mirror of
https://github.com/hwchase17/langchain
synced 2024-11-08 07:10:35 +00:00
68 lines
2.5 KiB
Python
68 lines
2.5 KiB
Python
from typing import Any, Dict, Iterator, List, Optional
|
|
|
|
from langchain_core.documents import Document
|
|
|
|
from langchain_community.document_loaders.base import BaseLoader
|
|
|
|
|
|
class TiDBLoader(BaseLoader):
|
|
"""Load documents from TiDB."""
|
|
|
|
def __init__(
|
|
self,
|
|
connection_string: str,
|
|
query: str,
|
|
page_content_columns: Optional[List[str]] = None,
|
|
metadata_columns: Optional[List[str]] = None,
|
|
engine_args: Optional[Dict[str, Any]] = None,
|
|
) -> None:
|
|
"""Initialize TiDB document loader.
|
|
|
|
Args:
|
|
connection_string (str): The connection string for the TiDB database,
|
|
format: "mysql+pymysql://root@127.0.0.1:4000/test".
|
|
query: The query to run in TiDB.
|
|
page_content_columns: Optional. Columns written to Document `page_content`,
|
|
default(None) to all columns.
|
|
metadata_columns: Optional. Columns written to Document `metadata`,
|
|
default(None) to no columns.
|
|
engine_args: Optional. Additional arguments to pass to sqlalchemy engine.
|
|
"""
|
|
self.connection_string = connection_string
|
|
self.query = query
|
|
self.page_content_columns = page_content_columns
|
|
self.metadata_columns = metadata_columns if metadata_columns is not None else []
|
|
self.engine_args = engine_args
|
|
|
|
def lazy_load(self) -> Iterator[Document]:
|
|
"""Lazy load TiDB data into document objects."""
|
|
|
|
from sqlalchemy import create_engine
|
|
from sqlalchemy.engine import Engine
|
|
from sqlalchemy.sql import text
|
|
|
|
# use sqlalchemy to create db connection
|
|
engine: Engine = create_engine(
|
|
self.connection_string, **(self.engine_args or {})
|
|
)
|
|
|
|
# execute query
|
|
with engine.connect() as conn:
|
|
result = conn.execute(text(self.query))
|
|
|
|
# convert result to Document objects
|
|
column_names = list(result.keys())
|
|
for row in result:
|
|
# convert row to dict{column:value}
|
|
row_data = {
|
|
column_names[index]: value for index, value in enumerate(row)
|
|
}
|
|
page_content = "\n".join(
|
|
f"{k}: {v}"
|
|
for k, v in row_data.items()
|
|
if self.page_content_columns is None
|
|
or k in self.page_content_columns
|
|
)
|
|
metadata = {col: row_data[col] for col in self.metadata_columns}
|
|
yield Document(page_content=page_content, metadata=metadata)
|