Accessing Lambda and S3 in deployments

You can configure Modelbit deployments to have access to your AWS resources. First, set up a Trust relationship with Modelbit by following the instructions in the Integrations tab within Settings. Then return here to use this sample code.

Calling AWS Lambda

Copy-paste the following sample code into your notebook:

import boto3

def invoke_remote_lambda(remote_role_arn, remote_func_arn, request_payload):
    sts_client = boto3.client('sts')
    remote_role = sts_client.assume_role(
        RoleArn=remote_role_arn,
        RoleSessionName="modelbit_deployment")
    remote_creds = remote_role['Credentials']
    remote_client = boto3.client(
        "lambda",
        region_name=remote_func_arn.split(":")[3],
        aws_access_key_id=remote_creds['AccessKeyId'],
        aws_secret_access_key=remote_creds['SecretAccessKey'],
        aws_session_token=remote_creds['SessionToken'])
    response = remote_client.invoke(
        FunctionName=remote_func_arn,
        InvocationType='RequestResponse',
        Payload=request_payload)
    return response['Payload'].read().decode()

Then write a deployment function, like what follows, to call your Lambda. Change the remote_role_arn to the ARN of the role with the Trust relationship, and remote_func_arn to the ARN of your Lambda.

import json

def predict_with_remote_lambda(input_values):
    lambda_response = invoke_remote_lambda(
        remote_role_arn="arn:aws:iam::111111111111:role/my-lambda-role",
        remote_func_arn="arn:aws:lambda:us-east-2:111111111111:function:my-lambda-func",
        request_payload=json.dumps(input_values))
    return json.loads(lambda_response)

# Test your lambda call
predict_with_remote_lambda({ "foo": 1, "bar": 2 })

# Deploy to Modelbit!
mb.deploy(predict_with_remote_lambda)

Now the predict_with_remote_lambda deployment is able to call your Lambda function! Pro tip: Set up your notebook to have access to AWS and allowed to assume the same role with the Trust relationship. That will make development and debugging much easier.

Reading from AWS S3

Copy-paste the following sample code into your notebook:

import boto3

def get_s3_data(remote_role_arn, s3bucket, s3key):
    sts_client = boto3.client('sts')
    remote_role = sts_client.assume_role(
        RoleArn=remote_role_arn,
        RoleSessionName="modelbit_deployment")
    remote_creds = remote_role['Credentials']
    remote_client = boto3.client(
        "s3",
        aws_access_key_id=remote_creds['AccessKeyId'],
        aws_secret_access_key=remote_creds['SecretAccessKey'],
        aws_session_token=remote_creds['SessionToken'])
    response = remote_client.get_object(
        Bucket=s3bucket,
        Key=s3key)
    return response['Body'].read().decode()

Then write a deployment function, like what follows, to read your S3 data. Change the remote_role_arn to the ARN of the role with the Trust relationship, and s3bucket and s3key to the values of your S3 data file.

def predict_with_s3_data():
    s3data = get_s3_data(
        remote_role_arn="arn:aws:iam::111111111111:role/my-s3-reading-role",
        s3bucket="my-s3-bucket",
        s3key="my-s3-key")
    return f"The s3 file's length is {len(s3data)}"

# Test your lambda call
predict_with_s3_data()

# Deploy to Modelbit!
mb.deploy(predict_with_s3_data)

Now the predict_with_s3_data deployment is able to read your S3 data! Pro tip: Set up your notebook to have access to AWS and allowed to assume the same role with the Trust relationship. That will make development and debugging much easier.