fix(server): prevent streaming hang during long agent sessions (#1088)

* fix(server): increase WS buffer/replay limits and add backpressure warning

During long-running agent sessions (e.g. ultrawork loops with many tool
calls), the browser WebSocket client can briefly fall behind the server.
When the outbound buffer exceeds the limit, the server force-disconnects
with close code 1013, and the small replay buffer (512 events) is
insufficient to recover all missed events — leaving the UI permanently
stalled.

Changes:
- Raise MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES from 4 MB to 16 MB to
  tolerate larger bursts without disconnecting
- Add MESSAGE_STREAM_WS_BACKPRESSURE_WARN_BYTES (12 MB) threshold that
  sends a one-shot "backpressure" frame to the client before the hard
  disconnect, giving it a chance to shed low-priority updates
- Raise MESSAGE_STREAM_GLOBAL_REPLAY_LIMIT from 512 to 2048 so more
  events survive brief reconnection gaps
- Add tests for the backpressure warning behavior (emit, dedup, reset)

* fix(ui): batch event flushes under backpressure

---------
Co-authored-by: Bohdan Triapitsyn <artmore@protonmail.com>

packages/web/server/lib/event-stream/global-hub.js

@@ -1,6 +1,8 @@
 import { createUpstreamSseReader } from './upstream-reader.js';
 
-export const MESSAGE_STREAM_GLOBAL_REPLAY_LIMIT = 512;
+// Raised from 512 → 2048 to improve recovery after brief disconnects during
+// long-running agent sessions where many events accumulate quickly.
+export const MESSAGE_STREAM_GLOBAL_REPLAY_LIMIT = 2048;
 
 export function createGlobalMessageStreamHub({
   buildOpenCodeUrl,

packages/web/server/lib/event-stream/protocol.js

@@ -3,7 +3,13 @@ export const MESSAGE_STREAM_DIRECTORY_WS_PATH = '/api/event/ws';
 export const MESSAGE_STREAM_WS_HEARTBEAT_INTERVAL_MS = 15 * 1000;
 // Per-client pending outbound WS buffer, not a payload or stream-size limit.
 // Healthy clients stay near 0; this only trips when a client is far behind.
-export const MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES = 4 * 1024 * 1024;
+// Raised from 4 MB → 16 MB to tolerate bursts during long agent sessions
+// (e.g. ultrawork / multi-tool loops) where the browser briefly falls behind.
+export const MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES = 16 * 1024 * 1024;
+
+// Threshold at which we emit a backpressure warning frame so the client can
+// proactively start shedding low-priority updates before the hard disconnect.
+export const MESSAGE_STREAM_WS_BACKPRESSURE_WARN_BYTES = 12 * 1024 * 1024;
 
 export function parseSseEventEnvelope(block) {
   if (!block || typeof block !== 'string') {
@@ -67,7 +73,9 @@ export function sendMessageStreamWsFrame(socket, payload) {
     return false;
   }
 
-  if (typeof socket.bufferedAmount === 'number' && socket.bufferedAmount > MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES) {
+  const buffered = typeof socket.bufferedAmount === 'number' ? socket.bufferedAmount : 0;
+
+  if (buffered > MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES) {
     try {
       socket.close(1013, 'Message stream client is too slow');
     } catch {
@@ -77,13 +85,36 @@ export function sendMessageStreamWsFrame(socket, payload) {
 
   try {
     socket.send(JSON.stringify(payload));
-    if (typeof socket.bufferedAmount === 'number' && socket.bufferedAmount > MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES) {
+    const bufferedAfter = typeof socket.bufferedAmount === 'number' ? socket.bufferedAmount : 0;
+    if (bufferedAfter > MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES) {
       try {
         socket.close(1013, 'Message stream client is too slow');
       } catch {
       }
       return false;
     }
+
+    // Emit a one-shot backpressure warning when the buffer is building up.
+    // The flag prevents sending repeated warnings that would themselves
+    // increase the buffer.  It resets once the buffer drains below the
+    // threshold.
+    if (bufferedAfter > MESSAGE_STREAM_WS_BACKPRESSURE_WARN_BYTES) {
+      if (!socket._ocBackpressureWarned) {
+        socket._ocBackpressureWarned = true;
+        try {
+          socket.send(JSON.stringify({
+            type: 'backpressure',
+            bufferedBytes: bufferedAfter,
+            maxBytes: MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES,
+          }));
+        } catch {
+          // Best-effort warning — ignore send failures.
+        }
+      }
+    } else if (socket._ocBackpressureWarned) {
+      socket._ocBackpressureWarned = false;
+    }
+
     return true;
   } catch {
     return false;

packages/web/server/lib/event-stream/protocol.test.js

@@ -4,6 +4,7 @@ import {
   MESSAGE_STREAM_DIRECTORY_WS_PATH,
   MESSAGE_STREAM_GLOBAL_WS_PATH,
   MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES,
+  MESSAGE_STREAM_WS_BACKPRESSURE_WARN_BYTES,
   parseSseEventEnvelope,
   sendMessageStreamWsEvent,
   sendMessageStreamWsFrame,
@@ -88,6 +89,73 @@ describe('event stream protocol helpers', () => {
     });
   });
 
+  it('emits a backpressure warning when buffer exceeds the warn threshold', () => {
+    const sentPayloads = [];
+    const socket = {
+      readyState: 1,
+      bufferedAmount: MESSAGE_STREAM_WS_BACKPRESSURE_WARN_BYTES + 1,
+      send(payload) {
+        sentPayloads.push(payload);
+      },
+    };
+
+    const sent = sendMessageStreamWsFrame(socket, { type: 'test' });
+
+    expect(sent).toBe(true);
+    expect(sentPayloads).toHaveLength(2);
+
+    const warning = JSON.parse(sentPayloads[1]);
+    expect(warning.type).toBe('backpressure');
+    expect(warning.bufferedBytes).toBeGreaterThan(0);
+    expect(warning.maxBytes).toBe(MESSAGE_STREAM_WS_MAX_BUFFERED_BYTES);
+  });
+
+  it('does not repeat backpressure warnings while still above threshold', () => {
+    const sentPayloads = [];
+    const socket = {
+      readyState: 1,
+      bufferedAmount: MESSAGE_STREAM_WS_BACKPRESSURE_WARN_BYTES + 1,
+      send(payload) {
+        sentPayloads.push(payload);
+      },
+    };
+
+    sendMessageStreamWsFrame(socket, { type: 'test1' });
+    sendMessageStreamWsFrame(socket, { type: 'test2' });
+
+    // First call: data + warning = 2 sends.  Second call: data only = 1 send.
+    expect(sentPayloads).toHaveLength(3);
+    expect(JSON.parse(sentPayloads[1]).type).toBe('backpressure');
+    expect(JSON.parse(sentPayloads[2])).toEqual({ type: 'test2' });
+  });
+
+  it('resets backpressure warning flag when buffer drains', () => {
+    const sentPayloads = [];
+    const socket = {
+      readyState: 1,
+      bufferedAmount: MESSAGE_STREAM_WS_BACKPRESSURE_WARN_BYTES + 1,
+      send(payload) {
+        sentPayloads.push(payload);
+      },
+    };
+
+    sendMessageStreamWsFrame(socket, { type: 'first' });
+    expect(socket._ocBackpressureWarned).toBe(true);
+
+    // Buffer drains
+    socket.bufferedAmount = 100;
+    sendMessageStreamWsFrame(socket, { type: 'recovered' });
+    expect(socket._ocBackpressureWarned).toBe(false);
+
+    // Buffer spikes again — warning should fire again
+    socket.bufferedAmount = MESSAGE_STREAM_WS_BACKPRESSURE_WARN_BYTES + 1;
+    sendMessageStreamWsFrame(socket, { type: 'again' });
+    const backpressureFrames = sentPayloads
+      .map((p) => JSON.parse(p))
+      .filter((p) => p.type === 'backpressure');
+    expect(backpressureFrames).toHaveLength(2);
+  });
+
   it('serializes event frames with routing metadata', () => {
     let rawPayload = null;
     const socket = {