httpMethod KeyError on app.resolve(event, context)

[kaizennet]

I'm quite certain the error exists on my inexperience:

Using any of the aws_lambda_powertools BedrockAgentResolver online examples (direct from site) I always get the error
from the line
return app.resolve(event, context)

Status: Failed
Test Event Name: a1

Response:
{
  "errorMessage": "'httpMethod'",
  "errorType": "KeyError",
  "requestId": "746a79fe-79e6-48b8-aa52-1bd71341fc0a",
  "stackTrace": [
    "  File \"/var/task/lambda_function.py\", line 12, in lambda_handler\n    return app.resolve(event, context)\n",
    "  File \"/opt/python/aws_lambda_powertools/event_handler/api_gateway.py\", line 2083, in resolve\n    response = self._resolve().build(self.current_event, self._cors)\n",
    "  File \"/opt/python/aws_lambda_powertools/event_handler/api_gateway.py\", line 2177, in _resolve\n    method = self.current_event.http_method.upper()\n",
    "  File \"/opt/python/aws_lambda_powertools/utilities/data_classes/bedrock_agent_event.py\", line 82, in http_method\n    return self[\"httpMethod\"]\n",
    "  File \"/opt/python/aws_lambda_powertools/utilities/data_classes/common.py\", line 89, in __getitem__\n    return self._data[key]\n"
  ]
}

Using the public ARN (very convenient - thank you). Using AWS function editor interface. Python 3.12. Any suggestions greatly appreciated.

[leandrodamascena]

Hey @kaizennet! This looks like you're sending an incorrect payload. Can you help me to understand a little bit more how you're invoking this Lambda function? Maybe the payload? Please remove any sensitive data.

Thanks

[kaizennet]

@leandrodamascena You taking the time to respond is greatly appreciated. I do have lofty dreams of using my Bedrock agent with my Athena dataset (populated from glue). But for now I'm just trying to get anything to work.

The same thing happens on any of the powertools-lambda-python examples but for example:

1. Click on the "powertools-lambda-python" git project.

2. Scroll down to Event Handler: Agents for Amazon Bedrock

3. Under "Your first Agent" Copy / Paste 100% verbatim into AWS new AWS Lambda Function (python 3.12 using the arn:aws:lambda:us-east-1:017000801446:layer:AWSLambdaPowertoolsPythonV3-python312-x86_64:9 for lambda layer x_86_64)

from time import time

from aws_lambda_powertools import Logger, Tracer
from aws_lambda_powertools.event_handler import BedrockAgentResolver
from aws_lambda_powertools.utilities.typing import LambdaContext

tracer = Tracer()
logger = Logger()
app = BedrockAgentResolver()


@app.get("/current_time", description="Gets the current time in seconds")  
@tracer.capture_method
def current_time() -> int:
    return int(time())


@logger.inject_lambda_context
@tracer.capture_lambda_handler
def lambda_handler(event: dict, context: LambdaContext):
    return app.resolve(event, context)  

4. Create default Test event, Deploy, Test

If it isn't painfully obvious I'm inexperienced with both lambda and python.

If i comment out the last line "return app.resolve(event, context) " and replace it with something like "return 'hello" , no errors are generated.

Thank you.

[leandrodamascena]

Hi @kaizennet! Thanks for adding more information. Just to understand the goal here: are you trying to invoke a Lambda function with Bedrock resolver and with a fake payload? How do you expect to connect this Bedrock agent to Lambda? What are the routes you will need? Are you going to generate OpenAPI from this?

I can help you understand a little better how Powertools and Bedrock Agents work, but I think I need more explanation here because I'm not getting the whole picture.

[kaizennet]

@leandrodamascena Thanks once again. Far from this goal currently but:

My ultimate goal is a Text-2-SQL solution. Where:

1. A relational dataset is built through Athena. This is already accomplished. Kind of long winded road but the relational store starts in SQL Server on our domain, tables exported to csv, csv files uploaded to S3, S3 -> Glue, Glue builds the JSON metadata OK and finally Athena exposes our glue tables to Lambda. I have that part working OK either through pyathena(sqlalchemy) or boto3 client('athena).

2. Vision is to expose the lambda function through URL making a REST POST call passing a natural language question (parameter) from the user (related to the data set). Like "In 2023 what was the average measurement range value for location X".

3. This parameter gets passed on to LLM through Bedrock. The LLM takes the NL query and using the Athena data set does it's best to come up with a SQL query to return back to the user. We'd save this query to build a repository and maybe execute it back on the SQL Server (even though it's returning other variant than T-SQL - we can create some sort of parsing function if needed). We want the query as accurate as possible so hoping not to vectorize the data (like in a knowledge base, etc)

4. SQL query response is returned back to the user that made the lambda function URL request.

The 3rd step is what i'm banging my head as to the best solution . In Bedrock we have enabled a few of the Anthropic, Cohere + Amazon models enabled. Honestly i don't care at all what we use OpenAI, haiku, sonnet, R as long as it returns a reasonably accurate SQL query back - any suggestions greatly appreciated! OpenAI , i registered an account and got API key although only free tier, unsure if there's something else i need to do there.

powertools-lambda-python project seems like a good way to go to communicate with a bedrock agent. Unable to get any of the examples to work so far.

Thank you very much

[leandrodamascena]

Hey @kaizennet! I really sorry for the long delay in replying this, but I was on PTO! Do you think that make we schedule a call to talk about Bedrock Agent and how we can help with that?