fix(channels): unify lifecycle ownership

src/channels/feishu.rs

@@ -8,7 +8,8 @@ use futures_util::{SinkExt, StreamExt};
 use prost::{Message as ProstMessage, bytes::Bytes};
 use regex::Regex;
 use serde::Deserialize;
-use tokio::sync::{RwLock, broadcast};
+use tokio::sync::{Mutex, RwLock, broadcast};
+use tokio::task::JoinHandle;
 
 use crate::bus::{MediaItem, MessageBus, OutboundMessage};
 use crate::channels::base::{Channel, ChannelError};
@@ -152,6 +153,7 @@ pub struct FeishuChannel {
     http_client: reqwest::Client,
     running: Arc<RwLock<bool>>,
     shutdown_tx: Arc<RwLock<Option<broadcast::Sender<()>>>>,
+    run_task: Arc<Mutex<Option<JoinHandle<()>>>>,
     connected: Arc<RwLock<bool>>,
     /// Cached tenant access token with proactive refresh.
     tenant_token: Arc<RwLock<Option<CachedTenantToken>>>,
@@ -185,6 +187,7 @@ impl FeishuChannel {
             http_client: reqwest::Client::new(),
             running: Arc::new(RwLock::new(false)),
             shutdown_tx: Arc::new(RwLock::new(None)),
+            run_task: Arc::new(Mutex::new(None)),
             connected: Arc::new(RwLock::new(false)),
             tenant_token: Arc::new(RwLock::new(None)),
             seen_message_ids: Arc::new(RwLock::new(HashMap::new())),
@@ -1953,6 +1956,14 @@ impl Channel for FeishuChannel {
             ));
         }
 
+        let mut run_task = self.run_task.lock().await;
+        if run_task.as_ref().is_some_and(|task| !task.is_finished()) {
+            return Ok(());
+        }
+        if let Some(finished) = run_task.take() {
+            let _ = finished.await;
+        }
+
         *self.running.write().await = true;
 
         let (shutdown_tx, _) = broadcast::channel(1);
@@ -1960,7 +1971,7 @@ impl Channel for FeishuChannel {
 
         let channel = self.clone();
         let bus = bus.clone();
-        tokio::spawn(async move {
+        *run_task = Some(tokio::spawn(async move {
             let mut consecutive_failures = 0;
             let max_failures = 3;
 
@@ -1994,7 +2005,7 @@ impl Channel for FeishuChannel {
 
             *channel.running.write().await = false;
             tracing::info!("Feishu channel stopped");
-        });
+        }));
 
         tracing::info!("Feishu channel started");
         Ok(())
@@ -2007,6 +2018,12 @@ impl Channel for FeishuChannel {
             let _ = tx.send(());
         }
 
+        if let Some(task) = self.run_task.lock().await.take() {
+            task.await.map_err(|error| {
+                ChannelError::Other(format!("Feishu channel task failed to join: {error}"))
+            })?;
+        }
+
         Ok(())
     }
 

src/channels/manager.rs

@@ -2,7 +2,7 @@ use std::collections::HashMap;
 use std::sync::Arc;
 use tokio::sync::RwLock;
 
-use crate::bus::{MessageBus, OutboundMessage};
+use crate::bus::MessageBus;
 use crate::channels::base::{Channel, ChannelError};
 use crate::channels::feishu::FeishuChannel;
 use crate::config::Config;
@@ -16,14 +16,6 @@ pub struct ChannelManager {
 }
 
 impl ChannelManager {
-    pub fn new(cli_chat_channel: Arc<crate::channels::CliChatChannel>) -> Self {
-        Self {
-            channels: Arc::new(RwLock::new(HashMap::new())),
-            cli_chat_channel,
-            bus: MessageBus::new(100),
-        }
-    }
-
     pub fn with_bus(
         cli_chat_channel: Arc<crate::channels::CliChatChannel>,
         bus: Arc<MessageBus>,
@@ -91,27 +83,53 @@ impl ChannelManager {
     }
 
     pub async fn start_all(&self) -> Result<(), ChannelError> {
-        let channels = self.channels.read().await;
+        let channels: Vec<_> = self
+            .channels
+            .read()
+            .await
+            .iter()
+            .map(|(name, channel)| (name.clone(), channel.clone()))
+            .collect();
         let bus = self.bus.clone();
-        for (name, channel) in channels.iter() {
+        let mut failures = Vec::new();
+        for (name, channel) in channels {
             tracing::info!(channel = %name, "Starting channel");
             if let Err(e) = channel.start(bus.clone()).await {
                 tracing::error!(channel = %name, error = %e, "Failed to start channel");
+                failures.push(format!("{name}: {e}"));
             }
         }
+        if failures.is_empty() {
             Ok(())
+        } else {
+            Err(ChannelError::Other(format!(
+                "failed to start channels: {}",
+                failures.join("; ")
+            )))
+        }
     }
 
     pub async fn stop_all(&self) -> Result<(), ChannelError> {
-        let mut channels = self.channels.write().await;
+        let channels: Vec<_> = {
-        for (name, channel) in channels.iter() {
+            let mut registered = self.channels.write().await;
+            registered.drain().collect()
+        };
+        let mut failures = Vec::new();
+        for (name, channel) in channels {
             tracing::info!(channel = %name, "Stopping channel");
             if let Err(e) = channel.stop().await {
                 tracing::error!(channel = %name, error = %e, "Error stopping channel");
+                failures.push(format!("{name}: {e}"));
             }
         }
-        channels.clear();
+        if failures.is_empty() {
             Ok(())
+        } else {
+            Err(ChannelError::Other(format!(
+                "failed to stop channels: {}",
+                failures.join("; ")
+            )))
+        }
     }
 
     pub async fn get_channel(&self, name: &str) -> Option<Arc<dyn Channel + Send + Sync>> {
@@ -122,17 +140,104 @@ impl ChannelManager {
     pub async fn list_channel_names(&self) -> Vec<String> {
         self.channels.read().await.keys().cloned().collect()
     }
+}
 
-    /// Dispatch an outbound message to the appropriate channel
+#[cfg(test)]
-    pub async fn dispatch(&self, msg: OutboundMessage) -> Result<(), ChannelError> {
+mod tests {
-        let channel_name = &msg.channel;
+    use super::*;
-        if let Some(channel) = self.get_channel(channel_name).await {
+    use async_trait::async_trait;
-            channel.send(msg).await
+    use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
-        } else {
+
-            Err(ChannelError::Other(format!(
+    struct TestChannel {
-                "Channel not found: {}",
+        name: &'static str,
-                channel_name
+        fail_start: bool,
-            )))
+        fail_stop: bool,
+        running: AtomicBool,
+        starts: AtomicUsize,
+        stops: AtomicUsize,
+    }
+
+    impl TestChannel {
+        fn new(name: &'static str, fail_start: bool, fail_stop: bool) -> Self {
+            Self {
+                name,
+                fail_start,
+                fail_stop,
+                running: AtomicBool::new(false),
+                starts: AtomicUsize::new(0),
+                stops: AtomicUsize::new(0),
             }
         }
     }
+
+    #[async_trait]
+    impl Channel for TestChannel {
+        fn name(&self) -> &str {
+            self.name
+        }
+
+        fn is_running(&self) -> bool {
+            self.running.load(Ordering::SeqCst)
+        }
+
+        async fn start(&self, _bus: Arc<MessageBus>) -> Result<(), ChannelError> {
+            self.starts.fetch_add(1, Ordering::SeqCst);
+            if self.fail_start {
+                return Err(ChannelError::ConnectionError("unavailable".into()));
+            }
+            self.running.store(true, Ordering::SeqCst);
+            Ok(())
+        }
+
+        async fn stop(&self) -> Result<(), ChannelError> {
+            self.stops.fetch_add(1, Ordering::SeqCst);
+            self.running.store(false, Ordering::SeqCst);
+            if self.fail_stop {
+                return Err(ChannelError::Other("stop failed".into()));
+            }
+            Ok(())
+        }
+
+        async fn send(&self, _msg: crate::bus::OutboundMessage) -> Result<(), ChannelError> {
+            Ok(())
+        }
+    }
+
+    fn manager() -> ChannelManager {
+        ChannelManager::with_bus(
+            Arc::new(crate::channels::CliChatChannel::new()),
+            MessageBus::new(8),
+        )
+    }
+
+    #[tokio::test]
+    async fn start_all_reports_channel_failures_and_starts_the_rest() {
+        let manager = manager();
+        let healthy = Arc::new(TestChannel::new("healthy", false, false));
+        let broken = Arc::new(TestChannel::new("broken", true, false));
+        manager.register_channel("healthy", healthy.clone()).await;
+        manager.register_channel("broken", broken.clone()).await;
+
+        let error = manager.start_all().await.unwrap_err().to_string();
+
+        assert!(error.contains("broken"));
+        assert_eq!(healthy.starts.load(Ordering::SeqCst), 1);
+        assert_eq!(broken.starts.load(Ordering::SeqCst), 1);
+    }
+
+    #[tokio::test]
+    async fn stop_all_reports_failures_and_unregisters_every_channel() {
+        let manager = manager();
+        let healthy = Arc::new(TestChannel::new("healthy", false, false));
+        let broken = Arc::new(TestChannel::new("broken", false, true));
+        manager.register_channel("healthy", healthy.clone()).await;
+        manager.register_channel("broken", broken.clone()).await;
+
+        let error = manager.stop_all().await.unwrap_err().to_string();
+
+        assert!(error.contains("broken"));
+        assert_eq!(healthy.stops.load(Ordering::SeqCst), 1);
+        assert_eq!(broken.stops.load(Ordering::SeqCst), 1);
+        assert!(manager.list_channel_names().await.is_empty());
+    }
+}

src/gateway/mod.rs

@@ -6,7 +6,7 @@ use std::sync::Arc;
 use tokio::net::TcpListener;
 
 use crate::bus::{ControlMessage, MessageBus, OutboundDispatcher};
-use crate::channels::base::{Channel, ChannelError};
+use crate::channels::base::ChannelError;
 use crate::channels::{ChannelManager, CliChatChannel};
 use crate::config::{Config, ensure_workspace_dir, expand_path};
 use crate::logging;
@@ -198,12 +198,6 @@ impl GatewayState {
         let bus_for_outbound = bus.clone();
         let session_manager = self.session_manager.clone();
 
-        // Start CLI Chat Channel (it's already registered in ChannelManager)
-        let cli_chat_channel = self.cli_chat_channel();
-        if let Err(e) = cli_chat_channel.start(bus.clone()).await {
-            tracing::error!(error = %e, "Failed to start CLI chat channel");
-        }
-
         // Spawn unified message processor
         // This handles both inbound AI messages and control messages in one loop
         self.task_supervisor.spawn("message-processor", async move {