iFood Data Architect Test

This is my solution for the iFood Test, whose proposal is to create a prototype datalake from json and csv files. Raw and trusted layers where required. My solution runs locally inside a docker container with Pyspark and all further necessary dependencies. It gets the data ingested, wrangled, processed and finally exported as parquet files. Partitioning was done based on test requirements.

The solution is split into two parts:

Development: Jupyter Notebooks with a development walkthrough can be found here.
Final script: main.py can be found here here

The raw data had some duplicated values which I've decided to drop after having a look into some of them and understanding it was safe to do so. Data validation was accomplished by casting data types after manually understanding each present column. For cases when I was unsure, the data was left as string in order to avoid possible crashes. Regarding anonymization I've just dropped all sensitive data columns since their owners, customers and merchants, can be identified via their unique ids anyways.

Regarding data persistency, all storage is done on the mapped docker-compose volume. In this case locally at the relative path ./dev/docker-volume inside the directory where the repo was cloned to - see below. Ideally data should be made avaiable at a shared storage file system, so multiple teams could have access to it. For simplicity purpouses it's left like stated here.

volumes:
    - ./dev/docker-volume:/home/jovyan

The complete solution was run on my local laptop, that's why the Spark session has modest configurations. But once the final application script gets deployed to a proper development environment, such as suggested Databricks, it should scale accordingly.

How to Run

Requirements

docker >= 19.03.9
docker-compose >= 1.25.0

Step-by-step

Place your AWS Credentials into ./dev/docker-volume/.aws/secrets.json according the following format:

{
    "PUBLIC_KEY": "<PUBLIC_KEY>",
    "SECRET_KEY": "<SECRET_KEY>"
}

Run docker-compose at root directory

docker-compose up --build

Access http://localhost:8888 at your browser
Then decide to run production script or developement notebooks:

Production: Open terminal window within Jupyter Notebook and run python main.py.
Development: Run notebooks.

Test Scope

Please find the code challenge here.

References

Folder Organization

├── README.md                   <- The top-level README.
│
├── media                       <- Folder for storing media such as images.
│
├── dev                         <- Development folder where  application scripts are stored.
│   ├── docker-volume           <- Shared volume with docker container.
│   │   ├── credentials         <- Folder for storing credentials, only one required so far is AWS account.
│   │   ├── data                <- Data for file system, here is where raw and trusted layers shall reside.
│   │   ├── notebooks           <- Development notebooks to have a walkthrough of development phase.
│   │   ├── src                 <- Custom library for storing required code.
│   │   └── main.py             <- Main application to run full pipeline from single script.
│   │
│   ├── Dockerfile              <- Defines docker container and installs dependencies.
│   ├── requirements.txt        <- Stores Python required libraries.
│   └── setup.py                <- Installs `src` custom library
│
├── .gitignore                  <- Used for ignoring uploading unecessary data to repo.
│
├── docker-compose.yml          <- Docker-compose file for runing docker containers with environemnt specs. 
│
└── TestScope.md                <- Clone of original test scope in case the original repo gets deleted.

brunofaustino / ifood-data-architect-test Goto Github PK