166 lines
6.9 KiB
Python
166 lines
6.9 KiB
Python
"""researcher_node：检索类型 `knowledge_base`（知识库检索）= 直连内网 ES 接口。
|
||
|
||
弱模型识别不了技能时用这条路：不编排任何技能、不跑通用 provider，
|
||
直接调 ai_writing.intranet_search_url 内网知识库 ES 接口检索。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
from types import SimpleNamespace
|
||
|
||
import pytest
|
||
|
||
from deerflow.agents.ai_writing.nodes import researcher as R
|
||
from deerflow.agents.ai_writing.search.base import SearchResult
|
||
|
||
|
||
class _FakeIntranet:
|
||
instances: list = []
|
||
|
||
def __init__(self, **kw):
|
||
self.errors = []
|
||
self.raw_responses = []
|
||
self.calls = []
|
||
self.kw = kw
|
||
_FakeIntranet.instances.append(self)
|
||
|
||
async def search(self, query, max_results=5, **_):
|
||
self.calls.append({"query": query, "status": "ok", "ok": True, "count": 1, "elapsed_ms": 12})
|
||
return [SearchResult(
|
||
title=f"内网-{query}", content="内网正文", source="ES",
|
||
published_at="", url=f"http://es/{query}", source_type="intranet",
|
||
)]
|
||
|
||
|
||
class _EmptyIntranet(_FakeIntranet):
|
||
async def search(self, query, max_results=5, **_):
|
||
self.calls.append({"query": query, "status": "empty", "ok": True, "count": 0, "elapsed_ms": 9})
|
||
return []
|
||
|
||
|
||
class _FailingIntranet(_FakeIntranet):
|
||
async def search(self, query, max_results=5, **_):
|
||
msg = f"ConnectError: cannot connect for {query}"
|
||
self.errors.append(msg)
|
||
self.raw_responses.append("{\"success\": false, \"error\": \"cannot connect\"}")
|
||
self.calls.append({"query": query, "status": "error", "ok": False, "count": 0, "elapsed_ms": 3, "error": msg})
|
||
return []
|
||
|
||
|
||
class _SkillProviderShouldNotRun:
|
||
"""knowledge_base 模式不应触碰通用 provider —— 调到就让测试爆。"""
|
||
errors: list = []
|
||
raw_responses: list = []
|
||
|
||
async def search(self, query, max_results=5, **_): # pragma: no cover - 不应被调用
|
||
raise AssertionError("knowledge_base 模式不应调用通用 provider / 技能检索")
|
||
|
||
|
||
def _patch(monkeypatch, intranet_url="http://es/api", intranet_cls=_FakeIntranet):
|
||
_FakeIntranet.instances = []
|
||
cfg = {
|
||
"max_materials": 20,
|
||
"intranet_search_url": intranet_url,
|
||
"intranet_knowledge_base": "samples",
|
||
}
|
||
monkeypatch.setattr(R, "get_ai_writing_config", lambda: cfg)
|
||
monkeypatch.setattr(R, "use_builtin_agent", lambda *_: False)
|
||
monkeypatch.setattr(R, "safe_get_stream_writer", lambda: (lambda *a, **k: None))
|
||
monkeypatch.setattr(R, "use_skill_search_agent", lambda: False)
|
||
|
||
async def fake_llm_json(*a, **k):
|
||
return {"keywords": ["特朗普竞选"]}
|
||
|
||
monkeypatch.setattr(R, "llm_json", fake_llm_json)
|
||
# knowledge_base 模式不该加载/编排技能，但仍打桩以防意外路径
|
||
monkeypatch.setattr(R, "_load_configured_skills",
|
||
lambda: [SimpleNamespace(name="knowledge-search-v2", description="内网检索")])
|
||
monkeypatch.setattr(R, "get_search_provider", lambda _cfg: _SkillProviderShouldNotRun())
|
||
monkeypatch.setattr(
|
||
"deerflow.agents.ai_writing.search.intranet_stub.IntranetSearchProvider",
|
||
intranet_cls,
|
||
)
|
||
|
||
|
||
def _state(material_source="knowledge_base"):
|
||
return {
|
||
"user_intent": "分析特朗普近期动向",
|
||
"article_type": "政治分析",
|
||
"target_audience": "普通读者",
|
||
"keyword_count": 1,
|
||
"material_source": material_source,
|
||
}
|
||
|
||
|
||
@pytest.mark.asyncio
|
||
async def test_knowledge_base_hits_intranet_directly(monkeypatch):
|
||
_patch(monkeypatch)
|
||
out = await R.researcher_node(_state(), {"configurable": {}})
|
||
mats = out["material_package"]["materials"]
|
||
assert mats, "知识库检索应返回内网素材"
|
||
assert all(m["source_type"] == "intranet" for m in mats)
|
||
assert mats[0]["title"] == "内网-特朗普竞选"
|
||
assert _FakeIntranet.instances, "应实例化 IntranetSearchProvider 直连内网接口"
|
||
# 走素材确认暂停点
|
||
assert out["status"] == "awaiting_material_confirm"
|
||
dbg = out["material_package"].get("intranet_debug")
|
||
assert dbg["mode"] == "knowledge_base"
|
||
assert dbg["status"] == "ok"
|
||
assert dbg["endpoint"] == "http://es/api"
|
||
assert dbg["result_count"] == 1
|
||
assert dbg["queries"][0]["query"] == "特朗普竞选"
|
||
assert dbg["queries"][0]["count"] == 1
|
||
assert dbg["calls"][0]["status"] == "ok"
|
||
|
||
|
||
@pytest.mark.asyncio
|
||
async def test_knowledge_base_no_url_falls_back_to_default_endpoint(monkeypatch):
|
||
"""未配置接口地址：仍实例化 IntranetSearchProvider（base_url 传空串，由 provider 内部
|
||
回退默认内网端点），内网直连依然可用——恢复旧版 search_provider=intranet 的行为。"""
|
||
_patch(monkeypatch, intranet_url="")
|
||
out = await R.researcher_node(_state(), {"configurable": {}})
|
||
mats = out["material_package"]["materials"]
|
||
assert mats, "无地址时应回退默认内网端点，仍返回素材"
|
||
assert all(m["source_type"] == "intranet" for m in mats)
|
||
assert _FakeIntranet.instances, "应实例化 IntranetSearchProvider 直连内网接口"
|
||
assert _FakeIntranet.instances[0].kw.get("base_url") == "", "空地址应原样下传，由 provider 回退默认端点"
|
||
|
||
|
||
@pytest.mark.asyncio
|
||
async def test_knowledge_base_does_not_orchestrate_skills(monkeypatch):
|
||
"""knowledge_base 不编排技能：通用 provider 一旦被调用就会 raise。"""
|
||
_patch(monkeypatch)
|
||
# 不抛异常即证明没走 _SkillProviderShouldNotRun
|
||
out = await R.researcher_node(_state(), {"configurable": {}})
|
||
assert out["material_package"]["materials"]
|
||
|
||
|
||
@pytest.mark.asyncio
|
||
async def test_knowledge_base_empty_result_is_visible(monkeypatch):
|
||
"""接口成功但 0 条也要在返回里能看出来，而不是只显示"无素材"。"""
|
||
_patch(monkeypatch, intranet_cls=_EmptyIntranet)
|
||
out = await R.researcher_node(_state(), {"configurable": {}})
|
||
pkg = out["material_package"]
|
||
assert pkg["materials"] == []
|
||
dbg = pkg.get("intranet_debug")
|
||
assert dbg["status"] == "empty"
|
||
assert dbg["result_count"] == 0
|
||
assert dbg["queries"][0]["count"] == 0
|
||
assert dbg["calls"][0]["status"] == "empty"
|
||
steps = pkg["skill_search_steps"]
|
||
assert any(step["type"] == "skill_result" and "0 条" in step["message"] for step in steps)
|
||
|
||
|
||
@pytest.mark.asyncio
|
||
async def test_knowledge_base_failure_is_visible(monkeypatch):
|
||
"""provider 吞掉异常返回 [] 时，前端仍应看到报错步骤与 notice。"""
|
||
_patch(monkeypatch, intranet_cls=_FailingIntranet)
|
||
out = await R.researcher_node(_state(), {"configurable": {}})
|
||
pkg = out["material_package"]
|
||
dbg = pkg.get("intranet_debug")
|
||
assert dbg["status"] == "error"
|
||
assert dbg["errors"]
|
||
assert dbg["raw_responses"]
|
||
steps = pkg["skill_search_steps"]
|
||
assert any(step["type"] == "skill_error" for step in steps)
|
||
assert any(event["type"] == "notice" and event.get("raw") for event in out["progress_events"])
|