fix: request no reasoning for liveness and handle empty text

app/main.py

@@ -638,6 +638,7 @@ def api_check(model_id: str):
             "model": model_id,
             "messages": [{"role": "user", "content": "What is 2+2? Reply with just the number."}],
             "max_tokens": 5,
+            "reasoning_effort": "none",
         }).encode()
         req = urllib.request.Request(
             f"{OMNIROUTE_URL}/v1/chat/completions",
@@ -652,8 +653,11 @@ def api_check(model_id: str):
             # Usage фиксируем ДО валидации контента: даже если парсинг choices
             # упадёт, реально потраченные токены сохраняются в checks.
             usage = extract_usage(resp)
-            content = resp["choices"][0]["message"]["content"]
-            ok = bool(content.strip())
+            choice = resp["choices"][0]
+            content = choice["message"].get("content")
+            ok = isinstance(content, str) and bool(content.strip())
+            if not ok:
+                error = f"Empty text response (finish_reason={choice.get('finish_reason')})"
     except Exception as e:
         error = f"{type(e).__name__}: {e}"
         print(f"[check] {model_id}: {error}")

app/probe.py

@@ -37,6 +37,7 @@ def check_model(model_id: str) -> tuple[bool, int | None, str | None, dict]:
             "model": model_id,
             "messages": [{"role": "user", "content": "2+2=?"}],
             "max_tokens": 5,
+            "reasoning_effort": "none",
         }).encode()
         req = urllib.request.Request(
             f"{OMNIROUTE_URL}/v1/chat/completions",
@@ -50,8 +51,13 @@ def check_model(model_id: str) -> tuple[bool, int | None, str | None, dict]:
             resp = json.loads(r.read())
             # Usage — до любой валидации контента.
             usage = extract_usage(resp)
-            content = resp["choices"][0]["message"]["content"]
-            ok = bool(content.strip())
+            # HTTP 200 может содержать только reasoning и content=null.
+            # Это не успешный текстовый ответ и не ошибка Python.
+            choice = resp["choices"][0]
+            content = choice["message"].get("content")
+            ok = isinstance(content, str) and bool(content.strip())
+            if not ok:
+                error = f"Empty text response (finish_reason={choice.get('finish_reason')})"
     except Exception as e:
         error = f"{type(e).__name__}: {e}"
         print(f"  ✗ {model_id}: {error}", file=sys.stderr)

tests/conftest.py

@@ -81,9 +81,10 @@ class OmniMock:
                 if not beh.get("no_usage"):
                     resp["usage"] = beh.get("usage")
                 if not beh.get("omit_choices"):
-                    resp["choices"] = [{"index": 0, "finish_reason": "stop",
+                    resp["choices"] = [{"index": 0, "finish_reason": beh.get("finish_reason", "stop"),
                                         "message": {"role": "assistant",
-                                                    "content": beh.get("content", "4")}}]
+                                                    "content": beh.get("content", "4"),
+                                                    "reasoning_content": beh.get("reasoning_content")}}]
                 self._send_json(200, resp)
 
             def do_GET(self):

tests/test_reasoning_probe.py

@@ -0,0 +1,34 @@
+"""Оба пути пробы запрашивают none и не падают на reasoning-only ответе."""
+import json
+import sqlite3
+
+
+def test_cron_disables_reasoning_and_reports_null_content(core, omni_server):
+    omni_server.set_completions(content=None, reasoning_content="thinking", finish_reason="length",
+                                usage={"prompt_tokens": 12, "completion_tokens": 5, "total_tokens": 17})
+    ok, latency, error, usage = core.probe.check_model("zai/test-model-a")
+    assert ok is False
+    assert error == "Empty text response (finish_reason=length)"
+    assert usage["total_tokens"] == 17
+    request = omni_server.requests[-1]["body"]
+    assert request["reasoning_effort"] == "none"
+    assert request["max_tokens"] == 5
+
+
+def test_manual_disables_reasoning_and_persists_empty_text(core, omni_server):
+    omni_server.set_completions(content=None, reasoning_content="thinking", finish_reason="length",
+                                usage={"prompt_tokens": 12, "completion_tokens": 5, "total_tokens": 17})
+    response = core.client.get("/api/check/zai/test-model-a")
+    assert response.status_code == 200
+    assert response.json()["ok"] is False
+    assert omni_server.requests[-1]["body"]["reasoning_effort"] == "none"
+    db = sqlite3.connect(core.db_path)
+    row = db.execute("SELECT ok,error,total_tokens FROM checks ORDER BY id DESC LIMIT 1").fetchone()
+    db.close()
+    assert row == (0, "Empty text response (finish_reason=length)", 17)
+
+
+def test_text_response_still_succeeds(core, omni_server):
+    omni_server.set_completions(content=" 4 ")
+    assert core.probe.check_model("zai/test-model-a")[0] is True
+    assert core.client.get("/api/check/zai/test-model-a").json()["ok"] is True