Api-Gateway on The Edge Cases

Raising Exceptions from Lambda to API Gateway Caller

Wed, 05 Oct 2022 17:09:29 +0000

You can send exception feedback from a Lambda to an API Gateway API either by raising a Python exception and handling it via a mapping template, or return a specific set of values from the main handler function.

Returning from the Handler Function

You can return a dictionary that includes at least:

the desired status code
the message body (JSONified)
and whether or not the body is base64 encoded

return {  
    "statusCode": 404,  
    "body": simplejson.dumps(  
        f"User {username} not found"  
    ),  
    "isBase64Encoded": False,  
}

Raising an Exceptions

Using a mapping template allows you to simply use regex to associate an error message from a plain Python exception with a certain HTTP code. I don’t have any examples because I haven’t done it, preferring the previous method.

API Gateway URL Construction

Tue, 07 Jun 2022 16:39:00 +0000

This is the construction of the base URL of an API Gateway API: https://{restapi_id}.execute-api.{region}.amazonaws.com/{stage_name}/

Sources

AWS Docs - How to call API

AWS Lambda Return Values for API Triggering

Mon, 07 Feb 2022 21:41:05 +0000

A Lambda triggered by an API has to return a dictionary with certain keys and values in order to not cause an error in execution (e.g. “malformed lambda proxy response”). These are the status code, body (JSONified), and if it’s base64 encoded. Headers can also be added to this structure, but they don’t seem required.

Code Example

    return {  
      "statusCode": 200,  
   "body": json.dumps(f"User {username} not found"),  
   "isBase64Encoded": False  
  }

Sources

AWS API Gateway Changes Break Deployment

Mon, 07 Feb 2022 15:43:47 +0000

If you try to rename a parameter in a deployed API Gateway object via CDK, deployment will break (you won’t be able to deploy the change). This seems to do with how CloudFormation diffs are applied, and it will cause a conflict during the deployment process (see here). The workaround is to comment out all the API code in the CDK stack, deploy, then uncomment and redeploy. This removes and remakes the API.