feat: multi-collection support — each project uses its own isolated collection

This commit is contained in:
2026-07-05 02:07:02 +08:00
parent 8042cf288b
commit 0aee167085
2 changed files with 59 additions and 18 deletions
+20 -12
View File
@@ -21,14 +21,20 @@ logger = logging.getLogger("md-vector-db")
class IngestRequest(BaseModel): class IngestRequest(BaseModel):
file_path: str | None = None file_path: str | None = None
content: str | None = None content: str | None = None
file_name: str | None = Field( file_name: str | None = Field(default=None, max_length=255)
default=None, max_length=255 collection: str | None = Field(
default=None, max_length=128,
description="目标 collection(默认使用配置文件中的 collection_name",
) )
class SearchRequest(BaseModel): class SearchRequest(BaseModel):
query: str = Field(..., min_length=1, max_length=2000) query: str = Field(..., min_length=1, max_length=2000)
top_k: int = Field(default=10, ge=1, le=100) top_k: int = Field(default=10, ge=1, le=100)
collection: str | None = Field(
default=None, max_length=128,
description="检索的 collection(默认使用配置文件中的 collection_name",
)
# -- 路径安全检查 -- # -- 路径安全检查 --
@@ -64,9 +70,7 @@ def health(state: AppState = Depends(get_state)):
@app.get("/api/v1/collections") @app.get("/api/v1/collections")
def list_collections(state: AppState = Depends(get_state)): def list_collections(state: AppState = Depends(get_state)):
info = state.searcher.get_collection_info() return {"collections": state.list_collections_with_stats()}
sources = state.searcher.list_sources()
return {"collections": [info], "sources": sources}
@app.post("/api/v1/ingest") @app.post("/api/v1/ingest")
@@ -75,6 +79,7 @@ def ingest_document(
state: AppState = Depends(get_state), state: AppState = Depends(get_state),
_: bool = Depends(verify_api_key), _: bool = Depends(verify_api_key),
): ):
ingestor = state.get_ingestor(req.collection)
try: try:
if req.file_path: if req.file_path:
if not _is_safe_path(req.file_path): if not _is_safe_path(req.file_path):
@@ -85,14 +90,14 @@ def ingest_document(
raise HTTPException(status_code=400, detail="不允许访问当前目录外的路径") raise HTTPException(status_code=400, detail="不允许访问当前目录外的路径")
if not path.exists(): if not path.exists():
raise HTTPException(status_code=404, detail=f"文件不存在: {path.name}") raise HTTPException(status_code=404, detail=f"文件不存在: {path.name}")
count = state.ingestor.ingest_file(str(path)) count = ingestor.ingest_file(str(path))
file_name = path.name file_name = path.name
elif req.content: elif req.content:
file_name = req.file_name or "untitled.md" file_name = req.file_name or "untitled.md"
count = state.ingestor.ingest_content(req.content, file_name) count = ingestor.ingest_content(req.content, file_name)
else: else:
raise HTTPException(status_code=400, detail="需要提供 file_path 或 content") raise HTTPException(status_code=400, detail="需要提供 file_path 或 content")
return {"status": "ok", "chunks": count, "file": file_name} return {"status": "ok", "chunks": count, "file": file_name, "collection": ingestor.collection_name}
except HTTPException: except HTTPException:
raise raise
except Exception: except Exception:
@@ -106,8 +111,9 @@ def search_documents(
state: AppState = Depends(get_state), state: AppState = Depends(get_state),
_: bool = Depends(verify_api_key), _: bool = Depends(verify_api_key),
): ):
results = state.searcher.search(req.query, top_k=req.top_k) searcher = state.get_searcher(req.collection)
return {"results": results} results = searcher.search(req.query, top_k=req.top_k)
return {"results": results, "collection": searcher.collection_name}
@app.delete("/api/v1/documents/{file_name}") @app.delete("/api/v1/documents/{file_name}")
@@ -115,8 +121,10 @@ def delete_document(
file_name: str, file_name: str,
state: AppState = Depends(get_state), state: AppState = Depends(get_state),
_: bool = Depends(verify_api_key), _: bool = Depends(verify_api_key),
collection: str | None = None,
): ):
deleted = state.searcher.delete_by_source(file_name) searcher = state.get_searcher(collection)
deleted = searcher.delete_by_source(file_name)
if not deleted: if not deleted:
raise HTTPException(status_code=404, detail=f"文档不存在: {file_name}") raise HTTPException(status_code=404, detail=f"文档不存在: {file_name}")
return {"status": "ok", "file": file_name} return {"status": "ok", "file": file_name, "collection": searcher.collection_name}
+39 -6
View File
@@ -12,7 +12,7 @@ logger = logging.getLogger(__name__)
class AppState: class AppState:
"""应用级共享状态.""" """应用级共享状态 — db 和 embedder 全局共享, searcher/ingestor 按 collection 懒创建."""
def __init__(self): def __init__(self):
config_path = os.environ.get("MD_VECTOR_CONFIG", "config.yaml") config_path = os.environ.get("MD_VECTOR_CONFIG", "config.yaml")
@@ -22,20 +22,48 @@ class AppState:
"MD_VECTOR_DB_DATA_DIR", self.config.chroma.persist_dir "MD_VECTOR_DB_DATA_DIR", self.config.chroma.persist_dir
) )
self.db = VectorDB(persist_dir=data_dir) self.db = VectorDB(persist_dir=data_dir)
self.embedder = create_embedder(self.config.embed) self.embedder = create_embedder(self.config.embed)
collection = os.environ.get( self.default_collection = os.environ.get(
"MD_VECTOR_DB_COLLECTION", self.config.chroma.collection_name "MD_VECTOR_DB_COLLECTION", self.config.chroma.collection_name
) )
self.searcher = Searcher(self.db, self.embedder, collection)
self.ingestor = DocumentIngestor(self.db, self.embedder, collection) # 按 collection 懒加载 searcher / ingestor
self._searchers: dict[str, Searcher] = {}
self._ingestors: dict[str, DocumentIngestor] = {}
def get_searcher(self, collection: str | None = None) -> Searcher:
name = collection or self.default_collection
if name not in self._searchers:
self._searchers[name] = Searcher(self.db, self.embedder, name)
return self._searchers[name]
def get_ingestor(self, collection: str | None = None) -> DocumentIngestor:
name = collection or self.default_collection
if name not in self._ingestors:
self._ingestors[name] = DocumentIngestor(self.db, self.embedder, name)
return self._ingestors[name]
def list_collections_with_stats(self) -> list[dict]:
"""列出所有 collection 及其统计."""
result = []
all_names = set(self._searchers.keys()) | set(self._ingestors.keys())
all_names.add(self.default_collection)
for name in sorted(all_names):
try:
coll = self.db.get_or_create_collection(name)
result.append({"name": name, "count": coll.count()})
except Exception:
result.append({"name": name, "count": 0})
return result
def is_healthy(self) -> dict: def is_healthy(self) -> dict:
"""真实健康检查: 验证 ChromaDB 和 Embedder 是否可用.""" """真实健康检查: 验证 ChromaDB 和 Embedder 是否可用."""
status = {"status": "ok", "checks": {}} status = {"status": "ok", "checks": {}}
try: try:
count = self.searcher.collection.count() count = self.db.get_or_create_collection(
self.default_collection
).count()
status["checks"]["chromadb"] = {"status": "ok", "count": count} status["checks"]["chromadb"] = {"status": "ok", "count": count}
except Exception as e: except Exception as e:
status["checks"]["chromadb"] = {"status": "error", "detail": str(e)} status["checks"]["chromadb"] = {"status": "error", "detail": str(e)}
@@ -59,3 +87,8 @@ def get_state() -> AppState:
logger.info("初始化应用状态...") logger.info("初始化应用状态...")
_state = AppState() _state = AppState()
return _state return _state
def get_default_collection() -> str:
"""获取默认 collection 名, 供 CLI 等非 HTTP 场景使用."""
return get_state().default_collection