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 BeginTo run this notebook, you will need to do the following:
After confirmed access to database in the runtime environment of this notebook, filling the following values and run the cell before running example scripts.
REGION = "us-central1"
INSTANCE = "test-instance"
DB_USER = "sqlserver"
DB_PASS = "password"
DATABASE = "test"
TABLE_NAME = "test-default"
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.
🔐 AuthenticationAuthenticate to Google Cloud as the IAM user logged into this notebook in order to access your Google Cloud Project.
from google.colab import auth
auth.authenticate_user()
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:
gcloud config list.gcloud projects list.
PROJECT_ID = "my-project-id"
!gcloud config set project {PROJECT_ID}
The langchain-google-cloud-sql-mssql package requires that you enable the Cloud SQL Admin API in your Google Cloud Project.
!gcloud services enable sqladmin.googleapis.com
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:
project_id : Project ID of the Google Cloud Project where the Cloud SQL instance is located.region : Region where the Cloud SQL instance is located.instance : The name of the Cloud SQL instance.database : The name of the database to connect to on the Cloud SQL instance.user : Database user to use for built-in database authentication and login.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 of default schema via MSSQLEngine.init_document_table(<table_name>). Table Columns:
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 langchain documents with MSSQLDocumentSaver.add_documents(<documents>). To initialize MSSQLDocumentSaver class you need to provide 2 things:
engine - An instance of a MSSQLEngine engine.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)
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:
engine - An instance of a MSSQLEngine engine.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)
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 documentsDelete 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())
First we prepare an example table with non-default schema, and populate it with some arbitrary 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.
content_columns: The columns to write into the page_content of the document.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()
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:
We can use the following parameters with MSSQLEngine.init_document_table() to create the table:
table_name: The name of the table within the Cloud SQL database to store langchain documents.metadata_columns: A list of sqlalchemy.Column indicating the list of metadata columns we need.content_column: The name of column to store page_content of langchain document. Default: page_content.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())
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]k in document.metadata
document.metadata[k] equals row[k] or document.metadata[k] equals row[langchain_metadata][k]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())
RetroSearch is an open source project built by @garambo | Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4