fix: memory leak hunting and reconnecting on data error

toju-app/src/app/infrastructure/realtime/peer-connection-manager/recovery/peer-recovery.spec.ts

@@ -13,7 +13,7 @@ describe('peer recovery', () => {
     vi.useRealTimers();
   });
 
-  it('waits a short grace period before replacing a closed data channel in place', () => {
+  it('recreates a peer immediately when the data channel is already closed', () => {
     vi.useFakeTimers();
 
     const channel = createDataChannel('closed');
@@ -24,29 +24,28 @@ describe('peer recovery', () => {
 
     scheduleDataChannelRecovery(context, 'bob', channel, 'close', handlers);
 
+    expect(handlers.removePeer).toHaveBeenCalledWith('bob', { preserveReconnectState: true });
+    expect(handlers.createPeerConnection).toHaveBeenCalledWith('bob', true);
+    expect(handlers.createAndSendOffer).toHaveBeenCalledWith('bob');
+    expect(context.state.dataChannelRecoveryTimers.has('bob')).toBe(false);
+  });
+
+  it('waits a short grace period before recreating a peer with a closing data channel', () => {
+    vi.useFakeTimers();
+
+    const channel = createDataChannel('closing');
+    const context = createContext('alice');
+    const handlers = createRecoveryHandlers(context);
+
+    context.state.activePeerConnections.set('bob', createPeerData(channel, 'connected'));
+
+    scheduleDataChannelRecovery(context, 'bob', channel, 'close', handlers);
+
     vi.advanceTimersByTime(DATA_CHANNEL_RECOVERY_GRACE_MS - 1);
     expect(handlers.removePeer).not.toHaveBeenCalled();
     expect(handlers.createPeerConnection).not.toHaveBeenCalled();
 
     vi.advanceTimersByTime(1);
-    expect(handlers.replaceDataChannel).toHaveBeenCalledWith('bob', channel);
-    expect(handlers.removePeer).not.toHaveBeenCalled();
-    expect(handlers.createPeerConnection).not.toHaveBeenCalled();
-    expect(handlers.createAndSendOffer).not.toHaveBeenCalled();
-  });
-
-  it('falls back to full peer recreation when in-place data channel replacement fails', () => {
-    vi.useFakeTimers();
-
-    const channel = createDataChannel('closed');
-    const context = createContext('alice');
-    const handlers = createRecoveryHandlers(context);
-
-    handlers.replaceDataChannel.mockReturnValueOnce(false);
-    context.state.activePeerConnections.set('bob', createPeerData(channel, 'connected'));
-
-    scheduleDataChannelRecovery(context, 'bob', channel, 'close', handlers);
-    vi.advanceTimersByTime(DATA_CHANNEL_RECOVERY_GRACE_MS);
 
     expect(handlers.removePeer).toHaveBeenCalledWith('bob', { preserveReconnectState: true });
     expect(handlers.createPeerConnection).toHaveBeenCalledWith('bob', true);
@@ -56,7 +55,7 @@ describe('peer recovery', () => {
   it('does not recreate a peer when a replacement data channel is adopted before the grace expires', () => {
     vi.useFakeTimers();
 
-    const staleChannel = createDataChannel('closed');
+    const staleChannel = createDataChannel('closing');
     const replacementChannel = createDataChannel(DATA_CHANNEL_STATE_OPEN);
     const context = createContext('alice');
     const handlers = createRecoveryHandlers(context);
@@ -90,7 +89,7 @@ describe('peer recovery', () => {
     expect(handlers.createPeerConnection).not.toHaveBeenCalled();
   });
 
-  it('preserves a connected non-initiator peer while waiting for the remote initiator to replace the channel', () => {
+  it('recreates a connected non-initiator peer and waits for the remote initiator offer', () => {
     vi.useFakeTimers();
 
     const channel = createDataChannel('closed');
@@ -99,11 +98,10 @@ describe('peer recovery', () => {
 
     context.state.activePeerConnections.set('bob', createPeerData(channel, 'connected', false));
     scheduleDataChannelRecovery(context, 'bob', channel, 'close', handlers);
-    vi.advanceTimersByTime(DATA_CHANNEL_RECOVERY_GRACE_MS);
 
-    expect(handlers.removePeer).not.toHaveBeenCalled();
+    expect(handlers.removePeer).toHaveBeenCalledWith('bob', { preserveReconnectState: true });
     expect(handlers.replaceDataChannel).not.toHaveBeenCalled();
-    expect(handlers.createPeerConnection).not.toHaveBeenCalled();
+    expect(handlers.createPeerConnection).toHaveBeenCalledWith('bob', false);
     expect(handlers.createAndSendOffer).not.toHaveBeenCalled();
   });
 

toju-app/src/app/infrastructure/realtime/peer-connection-manager/recovery/peer-recovery.ts

@@ -154,6 +154,18 @@ export function scheduleDataChannelRecovery(
   if (channel.readyState === DATA_CHANNEL_STATE_OPEN)
     return;
 
+  if (channel.readyState === 'closed') {
+    logger.warn('[data-channel] Control channel closed; reconnecting peer immediately', {
+      channelLabel: channel.label,
+      connectionState: peerData.connection.connectionState,
+      peerId,
+      reason
+    });
+
+    repairUnavailableDataChannel(context, peerId, channel, reason, handlers);
+    return;
+  }
+
   if (state.dataChannelRecoveryTimers.has(peerId))
     return;
 
@@ -183,35 +195,42 @@ export function scheduleDataChannelRecovery(
       reason
     });
 
-    if (latestPeerData.connection.connectionState === CONNECTION_STATE_CONNECTED) {
-      if (latestPeerData.isInitiator && handlers.replaceDataChannel(peerId, channel)) {
-        logger.info('[data-channel] Replaced control channel without recreating media transport', {
-          peerId,
-          reason
-        });
-
-        return;
-      }
-
-      if (!latestPeerData.isInitiator) {
-        logger.info('[data-channel] Waiting for initiator to replace control channel; preserving media transport', {
-          peerId,
-          reason
-        });
-
-        return;
-      }
-    }
-
-    trackDisconnectedPeer(state, peerId);
-    handlers.removePeer(peerId, { preserveReconnectState: true });
-    attemptPeerReconnect(context, peerId, handlers);
-    schedulePeerReconnect(context, peerId, handlers);
+    repairUnavailableDataChannel(context, peerId, channel, reason, handlers);
   }, DATA_CHANNEL_RECOVERY_GRACE_MS);
 
   state.dataChannelRecoveryTimers.set(peerId, timer);
 }
 
+function repairUnavailableDataChannel(
+  context: PeerConnectionManagerContext,
+  peerId: string,
+  channel: RTCDataChannel,
+  reason: string,
+  handlers: RecoveryHandlers
+): void {
+  const { logger, state } = context;
+  const peerData = state.activePeerConnections.get(peerId);
+
+  if (!peerData || peerData.dataChannel !== channel)
+    return;
+
+  if (peerData.dataChannel?.readyState === DATA_CHANNEL_STATE_OPEN)
+    return;
+
+  logger.warn('[data-channel] Recreating peer transport after control channel failure', {
+    channelLabel: channel.label,
+    connectionState: peerData.connection.connectionState,
+    peerId,
+    readyState: peerData.dataChannel?.readyState ?? null,
+    reason
+  });
+
+  trackDisconnectedPeer(state, peerId);
+  handlers.removePeer(peerId, { preserveReconnectState: true });
+  attemptPeerReconnect(context, peerId, handlers);
+  schedulePeerReconnect(context, peerId, handlers);
+}
+
 export function schedulePeerDisconnectRecovery(
   context: PeerConnectionManagerContext,
   peerId: string,