Databricks Community

Octavian1 · ‎03-20-2024

The following assignment:

from langchain.sql_database import SQLDatabase

dbase = SQLDatabase.from_databricks(
catalog=catalog, 
schema=db,
host=host, 
api_token=token,
)

fails with

ValueError: invalid literal for int() with base 10: ''

because of

cls._assert_port(port),

in /lib/python3.10/site-packages/sqlalchemy/engine/url.py:211, in URL.create(cls, drivername, username, password, host, port, database, query)
as shown in the error trace.

Couldn't find any information regarding this (for instance, in the example given in: https://docs.gcp.databricks.com/en/large-language-models/langchain.html)

The token and host are OK. Also the catalog and schema have correct values.

Using langchain==0.1.5.

Kaniz_Fatma · ‎03-21-2024

Hi @Octavian1, Ensure that the port parameter you’re passing to SQLDatabase.from_databricks is a valid integer. If it’s empty or contains non-numeric characters, that could be the root cause.

In a Stack Overflow post, someone faced a similar issue while connecting SQLAlchemy (SQLDatabaseChain from langchain) to SingleStoreDB. They provided a workaround using a custom connection URL.

Here’s an example snippet:

from sqlalchemy import create_engine
engine = create_engine(CONNECTION_URL)
db = SQLDatabase(engine)

You might need to adapt this approach to your specific use case ¹.

Ensure that your langchain version (0.1.5) is compatible with your environment and dependencies. Sometimes, upgrading to a newer version might resolve issues.
If you encounter any other issues, feel free to ask!