starting_life,
1871                        remote_prompt_tx,
1872                        remote_response_rxs,
1873                        game_over_tx,
1874                        cancel,
1875                    )
1876                }));
1877                finish_hosted_engine(
1878                    result,
1879                    &game_id,
1880                    num_players,
1881                    started,
1882                    &outbound_tx,
1883                    &session_handle,
1884                    &snapshot,
1885                );
1886            });
1887        }
1888    }
1889}
1890
1891fn clear_engine_session(engine_session: &SharedEngineSession) {
1892    match engine_session.lock() {
1893        Ok(mut guard) => *guard = None,
1894        Err(error) => warn!(%error, "engine session lock poisoned on reset"),
1895    }
1896}
1897
1898/// Usernames of engine players the game still needs — seated in the current
1899/// game and not yet eliminated per the last state broadcast. Everyone else
1900/// (eliminated players, non-playing members) is effectively a spectator whose
1901/// absence never matters. `None` = no hosted game.
1902fn active_player_usernames(snapshot: &SharedHostSnapshot) -> Option<HashSet<String>> {
1903    let snap = snapshot.lock().ok()?;
1904    let game = snap.game.as_ref()?;
1905    let mut active: HashSet<String> = game.player_order.iter().cloned().collect();
1906    let players = snap
1907        .last_state
1908        .as_ref()
1909        .and_then(|state| state.pointer("/state/gameView/players"))
1910        .and_then(Value::as_array);
1911    if let Some(players) = players {
1912        for player in players {
1913            let playing = player
1914                .get("status")
1915                .and_then(Value::as_str)
1916                .is_none_or(|status| status == "playing");
1917            if playing {
1918                continue;
1919            }
1920            let index = player
1921                .get("id")
1922                .and_then(Value::as_str)
1923                .and_then(parse_player_slot);
1924            if let Some(username) = index.and_then(|i| game.player_order.get(i)) {
1925                active.remove(username);
1926            }
1927        }
1928    }
1929    Some(active)
1930}
1931
1932fn seat_index_of(snapshot: &SharedHostSnapshot, username: &str) -> Option<usize> {
1933    let snap = snapshot.lock().ok()?;
1934    let game = snap.game.as_ref()?;
1935    game.player_order.iter().position(|name| name == username)
1936}
1937
1938fn concede_missing_active_seats(
1939    engine_session: &SharedEngineSession,
1940    snapshot: &SharedHostSnapshot,
1941    room: &RoomInfo,
1942) {
1943    let Some(active) = active_player_usernames(snapshot) else {
1944        return;
1945    };
1946    let present: HashSet<&str> = room
1947        .players
1948        .iter()
1949        .map(|player| player.username.as_str())
1950        .collect();
1951    for username in active {
1952        if !present.contains(username.as_str()) {
1953            if let Some(index) = seat_index_of(snapshot, &username) {
1954                concede_seat(engine_session, index);
1955            }
1956        }
1957    }
1958}
1959
1960fn concede_seat(engine_session: &SharedEngineSession, player_index: usize) {
1961    send_seat_message(
1962        engine_session,
1963        player_index,
1964        ClientToServerMessage::Directive {
1965            directive: manabrew_protocol::transport::DirectiveInput::Concede,
1966        },
1967    );
1968}
1969
1970fn send_seat_message(
1971    engine_session: &SharedEngineSession,
1972    player_index: usize,
1973    message: ClientToServerMessage,
1974) {
1975    let guard = match engine_session.lock() {
1976        Ok(guard) => guard,
1977        Err(error) => {
1978            warn!(%error, "engine session lock poisoned");
1979            return;
1980        }
1981    };
1982    let Some(session) = guard.as_ref() else {
1983        debug!(player_index, "no engine session for seat action");
1984        return;
1985    };
1986    let txs = match session {
1987        EngineSession::Manabrew {
1988            remote_response_txs,
1989            ..
1990        }
1991        | EngineSession::Forge {
1992            remote_response_txs,
1993            ..
1994        } => remote_response_txs,
1995    };
1996    let Some(tx) = txs.get(&player_index) else {
1997        debug!(player_index, "no response channel for player");
1998        return;
1999    };
2000    if let Err(error) = tx.send(message) {
2001        warn!(player_index, %error, "failed to route seat message");
2002    }
2003}
2004
2005fn finish_hosted_engine(
2006    result: std::thread::Result<Result<(), String>>,
2007    game_id: &str,
2008    players: usize,
2009    started: Instant,
2010    outbound_tx: &tokio_mpsc::UnboundedSender<ClientMessage>,
2011    session_handle: &SharedEngineSession,
2012    snapshot: &SharedHostSnapshot,
2013) {
2014    let fatal = match result {
2015        Ok(Ok(())) => {
2016            info!(game_id, "hosted engine thread finished");
2017            None
2018        }
2019        Ok(Err(message)) => {
2020            error!(game_id, message, "hosted engine exited with a fatal error");
2021            Some(message)
2022        }
2023        Err(panic) => {
2024            let message = if let Some(message) = panic.downcast_ref::<String>() {
2025                message.clone()
2026            } else if let Some(message) = panic.downcast_ref::<&str>() {
2027                message.to_string()
2028            } else {
2029                "the host engine panicked".to_string()
2030            };
2031            error!(game_id, message, "hosted engine thread panicked");
2032            Some(message)
2033        }
2034    };
2035    crate::metrics::record_engine_session_finished(players, started, fatal.as_deref());
2036    let still_owner = session_handle
2037        .lock()
2038        .map(|guard| guard.as_ref().is_some_and(|s| s.game_id() == game_id))
2039        .unwrap_or(false);
2040    if let Some(message) = fatal {
2041        if still_owner {
2042            if let Ok(mut snap) = snapshot.lock() {
2043                if snap
2044                    .game
2045                    .as_ref()
2046                    .is_some_and(|game| game.game_id == game_id)
2047                {
2048                    snap.pending_end_game = Some(game_id.to_string());
2049                }
2050            }
2051            let _ = outbound_tx.send(ClientMessage::ReportGameOutcome {
2052                game_id: game_id.to_string(),
2053                outcome: GameOutcomeReport {
2054                    fatal_message: Some(message.clone()),
2055                    ..GameOutcomeReport::default()
2056                },
2057            });
2058            if let Ok(state) = serde_json::to_value(StateEnvelope::Fatal { message }) {
2059                let _ = outbound_tx.send(ClientMessage::BroadcastState {
2060                    state,
2061                    target_player: None,
2062                });
2063            }
2064            let _ = outbound_tx.send(ClientMessage::EndGame {
2065                game_id: game_id.to_string(),
2066            });
2067        } else {
2068            warn!(game_id, message, "stale engine session finished with error");
2069        }
2070    }
2071}
2072
2073/// `from_player` is relay-attested; never client supplied.
2074fn route_seat_envelope(
2075    engine_session: &SharedEngineSession,
2076    snapshot: &SharedHostSnapshot,
2077    from_player: &str,
2078    state: &Value,
2079) {
2080    match serde_json::from_value::<StateEnvelope>(state.clone()) {
2081        Ok(StateEnvelope::Response { .. }) => {
2082            route_remote_response(engine_session, snapshot, from_player, state)
2083        }
2084        Ok(StateEnvelope::Directive {
2085            from_player: claimed_slot,
2086            directive,
2087        }) => route_remote_directive(
2088            engine_session,
2089            snapshot,
2090            from_player,
2091            &claimed_slot,
2092            &directive,
2093        ),
2094        _ => debug!(from_player, state = %state, "unroutable seat envelope"),
2095    }
2096}
2097
2098fn route_remote_response(
2099    engine_session: &SharedEngineSession,
2100    snapshot: &SharedHostSnapshot,
2101    authenticated_username: &str,
2102    state: &Value,
2103) {
2104    let envelope: StateEnvelope = match serde_json::from_value(state.clone()) {
2105        Ok(envelope) => envelope,
2106        Err(error) => {
2107            warn!(%error, state = %state, "relay response invalid envelope");
2108            return;
2109        }
2110    };
2111    let StateEnvelope::Response {
2112        from_player,
2113        prompt_id,
2114        action: action_value,
2115    } = envelope
2116    else {
2117        warn!(state = %state, "expected response envelope");
2118        return;
2119    };
2120    // Zero is reserved for in-process absent-player defaults, never client input.
2121    if prompt_id == 0 {
2122        warn!(from_player, "relay response has no prompt identity");
2123        return;
2124    }
2125    let Some(player_index) =
2126        authenticated_player_index(snapshot, authenticated_username, &from_player, "response")
2127    else {
2128        return;
2129    };
2130    let action: PromptOutput = match serde_json::from_value(action_value) {
2131        Ok(action) => action,
2132        Err(error) => {
2133            warn!(from_player, %error, "relay response has invalid action");
2134            return;
2135        }
2136    };
2137
2138    let guard = match engine_session.lock() {
2139        Ok(guard) => guard,
2140        Err(error) => {
2141            warn!(%error, "engine session lock poisoned");
2142            return;
2143        }
2144    };
2145    let Some(session) = guard.as_ref() else {
2146        debug!(from_player, "no engine session for relay response");
2147        return;
2148    };
2149    if session.is_local_bot_duplicate(player_index, prompt_id) {
2150        debug!(
2151            from_player,
2152            prompt_id, "relay response already answered locally"
2153        );
2154        return;
2155    }
2156    let tx = match session {
2157        EngineSession::Manabrew {
2158            remote_response_txs,
2159            ..
2160        } => remote_response_txs.get(&player_index),
2161        EngineSession::Forge {
2162            remote_response_txs,
2163            ..
2164        } => {
2165            debug!(
2166                from_player,
2167                player_index, "routing relay response to java engine"
2168            );
2169            remote_response_txs.get(&player_index)
2170        }
2171    };
2172    let Some(tx) = tx else {
2173        debug!(from_player, player_index, "no response channel for player");
2174        return;
2175    };
2176    session.engine_clock().mark_response();
2177    if let Err(error) = tx.send(ClientToServerMessage::Response { prompt_id, action }) {
2178        warn!(from_player, %error, "failed to route relay response");
2179        return;
2180    }
2181    drop(guard);
2182    if let Ok(mut snap) = snapshot.lock() {
2183        snap.pending_prompts.remove(&from_player);
2184    }
2185}
2186
2187fn route_remote_directive(
2188    engine_session: &SharedEngineSession,
2189    snapshot: &SharedHostSnapshot,
2190    authenticated_username: &str,
2191    claimed_slot: &str,
2192    directive: &Value,
2193) {
2194    let directive: DirectiveInput = match serde_json::from_value(directive.clone()) {
2195        Ok(directive) => directive,
2196        Err(error) => {
2197            warn!(claimed_slot, %error, "relay directive is invalid");
2198            return;
2199        }
2200    };
2201    let Some(player_index) =
2202        authenticated_player_index(snapshot, authenticated_username, claimed_slot, "directive")
2203    else {
2204        return;
2205    };
2206    info!(claimed_slot, player_index, ?directive, "routing directive");
2207    match directive {
2208        DirectiveInput::Concede => concede_seat(engine_session, player_index),
2209        DirectiveInput::RequestRestore { .. }
2210        | DirectiveInput::RestoreVote { .. }
2211        | DirectiveInput::StartRecording
2212        | DirectiveInput::StopRecording
2213        | DirectiveInput::RequestRepeat { .. }
2214        | DirectiveInput::RepeatVote { .. } => send_seat_message(
2215            engine_session,
2216            player_index,
2217            ClientToServerMessage::Directive { directive },
2218        ),
2219        DirectiveInput::SetSnapshotRecording { .. } => {
2220            warn!(claimed_slot, "only the engine host sets snapshot recording");
2221        }
2222    }
2223}
2224
2225fn authenticated_player_index(
2226    snapshot: &SharedHostSnapshot,
2227    authenticated_username: &str,
2228    claimed_slot: &str,
2229    message_kind: &str,
2230) -> Option<usize> {
2231    let Some(player_index) = seat_index_of(snapshot, authenticated_username) else {
2232        warn!(
2233            authenticated_username,
2234            message_kind, "relay sender has no engine seat"
2235        );
2236        return None;
2237    };
2238    let authenticated_slot = player_slot(player_index);
2239    if claimed_slot != authenticated_slot {
2240        warn!(
2241            authenticated_username,
2242            claimed_slot,
2243            authenticated_slot,
2244            message_kind,
2245            "relay sender claimed another engine seat"
2246        );
2247        return None;
2248    }
2249    Some(player_index)
2250}
2251
2252/// Add the fingerprint of the carried state to a full state envelope, so a
2253/// receiver can tell whether a later patch applies to what it holds.
2254fn stamp_fingerprint(mut envelope: Value) -> Value {
2255    let Some(state) = envelope.get("state") else {
2256        return envelope;
2257    };
2258    let fingerprint = manabrew_relay_protocol::state_delta::fingerprint(state);
2259    if let Some(object) = envelope.as_object_mut() {
2260        object.insert("fingerprint".to_string(), Value::String(fingerprint));
2261    }
2262    envelope
2263}
2264
2265/// Time spent building this envelope after `engine_ms` was sampled: serialising
2266/// the state, fingerprinting it, diffing it. Must be stamped last, or it reaches
2267/// the dedup comparisons and the diff base, neither of which may see it.
2268fn stamp_emit_ms(mut envelope: Value, elapsed: Duration) -> Value {
2269    if let Some(object) = envelope.as_object_mut() {
2270        object.insert(
2271            "emitMs".to_string(),
2272            Value::from(elapsed.as_millis().min(u128::from(u32::MAX)) as u32),
2273        );
2274    }
2275    envelope
2276}
2277
2278/// Patch form of a fingerprinted state envelope, against the last one sent to
2279/// this seat. `None` the first time a seat is served, or if anything is missing,
2280/// and the caller falls back to the full state, which is always correct.
2281/// When the engine host last received a player's response, as milliseconds on
2282/// its own monotonic clock. The response path writes it and the emit path
2283/// reads it, so an outgoing envelope can say how long the engine took without
2284/// anyone comparing clocks across hosts.
2285#[derive(Clone, Default)]
2286pub struct EngineClock(std::sync::Arc<std::sync::atomic::AtomicU64>);
2287
2288impl EngineClock {
2289    fn now() -> u64 {
2290        use std::sync::OnceLock;
2291        static START: OnceLock<Instant> = OnceLock::new();
2292        START.get_or_init(Instant::now).elapsed().as_millis() as u64
2293    }
2294
2295    pub fn mark_response(&self) {
2296        self.0
2297            .store(Self::now(), std::sync::atomic::Ordering::Relaxed);
2298    }
2299
2300    /// Cleared once the decision it was timing completes, so an envelope emitted
2301    /// later, while a player is thinking, cannot report that wait as engine time.
2302    pub fn mark_decision_complete(&self) {
2303        self.0.store(0, std::sync::atomic::Ordering::Relaxed);
2304    }
2305
2306    /// Milliseconds since the last response, or `None` if none has arrived yet
2307    /// or the decision it was timing has already completed.
2308    ///
2309    /// There is deliberately no upper bound. Idle time used to be excluded by
2310    /// discarding anything past 120s, which also discarded the slow decisions
2311    /// the metric exists to expose: a real 175s decision reported no `engineMs`
2312    /// at all, so the field was absent exactly when it mattered and every
2313    /// quantile built from it was understated. `mark_decision_complete` draws
2314    /// that line by state instead of by duration.
2315    pub fn elapsed_ms(&self) -> Option<u32> {
2316        let at = self.0.load(std::sync::atomic::Ordering::Relaxed);
2317        if at == 0 {
2318            return None;
2319        }
2320        let delta = Self::now().saturating_sub(at);
2321        Some(delta.min(u32::MAX as u64) as u32)
2322    }
2323}
2324
2325fn patch_against_last(
2326    bases: &mut HashMap<usize, (Value, String)>,
2327    player_index: usize,
2328    per_seat: bool,
2329    slot: &str,
2330    envelope: &Value,
2331    engine_ms: Option<u32>,
2332) -> Option<Value> {
2333    let next = envelope.get("state")?.clone();
2334    let fingerprint = envelope.get("fingerprint")?.as_str()?.to_string();
2335    let previous = bases.insert(player_index, (next.clone(), fingerprint.clone()));
2336    let (base_state, base) = previous?;
2337    let patch = manabrew_relay_protocol::state_delta::diff(&base_state, &next)?;
2338    serde_json::to_value(StateEnvelope::StateDelta {
2339        for_player: per_seat.then(|| slot.to_string()),
2340        base,
2341        fingerprint,
2342        patch,
2343        engine_ms,
2344        emit_ms: None,
2345    })
2346    .ok()
2347}
2348
2349/// This node's own bot seats, answered in-process with the same agent their
2350/// relay connection runs. Saves two relay hops per bot decision.
2351///
2352/// `config.bot_agent` is what makes that first sentence true rather than a
2353/// coincidence: `spawn_bots` gives the seat's relay connection the same value.
2354/// Building a fixed agent here instead would silently outrank whatever the
2355/// seat is actually configured to play, because the relay's answer arrives
2356/// second and is dropped by `is_local_bot_duplicate`.
2357fn local_bot_seats(
2358    config: &Config,
2359    forge_ai: bool,
2360    player_names: &[String],
2361    bot_usernames: &HashSet<String>,
2362) -> HashMap<usize, BotResponder> {
2363    if !config.bot_enabled || !config.bot_local_answers || forge_ai {
2364        return HashMap::new();
2365    }
2366    player_names
2367        .iter()
2368        .enumerate()
2369        .filter(|(_, name)| bot_usernames.contains(*name))
2370        .map(|(index, name)| {
2371            info!(
2372                player_index = index,
2373                seat = name.as_str(),
2374                agent = ?config.bot_agent,
2375                "answering this bot seat in-process; its relay answers are dropped"
2376            );
2377            (index, BotResponder::new(config.bot_agent.build()))
2378        })
2379        .collect()
2380}
2381
2382/// Answer a bot seat's prompt in-process. State and prompt still go to the
2383/// relay; its copy of the answer is dropped on arrival.
2384fn answer_bot_locally(
2385    session: &EngineSession,
2386    snapshot: &SharedHostSnapshot,
2387    player_index: usize,
2388    bot: &mut BotResponder,
2389    message: &AgentMessage,
2390) {
2391    match message {
2392        AgentMessage::State(update) => bot.observe(update.game_view.clone()),
2393        AgentMessage::Prompt(prompt) => {
2394            let response = bot.respond(prompt.clone());
2395            let ClientToServerMessage::Response { prompt_id, .. } = &response else {
2396                return;
2397            };
2398            if let Ok(mut answers) = session.local_bot_answers().lock() {
2399                answers.insert(player_index, *prompt_id);
2400            }
2401            let Some(tx) = session.response_tx(player_index) else {
2402                debug!(player_index, "no response channel for local bot");
2403                return;
2404            };
2405            session.engine_clock().mark_response();
2406            if let Err(error) = tx.send(response) {
2407                warn!(player_index, %error, "failed to route local bot response");
2408                return;
2409            }
2410            if let Ok(mut snap) = snapshot.lock() {
2411                snap.pending_prompts.remove(&player_slot(player_index));
2412            }
2413        }
2414        _ => {}
2415    }
2416}
2417
2418#[allow(clippy::too_many_arguments)]
2419fn spawn_remote_prompt_forwarder(
2420    outbound_tx: tokio_mpsc::UnboundedSender<ClientMessage>,
2421    snapshot: SharedHostSnapshot,
2422    engine_session: SharedEngineSession,
2423    game_id: String,
2424    remote_prompt_rx: std_mpsc::Receiver<(usize, AgentMessage)>,
2425    seat_usernames: Option<Vec<String>>,
2426    state_delta: bool,
2427    engine_clock: EngineClock,
2428    mut local_bots: HashMap<usize, BotResponder>,
2429) {
2430    thread::spawn(move || {
2431        let mut last_state_by_seat: HashMap<usize, Value> = HashMap::new();
2432        let mut delta_bases: HashMap<usize, (Value, String)> = HashMap::new();
2433        let mut last_state: Option<Value> = None;
2434        while let Ok((player_index, message)) = remote_prompt_rx.recv() {
2435            let Ok(session) = engine_session.lock() else {
2436                break;
2437            };
2438            let Some(current) = session
2439                .as_ref()
2440                .filter(|session| session.game_id() == game_id)
2441            else {
2442                break;
2443            };
2444            let per_seat = seat_usernames.is_some() && player_index != OBSERVER_SEAT;
2445            let slot = player_slot(player_index);
2446            let engine_ms = engine_clock.elapsed_ms();
2447            if matches!(message, AgentMessage::Prompt(_)) {
2448                engine_clock.mark_decision_complete();
2449            }
2450            let emit_started = Instant::now();
2451            let envelope = match &message {
2452                AgentMessage::State(state_update) if !per_seat => StateEnvelope::State {
2453                    for_player: None,
2454                    state: serde_json::to_value(state_update).unwrap_or(Value::Null),
2455                    fingerprint: None,
2456                    engine_ms,
2457                    emit_ms: None,
2458                },
2459                _ => StateEnvelope::for_agent_message_timed(slot.clone(), &message, engine_ms),
2460            };
2461            let Ok(state) = serde_json::to_value(envelope) else {
2462                continue;
2463            };
2464            let state = match &message {
2465                AgentMessage::State(_) if state_delta => stamp_fingerprint(state),
2466                _ => state,
2467            };
2468            match &message {
2469                AgentMessage::State(_) if per_seat => {
2470                    if last_state_by_seat.get(&player_index) == Some(&state) {
2471                        continue;
2472                    }
2473                    last_state_by_seat.insert(player_index, state.clone());
2474                }
2475                AgentMessage::State(_) if last_state.as_ref() == Some(&state) => continue,
2476                AgentMessage::State(_) => last_state = Some(state.clone()),
2477                AgentMessage::Prompt(_) | AgentMessage::Directive(_) | AgentMessage::Error(_) => {}
2478            }
2479            if let Ok(mut snap) = snapshot.lock() {
2480                match &message {
2481                    AgentMessage::State(_) if per_seat => {
2482                        snap.last_state_by_slot.insert(slot.clone(), state.clone());
2483                    }
2484                    AgentMessage::State(_) => snap.last_state = Some(state.clone()),
2485                    AgentMessage::Prompt(_) => {
2486                        snap.pending_prompts.insert(slot.clone(), state.clone());
2487                    }
2488                    AgentMessage::Directive(directive) => {
2489                        let kind = match directive {
2490                            EngineDirective::Recording { .. } => "recording",
2491                            EngineDirective::RepeatVote { .. } => "repeatVote",
2492                        };
2493                        snap.pending_directives
2494                            .insert((slot.clone(), kind), state.clone());
2495                    }
2496                    AgentMessage::Error(_) => {}
2497                }
2498            }
2499            if let Some(bot) = local_bots.get_mut(&player_index) {
2500                answer_bot_locally(current, &snapshot, player_index, bot, &message);
2501            }
2502            let state = match &message {
2503                AgentMessage::State(_) if state_delta => patch_against_last(
2504                    &mut delta_bases,
2505                    player_index,
2506                    per_seat,
2507                    &slot,
2508                    &state,
2509                    engine_ms,
2510                )
2511                .unwrap_or(state),
2512                _ => state,
2513            };
2514            let target_player = if per_seat {
2515                let Some(target_player) = seat_usernames
2516                    .as_ref()
2517                    .and_then(|names| names.get(player_index).cloned())
2518                else {
2519                    warn!(player_index, PRIVATE_MESSAGE_MISSING_PLAYER);
2520                    continue;
2521                };
2522                Some(target_player)
2523            } else {
2524                None
2525            };
2526            if outbound_tx
2527                .send(ClientMessage::BroadcastState {
2528                    state: stamp_emit_ms(state, emit_started.elapsed()),
2529                    target_player,
2530                })
2531                .is_err()
2532            {
2533                break;
2534            }
2535        }
2536    });
2537}
2538
2539fn outcome_report(view: &GameViewDto) -> GameOutcomeReport {
2540    GameOutcomeReport {
2541        game_over: view.game_over,
2542        winner_slot: view.winner_id.clone(),
2543        conceded_slots: view
2544            .players
2545            .iter()
2546            .filter(|player| player.status == PlayerStatus::Conceded)
2547            .map(|player| player.id.clone())
2548            .collect(),
2549        fatal_message: None,
2550        turns: Some(view.turn),
2551    }
2552}
2553
2554fn cached_outcome_report(envelope: &Value) -> Option<GameOutcomeReport> {
2555    let update: StateUpdate = serde_json::from_value(envelope.get("state")?.clone()).ok()?;
2556    Some(outcome_report(&update.game_view))
2557}
2558
2559fn spawn_game_over_forwarder(
2560    outbound_tx: tokio_mpsc::UnboundedSender<ClientMessage>,
2561    game_over_rx: std_mpsc::Receiver<HostedGameOver>,
2562    session_handle: SharedEngineSession,
2563    snapshot: SharedHostSnapshot,
2564    game_id: String,
2565    seat_usernames: Option<Vec<String>>,
2566) {
2567    thread::spawn(move || {
2568        while let Ok(game_over) = game_over_rx.recv() {
2569            let Ok(session) = session_handle.lock() else {
2570                return;
2571            };
2572            if session
2573                .as_ref()
2574                .is_none_or(|session| session.game_id() != game_id)
2575            {
2576                warn!(
2577                    game_id,
2578                    "stale engine session reached game over; not ending the relay game"
2579                );
2580                continue;
2581            }
2582            if let Ok(mut snap) = snapshot.lock() {
2583                if snap
2584                    .game
2585                    .as_ref()
2586                    .is_some_and(|game| game.game_id == game_id)
2587                {
2588                    snap.pending_end_game = Some(game_id.clone());
2589                }
2590            }
2591            let outcome = game_over
2592                .messages
2593                .iter()
2594                .rev()
2595                .find_map(|(_, message)| match message {
2596                    AgentMessage::State(update) => Some(outcome_report(&update.game_view)),
2597                    _ => None,
2598                })
2599                .or_else(|| {
2600                    let snap = snapshot.lock().ok()?;
2601                    snap.last_state
2602                        .as_ref()
2603                        .or_else(|| snap.last_state_by_slot.values().next())
2604                        .and_then(cached_outcome_report)
2605                        .filter(|outcome| outcome.game_over)
2606                });
2607            let mut last_state_by_seat: HashMap<usize, Value> = HashMap::new();
2608            let mut last_state: Option<Value> = None;
2609            for (player_index, message) in game_over.messages {
2610                let per_seat = seat_usernames.is_some() && player_index != OBSERVER_SEAT;
2611                let envelope = match &message {
2612                    AgentMessage::State(state_update) if !per_seat => StateEnvelope::State {
2613                        for_player: None,
2614                        state: serde_json::to_value(state_update).unwrap_or(Value::Null),
2615                        fingerprint: None,
2616                        engine_ms: None,
2617                        emit_ms: None,
2618                    },
2619                    _ => StateEnvelope::for_agent_message(player_slot(player_index), &message),
2620                };
2621                let Ok(state) = serde_json::to_value(envelope) else {
2622                    continue;
2623                };
2624                match &message {
2625                    AgentMessage::State(_) if per_seat => {
2626                        if last_state_by_seat.get(&player_index) == Some(&state) {
2627                            continue;
2628                        }
2629                        last_state_by_seat.insert(player_index, state.clone());
2630                    }
2631                    AgentMessage::State(_) if last_state.as_ref() == Some(&state) => continue,
2632                    AgentMessage::State(_) => last_state = Some(state.clone()),
2633                    AgentMessage::Prompt(_)
2634                    | AgentMessage::Directive(_)
2635                    | AgentMessage::Error(_) => {}
2636                }
2637                let target_player = if per_seat {
2638                    let Some(target_player) = seat_usernames
2639                        .as_ref()
2640                        .and_then(|names| names.get(player_index).cloned())
2641                    else {
2642                        warn!(player_index, PRIVATE_MESSAGE_MISSING_PLAYER);
2643                        continue;
2644                    };
2645                    Some(target_player)
2646                } else {
2647                    None
2648                };
2649                if outbound_tx
2650                    .send(ClientMessage::BroadcastState {
2651                        state,
2652                        target_player,
2653                    })
2654                    .is_err()
2655                {
2656                    return;
2657                }
2658            }
2659            let Ok(state) = serde_json::to_value(StateEnvelope::RoomRelay {
2660                protocol: SELF_HOSTED_NODE_PROTOCOL.to_string(),
2661                version: 1,
2662                message_id: Uuid::new_v4().to_string(),
2663                from_player: None,
2664                target_player: None,
2665                room_id: None,
2666                payload: json!({ "type": "gameOver", "gameId": game_over.game_id }),
2667            }) else {
2668                continue;
2669            };
2670            if outbound_tx
2671                .send(ClientMessage::BroadcastState {
2672                    state,
2673                    target_player: None,
2674                })
2675                .is_err()
2676            {
2677                return;
2678            }
2679            if let Some(outcome) = outcome {
2680                let _ = outbound_tx.send(ClientMessage::ReportGameOutcome {
2681                    game_id: game_id.clone(),
2682                    outcome,
2683                });
2684            }
2685            if outbound_tx
2686                .send(ClientMessage::EndGame {
2687                    game_id: game_id.clone(),
2688                })
2689                .is_err()
2690            {
2691                return;
2692            }
2693        }
2694    });
2695}
2696
2697async fn relay_writer(mut write: WsWrite, mut outbound: tokio_mpsc::UnboundedReceiver<Message>) {
2698    while let Some(message) = outbound.recv().await {
2699        let started = Instant::now();
2700        let closing = matches!(message, Message::Close(_));
2701        if let Err(error) = write.send(message).await {
2702            warn!(%error, "relay write failed");
2703            break;
2704        }
2705        crate::metrics::record_relay_send(started.elapsed());
2706        if closing {
2707            break;
2708        }
2709    }
2710    let _ = write.close().await;
2711}
2712
2713fn self_minted_identity_token(kind: &str, handle: &str) -> String {
2714    let iat = std::time::SystemTime::now()
2715        .duration_since(std::time::UNIX_EPOCH)
2716        .map(|elapsed| elapsed.as_secs() as i64)
2717        .unwrap_or_default();
2718    identity_token::mint_unsigned(&format!("{kind}:{handle}"), handle, iat, 24 * 60 * 60)
2719}
2720
2721impl RelayClient {
2722    async fn connect(
2723        relay_url: &str,
2724        username: &str,
2725        password: &str,
2726    ) -> Result<Self, Box<dyn std::error::Error + Send + Sync>> {
2727        info!(relay_url, username, "connecting relay client");
2728        let (socket, _) = connect_async_with_config(relay_url, None, true).await?;
2729        let (write, read) = socket.split();
2730        let (outbound, outbound_rx) = tokio_mpsc::unbounded_channel();
2731        let writer = tokio::spawn(relay_writer(write, outbound_rx));
2732        let mut client = Self {
2733            username: username.to_string(),
2734            outbound,
2735            read,
2736            writer,
2737        };
2738        client
2739            .send(&ClientMessage::Authenticate {
2740                username: username.to_string(),
2741                password: password.to_string(),
2742                service: true,
2743                identity: Some(IdentityProof {
2744                    token: Some(self_minted_identity_token("node", username)),
2745                    device: None,
2746                }),
2747                client_platform: ClientPlatform::Unknown,
2748                client_version: None,
2749                engine_gate: EngineGate::Unknown,
2750                acquisition_source: None,
2751            })
2752            .await?;
2753        client.wait_for_auth().await?;
2754        Ok(client)
2755    }
2756
2757    async fn wait_for_auth(&mut self) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
2758        loop {
2759            match self.recv().await? {
2760                Some(ServerMessage::AuthResult { success, error, .. }) => {
2761                    if success {
2762                        info!(username = %self.username, "authenticated");
2763                        return Ok(());
2764                    }
2765                    return Err(format!("authentication failed: {error:?}").into());
2766                }
2767                Some(other) => {
2768                    debug!(?other, username = %self.username, "ignored pre-auth message")
2769                }
2770                None => return Err("relay closed before authentication".into()),
2771            }
2772        }
2773    }
2774
2775    async fn send(
2776        &mut self,
2777        message: &ClientMessage,
2778    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
2779        self.enqueue(Message::Text(serde_json::to_string(message)?))
2780    }
2781
2782    fn enqueue(&self, message: Message) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
2783        self.outbound
2784            .send(message)
2785            .map_err(|_| "relay writer stopped".into())
2786    }
2787
2788    async fn recv(
2789        &mut self,
2790    ) -> Result<Option<ServerMessage>, Box<dyn std::error::Error + Send + Sync>> {
2791        loop {
2792            let Some(message) = self.read.next().await else {
2793                return Ok(None);
2794            };
2795            let message = message?;
2796            match message {
2797                Message::Text(text) => {
2798                    return Ok(Some(serde_json::from_str(&text)?));
2799                }
2800                Message::Ping(payload) => {
2801                    self.enqueue(Message::Pong(payload))?;
2802                }
2803                Message::Close(_) => return Ok(None),
2804                _ => {}
2805            }
2806        }
2807    }
2808
2809    async fn close(&mut self) {
2810        if self.enqueue(Message::Close(None)).is_err() {
2811            return;
2812        }
2813        let _ = time::timeout(CLOSE_DRAIN_TIMEOUT, async {
2814            while let Some(Ok(_)) = self.read.next().await {}
2815        })
2816        .await;
2817        self.writer.abort();
2818    }
2819
2820    async fn broadcast_room_message(
2821        &mut self,
2822        protocol: &str,
2823        payload: Value,
2824    ) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
2825        let envelope = StateEnvelope::RoomRelay {
2826            protocol: protocol.to_string(),
2827            version: 1,
2828            message_id: Uuid::new_v4().to_string(),
2829            from_player: Some(self.username.clone()),
2830            target_player: None,
2831            room_id: None,
2832            payload,
2833        };
2834        self.send(&ClientMessage::BroadcastState {
2835            state: serde_json::to_value(envelope)?,
2836            target_player: None,
2837        })
2838        .await
2839    }
2840}
2841
2842fn log_room_update(observer: &str, room: &RoomInfo) {
2843    let players = room
2844        .players
2845        .iter()
2846        .map(|player| {
2847            format!(
2848                "{}{}{}",
2849                player.username,
2850                if player.ready { ":ready" } else { "" },
2851                if player.connected { "" } else { ":offline" },
2852            )
2853        })
2854        .collect::<Vec<_>>()
2855        .join(",");
2856    info!(
2857        observer,
2858        room_id = %room.room_id,
2859        room_name = %room.room_name,
2860        host = %room.host,
2861        players,
2862        "room update"
2863    );
2864}
2865
2866#[cfg(test)]
2867mod local_bot_tests {
2868    use super::*;
2869    use manabrew_protocol::prompts::choose_action::ChooseActionInput;
2870    use manabrew_protocol::prompts::PromptInput;
2871    use manabrew_protocol::transport::AgentPrompt;
2872    use std::time::Duration;
2873
2874    const HUMAN: &str = "alice";
2875    const BOT: &str = "forge-bot";
2876
2877    fn config(bot_enabled: bool, local: bool, forge_ai: bool) -> Config {
2878        let mut config = Config::for_hosted_room(
2879            "ws://127.0.0.1:1".to_string(),
2880            String::new(),
2881            "t".to_string(),
2882            GameFormat::Any,
2883            2,
2884            None,
2885            None,
2886            None,
2887        );
2888        config.bot_enabled = bot_enabled;
2889        config.bot_local_answers = local;
2890        config.forge_ai = forge_ai;
2891        config
2892    }
2893
2894    fn prompt(seat: usize, prompt_id: u32) -> AgentPrompt {
2895        AgentPrompt {
2896            prompt_id,
2897            deciding_player_id: player_slot(seat),
2898            source_card: None,
2899            source_ability_text: None,
2900            input: PromptInput::ChooseAction(ChooseActionInput {
2901                actions: Vec::new(),
2902            }),
2903        }
2904    }
2905
2906    /// A hosted 2-seat game: seat 0 human, seat 1 this node's bot.
2907    struct Rig {
2908        engine_session: SharedEngineSession,
2909        snapshot: SharedHostSnapshot,
2910        outbound_rx: tokio_mpsc::UnboundedReceiver<ClientMessage>,
2911        prompt_tx: std_mpsc::Sender<(usize, AgentMessage)>,
2912        bot_rx: std_mpsc::Receiver<ClientToServerMessage>,
2913    }
2914
2915    fn rig(config: &Config) -> Rig {
2916        let names = vec![HUMAN.to_string(), BOT.to_string()];
2917        let bots: HashSet<String> = [BOT.to_string()].into_iter().collect();
2918        let (outbound_tx, outbound_rx) = tokio_mpsc::unbounded_channel();
2919        let (prompt_tx, prompt_rx) = std_mpsc::channel();
2920        let (bot_tx, bot_rx) = std_mpsc::channel();
2921        let (human_tx, _human_rx) = std_mpsc::channel();
2922        let mut txs = HashMap::new();
2923        txs.insert(0, human_tx);
2924        txs.insert(1, bot_tx);
2925        let engine_clock = EngineClock::default();
2926        let engine_session: SharedEngineSession =
2927            Arc::new(Mutex::new(Some(EngineSession::Forge {
2928                game_id: "g1".to_string(),
2929                remote_response_txs: txs,
2930                cancel: Arc::new(AtomicBool::new(false)),
2931                engine_clock: engine_clock.clone(),
2932                local_bot_answers: LocalBotAnswers::default(),
2933            })));
2934        let snapshot: SharedHostSnapshot = Arc::new(Mutex::new(HostSnapshot {
2935            game: Some(GameStart {
2936                game_id: "g1".to_string(),
2937                player_order: names.clone(),
2938                player_decks: Vec::new(),
2939                starting_life: 20,
2940            }),
2941            ..HostSnapshot::default()
2942        }));
2943        spawn_remote_prompt_forwarder(
2944            outbound_tx,
2945            snapshot.clone(),
2946            engine_session.clone(),
2947            "g1".to_string(),
2948            prompt_rx,
2949            Some(names.clone()),
2950            false,
2951            engine_clock,
2952            local_bot_seats(config, config.forge_ai, &names, &bots),
2953        );
2954        Rig {
2955            engine_session,
2956            snapshot,
2957            outbound_rx,
2958            prompt_tx,
2959            bot_rx,
2960        }
2961    }
2962
2963    fn relay_response(rig: &Rig, prompt_id: u32) {
2964        let state = serde_json::to_value(StateEnvelope::Response {
2965            from_player: player_slot(1),
2966            prompt_id,
2967            action: json!({ "type": "chooseAction", "output": { "type": "pass" } }),
2968        })
2969        .unwrap();
2970        route_remote_response(&rig.engine_session, &rig.snapshot, BOT, &state);
2971    }
2972
2973    #[test]
2974    fn remote_zero_prompt_cannot_bypass_engine_validation() {
2975        let rig = rig(&config(false, false, false));
2976        relay_response(&rig, 0);
2977        assert!(rig.bot_rx.try_recv().is_err());
2978        relay_response(&rig, 7);
2979        assert!(matches!(
2980            rig.bot_rx.try_recv(),
2981            Ok(ClientToServerMessage::Response { prompt_id: 7, .. })
2982        ));
2983    }
2984
2985    #[test]
2986    fn local_bot_answers_its_prompt_and_the_relay_copy_is_dropped() {
2987        let rig = rig(&config(true, true, false));
2988        rig.prompt_tx
2989            .send((1, AgentMessage::Prompt(prompt(1, 7))))
2990            .unwrap();
2991
2992        let answer = rig
2993            .bot_rx
2994            .recv_timeout(Duration::from_secs(5))
2995            .expect("bot seat answered in-process");
2996        assert!(matches!(
2997            answer,
2998            ClientToServerMessage::Response { prompt_id: 7, .. }
2999        ));
3000
3001        // The prompt still went out to the relay, unchanged. The local answer
3002        // lands first, so give the forwarder a moment to send.
3003        let mut rig = rig;
3004        let sent = tokio::runtime::Runtime::new()
3005            .unwrap()
3006            .block_on(async {
3007                tokio::time::timeout(Duration::from_secs(5), rig.outbound_rx.recv()).await
3008            })
3009            .expect("prompt forwarded to relay")
3010            .expect("outbound channel open");
3011        let ClientMessage::BroadcastState { target_player, .. } = sent else {
3012            panic!("expected a broadcast, got {sent:?}");
3013        };
3014        assert_eq!(target_player.as_deref(), Some(BOT));
3015
3016        // The bot's own relay connection answers the same prompt later: dropped.
3017        relay_response(&rig, 7);
3018        assert!(
3019            rig.bot_rx.try_recv().is_err(),
3020            "relay duplicate reached the engine"
3021        );
3022
3023        // A newer prompt the node did not answer still routes.
3024        relay_response(&rig, 8);
3025        assert!(matches!(
3026            rig.bot_rx.try_recv(),
3027            Ok(ClientToServerMessage::Response { prompt_id: 8, .. })
3028        ));
3029    }
3030
3031    #[test]
3032    fn local_answers_are_off_without_bots_or_when_forge_plays_them() {
3033        for config in [
3034            config(false, true, false),
3035            config(true, false, false),
3036            config(true, true, true),
3037        ] {
3038            let rig = rig(&config);
3039            rig.prompt_tx
3040                .send((1, AgentMessage::Prompt(prompt(1, 7))))
3041                .unwrap();
3042            assert!(
3043                rig.bot_rx.recv_timeout(Duration::from_millis(200)).is_err(),
3044                "bot seat was answered in-process"
3045            );
3046            relay_response(&rig, 7);
3047            assert!(rig.bot_rx.try_recv().is_ok(), "relay answer did not route");
3048        }
3049    }
3050}