Overview

A simple project to deploy a machine learning based API on Kubernetes. The machine learning part is handed out as complete, with a pre-trained model and a dataset readily taken from Kaggle. The API itself is written in Python Flask and is containerized with the Docker container runtime

Running

A Quick explanation of each file provided:

├── Dockerfile		# commands and directives used by the docker container runtime
├── Makefile		# targets for "make"
├── README.md		# project description and getting started
├── app.py		# main application entrance
├── deployment.yaml	# description of the deployment needed by Kubernetes
├── make_prediction.sh	# run this to test a local installation of yours on port 8000 (default)
├── model_data		# the dataset used by the pre-trained model
│   ├── boston_housing_prediction.joblib
│   └── housing.csv
├── output_txt_files	# output files required for submission
│   ├── docker_out.txt
│   └── kubernetes_out.txt
├── requirements.txt	# used by the python package manager to setup all dependencies
├── run_docker.sh	# run inside a docker container
├── run_kubernetes.sh	# run inside a kubernetes cluster
└── upload_docker.sh	# uploads an image built by running `run_docker.sh` to Docker Hub

The following instructions are provided solely for Linux. First, setup the environment with virtualenv, run the project, then in a separate terminal window type:

./make_predictions.sh

to query the API for a single prediction. The input data format is JSON. See the Kaggle link for explanation of each field

Setup the Environment

• Create a Python 3 virtualenv and activate it:
  • python3 -m venv venv
  • source venv/bin/activate
• Run make install to install the necessary dependencies

Running app.py

1. Standalone: python app.py
2. Run in a Docker container: ./run_docker.sh
3. Run in Kubernetes: ./run_kubernetes.sh

Testing

So far, no actual tests are provided other than linting. Run make lint from within the project directory. You need pylint (pip install pylint) if you would like to run python liniting tests

Notes

There are tiny deriviations from the expected project outcome mentioned at the course page, namely:

1. Private image at Docker Hub was used
2. A declarative deployment.yaml file for Kubernetes was created, instead of specifying --image parameter, as it wasn't possible to specify the imagePullSecrets parameter via the CLI. The parameter is needed to pull the image from the private docker hub repository
3. A deployment (not pod) was created