Skip to main content

Google alloydb for postgresql integration

还没翻译译自英文（docs.langchain.com）本页 48 段都还没译

AlloyDB is a fully managed relational database service that offers high performance, seamless integration, and impressive scalability. AlloyDB is 100% compatible with PostgreSQL. Extend your database application to build AI-powered experiences leveraging AlloyDB's LangChain integrations.

This notebook goes over how to use AlloyDB for PostgreSQL to store vector embeddings with the AlloyDBVectorStore class.

AlloyDB AI Features: This integration natively supports the alloydb_ai_nl and google_ml_integration extensions, providing database-native AI capabilities:

  • Zero-Knob ScaNN: Auto-tuning, self-managing vector indices via ScaNNIndex(mode="AUTO").
  • Columnar Engine & Vector Assist: Accelerate vector queries dynamically using the ColumnarEngine.
  • Semantic Reranking: Re-rank retrieved documents natively in the database using the AlloyDBDocumentCompressor (backed by google_ml.rank).
  • Natural Language to SQL: Build native SQL Agents using AlloyDBToolkit.
  • In-Database AI Tools: Utilize LLM functions directly in SQL (AlloyDBIfTool, AlloyDBSummaryTool, AlloyDBSentimentTool).
  • Multimodal Embeddings: Generate vectors natively in the database (e.g., using aembed_image()).

Learn more about the package on GitHub.

Open In Colab

Before you begin

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

🦜🔗 Library installation

Install the integration library, langchain-google-alloydb-pg, and the library for the embedding service, langchain-google-vertexai.

pip install -qU  langchain-google-alloydb-pg langchain-google-vertexai

Colab only: Uncomment the following cell to restart the kernel or use the button to restart the kernel. For Gemini Enterprise Agent Platform 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.

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:

# @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}

Basic usage

Set AlloyDB database values

Find your database values, in the AlloyDB Instances page.

# @title Set Your Values Here { display-mode: "form" }
REGION = "us-central1"  # @param {type: "string"}
CLUSTER = "my-cluster"  # @param {type: "string"}
INSTANCE = "my-primary"  # @param {type: "string"}
DATABASE = "my-database"  # @param {type: "string"}
TABLE_NAME = "vector_store"  # @param {type: "string"}

AlloyDBEngine connection pool

One of the requirements and arguments to establish AlloyDB as a vector store is a AlloyDBEngine object. The AlloyDBEngine configures a connection pool to your AlloyDB database, enabling successful connections from your application and following industry best practices.

To create a AlloyDBEngine using AlloyDBEngine.from_instance() you need to provide only 5 things:

  1. project_id : Project ID of the Google Cloud Project where the AlloyDB instance is located.
  2. region : Region where the AlloyDB instance is located.
  3. cluster: The name of the AlloyDB cluster.
  4. instance : The name of the AlloyDB instance.
  5. database : The name of the database to connect to on the AlloyDB instance.

By default, IAM database authentication will be used as the method of database authentication. This library uses the IAM principal belonging to the Application Default Credentials (ADC) sourced from the environment.

Optionally, built-in database authentication using a username and password to access the AlloyDB database can also be used. Just provide the optional user and password arguments to AlloyDBEngine.from_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.

Note: This tutorial demonstrates the async interface. All async methods have corresponding sync methods.

from langchain_google_alloydb_pg import AlloyDBEngine

engine = await AlloyDBEngine.afrom_instance(
    project_id=PROJECT_ID,
    region=REGION,
    cluster=CLUSTER,
    instance=INSTANCE,
    database=DATABASE,
)

Initialize a table

The AlloyDBVectorStore class requires a database table. The AlloyDBEngine engine has a helper method init_vectorstore_table() that can be used to create a table with the proper schema for you.

await engine.ainit_vectorstore_table(
    table_name=TABLE_NAME,
    vector_size=768,  # Vector size for Gemini Enterprise Agent Platform model (textembedding-gecko@latest)
)

Create an embedding class instance

You can use any LangChain embeddings model. You may need to enable Gemini Enterprise Agent Platform API to use VertexAIEmbeddings. We recommend setting the embedding model's version for production, learn more about the Text embeddings models.

# enable Gemini Enterprise Agent Platform API
!gcloud services enable aiplatform.googleapis.com
from langchain_google_vertexai import VertexAIEmbeddings

embedding = VertexAIEmbeddings(
    model_name="textembedding-gecko@latest", project=PROJECT_ID
)

Initialize a default AlloyDBVectorStore

from langchain_google_alloydb_pg import AlloyDBVectorStore

store = await AlloyDBVectorStore.create(
    engine=engine,
    table_name=TABLE_NAME,
    embedding_service=embedding,
)

Add texts

import uuid

all_texts = ["Apples and oranges", "Cars and airplanes", "Pineapple", "Train", "Banana"]
metadatas = [{"len": len(t)} for t in all_texts]
ids = [str(uuid.uuid4()) for _ in all_texts]

await store.aadd_texts(all_texts, metadatas=metadatas, ids=ids)

Delete texts

await store.adelete([ids[1]])

Search for documents

query = "I'd like a fruit."
docs = await store.asimilarity_search(query)
print(docs)

Search for documents by vector

query_vector = embedding.embed_query(query)
docs = await store.asimilarity_search_by_vector(query_vector, k=2)
print(docs)

Add an index

Speed up vector search queries by applying a vector index. Learn more about vector indexes.

from langchain_google_alloydb_pg.indexes import IVFFlatIndex

index = IVFFlatIndex()
await store.aapply_vector_index(index)

Re-index

await store.areindex()  # Re-index using default index name

Remove an index

await store.adrop_vector_index()  # Delete index using default name

Create a custom vector store

A Vector Store can take advantage of relational data to filter similarity searches.

Create a table with custom metadata columns.

from langchain_google_alloydb_pg import Column

# Set table name
TABLE_NAME = "vectorstore_custom"

await engine.ainit_vectorstore_table(
    table_name=TABLE_NAME,
    vector_size=768,  # Gemini Enterprise Agent Platform model: textembedding-gecko@latest
    metadata_columns=[Column("len", "INTEGER")],
)


# Initialize AlloyDBVectorStore
custom_store = await AlloyDBVectorStore.create(
    engine=engine,
    table_name=TABLE_NAME,
    embedding_service=embedding,
    metadata_columns=["len"],
    # Connect to an existing VectorStore by customizing the table schema:
    # id_column="uuid",
    # content_column="documents",
    # embedding_column="vectors",
)

Search for documents with metadata filter

import uuid

# Add texts to the Vector Store
all_texts = ["Apples and oranges", "Cars and airplanes", "Pineapple", "Train", "Banana"]
metadatas = [{"len": len(t)} for t in all_texts]
ids = [str(uuid.uuid4()) for _ in all_texts]
await store.aadd_texts(all_texts, metadatas=metadatas, ids=ids)

# Use filter on search
docs = await custom_store.asimilarity_search_by_vector(query_vector, filter="len >= 6")

print(docs)
AI 翻译查看英文原文