fix: retry transient model failures

tests/test_provider.py

@@ -0,0 +1,80 @@
+from __future__ import annotations
+
+from collections.abc import Awaitable, Callable
+
+import httpx
+import pytest
+
+from agent_platform.config import Settings
+from agent_platform.runtime.provider import ModelProvider
+
+
+@pytest.fixture(autouse=True)
+def disable_retry_sleep(monkeypatch: pytest.MonkeyPatch) -> None:
+    async def no_sleep(_: float) -> None:
+        return None
+
+    monkeypatch.setattr("agent_platform.runtime.provider.asyncio.sleep", no_sleep)
+
+
+async def _complete_with_handler(
+    settings: Settings,
+    handler: Callable[[httpx.Request], httpx.Response | Awaitable[httpx.Response]],
+) -> dict:
+    async with httpx.AsyncClient(transport=httpx.MockTransport(handler)) as client:
+        provider = ModelProvider(settings, client)
+        return await provider.complete(model="test-model", messages=[{"role": "user", "content": "hello"}])
+
+
+async def test_complete_retries_transient_gateway_response(settings: Settings) -> None:
+    calls = 0
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        nonlocal calls
+        calls += 1
+        if calls == 1:
+            return httpx.Response(524, request=request, headers={"Retry-After": "0"})
+        return httpx.Response(
+            200,
+            request=request,
+            json={"choices": [{"message": {"role": "assistant", "content": "ok"}}]},
+        )
+
+    result = await _complete_with_handler(settings, handler)
+
+    assert calls == 2
+    assert result["choices"][0]["message"]["content"] == "ok"
+
+
+async def test_complete_retries_transport_error(settings: Settings) -> None:
+    calls = 0
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        nonlocal calls
+        calls += 1
+        if calls == 1:
+            raise httpx.ConnectError("temporary disconnect", request=request)
+        return httpx.Response(
+            200,
+            request=request,
+            json={"choices": [{"message": {"role": "assistant", "content": "recovered"}}]},
+        )
+
+    result = await _complete_with_handler(settings, handler)
+
+    assert calls == 2
+    assert result["choices"][0]["message"]["content"] == "recovered"
+
+
+async def test_complete_does_not_retry_client_error(settings: Settings) -> None:
+    calls = 0
+
+    def handler(request: httpx.Request) -> httpx.Response:
+        nonlocal calls
+        calls += 1
+        return httpx.Response(400, request=request, json={"error": "bad request"})
+
+    with pytest.raises(httpx.HTTPStatusError):
+        await _complete_with_handler(settings, handler)
+
+    assert calls == 1