From fcc2f5c2142beca19fc8cb90652551b31d9a4b42 Mon Sep 17 00:00:00 2001
From: Pigbibi <20649888+Pigbibi@users.noreply.github.com>
Date: Mon, 8 Jun 2026 23:20:10 +0800
Subject: [PATCH] Fix probe timeout alert handling

---
 main.py                        | 36 +++++++++++++++++++++++
 notifications/telegram.py      |  2 ++
 tests/test_request_handling.py | 52 ++++++++++++++++++++++++++++++++++
 3 files changed, 90 insertions(+)

diff --git a/main.py b/main.py
index 44a4e98..be18b6f 100644
--- a/main.py
+++ b/main.py
@@ -618,6 +618,10 @@ def connect_ib():
     return build_broker_adapters().connect_ib()
 
 
+def _build_health_probe_connection_error_message(exc: Exception) -> str:
+    return f"{t('health_probe_title')}\n{t('ibkr_connection_error_prefix')}{str(exc)}"
+
+
 def log_runtime_event(log_context, event, **fields):
     return build_composer().build_reporting_adapters().log_event(log_context, event, **fields)
 
@@ -1158,6 +1162,38 @@ def _handle_probe(*, response_body: str = "Probe OK"):
             positions_count=len(positions),
         )
         return response_body, 200
+    except (ConnectionError, TimeoutError) as exc:
+        if report is not None:
+            append_runtime_report_error(
+                report,
+                stage="health_probe",
+                message=str(exc),
+                error_type=type(exc).__name__,
+                failure_category="ibkr_connection",
+            )
+            finalize_runtime_report(
+                report,
+                status="error",
+                diagnostics={"probe_failure_category": "ibkr_connection"},
+            )
+        if log_context is not None:
+            log_runtime_event(
+                log_context,
+                "health_probe_failed",
+                message="Health probe IBKR connection failed",
+                severity="ERROR",
+                execution_window="probe",
+                error_type=type(exc).__name__,
+                error_message=str(exc),
+                failure_category="ibkr_connection",
+            )
+        error_msg = _build_health_probe_connection_error_message(exc)
+        _publish_runtime_failure_notification(
+            detailed_text=error_msg,
+            compact_text=error_msg,
+            exc=exc,
+        )
+        return "Error", 500
     except Exception as exc:
         if report is not None:
             append_runtime_report_error(
diff --git a/notifications/telegram.py b/notifications/telegram.py
index 24ec785..a43b609 100644
--- a/notifications/telegram.py
+++ b/notifications/telegram.py
@@ -10,6 +10,7 @@
         "error_title": "🚨 【策略异常】",
         "health_probe_title": "🔎 【连接探针】",
         "health_probe_error_prefix": "健康探针异常:\n",
+        "ibkr_connection_error_prefix": "IBKR 连接异常:\n",
         "canary_title": "🐤 【金丝雀检查】",
         "strategy_label": "🧭 策略: {name}",
         "account_ids_detail": "🆔 账户: {account_ids}",
@@ -151,6 +152,7 @@
         "error_title": "🚨 【Strategy Error】",
         "health_probe_title": "🔎 【Health Probe】",
         "health_probe_error_prefix": "Health probe error:\n",
+        "ibkr_connection_error_prefix": "IBKR connection error:\n",
         "canary_title": "🐤 【Canary Check】",
         "strategy_label": "🧭 Strategy: {name}",
         "account_ids_detail": "🆔 Account: {account_ids}",
diff --git a/tests/test_request_handling.py b/tests/test_request_handling.py
index 5f8767b..3b90ace 100644
--- a/tests/test_request_handling.py
+++ b/tests/test_request_handling.py
@@ -210,6 +210,58 @@ def disconnect(self):
     assert observed["notifications"] == []
 
 
+def test_handle_probe_connect_timeout_sends_concise_connection_notification(strategy_module, monkeypatch):
+    observed = {"events": [], "notifications": []}
+    timeout_message = (
+        "IBKR API handshake timed out after TCP preflight succeeded "
+        "for 172.16.159.2:4001 clientId=231"
+    )
+
+    monkeypatch.setattr(strategy_module, "build_request_log_context", lambda: types.SimpleNamespace(run_id="run-001"))
+    monkeypatch.setattr(strategy_module, "build_execution_report", lambda log_context, **_kwargs: {"status": "pending"})
+    monkeypatch.setattr(
+        strategy_module,
+        "persist_execution_report",
+        lambda report, **_kwargs: observed.setdefault("report", dict(report)) or "/tmp/runtime-report.json",
+    )
+    monkeypatch.setattr(
+        strategy_module,
+        "log_runtime_event",
+        lambda context, event, **fields: observed["events"].append((event, fields)),
+    )
+    monkeypatch.setattr(strategy_module, "load_strategy_plugin_signals", lambda: ((), None))
+    monkeypatch.setattr(strategy_module, "attach_strategy_plugin_report", lambda *args, **kwargs: None)
+    monkeypatch.setattr(
+        strategy_module,
+        "connect_ib",
+        lambda: (_ for _ in ()).throw(TimeoutError(timeout_message)),
+    )
+    monkeypatch.setattr(
+        strategy_module,
+        "publish_notification",
+        lambda **kwargs: observed["notifications"].append(kwargs),
+    )
+
+    with strategy_module.app.test_request_context("/probe", method="POST"):
+        body, status = strategy_module.handle_probe()
+
+    assert status == 500
+    assert body == "Error"
+    assert observed["report"]["status"] == "error"
+    assert observed["report"]["errors"][0]["stage"] == "health_probe"
+    assert observed["report"]["errors"][0]["failure_category"] == "ibkr_connection"
+    assert [event for event, _fields in observed["events"]] == [
+        "health_probe_received",
+        "health_probe_failed",
+    ]
+    assert observed["events"][1][1]["failure_category"] == "ibkr_connection"
+    assert len(observed["notifications"]) == 1
+    detailed_text = observed["notifications"][0]["detailed_text"]
+    assert "IBKR" in detailed_text
+    assert timeout_message in detailed_text
+    assert "Traceback" not in detailed_text
+
+
 def test_handle_probe_failure_sends_notification(strategy_module, monkeypatch):
     observed = {"events": [], "notifications": []}