Google Cloud SQL for SQL server

Cloud SQL is a fully managed relational database service that offers high performance, seamless integration, and impressive scalability. It offers MySQL, PostgreSQL, and SQL Server database engines. Extend your database application to build AI-powered experiences leveraging Cloud SQL's Langchain integrations.

This notebook goes over how to use Cloud SQL for SQL server to save, load and delete langchain documents with MSSQLLoader and MSSQLDocumentSaver.

Learn more about the package on GitHub.

Before You Begin

To run this notebook, you will need to do the following:

• Create a Google Cloud Project
• Enable the Cloud SQL Admin API.
• Create a Cloud SQL for SQL server instance
• Create a Cloud SQL database
• Add an IAM database user to the database (Optional)

After confirmed access to database in the runtime environment of this notebook, filling the following values and run the cell before running example scripts.

# @markdown Please fill in the both the Google Cloud region and name of your Cloud SQL instance.
REGION = "us-central1"  # @param {type:"string"}
INSTANCE = "test-instance"  # @param {type:"string"}

# @markdown Please fill in user name and password of your Cloud SQL instance.
DB_USER = "sqlserver"  # @param {type:"string"}
DB_PASS = "password"  # @param {type:"string"}

# @markdown Please specify a database and a table for demo purpose.
DATABASE = "test"  # @param {type:"string"}
TABLE_NAME = "test-default"  # @param {type:"string"}

🦜🔗 Library Installation

The integration lives in its own langchain-google-cloud-sql-mssql package, so we need to install it.

%pip install --upgrade --quiet langchain-google-cloud-sql-mssql

Colab only: Uncomment the following cell to restart the kernel or use the button to restart the kernel. For Vertex AI Workbench you can restart the terminal using the button on top.

# # Automatically restart kernel after installs so that your environment can access the new packages
# import IPython

# app = IPython.Application.instance()
# app.kernel.do_shutdown(True)

🔐 Authentication

Authenticate to Google Cloud as the IAM user logged into this notebook in order to access your Google Cloud Project.

• If you are using Colab to run this notebook, use the cell below and continue.
• If you are using Vertex AI Workbench, check out the setup instructions here.

from google.colab import auth

auth.authenticate_user()

☁ Set Your Google Cloud Project

Set your Google Cloud project so that you can leverage Google Cloud resources within this notebook.

If you don't know your project ID, try the following:

• Run gcloud config list.
• Run gcloud projects list.
• See the support page: Locate the project ID.

# @markdown Please fill in the value below with your Google Cloud project ID and then run the cell.

PROJECT_ID = "my-project-id"  # @param {type:"string"}

# Set the project id
!gcloud config set project {PROJECT_ID}

💡 API Enablement

The langchain-google-cloud-sql-mssql package requires that you enable the Cloud SQL Admin API in your Google Cloud Project.

# enable Cloud SQL Admin API
!gcloud services enable sqladmin.googleapis.com

Basic Usage

MSSQLEngine Connection Pool

Before saving or loading documents from MSSQL table, we need first configures a connection pool to Cloud SQL database. The MSSQLEngine configures a SQLAlchemy connection pool to your Cloud SQL database, enabling successful connections from your application and following industry best practices.

To create a MSSQLEngine using MSSQLEngine.from_instance() you need to provide only 4 things:

1. project_id : Project ID of the Google Cloud Project where the Cloud SQL instance is located.
2. region : Region where the Cloud SQL instance is located.
3. instance : The name of the Cloud SQL instance.
4. database : The name of the database to connect to on the Cloud SQL instance.
5. user : Database user to use for built-in database authentication and login.
6. password : Database password to use for built-in database authentication and login.

from langchain_google_cloud_sql_mssql import MSSQLEngine

engine = MSSQLEngine.from_instance(
    project_id=PROJECT_ID,
    region=REGION,
    instance=INSTANCE,
    database=DATABASE,
    user=DB_USER,
    password=DB_PASS,
)

Initialize a table

Initialize a table of default schema via MSSQLEngine.init_document_table(<table_name>). Table Columns:

• page_content (type: text)
• langchain_metadata (type: JSON)

overwrite_existing=True flag means the newly initialized table will replace any existing table of the same name.

engine.init_document_table(TABLE_NAME, overwrite_existing=True)

Save documents

Save langchain documents with MSSQLDocumentSaver.add_documents(<documents>). To initialize MSSQLDocumentSaver class you need to provide 2 things:

1. engine - An instance of a MSSQLEngine engine.
2. table_name - The name of the table within the Cloud SQL database to store langchain documents.

from langchain_core.documents import Document
from langchain_google_cloud_sql_mssql import MSSQLDocumentSaver

test_docs = [
    Document(
        page_content="Apple Granny Smith 150 0.99 1",
        metadata={"fruit_id": 1},
    ),
    Document(
        page_content="Banana Cavendish 200 0.59 0",
        metadata={"fruit_id": 2},
    ),
    Document(
        page_content="Orange Navel 80 1.29 1",
        metadata={"fruit_id": 3},
    ),
]
saver = MSSQLDocumentSaver(engine=engine, table_name=TABLE_NAME)
saver.add_documents(test_docs)

API Reference:Document

Load documents

Load langchain documents with MSSQLLoader.load() or MSSQLLoader.lazy_load(). lazy_load returns a generator that only queries database during the iteration. To initialize MSSQLDocumentSaver class you need to provide:

1. engine - An instance of a MSSQLEngine engine.
2. table_name - The name of the table within the Cloud SQL database to store langchain documents.

from langchain_google_cloud_sql_mssql import MSSQLLoader

loader = MSSQLLoader(engine=engine, table_name=TABLE_NAME)
docs = loader.lazy_load()
for doc in docs:
    print("Loaded documents:", doc)

Load documents via query

Other than loading documents from a table, we can also choose to load documents from a view generated from a SQL query. For example:

from langchain_google_cloud_sql_mssql import MSSQLLoader

loader = MSSQLLoader(
    engine=engine,
    query=f"select * from \"{TABLE_NAME}\" where JSON_VALUE(langchain_metadata, '$.fruit_id') = 1;",
)
onedoc = loader.load()
onedoc

The view generated from SQL query can have different schema than default table. In such cases, the behavior of MSSQLLoader is the same as loading from table with non-default schema. Please refer to section Load documents with customized document page content & metadata.

Delete documents

Delete a list of langchain documents from MSSQL table with MSSQLDocumentSaver.delete(<documents>).

For table with default schema (page_content, langchain_metadata), the deletion criteria is:

A row should be deleted if there exists a document in the list, such that

• document.page_content equals row[page_content]
• document.metadata equals row[langchain_metadata]

from langchain_google_cloud_sql_mssql import MSSQLLoader

loader = MSSQLLoader(engine=engine, table_name=TABLE_NAME)
docs = loader.load()
print("Documents before delete:", docs)
saver.delete(onedoc)
print("Documents after delete:", loader.load())

Advanced Usage

Load documents with customized document page content & metadata

First we prepare an example table with non-default schema, and populate it with some arbitary data.

import sqlalchemy

with engine.connect() as conn:
    conn.execute(sqlalchemy.text(f'DROP TABLE IF EXISTS "{TABLE_NAME}"'))
    conn.commit()
    conn.execute(
        sqlalchemy.text(
            f"""
            IF NOT EXISTS (SELECT * FROM sys.objects WHERE object_id = OBJECT_ID(N'[dbo].[{TABLE_NAME}]') AND type in (N'U'))
                BEGIN
                    CREATE TABLE [dbo].[{TABLE_NAME}](
                        fruit_id INT IDENTITY(1,1) PRIMARY KEY,
                        fruit_name VARCHAR(100) NOT NULL,
                        variety VARCHAR(50),
                        quantity_in_stock INT NOT NULL,
                        price_per_unit DECIMAL(6,2) NOT NULL,
                        organic BIT NOT NULL
                    )
                END
            """
        )
    )
    conn.execute(
        sqlalchemy.text(
            f"""
            INSERT INTO "{TABLE_NAME}" (fruit_name, variety, quantity_in_stock, price_per_unit, organic)
            VALUES
                ('Apple', 'Granny Smith', 150, 0.99, 1),
                ('Banana', 'Cavendish', 200, 0.59, 0),
                ('Orange', 'Navel', 80, 1.29, 1);
            """
        )
    )
    conn.commit()

If we still load langchain documents with default parameters of MSSQLLoader from this example table, the page_content of loaded documents will be the first column of the table, and metadata will be consisting of key-value pairs of all the other columns.

loader = MSSQLLoader(
    engine=engine,
    table_name=TABLE_NAME,
)
loader.load()

We can specify the content and metadata we want to load by setting the content_columns and metadata_columns when initializing the MSSQLLoader.

1. content_columns: The columns to write into the page_content of the document.
2. metadata_columns: The columns to write into the metadata of the document.

For example here, the values of columns in content_columns will be joined together into a space-separated string, as page_content of loaded documents, and metadata of loaded documents will only contain key-value pairs of columns specified in metadata_columns.

loader = MSSQLLoader(
    engine=engine,
    table_name=TABLE_NAME,
    content_columns=[
        "variety",
        "quantity_in_stock",
        "price_per_unit",
        "organic",
    ],
    metadata_columns=["fruit_id", "fruit_name"],
)
loader.load()

Save document with customized page content & metadata

In order to save langchain document into table with customized metadata fields. We need first create such a table via MSSQLEngine.init_document_table(), and specify the list of metadata_columns we want it to have. In this example, the created table will have table columns:

• description (type: text): for storing fruit description.
• fruit_name (type text): for storing fruit name.
• organic (type tinyint(1)): to tell if the fruit is organic.
• other_metadata (type: JSON): for storing other metadata information of the fruit.

We can use the following parameters with MSSQLEngine.init_document_table() to create the table:

1. table_name: The name of the table within the Cloud SQL database to store langchain documents.
2. metadata_columns: A list of sqlalchemy.Column indicating the list of metadata columns we need.
3. content_column: The name of column to store page_content of langchain document. Default: page_content.
4. metadata_json_column: The name of JSON column to store extra metadata of langchain document. Default: langchain_metadata.

engine.init_document_table(
    TABLE_NAME,
    metadata_columns=[
        sqlalchemy.Column(
            "fruit_name",
            sqlalchemy.UnicodeText,
            primary_key=False,
            nullable=True,
        ),
        sqlalchemy.Column(
            "organic",
            sqlalchemy.Boolean,
            primary_key=False,
            nullable=True,
        ),
    ],
    content_column="description",
    metadata_json_column="other_metadata",
    overwrite_existing=True,
)

Save documents with MSSQLDocumentSaver.add_documents(<documents>). As you can see in this example,

• document.page_content will be saved into description column.
• document.metadata.fruit_name will be saved into fruit_name column.
• document.metadata.organic will be saved into organic column.
• document.metadata.fruit_id will be saved into other_metadata column in JSON format.

test_docs = [
    Document(
        page_content="Granny Smith 150 0.99",
        metadata={"fruit_id": 1, "fruit_name": "Apple", "organic": 1},
    ),
]
saver = MSSQLDocumentSaver(
    engine=engine,
    table_name=TABLE_NAME,
    content_column="description",
    metadata_json_column="other_metadata",
)
saver.add_documents(test_docs)

with engine.connect() as conn:
    result = conn.execute(sqlalchemy.text(f'select * from "{TABLE_NAME}";'))
    print(result.keys())
    print(result.fetchall())

Delete documents with customized page content & metadata

We can also delete documents from table with customized metadata columns via MSSQLDocumentSaver.delete(<documents>). The deletion criteria is:

A row should be deleted if there exists a document in the list, such that

• document.page_content equals row[page_content]
• For every metadata field k in document.metadata
  • document.metadata[k] equals row[k] or document.metadata[k] equals row[langchain_metadata][k]
• There no extra metadata field presents in row but not in document.metadata.

loader = MSSQLLoader(engine=engine, table_name=TABLE_NAME)
docs = loader.load()
print("Documents before delete:", docs)
saver.delete(docs)
print("Documents after delete:", loader.load())

Related

• Document loader conceptual guide
• Document loader how-to guides