Getting started with Spanner Vector Search

1. Introduction

Spanner is a fully managed horizontally scalable, globally distributed, database service that is great for both relational and non-relational operational workloads.

With the release of exact K-nearest neighbor functionality, Spanner is now also a highly scalable vector database, enabling you to perform similarity or semantic search and implement retrieval augmented generation (RAG) in GenAI applications at scale. Spanner's vector search queries return fresh real-time data as soon as transactions are committed, just like any other query on your operational data.

In this lab, you'll walk through setting up the basic features required to leverage Spanner to perform vector search, and access embedding and LLM models from VertexAI's model garden using SQL.

The architecture would look like this:

Note: This lab focuses on the right portion of the diagram, so you'll only be deploying a Spanner instance and the VertexAI integrations.

What you'll build

As part of this lab, you will:

• Create a Spanner instance
• Set up Spanner's database schema to integrate with embedding and LLM models in VertexAI
• Load a retail data set
• Issue similarity search queries against the dataset
• Provide context to the LLM model to generate product-specific recommendations.

What you'll learn

• How to setup a Spanner instance
• How to integrate with VertexAI
• How to use Spanner to perform vector search to find similar items in a retail dataset

What you'll need

• A Google Cloud project that is connected to a billing account.
• A web browser, such as Chrome or Firefox.

2. Setup and requirements

Create a project

If you don't already have a Google Account (Gmail or Google Apps), you must create one. Sign-in to Google Cloud Platform console ( console.cloud.google.com) and create a new project.

If you already have a project, click on the project selection pull down menu in the upper left of the console:

and click the ‘NEW PROJECT' button in the resulting dialog to create a new project:

If you don't already have a project, you should see a dialog like this to create your first one:

The subsequent project creation dialog allows you to enter the details of your new project:

Remember the project ID, which is a unique name across all Google Cloud projects (the name above has already been taken and will not work for you, sorry!). It will be referred to later in this codelab as PROJECT_ID.

Next, if you haven't already done so, you'll need to enable billing in the Developers Console in order to use Google Cloud resources and enable the Spanner API.

Running through this codelab shouldn't cost you more than a few dollars, but it could be more if you decide to use more resources or if you leave them running (see "cleanup" section at the end of this document). Google Cloud Spanner pricing is documented here.

New users of Google Cloud Platform are eligible for a $300 free trial, which should make this codelab entirely free of charge.

Google Cloud Shell Setup

While Google Cloud and Spanner can be operated remotely from your laptop, in this codelab we will be using Google Cloud Shell, a command line environment running in the Cloud.

This Debian-based virtual machine is loaded with all the development tools you'll need. It offers a persistent 5GB home directory and runs in Google Cloud, greatly enhancing network performance and authentication. This means that all you will need for this codelab is a browser (yes, it works on a Chromebook).

1. To activate Cloud Shell from the Cloud Console, simply click Activate Cloud Shell (it should only take a few moments to provision and connect to the environment).

Once connected to Cloud Shell, you should see that you are already authenticated and that the project is already set to your PROJECT_ID.

gcloud auth list

Command output

Credentialed accounts:
 - <myaccount>@<mydomain>.com (active)

Note: The gcloud command-line tool comes preinstalled in Cloud Shell. For more information, see gcloud command-line tool overview.

gcloud config list project

Command output

[core]
project = <PROJECT_ID>

If, for some reason, the project is not set, simply issue the following command:

gcloud config set project <PROJECT_ID>

Looking for your PROJECT_ID? Check out what ID you used in the setup steps or look it up in the Cloud Console dashboard:

Cloud Shell also sets some environment variables by default, which may be useful as you run future commands.

echo $GOOGLE_CLOUD_PROJECT

Command output

<PROJECT_ID>

Note: When you run gcloud on your own machine, the config settings would've been persisted across sessions. But in Cloud Shell, you will need to set this for every new session or reconnection.

Enable the Spanner API

gcloud services enable spanner.googleapis.com

Summary

In this step you have set up your project if you didn't already have one, activated cloud shell, and enabled the required APIs.

Next up

Next, you will set up the Spanner instance and database.

3. Create a Spanner instance and database

Create the Spanner instance

In this step we set up our Spanner Instance for the codelab. To do this, open up Cloud Shell and run this command:

export SPANNER_INSTANCE_ID=retail-demo
gcloud spanner instances create $SPANNER_INSTANCE_ID \
--config=regional-us-central1 \
--description="spanner AI retail demo" \
--nodes=1

Note: It can take a few moments to create the instance.

Note: This command uses regional-us-central1 as the region. You can select a different region based on region configurations available to Spanner that match regions VertexAI endpoints are located.

Command output:

$ gcloud spanner instances create $SPANNER_INSTANCE_ID \
--config=regional-us-central1 \
--description="spanner AI retail demo" \
--nodes=1
Creating instance...done.  

Create the database

Once your instance is running, you can create the database. Spanner allows for multiple databases on a single instance.

The database is where you define your schema. You can also control who has access to the database, set up custom encryption, configure the optimizer, and set the retention period.

To create the database, again use the gcloud command line tool:

export SPANNER_DATABASE=cymbal-bikes
gcloud spanner databases create $SPANNER_DATABASE \
 --instance=$SPANNER_INSTANCE_ID

Command output:

$ gcloud spanner databases create $SPANNER_DATABASE \
 --instance=$SPANNER_INSTANCE_ID
Creating database...done.

Summary

In this step you have created the Spanner instance and database.

Next up

Next, you will set up the Spanner schema and data.

4. Load Cymbal schema and data

Create the Cymbal schema

To set up the schema, navigate to Spanner Studio:

Note: Spanner Studio can be found by clicking into the retail-demo Spanner instance and then into the cymbal-bikes database.

There are two parts to the schema. First, you want to add the products table. Copy and paste this statement in the empty tab.

For the schema, copy and paste this DDL into the box:

CREATE TABLE products (
categoryId INT64 NOT NULL,
productId INT64 NOT NULL,
productName STRING(MAX) NOT NULL,
productDescription STRING(MAX) NOT NULL,
productDescriptionEmbedding ARRAY<FLOAT64>,
createTime TIMESTAMP NOT NULL OPTIONS (
allow_commit_timestamp = true
),
inventoryCount INT64 NOT NULL,
priceInCents INT64,
) PRIMARY KEY(categoryId, productId);

Then, click the run button and wait a few seconds for your schema to be created.

Next, you will create the two models and configure them to VertexAI model endpoints.

The first model is an Embedding model that is used to generate embeddings from text, and the second is an LLM model used to generate responses based on the data in Spanner.

Paste the following schema into a new tab in Spanner Studio:

Warning: Change the <PROJECT_ID> in both statements to your project ID you created at the beginning of the codelab.

Note: The schema below uses the us-central1 VertexAI endpoint. Change this in both EmbeddingsModel and LLMModel definitions if you chose a different Spanner region when creating the instance.

CREATE MODEL EmbeddingsModel INPUT(
content STRING(MAX),
) OUTPUT(
embeddings STRUCT<statistics STRUCT<truncated BOOL, token_count FLOAT64>, values ARRAY<FLOAT64>>,
) REMOTE OPTIONS (
endpoint = '//aiplatform.googleapis.com/projects/<PROJECT_ID>/locations/us-central1/publishers/google/models/textembedding-gecko@003'
);

CREATE MODEL LLMModel INPUT(
prompt STRING(MAX),
) OUTPUT(
content STRING(MAX),
) REMOTE OPTIONS (
endpoint = '//aiplatform.googleapis.com/projects/<PROJECT_ID>/locations/us-central1/publishers/google/models/text-bison@001',
default_batch_size = 1
);

Then, click the run button and wait a few seconds for your models to be created.

In the left pane of Spanner Studio you should see the following tables and models:

Load the data

Now, you will want to insert some products into your database. Open up a new tab in Spanner Studio, then copy and paste the following insert statements:

INSERT INTO products (categoryId, productId, productName, productDescription, createTime, inventoryCount, priceInCents)
VALUES (1, 1, "Cymbal Helios Helmet", "Safety meets style with the Cymbal children's bike helmet. Its lightweight design, superior ventilation, and adjustable fit ensure comfort and protection on every ride. Stay bright and keep your child safe under the sun with Cymbal Helios!", PENDING_COMMIT_TIMESTAMP(), 100, 10999),
(1, 2, "Cymbal Sprout", "Let their cycling journey begin with the Cymbal Sprout, the ideal balance bike for beginning riders ages 2-4 years. Its lightweight frame, low seat height, and puncture-proof tires promote stability and confidence as little ones learn to balance and steer. Watch them sprout into cycling enthusiasts with Cymbal Sprout!", PENDING_COMMIT_TIMESTAMP(), 10, 13999),
(1, 3, "Cymbal Spark Jr.", "Light, vibrant, and ready for adventure, the Spark Jr. is the perfect first bike for young riders (ages 5-8). Its sturdy frame, easy-to-use brakes, and puncture-resistant tires inspire confidence and endless playtime. Let the spark of cycling ignite with Cymbal!", PENDING_COMMIT_TIMESTAMP(), 34, 13900),
(1, 4, "Cymbal Summit", "Conquering trails is a breeze with the Summit mountain bike. Its lightweight aluminum frame, responsive suspension, and powerful disc brakes provide exceptional control and comfort for experienced bikers navigating rocky climbs or shredding downhill. Reach new heights with Cymbal Summit!", PENDING_COMMIT_TIMESTAMP(), 0, 79999),
(1, 5, "Cymbal Breeze", "Cruise in style and embrace effortless pedaling with the Breeze electric bike. Its whisper-quiet motor and long-lasting battery let you conquer hills and distances with ease. Enjoy scenic rides, commutes, or errands with a boost of confidence from Cymbal Breeze!", PENDING_COMMIT_TIMESTAMP(), 72, 129999),
(1, 6, "Cymbal Trailblazer Backpack", "Carry all your essentials in style with the Trailblazer backpack. Its water-resistant material, multiple compartments, and comfortable straps keep your gear organized and accessible, allowing you to focus on the adventure. Blaze new trails with Cymbal Trailblazer!", PENDING_COMMIT_TIMESTAMP(), 24, 7999),
(1, 7, "Cymbal Phoenix Lights", "See and be seen with the Phoenix bike lights. Powerful LEDs and multiple light modes ensure superior visibility, enhancing your safety and enjoyment during day or night rides. Light up your journey with Cymbal Phoenix!", PENDING_COMMIT_TIMESTAMP(), 87, 3999),
(1, 8, "Cymbal Windstar Pump", "Flat tires are no match for the Windstar pump. Its compact design, lightweight construction, and high-pressure capacity make inflating tires quick and effortless. Get back on the road in no time with Cymbal Windstar!", PENDING_COMMIT_TIMESTAMP(), 36, 24999),
(1, 9,"Cymbal Odyssey Multi-Tool","Be prepared for anything with the Odyssey multi-tool. This handy gadget features essential tools like screwdrivers, hex wrenches, and tire levers, keeping you ready for minor repairs and adjustments on the go. Conquer your journey with Cymbal Odyssey!", PENDING_COMMIT_TIMESTAMP(), 52, 999),
(1, 10,"Cymbal Nomad Water Bottle","Stay hydrated on every ride with the Nomad water bottle. Its sleek design, BPA-free construction, and secure lock lid make it the perfect companion for staying refreshed and motivated throughout your adventures. Hydrate and explore with Cymbal Nomad!", PENDING_COMMIT_TIMESTAMP(), 42, 1299);

Click the run button to insert the data.

Summary

In this step you created the schema and loaded some basic data into the cymbal-bikes database.

Next up

Next, you will integrate with the Embedding model to generate embeddings for the product descriptions, as well as convert a textual search request into an embedding to search for relevant products.

5. Work with embeddings

Generate vector embeddings for product descriptions

For similarity search to work on the products, you need to generate embeddings for the product descriptions.

With the EmbeddingsModel created in the schema, this is a simple UPDATE DML statement.

UPDATE products p1
SET productDescriptionEmbedding =
(SELECT embeddings.values from ML.PREDICT(MODEL EmbeddingsModel,
(SELECT productDescription as content FROM products p2 where p2.productId=p1.productId)))
WHERE categoryId=1;

Click the run button to update the product descriptions.

Note: For more performance sensitive workloads, it is best to generate the embeddings using ML.PREDICT in a read-only transaction, and then write them into Spanner using a read-write transaction.

Using vector search

In this example, you will provide a natural language search request via a SQL query. This query will turn the search request into an embedding, then search for similar results based on the stored embeddings of the product descriptions that were generated in the previous step.

-- Use Spanner's vector search, and integration with embedding and LLM models to
-- return items that are semantically relevant and available in inventory based on
-- real-time data.


SELECT productName, productDescription, inventoryCount, COSINE_DISTANCE(
productDescriptionEmbedding,
(   SELECT embeddings.values
FROM ML.PREDICT(
MODEL EmbeddingsModel,
(SELECT "I'd like to buy a starter bike for my 3 year old child" as content)
)
)
) as distance
FROM products
WHERE inventoryCount > 0
ORDER BY distance
LIMIT 5;

Click the run button to find the similar products. The results should look like this:

Notice that additional filters are used on the query, such as only being interested in products that are in stock (inventoryCount > 0).

Summary

In this step, you created product description embeddings and a search request embedding using SQL, leveraging Spanner's integration with models in VertexAI. You also performed a vector search to find similar products that match the search request.

Next Steps

Next, let's use the search results to feed into an LLM to generate a custom response for each product.

6. Work with an LLM

Spanner makes it easy to integrate with LLM models served from VertexAI. This allows developers to use SQL to interface with LLMs directly, rather than requiring the application to perform the logic.

For example, we have the results from the previous SQL query from the user "I'd like to buy a starter bike for my 3 year old child".

The developer would like to provide a response for each result on whether the product is a good fit for the user, using the following prompt:

"Answer with ‘Yes' or ‘No' and explain why: Is this a good fit for me? I'd like to buy a starter bike for my 3 year old child"

Here is the query you can use:

-- Use an LLM to analyze this list and provide a recommendation on whether each
-- product is a good fit for the user. We use the vector search and real time
-- inventory data to first filter the products to reduce the size of the prompt to
-- the LLM.
SELECT productName, productDescription, inventoryCount, content AS LLMResponse
FROM ML.PREDICT(
MODEL LLMModel,
(   SELECT
inventoryCount,
productName,
productDescription,
CONCAT(
"Answer with ‘Yes' or ‘No' and explain why: Is this a good fit for me?",
"I'd like to buy a starter bike for my 3 year old child \n",
"Product Name: ", productName, "\n",
"Product Description:", productDescription) AS prompt,
FROM products
WHERE inventoryCount > 0
ORDER by COSINE_DISTANCE(
productDescriptionEmbedding,
(   SELECT embeddings.values
FROM ML.PREDICT(
MODEL EmbeddingsModel,
( SELECT "I'd like to buy a starter bike for my 3 year old child" as content)
)
)
) LIMIT 5
),
STRUCT(256 AS maxOutputTokens)
);

Note: We provide the same query from the previous step to the LLM model to generate the list of products we want a response for.

Click the run button to issue the query. The results should look like this:

The first product is a fit for a 3-year old because of the age range in the product description (2-4 year olds). The other products are not great fits.

Summary

Next Steps

Next, it is time to clean up!

7. Cleaning up (optional)

To clean up, just go into the Cloud Spanner section of the Cloud Console and delete the 'retail-demo' instance we created in the codelab.

8. Congratulations!

Congratulations, you have successfully performed a similarity search using Spanner's built-in vector search. Additionally, you saw how easy it is to work with embedding and LLM models to provide generative AI functionality directly using SQL.

What's next?

Learn more about Spanner's exact nearest neighbor (KNN vector search) feature here: https://cloud.google.com/spanner/docs/find-k-nearest-neighbors

You can also read more about how to perform online predictions with SQL using Spanner's VertexAI integration here: https://cloud.google.com/spanner/docs/ml