V10.417 protect embedding fallback routing

tests/test_ai_insight_embedding_bridge.py

@@ -82,7 +82,7 @@ def test_process_one_embedding_writes_signature(monkeypatch):
     monkeypatch.setattr(
         learning.ollama_service,
         "generate_embedding",
-        lambda text, model="bge-m3:latest": [0.1] * 1024,
+        lambda text, model="bge-m3:latest", **_kwargs: [0.1] * 1024,
     )
 
     ok = learning._process_one_embedding(

tests/test_ollama_embedding.py

@@ -24,7 +24,11 @@ def test_generate_embedding_uses_current_embed_endpoint(monkeypatch):
 
     assert vec == [0.1, 0.2, 0.3]
     assert calls == [
-        ("http://ollama/api/embed", {"model": "bge-m3:latest", "input": "hello"}, 7),
+        (
+            "http://ollama/api/embed",
+            {"model": "bge-m3:latest", "input": "hello", "keep_alive": "1m"},
+            7,
+        ),
     ]
 
 
@@ -43,10 +47,42 @@ def test_generate_embedding_falls_back_to_legacy_embeddings_endpoint(monkeypatch
 
     assert vec == [0.4, 0.5]
     assert calls == [
-        ("http://ollama/api/embed", {"model": "bge-m3:latest", "input": "hello"}, 9),
+        (
+            "http://ollama/api/embed",
+            {"model": "bge-m3:latest", "input": "hello", "keep_alive": "1m"},
+            9,
+        ),
         ("http://ollama/api/embeddings", {"model": "bge-m3:latest", "prompt": "hello"}, 9),
     ]
 
 
 def test_extract_embedding_accepts_flat_embeddings_shape():
     assert OllamaService._extract_embedding({"embeddings": [0.1, 0.2]}) == [0.1, 0.2]
+
+
+def test_generate_embedding_caps_timeout_and_clips_input(monkeypatch):
+    calls = []
+
+    def fake_post(url, json, timeout):
+        calls.append((url, json, timeout))
+        return FakeResponse(200, {"embeddings": [[0.1, 0.2, 0.3]]})
+
+    monkeypatch.setattr("services.ollama_service.EMBED_MAX_TIMEOUT", 3)
+    monkeypatch.setattr("services.ollama_service.EMBED_MAX_CHARS", 5)
+    monkeypatch.setattr("services.ollama_service.requests.post", fake_post)
+
+    vec = OllamaService().generate_embedding(
+        "hello world",
+        model="bge-m3:latest",
+        host="http://ollama",
+        timeout=45,
+    )
+
+    assert vec == [0.1, 0.2, 0.3]
+    assert calls == [
+        (
+            "http://ollama/api/embed",
+            {"model": "bge-m3:latest", "input": "hello", "keep_alive": "1m"},
+            3,
+        ),
+    ]

tests/test_ollama_retry_chain.py

@@ -279,6 +279,51 @@ def test_embedding_all_three_hosts_fail_returns_empty():
     assert mock_post.call_count == 3
 
 
+def test_embedding_can_disable_111_fallback_for_background_rag_work():
+    """背景 embedding/RAG 任務只跑 GCP-A/GCP-B，避免 111 承接 bge-m3 長任務。"""
+    import requests
+    from services import ollama_service as oss
+    from services.ollama_service import OllamaService
+
+    svc = OllamaService()
+    hosts = [
+        oss.OLLAMA_HOST_SECONDARY,
+        oss.OLLAMA_HOST_FALLBACK,
+    ]
+
+    with patch('services.ollama_service.resolve_ollama_host', side_effect=hosts), \
+         patch.dict('os.environ', {}, clear=False), \
+         patch('services.ollama_service.requests.post',
+               side_effect=requests.Timeout('secondary timeout')) as mock_post:
+        import os
+        os.environ.pop('EMBEDDING_HOST', None)
+        vec = svc.generate_embedding('test text', allow_111_fallback=False)
+
+    posted_hosts = [call.args[0].split('/api/embed')[0] for call in mock_post.call_args_list]
+    assert vec == []
+    assert posted_hosts == [oss.OLLAMA_HOST_SECONDARY]
+    assert oss.OLLAMA_HOST_FALLBACK not in posted_hosts
+
+
+def test_embedding_ignores_111_embedding_host_when_fallback_disabled():
+    """EMBEDDING_HOST 若誤設 111，背景 embedding 仍回 GCP resolver，不直接棄跑。"""
+    from services import ollama_service as oss
+    from services.ollama_service import OllamaService
+
+    svc = OllamaService()
+    fake_ok = MagicMock(status_code=200)
+    fake_ok.json.return_value = {'embeddings': [[0.7, 0.8]]}
+
+    with patch('services.ollama_service.resolve_ollama_host', return_value=oss.OLLAMA_HOST_SECONDARY), \
+         patch.dict('os.environ', {'EMBEDDING_HOST': oss.OLLAMA_HOST_FALLBACK}, clear=False), \
+         patch('services.ollama_service.requests.post', return_value=fake_ok) as mock_post:
+        vec = svc.generate_embedding('test text', allow_111_fallback=False)
+
+    posted_hosts = [call.args[0].split('/api/embed')[0] for call in mock_post.call_args_list]
+    assert vec == [0.7, 0.8]
+    assert posted_hosts == [oss.OLLAMA_HOST_SECONDARY]
+
+
 # ═══════════════════════════════════════════════════════════════════════════
 # T4: mark_unhealthy 觸發 cache 失效（驗 self.host 取新主機）
 # ═══════════════════════════════════════════════════════════════════════════

tests/test_rag_service.py

@@ -99,7 +99,7 @@ class TestRagEnabledHits:
 
         monkeypatch.setattr(
             'services.ollama_service.ollama_service.generate_embedding',
-            lambda text, model="bge-m3:latest": _fake_embedding(),
+            lambda text, model="bge-m3:latest", **_kwargs: _fake_embedding(),
         )
 
         sig = rs.get_embedding_signature()
@@ -132,7 +132,7 @@ class TestRagEnabledHits:
 
         monkeypatch.setattr(
             'services.ollama_service.ollama_service.generate_embedding',
-            lambda text, model="bge-m3:latest": _fake_embedding(),
+            lambda text, model="bge-m3:latest", **_kwargs: _fake_embedding(),
         )
         monkeypatch.setattr(
             rs.rag_service,
@@ -170,7 +170,7 @@ class TestRagEnabledHits:
 
         monkeypatch.setattr(
             'services.ollama_service.ollama_service.generate_embedding',
-            lambda text, model="bge-m3:latest": _fake_embedding(),
+            lambda text, model="bge-m3:latest", **_kwargs: _fake_embedding(),
         )
         fake_session = MagicMock()
         fake_session.execute.return_value.fetchall.return_value = []
@@ -227,7 +227,7 @@ class TestEmbeddingSignature:
 
         monkeypatch.setattr(
             'services.ollama_service.ollama_service.generate_embedding',
-            lambda text, model="bge-m3:latest": _fake_embedding(),
+            lambda text, model="bge-m3:latest", **_kwargs: _fake_embedding(),
         )
         sig = rs.get_embedding_signature()
         fake_rows = [
@@ -261,7 +261,7 @@ class TestEmbeddingSignature:
 
         monkeypatch.setattr(
             'services.ollama_service.ollama_service.generate_embedding',
-            lambda text, model="bge-m3:latest": _fake_embedding(),
+            lambda text, model="bge-m3:latest", **_kwargs: _fake_embedding(),
         )
         fake_rows = [
             _make_row_obj(
@@ -288,7 +288,7 @@ class TestFireAndForgetLog:
 
         monkeypatch.setattr(
             'services.ollama_service.ollama_service.generate_embedding',
-            lambda text, model="bge-m3:latest": _fake_embedding(),
+            lambda text, model="bge-m3:latest", **_kwargs: _fake_embedding(),
         )
         # SELECT session 正常；INSERT session 故意 raise
         select_session = MagicMock()
@@ -344,7 +344,7 @@ class TestFireAndForgetLog:
 
         monkeypatch.setattr(
             'services.ollama_service.ollama_service.generate_embedding',
-            lambda text, model="bge-m3:latest": [],
+            lambda text, model="bge-m3:latest", **_kwargs: [],
         )
         # DB 不應被呼叫
         called = {'count': 0}
@@ -466,7 +466,7 @@ class TestParamGuards:
 
         monkeypatch.setattr(
             'services.ollama_service.ollama_service.generate_embedding',
-            lambda text, model="bge-m3:latest": _fake_embedding(),
+            lambda text, model="bge-m3:latest", **_kwargs: _fake_embedding(),
         )
         fake_session = MagicMock()
         fake_session.execute.return_value.fetchall.return_value = []
@@ -487,7 +487,7 @@ class TestParamGuards:
 
         monkeypatch.setattr(
             'services.ollama_service.ollama_service.generate_embedding',
-            lambda text, model="bge-m3:latest": _fake_embedding(),
+            lambda text, model="bge-m3:latest", **_kwargs: _fake_embedding(),
         )
         fake_session = MagicMock()
         captured = {}