fix(server): rebind message-stream upstreams after a managed OpenCode restart

When the managed OpenCode process exits but a server survives on the old
port (Windows: killProcessOnPort is a no-op, so the orphaned process tree
keeps the port), restartOpenCode() times out waiting for the port and
spawns a fresh server on a NEW port. HTTP/proxy traffic follows the new
port, but the global message-stream hub's upstream SSE reader stays pinned
to the old server's /global/event stream — that connection never closes —
so new events never reach the UI and the chat stops updating until the
app is restarted (#2638).

Lifecycle now fires an optional onOpenCodeRestarted hook after a
successful managed restart; index.js wires it to the new
messageStreamRuntime.rebindUpstream(), which restarts the shared hub
(its reader re-dials buildOpenCodeUrl → the current port) and closes
directory-scoped sockets so their per-connection readers rebuild against
the new port. External servers are untouched (their port cannot change).

Fixes #2638

packages/web/server/index.js

@@ -1079,6 +1079,19 @@ const openCodeLifecycleRuntime = createOpenCodeLifecycleRuntime({
     }
     return [...new Set(directories)];
   },
+  // A managed restart can move OpenCode to a NEW port (the old one may stay
+  // occupied by an orphaned process, e.g. killProcessOnPort is a no-op on
+  // Windows). Rebind the message-stream upstream readers to the current port
+  // so the UI keeps receiving events instead of staying pinned to the old
+  // process (#2638). The runtime is created later by the startup pipeline;
+  // by the time any restart runs, it is assigned.
+  onOpenCodeRestarted: () => {
+    try {
+      messageStreamRuntime?.rebindUpstream();
+    } catch (error) {
+      console.warn('Failed to rebind message stream after OpenCode restart:', error?.message ?? error);
+    }
+  },
   getManagedOpenCodeEnv: async () => {
     const settings = await readSettingsFromDiskMigrated().catch(() => null);
     const managedEnv = settings?.agentControlToolEnabled === false

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

@@ -0,0 +1,186 @@
+import { EventEmitter } from 'node:events';
+import { describe, expect, it, vi } from 'vitest';
+
+import { createGlobalMessageStreamHub } from './global-hub.js';
+import { createMessageStreamWsRuntime } from './runtime.js';
+
+class FakeSocket extends EventEmitter {
+  constructor() {
+    super();
+    this.readyState = 1;
+    this.sent = [];
+    this.closeCalls = [];
+  }
+
+  send(payload) {
+    this.sent.push(JSON.parse(payload));
+  }
+
+  ping() {
+    void 0;
+  }
+
+  close(code, reason) {
+    if (this.readyState === 3) {
+      return;
+    }
+    this.readyState = 3;
+    this.closeCalls.push({ code, reason });
+    this.emit('close');
+  }
+}
+
+function createSseResponse({ blocks = [], signal, holdOpen = false }) {
+  const encoder = new TextEncoder();
+  let index = 0;
+
+  return {
+    ok: true,
+    body: {
+      getReader() {
+        return {
+          async read() {
+            if (index < blocks.length) {
+              const next = blocks[index++];
+              return { value: encoder.encode(next), done: false };
+            }
+
+            if (!holdOpen) {
+              return { value: undefined, done: true };
+            }
+
+            return new Promise((resolve, reject) => {
+              const onAbort = () => {
+                signal.removeEventListener('abort', onAbort);
+                const error = new Error('Aborted');
+                error.name = 'AbortError';
+                reject(error);
+              };
+              signal.addEventListener('abort', onAbort, { once: true });
+            });
+          },
+        };
+      },
+    },
+  };
+}
+
+describe('rebindUpstream (#2638)', () => {
+  it('restarts the shared hub upstream so a connected client resumes receiving events on the new port', async () => {
+    const server = new EventEmitter();
+    const wsClients = new Set();
+    let port = 4096;
+    let fetchCalls = 0;
+
+    // Port changes after a managed restart: buildOpenCodeUrl resolves the
+    // CURRENT port on every attempt, exactly like production network-runtime.
+    const buildOpenCodeUrl = vi.fn(() => `http://127.0.0.1:${port}/global/event`);
+    const fetchImpl = vi.fn(async (_url, options) => {
+      fetchCalls += 1;
+      if (fetchCalls === 1) {
+        return createSseResponse({
+          signal: options.signal,
+          holdOpen: true,
+          blocks: ['id: evt-1\ndata: {"type":"server.connected","properties":{}}\n\n'],
+        });
+      }
+      return createSseResponse({
+        signal: options.signal,
+        holdOpen: true,
+        blocks: ['id: evt-2\ndata: {"type":"session.updated","properties":{"sessionID":"ses_1"}}\n\n'],
+      });
+    });
+
+    const globalHub = createGlobalMessageStreamHub({
+      buildOpenCodeUrl,
+      getOpenCodeAuthHeaders: () => ({}),
+      fetchImpl,
+      upstreamReconnectDelayMs: 0,
+    });
+
+    const runtime = createMessageStreamWsRuntime({
+      server,
+      uiAuthController: null,
+      isRequestOriginAllowed: async () => true,
+      rejectWebSocketUpgrade() {
+        throw new Error('upgrade should not be used in this test');
+      },
+      globalEventHub: globalHub,
+      buildOpenCodeUrl,
+      getOpenCodeAuthHeaders: () => ({}),
+      processForwardedEventPayload() {},
+      wsClients,
+      heartbeatIntervalMs: 5000,
+      upstreamReconnectDelayMs: 0,
+      fetchImpl,
+    });
+
+    const socket = new FakeSocket();
+    runtime.wsServer.emit('connection', socket, { url: '/api/global/event/ws' });
+    await new Promise((resolve) => setTimeout(resolve, 10));
+    expect(fetchCalls).toBe(1);
+    expect(socket.sent.some((frame) => frame.type === 'event' && frame.eventId === 'evt-1')).toBe(true);
+
+    // The managed process was restarted onto a new port while the old
+    // process's SSE stream stays open (orphaned survivor).
+    port = 5000;
+    runtime.rebindUpstream();
+    await new Promise((resolve) => setTimeout(resolve, 20));
+
+    // The hub dialed the new port and the connected client received events
+    // from the new upstream without reconnecting its own socket.
+    expect(fetchCalls).toBe(2);
+    expect(fetchImpl.mock.calls[1][0]).toContain(':5000/global/event');
+    expect(socket.sent.some((frame) => frame.type === 'event' && frame.eventId === 'evt-2')).toBe(true);
+
+    socket.close();
+    await runtime.close();
+  });
+
+  it('closes directory-scoped sockets so their pinned readers reconnect to the new port', async () => {
+    const server = new EventEmitter();
+    const wsClients = new Set();
+    let port = 4096;
+    let fetchCalls = 0;
+
+    const buildOpenCodeUrl = vi.fn(() => `http://127.0.0.1:${port}/event`);
+    const fetchImpl = vi.fn(async (_url, options) => {
+      fetchCalls += 1;
+      return createSseResponse({
+        signal: options.signal,
+        holdOpen: true,
+        blocks: ['id: evt-1\ndata: {"type":"server.connected","properties":{}}\n\n'],
+      });
+    });
+
+    const runtime = createMessageStreamWsRuntime({
+      server,
+      uiAuthController: null,
+      isRequestOriginAllowed: async () => true,
+      rejectWebSocketUpgrade() {
+        throw new Error('upgrade should not be used in this test');
+      },
+      buildOpenCodeUrl,
+      getOpenCodeAuthHeaders: () => ({}),
+      processForwardedEventPayload() {},
+      wsClients,
+      heartbeatIntervalMs: 5000,
+      upstreamReconnectDelayMs: 0,
+      fetchImpl,
+    });
+
+    const directorySocket = new FakeSocket();
+    runtime.wsServer.emit('connection', directorySocket, { url: '/api/event/ws?directory=%2Fproj' });
+    await new Promise((resolve) => setTimeout(resolve, 10));
+    expect(fetchCalls).toBe(1);
+
+    port = 5000;
+    runtime.rebindUpstream();
+
+    expect(directorySocket.readyState).toBe(3);
+    expect(directorySocket.closeCalls.length).toBeGreaterThan(0);
+
+    directorySocket.close();
+    await runtime.close();
+  });
+});

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

@@ -70,6 +70,12 @@ export function createMessageStreamWsRuntime({
     noServer: true,
   });
 
+  // Directory-scoped streams create one upstream reader per client
+  // connection. Track those sockets so a managed OpenCode restart can close
+  // them: each reader is pinned to the port it connected at and would
+  // otherwise keep streaming from an orphaned process on the old port (#2638).
+  const directorySockets = new Set();
+
   const ownsGlobalHub = !globalEventHub;
   const globalHub = globalEventHub ?? createGlobalMessageStreamHub({
     buildOpenCodeUrl,
@@ -103,6 +109,11 @@ export function createMessageStreamWsRuntime({
       return;
     }
 
+    directorySockets.add(socket);
+    socket.on('close', () => {
+      directorySockets.delete(socket);
+    });
+
     acceptDirectoryMessageStreamWsConnection({
       socket,
       requestedLastEventId,
@@ -156,6 +167,27 @@ export function createMessageStreamWsRuntime({
 
   return {
     wsServer,
+    /**
+     * Rebind all upstream readers to the current OpenCode port. Called after
+     * a managed process restart: the restart can land on a NEW port while
+     * the old process (or an orphaned survivor of it) still holds the
+     * previous one, and a healthy-but-pinned SSE connection never notices —
+     * so the UI would stop receiving events until the app restarts (#2638).
+     * Restarting the shared hub re-dials `buildOpenCodeUrl` (which reads the
+     * current port) on its next attempt; directory-scoped readers are
+     * rebuilt by closing their client sockets, which reconnect with
+     * `Last-Event-ID` and re-establish the stream against the new port.
+     */
+    rebindUpstream() {
+      globalHub.stop();
+      globalHub.start();
+      for (const socket of Array.from(directorySockets)) {
+        try {
+          socket.close(1012, 'OpenCode upstream restarted');
+        } catch {
+        }
+      }
+    },
     async close() {
       server.off('upgrade', upgradeHandler);
       globalBridge.close();

packages/web/server/lib/opencode/lifecycle.js

@@ -48,6 +48,7 @@ export const createOpenCodeLifecycleRuntime = (deps) => {
     getActiveSessionCount = () => 0,
     reapManagedOrphanedProcesses = reapOrphanedProcesses,
     getWarmupDirectories = async () => [],
+    onOpenCodeRestarted = null,
     now = Date.now,
   } = deps;
 
@@ -694,6 +695,17 @@ export const createOpenCodeLifecycleRuntime = (deps) => {
         setupProxy(state.expressApp);
         ensureOpenCodeApiPrefix();
       }
+
+      // The restart may have landed on a NEW port (the old one can remain
+      // occupied by an orphaned process, e.g. Windows killProcessOnPort is a
+      // no-op). Upstream event readers pinned to the old process would keep
+      // the UI silent forever, so rebind them to the current port. Best
+      // effort: a failure here must not fail the restart itself.
+      try {
+        onOpenCodeRestarted?.();
+      } catch (error) {
+        console.warn('Failed to rebind event stream after OpenCode restart:', error?.message ?? error);
+      }
     })();
 
     try {

packages/web/server/lib/opencode/lifecycle.test.js

@@ -286,6 +286,70 @@ describe('OpenCode lifecycle', () => {
     expect(spawnMock).toHaveBeenCalledTimes(1);
   });
 
+  it('calls onOpenCodeRestarted after a successful managed restart', async () => {
+    const close = vi.fn(async () => {});
+    const replacement = createMockChild();
+    const onOpenCodeRestarted = vi.fn();
+    globalThis.fetch = vi.fn(async () => ({
+      ok: false,
+      json: async () => null,
+    }));
+    spawnMock.mockImplementationOnce(() => {
+      queueMicrotask(() => {
+        replacement.stdout.emit('data', 'opencode server listening on http://127.0.0.1:45678\n');
+      });
+      return replacement;
+    });
+    const runtime = createRuntime({ onOpenCodeRestarted }, {
+      openCodePort: 45678,
+      openCodeProcess: {
+        pid: null,
+        exitCode: 1,
+        signalCode: null,
+        close,
+      },
+    });
+
+    await runtime.triggerHealthCheck();
+
+    expect(close).toHaveBeenCalledTimes(1);
+    expect(spawnMock).toHaveBeenCalledTimes(1);
+    // The restart completed on a (possibly new) port — the event-stream
+    // upstreams must rebind so the UI keeps receiving events (#2638).
+    expect(onOpenCodeRestarted).toHaveBeenCalledTimes(1);
+  });
+
+  it('does not call onOpenCodeRestarted when a managed restart fails', async () => {
+    const close = vi.fn(async () => {});
+    const onOpenCodeRestarted = vi.fn();
+    globalThis.fetch = vi.fn(async () => ({
+      ok: false,
+      json: async () => null,
+    }));
+    spawnMock.mockImplementation(() => {
+      const child = createMockChild();
+      queueMicrotask(() => {
+        child.emit('error', new Error('spawn failed'));
+      });
+      return child;
+    });
+    const runtime = createRuntime({ onOpenCodeRestarted }, {
+      openCodePort: 45678,
+      openCodeProcess: {
+        pid: null,
+        exitCode: 1,
+        signalCode: null,
+        close,
+      },
+    });
+
+    // triggerHealthCheck logs instead of rethrowing; call restartOpenCode
+    // directly to observe the failure result.
+    await expect(runtime.restartOpenCode()).rejects.toThrow();
+
+    expect(onOpenCodeRestarted).not.toHaveBeenCalled();
+  });
+
   it('launches managed OpenCode with the managed PATH', async () => {
     delete process.env.OPENCODE_BINARY;
     const child = createMockChild();