|
| 1 | +from __future__ import annotations |
| 2 | + |
| 3 | +import json |
| 4 | +from collections.abc import AsyncIterator |
| 5 | +from types import SimpleNamespace |
| 6 | +from typing import Any, cast |
| 7 | + |
| 8 | +import litellm |
| 9 | +import pytest |
| 10 | +from openai import AsyncOpenAI |
| 11 | +from openai.types.chat import ChatCompletion, ChatCompletionChunk |
| 12 | + |
| 13 | +from agents import Agent, ModelSettings, OpenAIChatCompletionsModel, RunConfig, Runner, trace |
| 14 | +from agents.extensions.models.litellm_model import LitellmModel |
| 15 | +from agents.models.interface import Model |
| 16 | +from tests.testing_processor import fetch_ordered_spans, fetch_traces |
| 17 | + |
| 18 | + |
| 19 | +@pytest.mark.allow_call_model_methods |
| 20 | +@pytest.mark.asyncio |
| 21 | +@pytest.mark.parametrize("adapter", ["openai", "litellm", "any-llm"]) |
| 22 | +@pytest.mark.parametrize("stream", [False, True]) |
| 23 | +@pytest.mark.parametrize("include_data", [False, True]) |
| 24 | +async def test_generation_metadata_respects_sensitive_capture( |
| 25 | + monkeypatch: pytest.MonkeyPatch, adapter: str, stream: bool, include_data: bool |
| 26 | +) -> None: |
| 27 | + metadata = {"request_label": "synthetic-request-metadata-sentinel"} |
| 28 | + settings = ModelSettings(temperature=0.25, metadata=metadata) |
| 29 | + request_calls: list[dict[str, Any]] = [] |
| 30 | + |
| 31 | + async def chunks() -> AsyncIterator[ChatCompletionChunk]: |
| 32 | + yield ChatCompletionChunk.model_validate( |
| 33 | + { |
| 34 | + "id": "chatcmpl_test", |
| 35 | + "created": 0, |
| 36 | + "model": "test-model", |
| 37 | + "object": "chat.completion.chunk", |
| 38 | + "choices": [ |
| 39 | + { |
| 40 | + "index": 0, |
| 41 | + "delta": {"role": "assistant", "content": "ok"}, |
| 42 | + "finish_reason": "stop", |
| 43 | + } |
| 44 | + ], |
| 45 | + } |
| 46 | + ) |
| 47 | + |
| 48 | + async def completion(**kwargs: Any) -> Any: |
| 49 | + request_calls.append(kwargs) |
| 50 | + if kwargs["stream"]: |
| 51 | + return chunks() |
| 52 | + payload = { |
| 53 | + "id": "chatcmpl_test", |
| 54 | + "created": 0, |
| 55 | + "model": "test-model", |
| 56 | + "object": "chat.completion", |
| 57 | + "choices": [ |
| 58 | + { |
| 59 | + "index": 0, |
| 60 | + "message": {"role": "assistant", "content": "ok"}, |
| 61 | + "finish_reason": "stop", |
| 62 | + } |
| 63 | + ], |
| 64 | + } |
| 65 | + if adapter == "litellm": |
| 66 | + return litellm.ModelResponse(**payload) |
| 67 | + return ChatCompletion.model_validate(payload) |
| 68 | + |
| 69 | + model: Model |
| 70 | + if adapter == "openai": |
| 71 | + client = SimpleNamespace( |
| 72 | + base_url="https://example.com/v1", |
| 73 | + chat=SimpleNamespace(completions=SimpleNamespace(create=completion)), |
| 74 | + ) |
| 75 | + model = OpenAIChatCompletionsModel("test-model", cast(AsyncOpenAI, client)) |
| 76 | + elif adapter == "litellm": |
| 77 | + monkeypatch.setattr(litellm, "acompletion", completion) |
| 78 | + model = LitellmModel("test-model") |
| 79 | + else: |
| 80 | + pytest.importorskip("any_llm", reason="any-llm-sdk requires Python 3.11+.") |
| 81 | + from agents.extensions.models.any_llm_model import AnyLLM, AnyLLMModel |
| 82 | + |
| 83 | + provider = SimpleNamespace(SUPPORTS_RESPONSES=False, acompletion=completion) |
| 84 | + monkeypatch.setattr(AnyLLM, "create", lambda *args, **kwargs: provider) |
| 85 | + model = AnyLLMModel("openai/test-model") |
| 86 | + |
| 87 | + agent = Agent(name="test", model=model, model_settings=settings) |
| 88 | + trace_metadata = {"metadata": "intentional-trace-label"} |
| 89 | + config = RunConfig(trace_include_sensitive_data=include_data, trace_metadata=trace_metadata) |
| 90 | + if stream: |
| 91 | + result = Runner.run_streamed(agent, "hello", run_config=config) |
| 92 | + async for _ in result.stream_events(): |
| 93 | + pass |
| 94 | + else: |
| 95 | + result = await Runner.run(agent, "hello", run_config=config) |
| 96 | + assert result.final_output == "ok" |
| 97 | + |
| 98 | + assert len(request_calls) == 1 |
| 99 | + assert request_calls[0]["metadata"] == metadata |
| 100 | + assert settings.metadata == metadata |
| 101 | + assert metadata == {"request_label": "synthetic-request-metadata-sentinel"} |
| 102 | + assert trace_metadata == {"metadata": "intentional-trace-label"} |
| 103 | + traces = fetch_traces() |
| 104 | + assert len(traces) == 1 |
| 105 | + exported_trace = traces[0].export() |
| 106 | + assert exported_trace is not None |
| 107 | + assert exported_trace["metadata"] == trace_metadata |
| 108 | + |
| 109 | + spans = fetch_ordered_spans() |
| 110 | + generations = [span for span in spans if span.span_data.type == "generation"] |
| 111 | + assert len(generations) == 1 |
| 112 | + exported = generations[0].export() |
| 113 | + assert exported is not None |
| 114 | + config_data = exported["span_data"]["model_config"] |
| 115 | + assert config_data["temperature"] == 0.25 |
| 116 | + if include_data: |
| 117 | + assert config_data["metadata"] == metadata |
| 118 | + else: |
| 119 | + assert "metadata" not in config_data |
| 120 | + assert "synthetic-request-metadata-sentinel" not in json.dumps( |
| 121 | + [span.export() for span in spans] |
| 122 | + ) |
| 123 | + |
| 124 | + |
| 125 | +@pytest.mark.allow_call_model_methods |
| 126 | +@pytest.mark.asyncio |
| 127 | +async def test_caller_authored_trace_metadata_remains_explicit() -> None: |
| 128 | + async def completion(**kwargs: Any) -> ChatCompletion: |
| 129 | + return ChatCompletion.model_validate( |
| 130 | + { |
| 131 | + "id": "chatcmpl_test", |
| 132 | + "created": 0, |
| 133 | + "model": "test-model", |
| 134 | + "object": "chat.completion", |
| 135 | + "choices": [ |
| 136 | + { |
| 137 | + "index": 0, |
| 138 | + "message": {"role": "assistant", "content": "ok"}, |
| 139 | + "finish_reason": "stop", |
| 140 | + } |
| 141 | + ], |
| 142 | + } |
| 143 | + ) |
| 144 | + |
| 145 | + client = SimpleNamespace( |
| 146 | + base_url="https://example.com/v1", |
| 147 | + chat=SimpleNamespace(completions=SimpleNamespace(create=completion)), |
| 148 | + ) |
| 149 | + agent = Agent( |
| 150 | + name="test", model=OpenAIChatCompletionsModel("test-model", cast(AsyncOpenAI, client)) |
| 151 | + ) |
| 152 | + metadata = {"metadata": "caller-owned-label"} |
| 153 | + with trace("caller-owned", metadata=metadata): |
| 154 | + await Runner.run(agent, "hello", run_config=RunConfig(trace_include_sensitive_data=False)) |
| 155 | + exported = fetch_traces()[0].export() |
| 156 | + assert exported is not None |
| 157 | + assert exported["metadata"] == metadata == {"metadata": "caller-owned-label"} |
0 commit comments