SageMaker Inference Timeout Configuration

Question

Hello,

I was trying to deploy this [CartoonGAN](https://github.com/Yijunmaverick/CartoonGAN-Test-Pytorch-Torch) solution on an *ml.x5.xlage* instance using AWS SageMaker endpoint. The deployment was successful and I can invoke the endpoint, passing a base64-encoded image, without any problems. But the responses said my inference task was terminated and timed out after 10.01 seconds. I am wondering if there is some sort of timeout variable that I can set to increase how long it waits for the model to respond to the query. Or is there anything I can do to solve this?

Here is my deployment code

```
from sagemaker.pytorch.model import PyTorchModel

pytorch_model = PyTorchModel(
        model_data=model_uri,
        role=role_arn,
        framework_version=torch_version,
        py_version=py_version,
        entry_point='serve.py',
        source_dir='.'
)
```

```
pytorch_model.deploy(
        endpoint_name=endpoint_name,
        instance_type=instance_type,
        initial_instance_count=1
    )
```
The log from CloudWatch

```
2024-04-02T22:48:54.258+07:00	INIT_START Runtime Version: python:3.10.v29 
2024-04-02T22:48:54.662+07:00	START RequestId: d57785ae-bc40-40f4-8f8b-dc1f77a84de4 Version: $LATEST
2024-04-02T22:49:04.677+07:00	2024-04-02T15:49:04.675Z d57785ae-bc40-40f4-8f8b-dc1f77a84de4 Task timed out after 10.01 seconds
2024-04-02T22:49:04.677+07:00	END RequestId: d57785ae-bc40-40f4-8f8b-dc1f77a84de4
2024-04-02T22:49:04.677+07:00	REPORT RequestId: d57785ae-bc40-40f4-8f8b-dc1f77a84de4	Duration: 10013.20 ms	Billed Duration: 10000 ms	Memory Size: 512 MB	Max Memory Used: 74 MB	Init Duration: 403.33 ms
```

Thank you,

Answer

I suspect that the issue might be related to the model's initialization or inference code in the `serve.py` script. Enable debugging logs to identify any specific bottlenecks in the inference process.

> A customer's model containers must respond to requests within 60 seconds. The model itself can have a maximum processing time of 60 seconds before responding to invocations. If your model is going to take 50-60 seconds of processing time, the SDK socket timeout should be set to be 70 seconds. [Source](https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_runtime_InvokeEndpoint.html)

Also if you're using AWS Lambda to invoke the SageMaker endpoint, check the timeout setting of your Lambda function.  the default timeout for a new Lambda function is 3 seconds, but it can be extended up to 15 minutes.

SageMaker Inference Timeout Configuration

관련 콘텐츠