refactor: streamline master message handling by removing unnecessary Arc and Mutex usage, and reorganizing handler structure

This commit is contained in:
GW_MC
2026-05-31 09:41:33 +00:00
parent 93a36cb101
commit 8dbbeb4b24
4 changed files with 226 additions and 276 deletions

View File

@@ -0,0 +1,62 @@
use std::sync::Arc;
use nxmesh_proto::{AgentMessage, ConfigUpdate, MasterMessage, master_message::Payload};
use crate::service::master_handler::{MasterHandlerError, MessageResult};
#[async_trait::async_trait]
pub trait MasterMessageHandler: Send + Sync + 'static {
async fn handle_master_message(&self, message: MasterMessage) -> MessageResult<()>;
}
#[async_trait::async_trait]
pub trait OnConfigUpdateHandler: Send + Sync + 'static {
// Handle the config update message from master, write the config content to files, validate the new config and reload nginx
async fn on_config_update(&self, config_info: ConfigUpdate) -> MessageResult<()>;
}
pub struct HandlerImpl<OCH>
where
OCH: OnConfigUpdateHandler,
{
on_config_update_handler: Arc<OCH>,
}
impl<OCH> HandlerImpl<OCH>
where
OCH: OnConfigUpdateHandler,
{
pub fn new(on_config_update_handler: Arc<OCH>) -> Self {
Self {
on_config_update_handler,
}
}
}
#[async_trait::async_trait]
impl<OCH> MasterMessageHandler for HandlerImpl<OCH>
where
OCH: OnConfigUpdateHandler,
{
async fn handle_master_message(&self, message: MasterMessage) -> MessageResult<()> {
match message.payload {
Some(Payload::ConfigUpdate(config_info)) => {
self.on_config_update_handler
.on_config_update(config_info)
.await
}
Some(_) => {
// We should never receive other types of messages from the master, but we should handle it anyway
Err(MasterHandlerError::MessageHandlingError(
"Received unsupported master message type".to_string(),
))
}
None => {
// This should never happen as the master should always send a valid message, but we should handle it anyway
return Err(MasterHandlerError::MessageHandlingError(
"Received master message with empty payload".to_string(),
));
}
}
}
}

View File

@@ -0,0 +1,163 @@
use std::sync::Arc;
use nxmesh_proto::AgentMessage;
use tokio::sync::mpsc;
use tokio_stream::wrappers::ReceiverStream;
use tracing::{error, info, warn};
use crate::{
connector::master::{MasterConnector, MasterConnectorTrait},
service::master_handler::handlers::MasterMessageHandler,
};
pub mod handlers;
#[derive(Debug)]
pub enum MasterHandlerError {
ConnectionError(String),
// TODO: should be protobuf error to transmit the error to master
MessageHandlingError(String),
RetryLimitExceeded(String),
SendMessageError(String),
}
pub type MessageResult<T> = std::result::Result<T, MasterHandlerError>;
#[async_trait::async_trait]
pub trait MasterHandler: Send + Sync + 'static {
// Create a new routine to handle incoming messages from the master
// This method will auto-reconnect if the connection is lost, so it should run indefinitely until the agent is shut down
async fn start_handle_master_message(&self) -> MessageResult<()>;
async fn stop_handle_master_message(&self) -> MessageResult<()>;
// Send a message to the master, response should be handled by the agent message handler registered
async fn send_message_to_master(&self, message: AgentMessage) -> MessageResult<()>;
}
struct MessageHandleInfo {
handle: tokio::task::JoinHandle<()>,
tx: mpsc::Sender<AgentMessage>,
}
pub struct MasterHandlerImpl<MMH>
where
MMH: MasterMessageHandler,
{
connector: Arc<MasterConnector>,
message_handler: Arc<MMH>,
message_handle_lock: tokio::sync::RwLock<Option<MessageHandleInfo>>,
}
impl<MMH> MasterHandlerImpl<MMH>
where
MMH: MasterMessageHandler,
{
pub fn new(connector: Arc<MasterConnector>, message_handler: Arc<MMH>) -> Self {
Self {
connector,
message_handler,
message_handle_lock: tokio::sync::RwLock::new(None),
}
}
}
#[async_trait::async_trait]
impl<MMH> MasterHandler for MasterHandlerImpl<MMH>
where
MMH: MasterMessageHandler,
{
async fn start_handle_master_message(&self) -> MessageResult<()> {
let mut client = self.connector.get_client();
'connection_loop: loop {
let guard_result = self.message_handle_lock.try_write();
let mut guard = match guard_result {
Ok(g) if g.is_none() => g,
Ok(_) => {
warn!("Master message handler is already running");
return Ok(());
}
Err(e) => {
return Err(MasterHandlerError::MessageHandlingError(format!(
"Failed to acquire lock for message handler: {}",
e
)));
}
};
let (tx, rx) = mpsc::channel(32);
let outbound_stream = ReceiverStream::new(rx);
// 2. Connect to the master and start the bi-directional streaming RPC
let mut stream = match client.stream(outbound_stream).await {
Ok(s) => s.into_inner(),
Err(e) => {
error!(
"Failed to connect to master: {}. Retrying in 5 seconds...",
e
);
tokio::time::sleep(std::time::Duration::from_secs(5)).await;
continue 'connection_loop;
}
};
// 3. Spawn a task to handle incoming messages from the master
let message_handler = self.message_handler.clone();
let handle = tokio::spawn(async move {
'message_loop: loop {
let message = stream.message().await;
match message {
Ok(Some(msg)) => {
if let Err(e) = message_handler.handle_master_message(msg).await {
error!("Failed to handle master message: {:?}", e);
}
continue 'message_loop;
}
Ok(None) => {
warn!("Master closed the connection");
return;
}
Err(e) => {
error!("Error receiving message from master: {:?}", e);
return;
}
}
}
});
*guard = Some(MessageHandleInfo { handle, tx });
}
}
async fn stop_handle_master_message(&self) -> MessageResult<()> {
// 1. signal the task to stop (e.g. using a cancellation token or channel)
// 2. wait for the task to finish and handle any errors
// 3. set handle_master_message_task to None
let mut guard = self.message_handle_lock.write().await;
if let Some(handle_info) = guard.take() {
handle_info.handle.abort();
match handle_info.handle.await {
Ok(_) => info!("Master message handler task stopped successfully"),
Err(e) => error!("Failed to stop master message handler task: {:?}", e),
}
} else {
warn!("Master message handler is not running");
}
Ok(())
}
async fn send_message_to_master(&self, message: AgentMessage) -> MessageResult<()> {
let guard = self.message_handle_lock.read().await;
if let Some(handle_info) = guard.as_ref() {
handle_info.tx.send(message).await.map_err(|e| {
MasterHandlerError::SendMessageError(format!(
"Failed to send message to master: {}",
e
))
})?;
} else {
return Err(MasterHandlerError::SendMessageError(
"Master message handler is not running".to_string(),
));
}
Ok(())
}
}