Elsayed91 / taxi-data-pipeline

Automated data & MLOps pipeline leveraging Kubernetes and Apache Airflow. Integrates Spark, Kafka, and DBT with a focus on data quality. Tailors solutions for diverse user needs.

Overview

This project showcases the orchestration of a scalable data pipeline using Kubernetes and Apache Airflow. It makes use of well known data solutions like Spark, Kafka, and DBT, while maintaining a strong focus on data quality.
With NYC Taxi data stored in S3 as the starting point, this pipeline goes beyond generating a standard BI dashboard. It also delivers a prediction service and data models tailored for data scientists' needs.

Table of Contents

1. Explore Some Project Deliverables
2. Architecture
   • Visually Summarized Architecture
   • Data Stack
   • Airflow Job DAG Flowchart
   • CICD Pipelines
3. Extra Notes on Technologies Used
4. Testing
5. How to Install

Explore Some Project Deliverables:

**Kindly note that the services are hosted using free GCP credits and may not be available once the credits are exhausted.

• Static Docs:
  • Great Expectations: Link | Screenshot
  • DBT Docs: Link | Screenshot
  • Elementary Docs: Link | Elementary Dashboard
• ML Serving/Prediction Service: Link | Screenshot
• MLFlow: Screenshot

You can find extra documentation in the docs/ dir.

Architecture

Visually Summarized Architecture

Data Stack

Airflow Job DAG Flowchart

CICD Pipelines

flowchart LR

    A[GitHub Push / Pull Request]
    A --> |Python File?| B(PyApps CI)
    A --> |Docker Folder Content/ K8 Manifests Changed?| C(Build & Deploy Components)
    A --> |Terraform File Changed?| D(Terraform Infrastructure CD)


    B -- Yes --> E[Run Tests]
    C -- Yes --> F[Build Images]
    D -- Yes --> G[Terraform Actions]

    E --> |Test Successful?| H[Format, Lint, and Sort]
    F --> |Deployment Restart Required?| I[Kubectl rollout restart deployment]
    G --> |Terraform Plan Successful?| J[Terraform Apply]

    E --> K[End]
    H --> K
    F --> |No Deployment Restart Required| K
    I --> K
    G --> |Terraform Plan Unsuccessful| K
    J --> K

    B -- No --> K
    C -- No --> K
    D -- No --> K

Loading

Extra Notes on Technologies Used:

• Flask: Used to serve static documentation websites for DBT, Great Expectation, and Elementary.
• Prometheus: Collects metrics from various sources including Airflow (via Statsd), SparkApplications (via JMX Prometheus Java Agent), Postgres (via prometheus-postgres-exporter), and Kubernetes (via kube-state-metrics).
• Grafana Utilized to visualize the metrics received by Prometheus. The Grafana folder contains JSON dashboard definitions created to monitor Airflow and Spark.
• Kafka: Integrated as a streaming solution in the project. It reads a file and utilizes its rows as streaming material.
• Streamlit: Used to serve the best xgboost model from the MLFlow Artifact Repository.
• gitsync: Implemented to allow components to directly read their respective code from GitHub without needing to include the code in the Docker image.
• Spark-on-K8s: Utilized to run Spark jobs on Kubernetes.

Testing

Other than standard unit testing, the below was also done.

Data Quality Tests

• Data quality is validated on ingestion using Great Expectations.
• All data source and models have dbt expectations data quality tests.
• There are also unit tests where a sample input and expected output are used to test the model.

Integration Tests

An integration test for the lambda function can be found in the tests/integration directory.

How to Install:

Prerequisites

• Google Cloud SDK with Kubectl.

Installation

1. Clone the repository:

git clone https://github.com/Elsayed91/taxi-data-pipeline

2. Rename template.env to .env and fill out the values, you dont need to fill out buckets or AUTH_TOKEN values.
3. Run the project setup script it will prompt you to login to your gcloud account, do so and it will do the rest.

make setup

• to manually trigger the batch-dag:

make trigger_batch_dag

• to run kafka

make run_kafka

• to destroy kafka instance

make destroy_kafka

• to run the lambda integration test

make run_lambda_integration_test

Note: to use the workflows, a GCP service account as well as the content of your .env file need to be added to your Github Secrets.