Topics Covered

On this page

1. NOT EXECUTED
2. Deploy
3. Invoke it
4. The container contract
5. Real-time, serverless or batch
6. Then delete it
7. Monitoring the deployed model

NOT EXECUTED

This repository has no cloud account, and none will be created. Creating one requires a payment card and accepts a billing relationship, which is not something a study repository should do on anyone's behalf.

So this file is what you type, with the traps marked. It has never been run here, and nothing in the notes claims an output for it.

The runnable half is 15_deploy_endpoint.py, which starts a REAL HTTP server, serves a REAL model and calls it.

---

Deploy

predictor = estimator.deploy(
    initial_instance_count=1,
    instance_type="ml.m5.large",
    endpoint_name="churn-endpoint",
)
predictor.predict([[1.2, 0.4, ...]])

Or from the CLI, in the three steps the SDK hides:

aws sagemaker create-model --model-name churn-model \
  --primary-container Image=<ecr-uri>,ModelDataUrl=s3://bucket/models/model.tar.gz \
  --execution-role-arn arn:aws:iam::<acct>:role/SageMakerExecutionRole

aws sagemaker create-endpoint-config --endpoint-config-name churn-config \
  --production-variants VariantName=AllTraffic,ModelName=churn-model,\
InitialInstanceCount=1,InstanceType=ml.m5.large,InitialVariantWeight=1

aws sagemaker create-endpoint --endpoint-name churn-endpoint \
  --endpoint-config-name churn-config

Model, endpoint config, endpoint — three objects, not one. That is what makes blue/green possible: create a second config and update the endpoint, and traffic shifts without downtime.

Invoke it

aws sagemaker-runtime invoke-endpoint \
  --endpoint-name churn-endpoint \
  --content-type application/json \
  --body '{"instances": [[1.2, 0.4, 3.3, ...]]}' \
  /dev/stdout

The request is IAM-signed. There is no API key to leak, and access is governed by the same policy evaluation as everything else.

The container contract

Your container must answer two routes:

Route	Must
GET /ping	return 200 quickly, without running the model
POST /invocations	run inference

A health check that does real inference marks the container unhealthy whenever the model is merely slow — and the platform then kills a container that was working. The runnable half implements both routes and says so.

Real-time, serverless or batch

	Real-time	Serverless	Batch transform
Latency	ms	ms, after a cold start	minutes to hours
Billed	per hour, always	per request	per job
Idle cost	the full instance	zero	zero
Good for	steady traffic	spiky or occasional	scoring a whole file

An ml.m5.large endpoint is about $70/month whether or not anything calls it. If traffic is occasional, serverless inference costs a fraction; if you are scoring a file, batch transform is the right tool and an endpoint is the expensive way to do arithmetic — the runnable half measures a 37x difference between one batched request and 100 single ones.

Then delete it

aws sagemaker delete-endpoint --endpoint-name churn-endpoint
aws sagemaker delete-endpoint-config --endpoint-config-name churn-config
aws sagemaker delete-model --model-name churn-model

Deleting the endpoint is a step in the experiment, not an afterthought. Every "surprise AWS bill" story is a resource nobody switched off.

Monitoring the deployed model

Data drift is the failure that has no error message: the endpoint keeps returning 200 and the predictions quietly stop being right.

from sagemaker.model_monitor import DefaultModelMonitor
monitor = DefaultModelMonitor(role=role, instance_type="ml.m5.xlarge")
monitor.suggest_baseline(baseline_dataset=f"s3://{bucket}/train/train.csv")
monitor.create_monitoring_schedule(endpoint_input=predictor.endpoint_name,
                                   schedule_cron_expression="cron(0 * ? * * *)")

Baseline the training distribution, then compare production inputs against it hourly. A drift alarm is the only thing that catches a model that has stopped working while every infrastructure metric stays green.