Building Real-Time Analytics Systems

Foreword

When I started my career in software engineering in the early 2000s, data analytics oftentimes was an afterthought when designing software systems. Batch jobs running once per day would extract data from operational databases and load it into data warehouses, and business analysts typically were happy when they could look at the data from yesterday or last week, creating reports, running once-off queries, etc.

Apart from perhaps a few handcrafted, highly optimized queries running within operational databases, the idea of user-facing analytics was pretty much unheard of: serving analytics workloads to thousands of concurrent users, based on the freshest data possible. Since then, the appetite for real-time analytics has substantially increased. Use cases like fraud detection, resource planning, content recommendations, predictive analytics, and many others require the latest data in order to provide value. If, for instance, your bank detects a pattern of misuse for your credit card because it got stolen, you’d want your card to be blocked right now and not tomorrow, right?

Tools and platforms such as Apache Kafka (for data streaming), Apache Flink (stream processing), Apache Pinot (data analytics) and Apache Superset (data visualization) provide an excellent foundation for real-time analytics and have seen a tremendous uptake over the last years. At the same time, getting started with implementing your first use cases can be challenging, and you might ask yourself questions ...