Okay, you've successfully packaged your Flask prediction service and all its dependencies into a Docker image. Think of this image as a self-contained blueprint. Now, it's time to bring that blueprint to life by running it as a container. A container is a live, running instance of your image, an isolated environment where your application executes.

Launching Your Container with `docker run`

The fundamental command to start a container from an image is docker run. Its basic structure looks like this:

docker run [OPTIONS] IMAGE[:TAG] [COMMAND] [ARG...]

OPTIONS: Flags that modify the container's behavior (we'll cover important ones shortly).
IMAGE[:TAG]: The name of the image you want to run. If you followed the previous section, this might be something like prediction-service. Docker defaults to the :latest tag if none is specified.
COMMAND and ARG...: Optional commands and arguments to execute inside the container, overriding the default command specified in the Dockerfile (like the CMD ["python", "app.py"] instruction).

Let's try running the container for our Flask application. If your image is named prediction-service, a simple execution would be:

# This will likely run in the foreground and print Flask logs to your terminal
docker run prediction-service

You'll probably see output from Flask indicating the server has started. However, running it this way occupies your terminal. To stop it, you'd press Ctrl+C. For a web service, we typically want it to run in the background.

Running in Detached Mode

To run the container in the background (detached mode), use the -d option:

docker run -d prediction-service

This command starts the container and immediately returns your terminal prompt, printing a long container ID. But how do we interact with the Flask application running inside the container? By default, container ports are not accessible from your host machine. We need to map them.

Mapping Ports for Access

Our Flask application inside the container is listening on a specific port (likely port 5000, as configured in the Flask app or Dockerfile). To access it from our host machine's browser or tools like curl, we need to map a port on the host to the container's port. This is done using the -p or --publish option, formatted as host_port:container_port.

For example, to map port 5000 on your host machine to port 5000 inside the container, you'd use -p 5000:5000.

This diagram illustrates how the -p flag connects a port on your host machine to the port where the Flask application is running inside the container.

Naming Your Container

Docker assigns random names to containers (like focused_turing). While functional, these aren't very memorable. You can assign a specific name using the --name option, which makes managing the container easier.

Putting It All Together

Let's combine these options to run our Flask service container:

Run in detached mode (-d).
Publish port 5000 on the host to port 5000 in the container (-p 5000:5000).
Give it a memorable name, like my-prediction-app (--name my-prediction-app).
Specify the image name (prediction-service).

docker run -d -p 5000:5000 --name my-prediction-app prediction-service

If successful, Docker will print the container ID, and the container will be running in the background.

Verifying the Running Container

How do you know it's actually running? The docker ps command lists all currently running containers:

docker ps

You should see output similar to this (details might vary):

CONTAINER ID   IMAGE                COMMAND                  CREATED          STATUS          PORTS                    NAMES
a1b2c3d4e5f6   prediction-service   "python app.py"          15 seconds ago   Up 14 seconds   0.0.0.0:5000->5000/tcp   my-prediction-app

This confirms:

Your container (my-prediction-app) is running.
It's based on the prediction-service image.
Port 5000 on your host (0.0.0.0:5000) is mapped to port 5000 inside the container.

You can also check the logs produced by the application inside the container using docker logs:

docker logs my-prediction-app

This should show the startup messages from your Flask application.

Accessing Your Service

Now that the container is running and the port is mapped, you can send requests to your prediction service just like you did when testing locally, but targeting localhost on the host port you published (5000 in our example).

Using curl (assuming your API expects JSON input at a /predict endpoint):

curl -X POST -H "Content-Type: application/json" \
     -d '{"features": [1.0, 2.5, 3.0, 4.5]}' \
     http://localhost:5000/predict

Or using Python's requests library:

import requests
import json

# Example input data (adjust according to your model's needs)
data = {"features": [1.0, 2.5, 3.0, 4.5]}

# URL of the service running in Docker
url = "http://localhost:5000/predict" # Use the host port

try:
    response = requests.post(url, json=data)
    response.raise_for_status() # Raise an exception for bad status codes (4xx or 5xx)
    
    prediction = response.json()
    print(f"Prediction response: {prediction}")

except requests.exceptions.RequestException as e:
    print(f"Error connecting to the service: {e}")
except json.JSONDecodeError:
    print(f"Could not decode JSON response: {response.text}")

You should receive the prediction output from your model, served by the Flask app running entirely within the isolated Docker container!

Stopping and Cleaning Up

Once you are finished experimenting, you can stop and remove the container to free up resources.

Stopping the Container: Use the docker stop command followed by the container's name or ID:

docker stop my-prediction-app

Removing the Container: A stopped container still exists. To remove it completely, use docker rm:

docker rm my-prediction-app

You can only remove stopped containers. If you want to stop and remove in one go, you can use the -f (force) flag with docker rm, but it's generally better practice to stop it first.

To see all containers, including stopped ones, use docker ps -a. Regularly removing stopped containers you no longer need is good housekeeping.

You have now successfully run your machine learning prediction service inside a Docker container, making it portable and ensuring a consistent runtime environment. This is a foundational step in deploying applications reliably.

Running the Application in a Docker Container

Launching Your Container with docker run