Skip to content

Latest commit

 

History

History

opensearch-serverless

README.md

Knowledge Base for Amazon Bedrock using Amazon OpenSearch Serverless

This is a complete setup for automatic deployment of Knowledge Bases for Amazon Bedrock using Amazon OpenSearch Serverless as a vector store.

Following resources will get created and deployed:

  • AWS IAM role
  • Amazon Open Search Serverless Collection and Index
  • Set up Data Source and Knowledge Base for Amazon Bedrock

The cdk.json file tells the CDK Toolkit how to execute your app.

This project is set up like a standard Python project. The initialization process also creates a virtualenv within this project, stored under the .venv directory. To create the virtualenv it assumes that there is a python3 (or python for Windows) executable in your path with access to the venv package. If for any reason the automatic creation of the virtualenv fails, you can create the virtualenv manually.

To manually create a virtualenv on MacOS and Linux:

$ python3 -m venv .venv

After the init process completes and the virtualenv is created, you can use the following step to activate your virtualenv.

$ source .venv/bin/activate

If you are a Windows platform, you would activate the virtualenv like this:

% .venv\Scripts\activate.bat

Once the virtualenv is activated, you can install the required dependencies.

(.venv) $ pip install -r requirements.txt

To add additional dependencies, for example other CDK libraries, just add them to your setup.py file and rerun the pip install -r requirements.txt command.

Prerequisites

Amazon S3 Bucket for a data source for Knowledge Base

  • You already have s3 bucket where your documents are stored.
    (e.g., aws s3 mb s3://bedrock-kb-us-east-1-123456789012 --region us-east-1)
  • The documents must be in one of the formats listed in here.

Prepare Python Packages for AWS Lambda Layer

Before synthesizing the CloudFormation, you first create a python package to regisiter with AWS Lambda Layer. Then you upload the python package into S3 (e.g., s3-bucket-lambda-layer-lib)

For more information about how to create a python package for AWS Lambda Layer, see References.

Set up cdk.context.json

Then, you should set approperly the cdk context configuration file, cdk.context.json.

For example,

{
  "knowledge_base_for_bedrock": {
    "name": "kb-demo",
    "description": "knowledge base description",
    "knowledge_base_configuration": {
      "vector_knowledge_base_configuration": {
        "embedding_model_arn": "arn:aws:bedrock:us-east-1::foundation-model/amazon.titan-embed-text-v1"
      }
    },
    "storage_configuration": {
      "opensearch_serverless_configuration": {
        "field_mapping": {
          "metadata_field": "metadata_field",
          "text_field": "text",
          "vector_field": "vector_field"
        },
        "vector_index_name": "embedding_vectors"
      }
    }
  },
  "knowledge_base_data_source_configuration": {
    "name": "kb-vector-db",
    "data_deletion_policy": "RETAIN",
    "description": "data source description",
    "s3_configuration": {
      "bucket_arn": "arn:aws:s3:::bedrock-kb-us-east-1-123456789012"
    },
    "chunking_configuration": {
      "chunking_strategy": "FIXED_SIZE",
      "fixed_size_chunking_configuration": {
        "max_tokens": 512,
        "overlap_percentage": 20
      }
    }
  },
  "opensearch_collection_name": "kb-vector-db",
  "lambda_layer_lib_s3_path": "s3://lambda-layer-resources/pylambda-layer/opensearch-py-sdk-lib.zip"
}

⚠️ It would be better NOT TO USE metadata for metadata_field in OpenSearch serverless field mapping. The popular LLM application frameworks like LangChain, LlamaIndex use metadata with data type other than text for OpenSearch field mapping. So to avoid conflicts when using the popular LLM frameworks, be careful to use metadata field name.

At this point you can now synthesize the CloudFormation template for this code.

(.venv) $ export CDK_DEFAULT_ACCOUNT=$(aws sts get-caller-identity --query Account --output text)
(.venv) $ export CDK_DEFAULT_REGION=$(aws configure get region)
(.venv) $ cdk synth --all

Now you can deploy all the CDK stacks at once like this:

(.venv) $ cdk deploy --require-approval never --all

Clean Up

Delete the CloudFormation stacks by running the below command.

(.venv) $ cdk destroy --all

Useful commands

  • cdk ls list all stacks in the app
  • cdk synth emits the synthesized CloudFormation template
  • cdk deploy deploy this stack to your default AWS account/region
  • cdk diff compare deployed stack with current state
  • cdk docs open CDK documentation

Enjoy!

References

  • Amazon Bedrock Workshop - Hands-on labs using Amazon Bedrock APIs, SDKs, and open-source software, such as LangChain and FAISS, to implement the most common Generative AI usage patterns (e.g., summarizing text, answering questions, building chatbots, creating images, and generating code).
  • Building with Amazon Bedrock and LangChain - Hands-on labs using LangChain to build generative AI prototypes with Amazon Bedrock.
  • Amazon Bedrock Samples - Pre-built examples to help customers get started with the Amazon Bedrock service.
  • AWS Lambda Layers

    • How to create a python package to register with AWS Lambda layer (e.g., elasticsearch, pytz) on Amazon Linux

      ⚠️ You should create the python package on Amazon Linux, otherwise create it using a simulated Lambda environment with Docker.

       $ python3 -m venv opensearch-py-lib
 $ cd opensearch-py-lib
 $ source bin/activate
 (opensearch-py-lib) $ mkdir -p python_modules
 (opensearch-py-lib) $ pip install opensearch-py==2.3.1 cfnresponse==1.1.2 urllib3==1.26.18 -t python_modules
 (opensearch-py-lib) $ mv python_modules python
 (opensearch-py-lib) $ zip -r opensearch-py-lib.zip python/
 (opensearch-py-lib) $ aws s3 mb s3://my-bucket-for-lambda-layer-packages
 (opensearch-py-lib) $ aws s3 cp opensearch-py-lib.zip s3://my-bucket-for-lambda-layer-packages/var/
 (opensearch-py-lib) $ deactivate

    • How to create a Lambda layer using a simulated Lambda environment with Docker

       $ cat < requirements.txt
 > opensearch-py==2.3.1
 > cfnresponse==1.1.2
 > urllib3==1.26.18
 > EOF
 $ docker run -v "$PWD":/var/task "public.ecr.aws/sam/build-python3.10" /bin/sh -c "pip install -r requirements.txt -t python/lib/python3.10/site-packages/; exit"
 $ zip -r opensearch-py-lib.zip python > /dev/null
 $ aws s3 mb s3://my-bucket-for-lambda-layer-packages
 $ aws s3 cp opensearch-py-lib.zip s3://my-bucket-for-lambda-layer-packages/var/