Why ReadTimeout are not arising with the sagemaker-runtime?

[stephenhutch]

Using the sagemaker-runtime client to invoke_endpoint in the following way:

import boto3
from botocore.config import Config

config_dict = {
    "retries": {
        "total_max_attempts": 1
    },
    "read_timeout": 1,
    "connect_timeout": 1
}

client = boto3.client(
    'sagemaker-runtime',
    config=Config(**config_dict)
)

response = client.invoke_endpoint(
    EndpointName=endpoint_name,
    Body=body
)

Running this code so far there have been no ReadTimeout exceptions raised even though metrics are showing that the call is taking greater than 1 second.

Do you know why ReadTimeout exceptions are not arising?