穩定 Ollama embedding GCP 失敗熔斷

tests/test_ollama_resolve.py

@@ -200,6 +200,21 @@ def test_get_ollama_host_falls_back_to_resolve_without_env(monkeypatch):
     assert host.startswith('http://')
 
 
+def test_config_ollama_compat_constants_do_not_probe_network(monkeypatch):
+    monkeypatch.setenv('OLLAMA_HOST', 'http://192.168.0.188:11434')
+    monkeypatch.setenv('HERMES_URL', 'http://192.168.0.188:11434')
+    monkeypatch.setenv('EMBEDDING_HOST', 'http://192.168.0.188:11434')
+    monkeypatch.setenv('OLLAMA_HOST_PRIMARY', 'http://34.143.170.20:11434')
+    with patch('services.ollama_service.requests.get') as mock_get:
+        import config
+        importlib.reload(config)
+
+    mock_get.assert_not_called()
+    assert config.OLLAMA_HOST == 'http://34.143.170.20:11434'
+    assert config.HERMES_URL == 'http://34.143.170.20:11434'
+    assert config.EMBEDDING_HOST == 'http://34.143.170.20:11434'
+
+
 def test_get_embedding_host_prefers_env(monkeypatch):
     monkeypatch.setenv('EMBEDDING_HOST', 'http://192.168.0.111:11434')
     import config

tests/test_ollama_retry_chain.py

@@ -31,11 +31,13 @@ def _reset_state():
     oss._resolved_host_cache['host'] = None
     oss._resolved_host_cache['ts'] = 0
     oss._fallback_111_circuit_cache.update({'blocked': False, 'reason': '', 'ts': 0})
+    oss._embedding_gcp_failure_circuit.update({'blocked_until': 0.0, 'notice_ts': 0.0, 'tried': ()})
     yield
     oss._unhealthy_marks.clear()
     oss._resolved_host_cache['host'] = None
     oss._resolved_host_cache['ts'] = 0
     oss._fallback_111_circuit_cache.update({'blocked': False, 'reason': '', 'ts': 0})
+    oss._embedding_gcp_failure_circuit.update({'blocked_until': 0.0, 'notice_ts': 0.0, 'tried': ()})
 
 
 # ═══════════════════════════════════════════════════════════════════════════
@@ -407,6 +409,37 @@ def test_embedding_fallback_disabled_uses_gcp_chain_when_resolver_returns_111():
     assert oss.OLLAMA_HOST_FALLBACK not in posted_hosts
 
 
+def test_embedding_fallback_disabled_opens_short_gcp_failure_circuit():
+    """GCP-A/GCP-B 全掛時，背景 embedding 短暫熔斷，避免下一筆立刻重打兩台。"""
+    import requests
+    from services import ollama_service as oss
+    from services.ollama_service import OllamaService
+
+    svc = OllamaService()
+
+    with patch('services.ollama_service.EMBED_GCP_FAILURE_COOLDOWN_SEC', 60), \
+         patch('services.ollama_service.resolve_ollama_host', side_effect=[
+             oss.OLLAMA_HOST_PRIMARY,
+             oss.OLLAMA_HOST_SECONDARY,
+             oss.OLLAMA_HOST_PRIMARY,
+         ]), \
+         patch.dict('os.environ', {}, clear=False), \
+         patch(
+             'services.ollama_service.requests.post',
+             side_effect=requests.Timeout('gcp timeout'),
+         ) as mock_post:
+        import os
+        os.environ.pop('EMBEDDING_HOST', None)
+        first = svc.generate_embedding('test text', allow_111_fallback=False)
+        second = svc.generate_embedding('another text', allow_111_fallback=False)
+
+    posted_hosts = [call.args[0].split('/api/embed')[0] for call in mock_post.call_args_list]
+    assert first == []
+    assert second == []
+    assert posted_hosts == [oss.OLLAMA_HOST_PRIMARY, oss.OLLAMA_HOST_SECONDARY]
+    assert oss._embedding_gcp_failure_circuit['blocked_until'] > 0
+
+
 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