object-lambda-transform-handler.py

obj_get_ctx = event['getObjectContext']
request_route = obj_get_ctx['outputRoute']
request_token = obj_get_ctx['outputToken']
obj_url = obj_get_ctx['inputS3Url']
requested_url = event['userRequest']['url']
path = Path(urlparse(requested_url).path).relative_to('/')

# First attempt to read directly using the signed URL provided
response = requests.get(obj_url)
resp = {'StatusCode': response.status_code}

if response.status_code == 404 and path.suffix == '.parquet':
    # Load CSV and convert to Parquet.
    csv_key = str(path.with_suffix('.csv'))
    try:
        csv_body = s3_client.get_object(Bucket=bucket_name, Key=csv_key)['Body']
        resp['Body'] = pd.read_csv(csv_body).to_parquet()
        resp['StatusCode'] = 200
    except botocore.exceptions.ClientError as error:
        # If the CSV could not be read, propagate that error
        resp['ErrorCode'] = error.response['Error']['Code']
        resp['StatusCode'] = error.response['ResponseMetadata']['HTTPStatusCode']
        resp['ErrorMessage'] = error.response['Error']['Message']
else:
    resp['Body'] = response.content

s3_client.write_get_object_response(
    RequestRoute=request_route,
    RequestToken=request_token,
    **resp
)