From 3b762ba6957c73a6a20216758e553face5cc92bd Mon Sep 17 00:00:00 2001
From: teknium1 <127238744+teknium1@users.noreply.github.com>
Date: Fri, 24 Jul 2026 12:06:11 -0700
Subject: [PATCH] fix: serialize on-demand slash-worker spawn per session
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

With the eager pre-warm removed (PR #66783), slash.exec is the only spawn
path — and it runs on the RPC thread pool, so two concurrent worker-routed
commands on a fresh session could both see slash_worker=None and each fork
a full stdio-MCP-fleet worker (the _attach_worker race loser leaking
unclosed). Add a per-session spawn lock with a double-check, plus a
regression test racing two slash.exec calls through handle_request.

Also maps Ne0teric's contributor email.
---
 contributors/emails/mpetrelli@gmail.com |  1 +
 tests/test_tui_gateway_server.py        | 54 +++++++++++++++++++++++++
 tui_gateway/server.py                   | 29 ++++++++-----
 3 files changed, 75 insertions(+), 9 deletions(-)
 create mode 100644 contributors/emails/mpetrelli@gmail.com

diff --git a/contributors/emails/mpetrelli@gmail.com b/contributors/emails/mpetrelli@gmail.com
new file mode 100644
index 000000000000..0a943bf232e4
--- /dev/null
+++ b/contributors/emails/mpetrelli@gmail.com
@@ -0,0 +1 @@
+Ne0teric
diff --git a/tests/test_tui_gateway_server.py b/tests/test_tui_gateway_server.py
index d0df4b027712..5aa76b437315 100644
--- a/tests/test_tui_gateway_server.py
+++ b/tests/test_tui_gateway_server.py
@@ -12045,6 +12045,60 @@ def test_restart_slash_worker_noop_without_worker(monkeypatch):
         server._sessions.pop("lazy-noop", None)
 
 
+def test_slash_exec_concurrent_first_use_spawns_single_worker(monkeypatch):
+    """With eager pre-warm removed, slash.exec is the only spawn path — two
+    concurrent worker-routed commands on a fresh session must not each fork a
+    full MCP-fleet worker. The per-session spawn lock serializes first use."""
+    import time as _time
+
+    spawned = []
+    barrier = threading.Barrier(2, timeout=5)
+
+    class _SlowWorker:
+        def __init__(self, *a, **k):
+            spawned.append(self)
+            _time.sleep(0.05)  # widen the None-observation window
+
+        def run(self, cmd):
+            return f"ran {cmd}"
+
+        def close(self):
+            pass
+
+    monkeypatch.setattr(server, "_SlashWorker", _SlowWorker)
+    monkeypatch.setattr(server, "_mirror_slash_side_effects", lambda *a, **k: None)
+    session = _session(slash_worker=None)
+    server._sessions["race-spawn"] = session
+
+    results = []
+
+    def _exec(n):
+        barrier.wait()
+        resp = server.handle_request(
+            {
+                "id": str(n),
+                "method": "slash.exec",
+                "params": {"command": "/context", "session_id": "race-spawn"},
+            }
+        )
+        results.append(resp)
+
+    try:
+        threads = [threading.Thread(target=_exec, args=(i,)) for i in range(2)]
+        for t in threads:
+            t.start()
+        for t in threads:
+            t.join(timeout=10)
+        assert len(spawned) == 1, (
+            f"concurrent slash.exec spawned {len(spawned)} workers — first-use "
+            f"spawn must be serialized per session"
+        )
+        assert session["slash_worker"] is spawned[0]
+        assert all("result" in r for r in results), results
+    finally:
+        server._sessions.pop("race-spawn", None)
+
+
 def test_session_close_rpc_claims_then_tears_down(monkeypatch):
     seen = []
     claimed = {"session_key": "k"}
diff --git a/tui_gateway/server.py b/tui_gateway/server.py
index d3b6b152c8b0..0c48059bf241 100644
--- a/tui_gateway/server.py
+++ b/tui_gateway/server.py
@@ -16064,15 +16064,26 @@ def _(rid, params: dict) -> dict:
 
     worker = session.get("slash_worker")
     if not worker:
-        try:
-            worker = _SlashWorker(
-                session["session_key"],
-                getattr(session.get("agent"), "model", _resolve_model()),
-                profile_home=session.get("profile_home"),
-            )
-            _attach_worker(params.get("session_id", ""), session, worker)
-        except Exception as e:
-            return _err(rid, 5030, f"slash worker start failed: {e}")
+        # On-demand spawn is now the ONLY spawn path for a fresh session
+        # (eager pre-warm removed), and slash.exec handlers run on the RPC
+        # thread pool — two concurrent slash commands on the same session
+        # could both observe slash_worker=None and each fork a full
+        # MCP-fleet worker (the loser of the _attach_worker race would leak
+        # unclosed). Serialize first-use spawn per session.
+        with _sessions_lock:
+            spawn_lock = session.setdefault("_slash_spawn_lock", threading.Lock())
+        with spawn_lock:
+            worker = session.get("slash_worker")
+            if not worker:
+                try:
+                    worker = _SlashWorker(
+                        session["session_key"],
+                        getattr(session.get("agent"), "model", _resolve_model()),
+                        profile_home=session.get("profile_home"),
+                    )
+                    _attach_worker(params.get("session_id", ""), session, worker)
+                except Exception as e:
+                    return _err(rid, 5030, f"slash worker start failed: {e}")
 
     try:
         output = worker.run(cmd)