fix(kb): 完善知识库入库和重建索引

tests/test_knowledge_base.py

@@ -3,6 +3,7 @@ from django.core.files.uploadedfile import SimpleUploadedFile
 from django.urls import reverse
 
 from review_agent.knowledge_base import build_knowledge_base_context, delete_document, search_knowledge_base
+from review_agent.views import rebuild_knowledge_base_index
 from review_agent.models import KnowledgeBaseDocument
 
 
@@ -16,6 +17,7 @@ def test_knowledge_base_context_reports_rule_and_sources():
     assert context["rule"]["requirement_count"] > 0
     assert context["source_count"] > 0
     assert context["collection_name"] == "nmpa_ivd_registration_v1"
+    assert not any("模拟题二" in source["relative_path"] for source in context["sources"])
 
 
 def test_knowledge_base_page_requires_login(client):
@@ -36,6 +38,11 @@ def test_knowledge_base_page_renders_for_user(client, django_user_model):
     content = response.content.decode("utf-8")
     tabbar = content[content.index('<div class="tabbar"') : content.index("</div>", content.index('<div class="tabbar"'))]
     assert tabbar.index("审核智能体") < tabbar.index("知识库管理") < tabbar.index("附件管理")
+    assert "data-rebuild-url=" in content
+    assert 'id="knowledgeRebuildIndexButton"' in content
+    assert "重建索引" in content
+    assert 'data-source-action="index"' in content
+    assert "手动入库" in content
 
 
 def test_knowledge_base_status_api(client, django_user_model):
@@ -48,6 +55,53 @@ def test_knowledge_base_status_api(client, django_user_model):
     assert response.json()["rule"]["code"] == "nmpa_ivd_registration_v1"
 
 
+def test_knowledge_base_rebuild_index_api(client, django_user_model, monkeypatch):
+    user = django_user_model.objects.create_user(username="owner", password="pass")
+    client.force_login(user)
+    calls = []
+
+    monkeypatch.setattr(
+        "review_agent.views.rebuild_knowledge_base_index",
+        lambda: calls.append("rebuild") or {"chunk_count": 12},
+    )
+
+    response = client.post(reverse("knowledge_base_rebuild_index"))
+
+    assert response.status_code == 200
+    assert response.json()["chunk_count"] == 12
+    assert response.json()["knowledge_base"]["collection"]["count"] >= 0
+    assert calls == ["rebuild"]
+
+
+def test_rebuild_knowledge_base_index_requests_reset(settings, tmp_path, monkeypatch):
+    settings.MEDIA_ROOT = tmp_path
+    settings.REGULATORY_RAG_CHROMA_PATH = tmp_path / "chroma"
+    settings.REGULATORY_RAG_CHROMA_PATH.mkdir()
+    stale_file = settings.REGULATORY_RAG_CHROMA_PATH / "chroma.sqlite3"
+    stale_file.write_text("stale", encoding="utf-8")
+    calls = []
+
+    monkeypatch.setattr("review_agent.views.load_rule_file", lambda: {"source_material_dir": "docs/0.原始材料"})
+    monkeypatch.setattr("review_agent.views.get_embedding_provider", lambda: "provider")
+    monkeypatch.setattr(
+        "review_agent.views.build_chroma_index",
+        lambda source_dir, embedding_provider, reset=False: calls.append(
+            {
+                "source_dir": source_dir,
+                "embedding_provider": embedding_provider,
+                "reset": reset,
+            }
+        )
+        or 8,
+    )
+
+    payload = rebuild_knowledge_base_index()
+
+    assert payload["chunk_count"] == 8
+    assert calls[0]["embedding_provider"] == "provider"
+    assert calls[0]["reset"] is True
+
+
 def test_knowledge_base_search_rejects_blank_query():
     payload = search_knowledge_base("")
 
@@ -103,6 +157,8 @@ def test_knowledge_base_search_api_returns_payload(client, django_user_model):
 
 def test_knowledge_base_document_crud_api(client, settings, tmp_path, django_user_model):
     settings.MEDIA_ROOT = tmp_path
+    settings.REGULATORY_RAG_CHROMA_PATH = tmp_path / "chroma"
+    settings.REGULATORY_RAG_PROVIDER = "deterministic"
     user = django_user_model.objects.create_user(username="owner", password="pass")
     client.force_login(user)
 
@@ -199,6 +255,8 @@ def test_knowledge_base_document_api_is_scoped_to_owner(client, django_user_mode
 
 def test_knowledge_base_document_manual_index_api(client, settings, tmp_path, django_user_model):
     settings.MEDIA_ROOT = tmp_path
+    settings.REGULATORY_RAG_CHROMA_PATH = tmp_path / "chroma"
+    settings.REGULATORY_RAG_PROVIDER = "deterministic"
     user = django_user_model.objects.create_user(username="owner", password="pass")
     client.force_login(user)
     source_path = tmp_path / "manual.md"

tests/test_regulatory_rag.py

@@ -1,3 +1,5 @@
+import sys
+
 import pytest
 
 from review_agent.regulatory_review.services.rag_citation import (
@@ -7,6 +9,7 @@ from review_agent.regulatory_review.services.rag_citation import (
 from review_agent.regulatory_review.services.rag_embedding import SiliconFlowEmbeddingProvider
 from review_agent.regulatory_review.services.rag_index import chunk_text
 from review_agent.regulatory_review.services.rag_index import collect_source_chunks
+from review_agent.regulatory_review.services.rag_index import build_chroma_index
 
 
 def test_siliconflow_embedding_provider_posts_expected_payload(monkeypatch):
@@ -86,3 +89,141 @@ def test_collect_source_chunks_requires_attachment4_extraction(monkeypatch, tmp_
 
     with pytest.raises(RuntimeError, match="附件 4"):
         collect_source_chunks(source_dir)
+
+
+def test_collect_source_chunks_excludes_demo_agent_materials(monkeypatch, tmp_path):
+    source_dir = tmp_path / "sources"
+    source_dir.mkdir()
+    demo_dir = source_dir / "【模拟题二】试剂盒临床注册文件准备与审核Agent"
+    demo_dir.mkdir()
+    (demo_dir / "【模拟题二】试剂盒临床注册文件准备与审核Agent.md").write_text("题目材料", encoding="utf-8")
+    (source_dir / "【模拟题二】试剂盒临床注册文件准备与审核Agent.docx").write_bytes(b"demo")
+    real_source = source_dir / "附件 4 体外诊断试剂注册申报资料要求及说明.doc"
+    real_source.write_bytes(b"rule")
+
+    def fake_extract(path):
+        return "附件4 正文" if path == real_source else "不应被抽取"
+
+    monkeypatch.setattr("review_agent.regulatory_review.services.rag_index.extract_text_from_path", fake_extract)
+
+    chunks = collect_source_chunks(source_dir)
+
+    assert chunks
+    assert all("模拟题二" not in chunk.metadata["source"] for chunk in chunks)
+
+
+def test_build_chroma_index_reset_recreates_collection_without_deleting_index_dir(settings, monkeypatch, tmp_path):
+    settings.MEDIA_ROOT = tmp_path
+    persist_path = tmp_path / "chroma"
+    persist_path.mkdir()
+    stale_file = persist_path / "chroma.sqlite3"
+    stale_file.write_text("stale", encoding="utf-8")
+    source_dir = tmp_path / "sources"
+    source_dir.mkdir()
+    (source_dir / "rule.md").write_text("注册检验报告要求", encoding="utf-8")
+    client_states = []
+    deleted_collections = []
+
+    class FakeCollection:
+        def upsert(self, **kwargs):
+            return None
+
+    class FakeClient:
+        def __init__(self, path):
+            client_states.append({"path": path, "stale_exists": stale_file.exists()})
+
+        def delete_collection(self, name):
+            deleted_collections.append(name)
+
+        def get_or_create_collection(self, name):
+            return FakeCollection()
+
+    class FakeSharedSystemClient:
+        @staticmethod
+        def clear_system_cache():
+            client_states.append({"path": "cache-cleared", "stale_exists": stale_file.exists()})
+
+    monkeypatch.setitem(sys.modules, "chromadb", type("FakeChromaModule", (), {"PersistentClient": FakeClient}))
+    monkeypatch.setitem(
+        sys.modules,
+        "chromadb.api.shared_system_client",
+        type("FakeSharedSystemClientModule", (), {"SharedSystemClient": FakeSharedSystemClient}),
+    )
+
+    count = build_chroma_index(
+        source_dir=source_dir,
+        embedding_provider=lambda texts: [[0.1, 0.2] for _ in texts],
+        persist_path=persist_path,
+        collection_name="test",
+        reset=True,
+    )
+
+    assert count == 1
+    assert client_states == [
+        {"path": str(persist_path), "stale_exists": True},
+        {"path": "cache-cleared", "stale_exists": True},
+        {"path": str(persist_path), "stale_exists": True},
+    ]
+    assert stale_file.exists()
+    assert deleted_collections == ["test"]
+
+
+def test_build_chroma_index_reset_clears_bad_index_dir_after_chroma_cache_reset(settings, monkeypatch, tmp_path):
+    settings.MEDIA_ROOT = tmp_path
+    persist_path = tmp_path / "chroma"
+    persist_path.mkdir()
+    stale_file = persist_path / "chroma.sqlite3"
+    stale_file.write_text("stale", encoding="utf-8")
+    source_dir = tmp_path / "sources"
+    source_dir.mkdir()
+    (source_dir / "rule.md").write_text("注册检验报告要求", encoding="utf-8")
+    events = []
+
+    class FakeCollection:
+        def upsert(self, **kwargs):
+            return None
+
+    class BrokenThenFreshClient:
+        attempts = 0
+
+        def __init__(self, path):
+            BrokenThenFreshClient.attempts += 1
+            events.append(("client", BrokenThenFreshClient.attempts, stale_file.exists()))
+            if BrokenThenFreshClient.attempts == 1:
+                raise ValueError("Could not connect to tenant default_tenant")
+
+        def get_or_create_collection(self, name):
+            return FakeCollection()
+
+    class FakeSharedSystemClient:
+        @staticmethod
+        def clear_system_cache():
+            events.append(("clear_cache", stale_file.exists()))
+
+    fake_chromadb = type(
+        "FakeChromaModule",
+        (),
+        {"PersistentClient": BrokenThenFreshClient},
+    )
+    monkeypatch.setitem(sys.modules, "chromadb", fake_chromadb)
+    monkeypatch.setitem(
+        sys.modules,
+        "chromadb.api.shared_system_client",
+        type("FakeSharedSystemClientModule", (), {"SharedSystemClient": FakeSharedSystemClient}),
+    )
+
+    count = build_chroma_index(
+        source_dir=source_dir,
+        embedding_provider=lambda texts: [[0.1, 0.2] for _ in texts],
+        persist_path=persist_path,
+        collection_name="test",
+        reset=True,
+    )
+
+    assert count == 1
+    assert events == [
+        ("client", 1, True),
+        ("clear_cache", True),
+        ("client", 2, False),
+    ]
+    assert not stale_file.exists()