Appending a local Pandas dataframe to a Delta table in Azure is slow

リポジトリのファイルもテストも指定されていません。まず、指定された SQL warehouse に対して databricks-sql-connector 3.4.0 を使い、提供された pandas DataFrame.to_sql の例を再現します。次に、SQLAlchemy と connector の書き込みパスを追跡して、行ごとの遅延箇所を特定します。サポートされている、より高速な書き込みパスを特定するか、制限とその原因を文書化できれば完了です。

I followed the instructions on this page to create a SQLAlchemy engine and used it with the Pandas to_sql() method. It's taking around 2 seconds to append one data point to a Delta table in Azure Databricks, and it seems to be scaling linearly. A dataframe containing 1 column and 10 rows is taking ~20 seconds to push to Azure.

Is there a way to make writing back to Databricks from a local machine faster?

Example code:

from sqlalchemy import create_engine
import pandas as pd
import os

server = 'my_server.azuredatabricks.net'
http_path = "/sql/1.0/warehouses/my_warehouse"
access_token = "MY_TOKEN"
catalog = "my_catalog"
schema = "my_schema"

if "NO_PROXY" in os.environ:
    os.environ["NO_PROXY"] = os.environ["NO_PROXY"] + "," + server
else:
    os.environ["NO_PROXY"] = server

if "no_proxy" in os.environ:
    os.environ["no_proxy"] = os.environ["no_proxy"] + "," + server
else:
    os.environ["no_proxy"] = server

engine = create_engine(f"databricks://token:{access_token}@{server}?" +
        f"http_path={http_path}&catalog={catalog}&schema={schema}"
)

df = pd.DataFrame({'name' : ['User 1', 'User 2', 'User 3', User 4', 'User 5', 'User 6', User 7', 'User 8', 'User 9', 'User 10']})

# The next command takes around 20 seconds to complete
df.to_sql(name='my_test_table', con=engine, if_exists='append', index=False)

Local specs:
databricks-sql-connector==3.4.0
pandas==2.2.2
Python 3.10.14

Azure specs:
Databricks SQL warehouse cluster
Runtime==13.3 LTS