Files
marathon/crates/libmarathon/src/engine/networking.rs
2025-12-28 17:39:27 +00:00

410 lines
16 KiB
Rust

//! Networking Manager - handles iroh networking and CRDT state outside Bevy
use std::time::Duration;
use tokio::sync::mpsc;
use tokio::time;
use bytes::Bytes;
use futures_lite::StreamExt;
use crate::networking::{
EntityLockRegistry, NodeId, OperationLog, SessionId, TombstoneRegistry, VectorClock,
VersionedMessage, SyncMessage, LockMessage,
};
use super::EngineEvent;
use super::events::NetworkingInitStatus;
pub struct NetworkingManager {
session_id: SessionId,
node_id: NodeId,
// Iroh networking
sender: iroh_gossip::api::GossipSender,
receiver: iroh_gossip::api::GossipReceiver,
// Keep these alive for the lifetime of the manager
_endpoint: iroh::Endpoint,
_router: iroh::protocol::Router,
_gossip: iroh_gossip::net::Gossip,
// Bridge to Bevy for message passing
bridge: crate::networking::GossipBridge,
// CRDT state
vector_clock: VectorClock,
operation_log: OperationLog,
tombstones: TombstoneRegistry,
locks: EntityLockRegistry,
// Track locks we own for heartbeat broadcasting
our_locks: std::collections::HashSet<uuid::Uuid>,
}
impl NetworkingManager {
pub async fn new(
session_id: SessionId,
progress_tx: Option<tokio::sync::mpsc::UnboundedSender<NetworkingInitStatus>>,
cancel_token: tokio_util::sync::CancellationToken,
) -> anyhow::Result<(Self, crate::networking::GossipBridge)> {
let send_progress = |status: NetworkingInitStatus| {
if let Some(ref tx) = progress_tx {
let _ = tx.send(status.clone());
}
tracing::info!("Networking init: {:?}", status);
};
use iroh::{
discovery::pkarr::dht::DhtDiscovery,
protocol::Router,
Endpoint,
};
use iroh_gossip::{
net::Gossip,
proto::TopicId,
};
// Check for cancellation at start
if cancel_token.is_cancelled() {
return Err(anyhow::anyhow!("Initialization cancelled before start"));
}
send_progress(NetworkingInitStatus::CreatingEndpoint);
// Create iroh endpoint with DHT discovery
// This allows peers to discover each other over the internet via Mainline DHT
// Security comes from the secret session-derived ALPN, not network isolation
let dht_discovery = DhtDiscovery::builder().build()?;
let endpoint = Endpoint::builder()
.discovery(dht_discovery)
.bind()
.await?;
send_progress(NetworkingInitStatus::EndpointReady);
let endpoint_id = endpoint.addr().id;
// Convert endpoint ID to NodeId (using first 16 bytes)
let id_bytes = endpoint_id.as_bytes();
let mut node_id_bytes = [0u8; 16];
node_id_bytes.copy_from_slice(&id_bytes[..16]);
let node_id = NodeId::from_bytes(node_id_bytes);
// Create pkarr client for DHT peer discovery
let pkarr_client = pkarr::Client::builder()
.no_default_network()
.dht(|x| x)
.build()?;
// Discover existing peers from DHT with retries
// Retry immediately without delays - if peers aren't in DHT yet, they'll appear soon
let mut peer_endpoint_ids = vec![];
for attempt in 1..=3 {
// Check for cancellation before each attempt
if cancel_token.is_cancelled() {
tracing::info!("Networking initialization cancelled during DHT discovery");
return Err(anyhow::anyhow!("Initialization cancelled"));
}
send_progress(NetworkingInitStatus::DiscoveringPeers {
session_code: session_id.to_code().to_string(),
attempt,
});
match crate::engine::peer_discovery::discover_peers_from_dht(&session_id, &pkarr_client).await {
Ok(peers) if !peers.is_empty() => {
let count = peers.len();
peer_endpoint_ids = peers;
send_progress(NetworkingInitStatus::PeersFound {
count,
});
break;
}
Ok(_) if attempt == 3 => {
// Last attempt and no peers found
send_progress(NetworkingInitStatus::NoPeersFound);
}
Ok(_) => {
// No peers found, but will retry immediately
}
Err(e) => {
tracing::warn!("DHT query attempt {} failed: {}", attempt, e);
}
}
}
// Check for cancellation before publishing
if cancel_token.is_cancelled() {
tracing::info!("Networking initialization cancelled before DHT publish");
return Err(anyhow::anyhow!("Initialization cancelled"));
}
// Publish our presence to DHT
send_progress(NetworkingInitStatus::PublishingToDHT);
if let Err(e) = crate::engine::peer_discovery::publish_peer_to_dht(
&session_id,
endpoint_id,
&pkarr_client,
)
.await
{
tracing::warn!("Failed to publish to DHT: {}", e);
}
// Check for cancellation before gossip initialization
if cancel_token.is_cancelled() {
tracing::info!("Networking initialization cancelled before gossip init");
return Err(anyhow::anyhow!("Initialization cancelled"));
}
// Derive session-specific ALPN for network isolation
let session_alpn = session_id.to_alpn();
// Create gossip protocol with custom session ALPN
send_progress(NetworkingInitStatus::InitializingGossip);
let gossip = Gossip::builder()
.alpn(&session_alpn)
.spawn(endpoint.clone());
// Set up router to accept session ALPN
let router = Router::builder(endpoint.clone())
.accept(session_alpn.as_slice(), gossip.clone())
.spawn();
// Subscribe to topic with discovered peers as bootstrap
let topic_id = TopicId::from_bytes(session_alpn);
let subscribe_handle = gossip.subscribe(topic_id, peer_endpoint_ids).await?;
let (sender, receiver) = subscribe_handle.split();
tracing::info!(
"NetworkingManager started for session {} with node {}",
session_id.to_code(),
node_id
);
// Create GossipBridge for Bevy integration
let bridge = crate::networking::GossipBridge::new(node_id);
// Spawn background task to maintain DHT presence
let session_id_clone = session_id.clone();
tokio::spawn(crate::engine::peer_discovery::maintain_dht_presence(
session_id_clone,
endpoint_id,
pkarr_client,
));
let manager = Self {
session_id,
node_id,
sender,
receiver,
_endpoint: endpoint,
_router: router,
_gossip: gossip,
bridge: bridge.clone(),
vector_clock: VectorClock::new(),
operation_log: OperationLog::new(),
tombstones: TombstoneRegistry::new(),
locks: EntityLockRegistry::new(),
our_locks: std::collections::HashSet::new(),
};
Ok((manager, bridge))
}
pub fn node_id(&self) -> NodeId {
self.node_id
}
pub fn session_id(&self) -> SessionId {
self.session_id.clone()
}
/// Process gossip events (unbounded) and periodic tasks (heartbeats, lock cleanup)
/// Also bridges messages between iroh-gossip and Bevy's GossipBridge
pub async fn run(mut self, event_tx: mpsc::UnboundedSender<EngineEvent>, cancel_token: tokio_util::sync::CancellationToken) {
let mut heartbeat_interval = time::interval(Duration::from_secs(1));
let mut bridge_poll_interval = time::interval(Duration::from_millis(10));
loop {
tokio::select! {
// Listen for shutdown signal
_ = cancel_token.cancelled() => {
tracing::info!("NetworkingManager received shutdown signal");
break;
}
// Process incoming gossip messages and forward to GossipBridge
Some(result) = self.receiver.next() => {
match result {
Ok(event) => {
use iroh_gossip::api::Event;
match event {
Event::Received(msg) => {
// Deserialize and forward to GossipBridge for Bevy systems
if let Ok(versioned) = rkyv::from_bytes::<VersionedMessage, rkyv::rancor::Failure>(&msg.content) {
if let Err(e) = self.bridge.push_incoming(versioned) {
tracing::error!("Failed to push message to GossipBridge: {}", e);
} else {
tracing::debug!("Forwarded message to Bevy via GossipBridge");
}
}
}
Event::NeighborUp(peer) => {
tracing::info!("Peer connected: {}", peer);
// Convert PublicKey to NodeId for Bevy
let peer_bytes = peer.as_bytes();
let mut node_id_bytes = [0u8; 16];
node_id_bytes.copy_from_slice(&peer_bytes[..16]);
let peer_node_id = NodeId::from_bytes(node_id_bytes);
// Notify Bevy of peer join
let _ = event_tx.send(EngineEvent::PeerJoined {
node_id: peer_node_id,
});
}
Event::NeighborDown(peer) => {
tracing::warn!("Peer disconnected: {}", peer);
// Convert PublicKey to NodeId for Bevy
let peer_bytes = peer.as_bytes();
let mut node_id_bytes = [0u8; 16];
node_id_bytes.copy_from_slice(&peer_bytes[..16]);
let peer_node_id = NodeId::from_bytes(node_id_bytes);
// Notify Bevy of peer leave
let _ = event_tx.send(EngineEvent::PeerLeft {
node_id: peer_node_id,
});
}
Event::Lagged => {
tracing::warn!("Event stream lagged");
}
}
}
Err(e) => {
tracing::warn!("Gossip receiver error: {}", e);
}
}
}
// Poll GossipBridge for outgoing messages and broadcast via iroh
_ = bridge_poll_interval.tick() => {
while let Some(msg) = self.bridge.try_recv_outgoing() {
if let Ok(bytes) = rkyv::to_bytes::<rkyv::rancor::Failure>(&msg).map(|b| b.to_vec()) {
if let Err(e) = self.sender.broadcast(Bytes::from(bytes)).await {
tracing::error!("Failed to broadcast message: {}", e);
} else {
tracing::debug!("Broadcast message from Bevy via iroh-gossip");
}
}
}
}
// Periodic tasks: heartbeats and lock cleanup
_ = heartbeat_interval.tick() => {
self.broadcast_lock_heartbeats(&event_tx).await;
self.cleanup_expired_locks(&event_tx);
}
}
}
}
async fn handle_sync_message(&mut self, msg_bytes: &[u8], event_tx: &mpsc::UnboundedSender<EngineEvent>) {
// Deserialize SyncMessage
let versioned: VersionedMessage = match rkyv::from_bytes::<VersionedMessage, rkyv::rancor::Failure>(msg_bytes) {
Ok(v) => v,
Err(e) => {
tracing::warn!("Failed to deserialize sync message: {}", e);
return;
}
};
match versioned.message {
SyncMessage::Lock(lock_msg) => {
self.handle_lock_message(lock_msg, event_tx);
}
_ => {
// TODO: Handle other message types (ComponentOp, EntitySpawn, etc.)
tracing::debug!("Unhandled sync message type");
}
}
}
fn handle_lock_message(&mut self, msg: LockMessage, event_tx: &mpsc::UnboundedSender<EngineEvent>) {
match msg {
LockMessage::LockRequest { entity_id, node_id } => {
match self.locks.try_acquire(entity_id, node_id) {
Ok(()) => {
// Track if this is our lock
if node_id == self.node_id {
self.our_locks.insert(entity_id);
}
let _ = event_tx.send(EngineEvent::LockAcquired {
entity_id,
holder: node_id,
});
}
Err(current_holder) => {
let _ = event_tx.send(EngineEvent::LockDenied {
entity_id,
current_holder,
});
}
}
}
LockMessage::LockHeartbeat { entity_id, holder } => {
self.locks.renew_heartbeat(entity_id, holder);
}
LockMessage::LockRelease { entity_id, node_id } => {
self.locks.release(entity_id, node_id);
// Remove from our locks tracking
if node_id == self.node_id {
self.our_locks.remove(&entity_id);
}
let _ = event_tx.send(EngineEvent::LockReleased { entity_id });
}
_ => {}
}
}
async fn broadcast_lock_heartbeats(&mut self, _event_tx: &mpsc::UnboundedSender<EngineEvent>) {
// Broadcast heartbeats for locks we hold
for entity_id in self.our_locks.iter().copied() {
self.locks.renew_heartbeat(entity_id, self.node_id);
let msg = VersionedMessage::new(SyncMessage::Lock(LockMessage::LockHeartbeat {
entity_id,
holder: self.node_id,
}));
if let Ok(bytes) = rkyv::to_bytes::<rkyv::rancor::Failure>(&msg).map(|b| b.to_vec()) {
let _ = self.sender.broadcast(Bytes::from(bytes)).await;
}
}
}
fn cleanup_expired_locks(&mut self, event_tx: &mpsc::UnboundedSender<EngineEvent>) {
// Get expired locks from registry
let expired = self.locks.get_expired_locks();
for entity_id in expired {
// Only cleanup if it's not our lock
if let Some(holder) = self.locks.get_holder(entity_id, self.node_id) {
if holder != self.node_id {
self.locks.force_release(entity_id);
let _ = event_tx.send(EngineEvent::LockExpired { entity_id });
tracing::info!("Lock expired for entity {}", entity_id);
}
}
}
}
pub async fn shutdown(self) {
tracing::info!("NetworkingManager shut down");
// endpoint and gossip will be dropped automatically
}
}