AWS Lambda调用TensorFlow SageMaker端点请求体格式验证错误排查

在AWS上训练并部署TensorFlow模型为SageMaker推理端点后，通过Lambda+API Gateway触发8×8矩阵推理时，始终出现请求体格式验证错误，预期输出单个数值。

错误信息 #

{"errorMessage": "Parameter validation failed:\nInvalid type for parameter Body, value: {'instances': [[[0.30050477, 0.31565664, 0.03535339, 0.25252531, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.26515138, 0.32828271, 0.11616141, 0.25757545, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.29292932, 0.45707068, 0.20959599, 0.44696943, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.4242426, 0.44949475, 0.05808071, 0.06565664, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.09595941, 0.44696943, 0.0, 0.36868672, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.33838396, 0.89898997, 0.3055554, 0.85101003, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.84090927, 0.89646466, 0.77272732, 0.85858596, 0.51226546, 0.62584169, 0.66940236, 0.0], [0.85606065, 0.85606065, 0.76262607, 0.7878787, 0.5910121, 0.62584169, 0.66940236, 0.0]]]}, type: <class 'dict'>, valid types: <class 'bytes'>, <class 'bytearray'>, file-like object", "errorType": "ParamValidationError", "requestId": "a1d05ff9-86fb-4a30-9dcb-2651bb001b6c", "stackTrace": ["  File \"/var/task/lambda_function.py\", line 19, in lambda_handler\n    response = runtime.invoke_endpoint(EndpointName=ENDPOINT_NAME,\n", "  File \"/var/runtime/botocore/client.py\", line 391, in _api_call\n    return self._make_api_call(operation_name, kwargs)\n", "  File \"/var/runtime/botocore/client.py\", line 691, in _make_api_call\n    request_dict = self._convert_to_request_dict(\n", "  File \"/var/runtime/botocore/client.py\", line 739, in _convert_to_request_dict\n    request_dict = self._serializer.serialize_to_request(\n", "  File \"/var/runtime/botocore/validate.py\", line 360, in serialize_to_request\n    raise ParamValidationError(report=report.generate_report())\n"]}

测试用请求体 #

{"instances": [[[0.30050477, 0.31565664, 0.03535339, 0.25252531, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.26515138, 0.32828271, 0.11616141, 0.25757545, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.29292932, 0.45707068, 0.20959599, 0.44696943, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.4242426, 0.44949475, 0.05808071, 0.06565664, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.09595941, 0.44696943, 0.0, 0.36868672, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.33838396, 0.89898997, 0.3055554, 0.85101003, 0.37373726, 0.62584169, 0.66940236, 0.0], [0.84090927, 0.89646466, 0.77272732, 0.85858596, 0.51226546, 0.62584169, 0.66940236, 0.0], [0.85606065, 0.85606065, 0.76262607, 0.7878787, 0.5910121, 0.62584169, 0.66940236, 0.0]]]}

原Lambda函数代码 #

import os
import json
import boto3

ENDPOINT_NAME = os.environ['ENDPOINT_NAME']
runtime = boto3.client('runtime.sagemaker')

def lambda_handler(event, context):
    data = json.loads(json.dumps(event))
    payload = data['instances']
    response = runtime.invoke_endpoint(EndpointName=ENDPOINT_NAME,
                                       ContentType='application/json',
                                       Body=data)
    result = json.loads(response['Body'].read().decode())

    return result

成功的Notebook调用示例 #

在SageMaker Jupyter Notebook中使用以下代码可成功触发端点，得到预测结果{'predictions': [[0.875023425]]}：

import numpy as np
import json
import boto3

runtime = boto3.client('runtime.sagemaker')
payload = {'instances': np.array([[[0.30050477, 0.31565664, 0.03535339, 0.25252531, 0.37373726,
      0.62584169, 0.66940236, 0.        ],
     [0.26515138, 0.32828271, 0.11616141, 0.25757545, 0.37373726,
      0.62584169, 0.66940236, 0.        ],
     [0.29292932, 0.45707068, 0.20959599, 0.44696943, 0.37373726,
      0.62584169, 0.66940236, 0.        ],
     [0.4242426 , 0.44949475, 0.05808071, 0.06565664, 0.37373726,
      0.62584169, 0.66940236, 0.        ],
     [0.09595941, 0.44696943, 0.        , 0.36868672, 0.37373726,
      0.62584169, 0.66940236, 0.        ],
     [0.33838396, 0.89898997, 0.3055554 , 0.85101003, 0.37373726,
      0.62584169, 0.66940236, 0.        ],
     [0.84090927, 0.89646466, 0.77272732, 0.85858596, 0.51226546,
      0.62584169, 0.66940236, 0.        ],
     [0.85606065, 0.85606065, 0.76262607, 0.7878787 , 0.5910121 ,
      0.62584169, 0.66940236, 0.        ]]])}
response = runtime.invoke_endpoint(EndpointName='your-endpoint-name', ContentType='application/json', Body=json.dumps(payload))
result = json.loads(response['Body'].read().decode())
print(result)

解决方案 #

1. 修正Lambda函数的核心错误 #

错误根源是invoke_endpoint的Body参数要求传入bytes/bytearray/file-like对象，不能直接传Python字典。同时需处理API Gateway的请求结构：

• 如果API Gateway使用Lambda代理集成，请求体在event['body']中，需先解析为字典
• 最终要把请求体序列化为JSON字符串再转成bytes

修正后的Lambda代码：

import os
import json
import boto3

ENDPOINT_NAME = os.environ['ENDPOINT_NAME']
runtime = boto3.client('runtime.sagemaker')

def lambda_handler(event, context):
    # 处理API Gateway代理集成的请求体
    if 'body' in event:
        data = json.loads(event['body'])
    else:
        # 本地测试时直接使用event作为请求体
        data = event
    
    # 将字典序列化为JSON字节流
    payload_bytes = json.dumps(data).encode('utf-8')
    
    response = runtime.invoke_endpoint(
        EndpointName=ENDPOINT_NAME,
        ContentType='application/json',
        Body=payload_bytes
    )
    
    result = json.loads(response['Body'].read().decode())
    # 返回符合API Gateway代理集成要求的格式
    return {
        'statusCode': 200,
        'body': json.dumps(result)
    }

2. API Gateway请求格式要求 #

发送请求时需满足：

• 请求方法：POST
• 请求头：Content-Type: application/json
• 请求体：使用测试用的JSON结构（嵌套列表，无需numpy数组），和Notebook中的instances结构一致

3. 验证关键点 #

• 确保Lambda执行角色拥有sagemaker:InvokeEndpoint权限
• 确认SageMaker端点状态为InService
• 测试时可先在Lambda控制台直接传入测试用JSON（无需body包装），验证端点调用成功后再通过API Gateway测试

内容的提问来源于stack exchange，提问作者Philippe