Show Off Your Machine Learning Project on Web Part 3: Deploy onto Heroku with Docker

What client needs are a domain name and a browser.

Visit part 1 and part 2 by clicking them.
Click this to go through paywall.

TL;DR: You can get the example project here.

Demo of my app on Heroku. You can visit the app at https://fish-yolo-grabcut.herokuapp.com/.

Motivation

After you create a dockerized environment, you think you should deploy your app on cloud so that you can show off your machine learning project, web development and DevOps skills. In addition, others can access your project with just one click on URL you provide without the knowledge of setting up the environment.

Goal of This Part

This part will teach you how to deploy your dockerized app onto Heroku, and tell you some specialized settings of Heroku when deploying with Docker.

Before Deploy to Heroku

Heroku provides three ways to deploy your dockerized app:

1. Container Registry & Runtime (Docker Deploys)
2. Building Docker Images with heroku.yml
3. Local Development with Docker Compose

As I don’t need Docker components for my app (only one streamlit service), there is no need to use Docker Compose.

Thought Most of posts use Method #1, I will choose Method #2 for these two reasons:

1. Rarely the posts mention about this method.
2. I write a specialized Dockerfile for Heroku deployment.

The Components We Need

For Method #2, we need at least two files:

1. Dockerfile
2. heroku.yml: to indicate Heroku what you want for building and deploying.

In my case, I need three files because I like modularization:

1. Dockerfile
2. heroku.yml
3. heroku_startup.sh

Dockerfile

Just like part 2 with some changes for Heroku deployment:

You can see the changes on line 42 and 48. The reasons are that:

1. Change the permission of heroku_startup.sh or you will receive permission denied error while executingheroku_startup.sh.
2. ENTRYPOINT "./heroku_startup.sh" to tell Docker to execute heroku_startup.sh when starting the container.

Heroku.yml

Heroku needs heroku.yml to build and deploy Docker Images, so I create one for my usage:

build:
  docker:
    web: heroku.Dockerfile
run:
  web: ./heroku_startup.sh

Obviously, the build and run sections indicate what you want to do in build and run stage on Heroku, respectively.

heroku_startup.sh

Mentioned here, Heroku uses a$PORT environment variable for port exposure. So the heroku_startup.sh will look like this:

#!/bin/bash

echo PORT $PORT
streamlit run --server.port $PORT app.py

You have to set the $PORT variable to streamlit or your app will not appear.

Deploy!

This step is rather easy, what you need is to deploy with Git. Then Heroku will use Heroku.yml rather Procfile to deploy your app with Docker.

Recap

This part shows how to deploy your machine learning app onto Heroku and mention some special setting on Heroku.

I hope this series can help you to combine your machine learning app with web and DevOps development cycle so that you can catch up the trend and won’t be mocked by web development company!

Originally published at https://cuda-chen.github.io on June 29, 2020.

If you have any thoughts and questions to share, please contact me at clh960524[at]gmail.com. Also, you can check my GitHub repositories for other works. If you are, like me passionate about machine learning, image processing and parallel computing, feel free to add me on LinkedIn.