invoke_endpoint error in Lambda: StreamingBody is not JSON serializable

Was this because I didn't attach the necessary SageMaker policies to the IAM role? I only added this policy:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Sid": "VisualEditor0",
            "Effect": "Allow",
            "Action": "sagemaker:InvokeEndpoint",
            "Resource": "*"
        }
    ]
}

Hi aurelius,
Do you have further code after the response = runtime.invoke_endpoint(..) line? the error message says problem with serialization of the response object.

Your role permission looks fine.

Thank you,
Arun

I am facing similar issue, in sagemaker jupyter notebook instance the endpoint is invoked successfully and I am able to get back the inference results.

Below is my Lambda function that i used to invoke the endpoint, but I am facing the following error,

import json 
import io
import boto3 

client = boto3.client('runtime.sagemaker')

def lambda_handler(event, context):
    print("Received event: " + json.dumps(event, indent=2))
    
    #data = json.loads(json.dumps(event))
    #payload = data['data']
    print(json.dumps(event))
    
    response = client.invoke_endpoint(EndpointName='linear-learner-2019-12-12-16-20-56-788',
                                  ContentType='application/json',
                                  Body=(json.dumps(event)))
    return response

Output:

Function Logs:
START RequestId: 7f4c7589-b70f-4af8-834c-89a1a1fbe5e5 Version: $LATEST
Received event: {
  "instances": [
    {
      "features": [
        0.1,
        0.2
      ]
    }
  ]
}
{"instances": [{"features": [0.1, 0.2]}]}
An error occurred during JSON serialization of response: <botocore.response.StreamingBody object at 0x7f53918e2828> is not JSON serializable

Please help me resolve the issue, you can see the JSON input passed to the function. Not sure what is going wrong here. I even checked the cloudwatch logs not able to identify the origin of the issue.

Thanks in advance,
Arun

Edited by: NMAK on Dec 18, 2019 5:54 AM

I call Sagemaker from Lambda using a slightly different approach in terms of data structures:

final_data = ','.join(ordered_data.iloc[0].astype(str).values.tolist())
runtime = boto3.client('runtime.sagemaker')
response = runtime.invoke_endpoint(EndpointName='whatever-endpoint', 
                                           ContentType='text/csv',
                                           Body=final_data)
result = json.loads(response['Body'].read().decode())

I start with a dataframe of one row containing all the data.

hello Javierlopez,

Thanks for your response, I really appreciate it.

I tried the below code and still it gives me same error

    payload = ','.join(str(item) for item in data['instances'][0]['data']['features'])
    #payload=bytearray(payload)
    print(payload)
    
    response = client.invoke_endpoint(EndpointName='linear-learner-2019-12-12-16-20-56-788',
                                  ContentType='text/csv',
                                  Body=payload)
    return response

below is output/error for the above code.

Function Logs:
START RequestId: 73daa03e-dec2-4d37-b779-72c1f70a7142 Version: $LATEST
Received event: {
  "instances": [
    {
      "data": {
        "features": [
          0.1,
          0.2
        ]
      }
    }
  ]
}
0.1,0.2 # this is the payload sent for inference.
An error occurred during JSON serialization of response: <botocore.response.StreamingBody object at 0x7f77555e59e8> is not JSON serializable

I really don't understand the logic behind this on what format the function expects the input, my colleague ran a different model with same JSON format I used and it works fine. I believe you used this is inference format for XGBoost. I am not able to find any documentation on linear-learner sample inference requests using lambda.

Could you please point me in the right direction.

Regards,
Arun

Edited by: NMAK on Dec 20, 2019 3:04 AM