fix(api): enforce global ollama endpoint order

apps/api/tests/test_local_code_review_cloud_fallback.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+from types import SimpleNamespace
 from typing import Any
 
 import httpx
@@ -32,6 +33,14 @@ async def _noop_save(*args: Any, **kwargs: Any) -> None:
     return None
 
 
+def _fake_ollama_order(_workload_type: str) -> tuple[SimpleNamespace, ...]:
+    return (
+        SimpleNamespace(url="http://gcp-a:11434", provider_name="ollama_gcp_a"),
+        SimpleNamespace(url="http://gcp-b:11434", provider_name="ollama_gcp_b"),
+        SimpleNamespace(url="http://local-111:11434", provider_name="ollama_local"),
+    )
+
+
 @pytest.mark.asyncio
 async def test_large_pr_uses_local_ollama_when_gemini_fallback_disabled(
     monkeypatch: pytest.MonkeyPatch,
@@ -43,8 +52,8 @@ async def test_large_pr_uses_local_ollama_when_gemini_fallback_disabled(
     )
     monkeypatch.setattr(
         review_module,
-        "resolve_ollama_endpoint",
-        lambda workload_type: "http://local-111:11434",
+        "resolve_ollama_order",
+        _fake_ollama_order,
     )
 
     client = _FakeClient()
@@ -68,8 +77,8 @@ async def test_large_pr_uses_local_ollama_when_gemini_fallback_disabled(
     )
 
     assert result is not None
-    assert result["provider"] == "ollama"
-    assert client.posted_urls == ["http://local-111:11434/api/generate"]
+    assert result["provider"] == "ollama_gcp_a"
+    assert client.posted_urls == ["http://gcp-a:11434/api/generate"]
 
 
 @pytest.mark.asyncio
@@ -83,8 +92,8 @@ async def test_ollama_failure_does_not_fall_back_to_gemini_by_default(
     )
     monkeypatch.setattr(
         review_module,
-        "resolve_ollama_endpoint",
-        lambda workload_type: "http://local-111:11434",
+        "resolve_ollama_order",
+        _fake_ollama_order,
     )
 
     client = _FakeClient(fail=True)
@@ -110,7 +119,11 @@ async def test_ollama_failure_does_not_fall_back_to_gemini_by_default(
     assert result is not None
     assert result["provider"] == "ollama_unavailable"
     assert result["cloud_fallback_skipped"] is True
-    assert client.posted_urls == ["http://local-111:11434/api/generate"]
+    assert client.posted_urls == [
+        "http://gcp-a:11434/api/generate",
+        "http://gcp-b:11434/api/generate",
+        "http://local-111:11434/api/generate",
+    ]
 
 
 @pytest.mark.asyncio