omikron/src/omega/omega_connection.rs
2026-06-03 20:55:22 +02:00

783 lines
27 KiB
Rust

use crate::{
data::user::UserStatus,
get_private_key, log, log_cv_in, log_cv_out, log_err, log_in,
rho::rho_manager::{self, RHO_CONNECTIONS, connection_count},
util::{
crypto_helper::{decrypt_b64, secret_key_to_base64},
file_util::load_file_vec,
logger::PrintType,
},
};
use dashmap::DashMap;
use once_cell::sync::Lazy;
use std::{collections::HashMap, env, sync::Arc, time::Duration};
use tokio::{
sync::{Mutex, RwLock, mpsc, watch},
task::JoinHandle,
time::{Instant, sleep},
};
use ttp_core::{CommunicationType, CommunicationValue, DataTypes, DataValue, rand_u32};
use ttp_native::{Policy, Receiver, SendMode, Sender};
use uuid::Uuid;
// ============================================================================
// Configuration
// ============================================================================
const OMEGA_HOST_DEFAULT: &str = "tensamin.net";
const OMEGA_PORT_DEFAULT: u16 = 9187;
fn omega_host_and_port() -> (String, u16) {
let host = env::var("OMEGA_HOST")
.map(|s| s.trim().to_string())
.unwrap_or_else(|_| OMEGA_HOST_DEFAULT.to_string());
let port = env::var("OMEGA_PORT")
.ok()
.and_then(|s| s.trim().parse().ok())
.unwrap_or(OMEGA_PORT_DEFAULT);
(host, port)
}
const RECONNECT_DELAY: Duration = Duration::from_secs(5);
const MAX_RECONNECT_DELAY: Duration = Duration::from_secs(300);
const CONNECTION_TIMEOUT: Duration = Duration::from_secs(10);
const HEARTBEAT_INTERVAL: Duration = Duration::from_secs(5);
const TASK_CLEANUP_INTERVAL: Duration = Duration::from_secs(60);
const TASK_MAX_AGE: Duration = Duration::from_secs(60);
// ============================================================================
// Waiting Task System
// ============================================================================
pub struct WaitingTask {
pub task: Box<dyn Fn(Arc<OmegaConnection>, CommunicationValue) -> bool + Send + Sync>,
pub inserted_at: Instant,
}
pub static WAITING_TASKS: Lazy<DashMap<u32, WaitingTask>> = Lazy::new(DashMap::new);
pub fn start_task_cleanup_loop() {
tokio::spawn(async {
loop {
sleep(TASK_CLEANUP_INTERVAL).await;
WAITING_TASKS.retain(|_, v| v.inserted_at.elapsed() < TASK_MAX_AGE);
}
});
}
// ============================================================================
// Connection State
// ============================================================================
#[derive(Clone, Copy, PartialEq, Eq, Debug)]
pub enum ConnectionState {
Disconnected,
Connecting,
Connected { identified: bool },
}
#[allow(unused_variables)]
impl ConnectionState {
pub fn is_connected(&self) -> bool {
match self {
ConnectionState::Connected { identified } => true,
_ => false,
}
}
#[allow(dead_code)]
pub fn is_identified(&self) -> bool {
match self {
ConnectionState::Connected { identified: true } => true,
_ => false,
}
}
}
// ============================================================================
// Omega Connection (Client-side with auto-reconnect)
// ============================================================================
#[allow(dead_code)]
pub struct OmegaConnection {
state: Arc<RwLock<ConnectionState>>,
sender: Arc<RwLock<Option<Arc<Sender>>>>,
connection_loop_handle: Arc<Mutex<Option<JoinHandle<()>>>>,
host: String,
port: u16,
server_cert: Vec<u8>,
last_ping: Arc<Mutex<i64>>,
heartbeat_handle: Arc<Mutex<Option<JoinHandle<()>>>>,
message_send_times: Arc<Mutex<HashMap<Uuid, Instant>>>,
pub connection_id: Uuid,
shutdown_tx: Arc<Mutex<Option<watch::Sender<bool>>>>,
// Track if we should reconnect on close
reconnect_on_close: Arc<RwLock<bool>>,
}
impl OmegaConnection {
pub fn new() -> Self {
let (host, port) = omega_host_and_port();
Self::with_host(&host, port)
}
pub fn with_host(host: &str, port: u16) -> Self {
// Load server certificate from default location
let server_cert =
load_file_vec("certs", "cert.pem").expect("Failed to load server certificate");
Self::with_host_and_cert(host, port, server_cert)
}
// New constructor that accepts certificate directly
pub fn with_host_and_cert(host: &str, port: u16, server_cert: Vec<u8>) -> Self {
let (shutdown_tx, _) = watch::channel(false);
OmegaConnection {
state: Arc::new(RwLock::new(ConnectionState::Disconnected)),
sender: Arc::new(RwLock::new(None)),
connection_loop_handle: Arc::new(Mutex::new(None)),
host: host.to_string(),
port,
server_cert,
last_ping: Arc::new(Mutex::new(-1)),
heartbeat_handle: Arc::new(Mutex::new(None)),
message_send_times: Arc::new(Mutex::new(HashMap::new())),
connection_id: Uuid::new_v4(),
shutdown_tx: Arc::new(Mutex::new(Some(shutdown_tx))),
reconnect_on_close: Arc::new(RwLock::new(true)),
}
}
// -------------------------------------------------------------------------
// Connection Management
// -------------------------------------------------------------------------
pub async fn start(self: Arc<Self>) {
// Cancel any existing connection loop
if let Some(handle) = self.connection_loop_handle.lock().await.take() {
handle.abort();
}
// Recreate shutdown_tx if it was taken by stop()
if self.shutdown_tx.lock().await.is_none() {
let (shutdown_tx, _) = watch::channel(false);
*self.shutdown_tx.lock().await = Some(shutdown_tx);
}
// Set reconnect flag
*self.reconnect_on_close.write().await = true;
let self_clone = self.clone();
let handle = tokio::spawn(async move {
self_clone.connection_loop().await;
});
*self.connection_loop_handle.lock().await = Some(handle);
}
#[allow(dead_code)]
pub async fn stop(&self) {
// Disable reconnection
*self.reconnect_on_close.write().await = false;
if let Some(tx) = self.shutdown_tx.lock().await.take() {
let _ = tx.send(true);
}
if let Some(handle) = self.connection_loop_handle.lock().await.take() {
handle.abort();
}
if let Some(handle) = self.heartbeat_handle.lock().await.take() {
handle.abort();
}
// Close sender if connected
if let Some(sender) = self.sender.read().await.as_ref() {
sender.close();
}
*self.state.write().await = ConnectionState::Disconnected;
*self.sender.write().await = None;
}
async fn connection_loop(self: Arc<Self>) {
let mut reconnect_delay = RECONNECT_DELAY;
let shutdown_rx = self.shutdown_tx.lock().await.as_ref().unwrap().subscribe();
let mut shutdown_rx = shutdown_rx;
loop {
if *shutdown_rx.borrow() {
log_in!(0, PrintType::Omega, "Connection loop shutting down");
break;
}
// Check if reconnection is enabled
if !*self.reconnect_on_close.read().await {
log_in!(0, PrintType::Omega, "Reconnection disabled, exiting loop");
break;
}
match self.clone().connect_once().await {
Ok(()) => {
// Connection closed gracefully, check if we should reconnect
if *self.reconnect_on_close.read().await {
log_err!(
0,
PrintType::Omega,
"Connection lost, reconnecting in {:?}...",
reconnect_delay
);
} else {
log_in!(0, PrintType::Omega, "Connection closed, not reconnecting");
break;
}
}
Err(e) => {
log_err!(
0,
PrintType::Omega,
"Connection failed: {}, retrying in {:?}...",
e,
reconnect_delay
);
}
}
tokio::select! {
_ = sleep(reconnect_delay) => {}
_ = shutdown_rx.changed() => {
if *shutdown_rx.borrow() {
break;
}
}
}
reconnect_delay = std::cmp::min(reconnect_delay * 2, MAX_RECONNECT_DELAY);
}
}
async fn connect_once(self: Arc<Self>) -> Result<(), String> {
*self.state.write().await = ConnectionState::Connecting;
let addr_str = format!("https://{}:{}", self.host, self.port);
let (sender, mut receiver) = ttp_native::client::connect(
&addr_str,
None,
Policy {
send_mode: SendMode::SingleStreamPerMessage,
max_message_size: 1_000_000_000,
close_frame_len: u32::MAX,
application_close_code: 0,
open_stream_timeout: Duration::from_millis(2_000),
write_timeout: Duration::from_millis(2_000),
accept_stream_timeout: Duration::from_millis(10_000),
read_timeout: Duration::from_millis(30_000),
keep_alive_interval: Some(Duration::from_secs(6)),
max_idle_timeout: Some(Duration::from_secs(30)),
force_close_delay: Duration::from_millis(300),
max_transient_recv_errors: 20,
transient_recv_backoff: Duration::from_millis(100),
receiver_queue_capacity: 1000,
},
)
.await
.map_err(|e| format!("Connection failed: {}", e))?;
log_in!(
0,
PrintType::Omega,
"QUIC connection established to {}",
addr_str
);
// Store sender
let sender_arc = Arc::new(sender);
*self.sender.write().await = Some(sender_arc.clone());
*self.state.write().await = ConnectionState::Connected { identified: false };
// Get handle for close monitoring
let sender_handle = sender_arc.handle().clone();
// Start read loop
let read_self = self.clone();
let read_handle = tokio::spawn(async move {
read_self.read_loop(&mut receiver, sender_handle).await;
});
// Send identification
self.send_identification().await;
// Start heartbeat
let heartbeat_self = self.clone();
let heartbeat_handle = tokio::spawn(async move {
heartbeat_self.heartbeat_loop().await;
});
*self.heartbeat_handle.lock().await = Some(heartbeat_handle);
// Wait for read loop to complete (connection closed)
let result = read_handle.await;
// Cleanup
*self.sender.write().await = None;
*self.state.write().await = ConnectionState::Disconnected;
if let Some(handle) = self.heartbeat_handle.lock().await.take() {
handle.abort();
}
match result {
Ok(()) => {
// Check if we should reconnect
if *self.reconnect_on_close.read().await {
Err("Connection closed, will reconnect".to_string())
} else {
Ok(())
}
}
Err(e) => Err(format!("Read loop error: {}", e)),
}
}
// -------------------------------------------------------------------------
// Identification Handshake
// -------------------------------------------------------------------------
async fn send_identification(&self) {
let id = rand_u32();
let omikron_id = env::var("ID")
.unwrap_or("0".to_string())
.parse::<i64>()
.unwrap_or(0);
let identify_msg = CommunicationValue::new(CommunicationType::identification)
.with_id(id)
.add_data(DataTypes::omikron_id, DataValue::Number(omikron_id));
WAITING_TASKS.insert(
id,
WaitingTask {
task: Box::new(|selfc, cv| {
if cv.is_type(CommunicationType::error_not_found) {
log_err!(
0,
PrintType::Omega,
"Identification failed: Omikron ID not found"
);
return false;
}
if !cv.is_type(CommunicationType::challenge) {
return false;
}
tokio::spawn(async move {
if let Err(e) = selfc.handle_challenge(cv).await {
log_err!(0, PrintType::Omega, "Challenge handling failed: {}", e);
}
});
true
}),
inserted_at: Instant::now(),
},
);
self.send_message(&identify_msg).await;
}
async fn handle_challenge(&self, cv: CommunicationValue) -> Result<(), String> {
let challenge = cv
.get_data(DataTypes::challenge)
.as_str()
.ok_or("Challenge not found")?;
let server_pub_key = cv
.get_data(DataTypes::public_key)
.as_str()
.ok_or("Public key not found")?;
let decrypted_challenge = decrypt_b64(
&secret_key_to_base64(&get_private_key()),
server_pub_key,
challenge,
)
.map_err(|e| format!("Decryption failed: {:?}", e))?;
let response_msg = CommunicationValue::new(CommunicationType::challenge_response)
.with_id(cv.get_id())
.add_data(DataTypes::challenge, DataValue::Str(decrypted_challenge));
let response_id = response_msg.get_id();
WAITING_TASKS.insert(
response_id,
WaitingTask {
task: Box::new(|selfc, final_cv| {
if !final_cv.is_type(CommunicationType::identification_response) {
log_err!(0, PrintType::Omega, "Expected identification_response");
return false;
}
let accepted = final_cv
.get_data(DataTypes::accepted)
.as_bool()
.unwrap_or(false);
if !accepted {
log_err!(0, PrintType::Omega, "Omega did not accept identification");
return false;
}
tokio::spawn(async move {
let mut state = selfc.state.write().await;
if let ConnectionState::Connected { identified: _ } = *state {
*state = ConnectionState::Connected { identified: true };
}
drop(state);
selfc.sync_client_iota_status().await;
});
log!(0, PrintType::Omega, "Successfully identified with Omega");
true
}),
inserted_at: Instant::now(),
},
);
self.send_message(&response_msg).await;
Ok(())
}
async fn sync_client_iota_status(self: Arc<Self>) {
let mut connected_iota_ids: Vec<DataValue> = Vec::new();
let mut connected_user_ids: Vec<DataValue> = Vec::new();
let rho_connections_reader = RHO_CONNECTIONS.read().await;
for iota_id in rho_connections_reader.keys() {
connected_iota_ids.push(DataValue::Number(*iota_id));
}
for rho in rho_connections_reader.values() {
for client_conn in rho.get_client_connections().await {
connected_user_ids.push(DataValue::Number(client_conn.get_user_id().await as i64));
}
}
drop(rho_connections_reader);
let sync_msg = CommunicationValue::new(CommunicationType::sync_client_iota_status)
.add_data(DataTypes::iota_ids, DataValue::Array(connected_iota_ids))
.add_data(DataTypes::user_ids, DataValue::Array(connected_user_ids))
.add_data(
DataTypes::rho_connections,
DataValue::Number(connection_count().await as i64),
);
self.send_message(&sync_msg).await;
}
// -------------------------------------------------------------------------
// Read Loop & Heartbeat
// -------------------------------------------------------------------------
async fn read_loop(
self: Arc<Self>,
receiver: &mut Receiver,
sender_handle: Arc<ttp_native::ConnectionHandle>,
) {
// Monitor both receiver and sender handle for close
let mut close_rx = sender_handle.subscribe_close();
loop {
tokio::select! {
result = receiver.receive() => {
match result {
Ok(cv) => {
if !cv.is_type(CommunicationType::pong) && !cv.is_type(CommunicationType::ping) {
log_cv_in!(PrintType::Omega, &cv);
}
if cv.is_type(CommunicationType::pong) || cv.is_type(CommunicationType::ping) {
self.handle_pong(&cv).await;
continue;
}
let msg_id = cv.get_id();
if let Some((_, task)) = WAITING_TASKS.remove(&msg_id) {
if (task.task)(self.clone(), cv.clone()) {
continue;
}
}
if cv.is_type(CommunicationType::iota_user_data) {
if let DataValue::Array(users) = cv.get_data(DataTypes::user_ids) {
let mut user_ids: Vec<u64> = Vec::new();
for value in users {
if let DataValue::Number(user_id) = value {
user_ids.push(*user_id as u64);
}
}
let connections = crate::rho::rho_manager::RHO_CONNECTIONS.read().await;
if let Some(iota_id) = cv.get_data(DataTypes::iota_id).as_number() {
if let Some(rho) = connections.get(&iota_id) {
rho.get_iota_connection().set_user_ids(user_ids).await;
}
} else {
for rho in connections.values() {
rho.get_iota_connection().set_user_ids(user_ids.clone()).await;
}
}
}
}
}
Err(e) => {
log_err!(0, PrintType::Omega, "Receive error: {}", e);
break;
}
}
}
_ = close_rx.changed() => {
// Connection was closed by either side
if let Some(reason) = close_rx.borrow().clone() {
log_err!(0, PrintType::Omega, "Connection closed: {:?}", reason);
} else {
log_in!(0, PrintType::Omega, "Connection closed cleanly");
}
break;
}
}
}
}
async fn heartbeat_loop(self: Arc<Self>) {
loop {
sleep(HEARTBEAT_INTERVAL).await;
// Check if still connected
if !self.state.read().await.is_connected() {
break;
}
// Check if sender is closed
if let Some(sender) = self.sender.read().await.as_ref() {
if sender.is_closed() {
log_err!(0, PrintType::Omega, "Sender closed, stopping heartbeat");
break;
}
} else {
break;
}
self.send_ping().await;
}
}
async fn send_ping(&self) {
let ping = CommunicationValue::new(CommunicationType::ping).add_data(
DataTypes::send_time,
DataValue::Number(
std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.unwrap_or_default()
.as_secs() as i64,
),
);
self.send_message(&ping).await;
}
async fn handle_pong(&self, cv: &CommunicationValue) {
let timestamp = cv
.get_data(DataTypes::send_time)
.as_number()
.unwrap_or_else(|| {
std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.unwrap_or_default()
.as_secs() as i64
});
*self.last_ping.lock().await = timestamp;
}
// -------------------------------------------------------------------------
// Public API
// -------------------------------------------------------------------------
pub async fn send_message(&self, cv: &CommunicationValue) {
if !cv.is_type(CommunicationType::pong) && !cv.is_type(CommunicationType::ping) {
log_cv_out!(PrintType::Omega, &cv);
}
let sender_guard = self.sender.read().await;
if let Some(sender) = sender_guard.as_ref() {
// Check if closed before sending
if sender.is_closed() {
log_err!(0, PrintType::Omega, "Cannot send: connection closed");
drop(sender_guard);
// Trigger reconnection by closing the connection state
if let Some(sender) = self.sender.write().await.take() {
sender.close();
}
return;
}
let sender_clone = Arc::clone(sender);
drop(sender_guard);
if let Err(e) = sender_clone.send(cv).await {
log_err!(0, PrintType::Omega, "Send failed: {}", e);
}
} else {
log_err!(0, PrintType::Omega, "Cannot send: not connected");
}
}
pub async fn await_connection(&self, timeout_duration: Option<Duration>) -> Result<(), String> {
if self.state.read().await.is_connected() {
return Ok(());
}
let timeout = timeout_duration.unwrap_or(CONNECTION_TIMEOUT);
let start = Instant::now();
loop {
if self.state.read().await.is_connected() {
return Ok(());
}
if start.elapsed() >= timeout {
return Err(format!(
"Connection not established within {} seconds",
timeout.as_secs()
));
}
sleep(Duration::from_millis(100)).await;
}
}
pub async fn await_response(
&self,
cv: &CommunicationValue,
timeout_duration: Option<Duration>,
) -> Result<CommunicationValue, String> {
self.await_connection(timeout_duration).await?;
let (tx, mut rx) = mpsc::channel(1);
let msg_id = cv.get_id();
WAITING_TASKS.insert(
msg_id,
WaitingTask {
task: Box::new(move |_, response_cv| {
let inner_tx = tx.clone();
tokio::spawn(async move {
let _ = inner_tx.send(response_cv).await;
});
true
}),
inserted_at: Instant::now(),
},
);
self.send_message(cv).await;
let timeout = timeout_duration.unwrap_or(Duration::from_secs(10));
match tokio::time::timeout(timeout, rx.recv()).await {
Ok(Some(response_cv)) => Ok(response_cv),
Ok(_) => Err("Channel closed".to_string()),
Err(_) => {
WAITING_TASKS.remove(&msg_id);
Err("Request timed out".to_string())
}
}
}
#[allow(dead_code)]
pub async fn is_connected(&self) -> bool {
self.state.read().await.is_connected()
}
#[allow(dead_code)]
pub async fn is_identified(&self) -> bool {
self.state.read().await.is_identified()
}
#[allow(dead_code)]
pub async fn close_iota(iota_id: i64) {
let cv = CommunicationValue::new(CommunicationType::iota_disconnected)
.add_data(DataTypes::iota_id, DataValue::Number(iota_id));
OMEGA_CONNECTION.send_message(&cv).await;
}
pub async fn client_changed(_iota_id: i64, user_id: i64, state: UserStatus) {
let msg_type = match state {
UserStatus::iota_offline => CommunicationType::user_disconnected,
UserStatus::user_offline => CommunicationType::user_disconnected,
UserStatus::user_invisible => CommunicationType::user_disconnected,
_ => CommunicationType::user_connected,
};
let cv = CommunicationValue::new(msg_type)
.add_data(DataTypes::user_id, DataValue::Number(user_id))
.add_data(DataTypes::user_state, DataValue::Str(state.to_string()));
OMEGA_CONNECTION.send_message(&cv).await;
}
pub async fn user_states(user_id: i64, user_ids: Vec<i64>) {
let user_ids = user_ids.iter().map(|v| DataValue::Number(*v)).collect();
let cv = CommunicationValue::new(CommunicationType::get_states)
.add_data(DataTypes::user_ids, DataValue::Array(user_ids));
let msg_id = cv.get_id();
WAITING_TASKS.insert(
msg_id,
WaitingTask {
task: Box::new(
move |_: Arc<OmegaConnection>, response: CommunicationValue| {
tokio::spawn(async move {
let rho = rho_manager::get_rho_con_for_user(user_id).await;
if let Some(rho) = rho {
for client in rho.get_client_connections_for_user(user_id).await {
client.send_message(&response).await;
}
}
});
true
},
),
inserted_at: Instant::now(),
},
);
OMEGA_CONNECTION.send_message(&cv).await;
}
}
// ============================================================================
// Global Instance
// ============================================================================
static OMEGA_CONNECTION: Lazy<Arc<OmegaConnection>> = Lazy::new(|| {
let conn = Arc::new(OmegaConnection::new());
// Start the connection manager immediately
let conn_clone = conn.clone();
tokio::spawn(async move {
conn_clone.start().await;
});
start_task_cleanup_loop();
conn
});
pub fn get_omega_connection() -> Arc<OmegaConnection> {
OMEGA_CONNECTION.clone()
}