test: guard reasoning-free telemetry through real client paths

2026-09-09 01:34:36 -04:00
parent d0078c1be5
commit 47488ee4fd
4 changed files with 233 additions and 0 deletions
@@ -2839,3 +2839,90 @@ class TestPostgresFailureClassification:

        assert [s for s in conn.statements if s.startswith("INSERT INTO llm_calls")]
        assert recorder.telemetry_status.degraded is False
+
+
+class TestConcurrentReasoningPathContracts:
+    """真实治理链路并发时，按逻辑调用归组且 attempts 不串档。"""
+
+    async def test_chat_and_reasonless_clients_keep_distinct_rows(self):
+        from polygateway.errors import TransientError
+        from tests.unit.test_client import _client as chat_client
+        from tests.unit.test_client import _source, _sse
+        from tests.unit.test_embedding import _embed_client
+        from tests.unit.test_embedding import _src as embed_source
+        from tests.unit.test_ocr_client import _client as ocr_client
+        from tests.unit.test_ocr_client import _src as ocr_source
+
+        recorder = _MemoryRecorder()
+        calls = 0
+
+        def handler(request):
+            nonlocal calls
+            calls += 1
+            if calls == 1:
+                import httpx
+
+                return httpx.Response(503, json={"error": {"message": "retry"}})
+            return _sse()
+
+        chat = chat_client(
+            sources=[_source(provider="zhipu", model="glm-5.3", effort_fallback="nearest")],
+            handler=handler,
+            telemetry=recorder,
+            retry=RetryPolicy(3, 0.001, 0.01),
+        )
+        embed, _ = _embed_client(
+            [embed_source(enable_thinking=True)],
+            [TransientError("retry"), "ok"],
+            telemetry=recorder,
+        )
+        ocr, _, _ = ocr_client(
+            [ocr_source(enable_thinking=True)],
+            [TransientError("retry"), "text"],
+            telemetry=recorder,
+        )
+        try:
+            await asyncio.gather(
+                chat.chat([], reasoning_effort="medium", session_id="run", parent_call_id="chat"),
+                embed.embed(["text"], session_id="run", parent_call_id="embed"),
+                ocr.recognize_text(b"image", session_id="run", parent_call_id="ocr"),
+            )
+            groups = {
+                name: [r for r in recorder.rows if r["parent_call_id"] == name]
+                for name in ("chat", "embed", "ocr")
+            }
+            assert len(recorder.rows) == 6
+            assert len({r["call_id"] for r in recorder.rows}) == 6
+            assert all(r["session_id"] == "run" for r in recorder.rows)
+            assert [r["reasoning_effort"] for r in groups["chat"]] == ["medium", "low"]
+            for name in ("embed", "ocr"):
+                assert len(groups[name]) == 2
+                assert [r["reasoning_effort"] for r in groups[name]] == [None, None]
+        finally:
+            await chat._transport.aclose()
+            await chat.aclose()
+            await embed.aclose()
+            await ocr.aclose()
+
+    async def test_chat_sugar_failure_records_auto_through_retry(self):
+        import httpx
+
+        from polygateway.errors import AllSourcesExhausted
+        from tests.unit.test_client import _client, _source
+
+        recorder = _MemoryRecorder()
+        client = _client(
+            sources=[_source(model="qwen3.7-plus", enable_thinking=True)],
+            handler=lambda request: httpx.Response(503),
+            telemetry=recorder,
+            retry=RetryPolicy(1, 0.001, 0.01),
+        )
+        try:
+            with pytest.raises(AllSourcesExhausted):
+                await client.chat([])
+            attempts = [r for r in recorder.rows if r["source_name"]]
+            assert len(attempts) == 1
+            assert attempts[0]["reasoning_effort"] == "auto"
+            assert attempts[0]["error"]
+        finally:
+            await client._transport.aclose()