Our Engineering
Methodology

We have learned important lessons during our decades of experience, and here are some of our beliefs:

• We embrace polyglot databases to leverage the benefits of both relational and NoSQL worlds
• We prioritize performance from day one and favor denormalization when appropriate. Although normalization is generally good practice, denormalization can improve performance in certain cases. We permit data duplication to avoid accessing multiple tables or documents and try to avoid joins or lookups as much as possible
• We carefully choose data types for each field to ensure data accuracy, storage efficiency, and performance. For example, using integer data types for numeric data can improve performance and storage efficiency. Changing them in production as we evolve can be a nightmare
• We avoid stored procedures as much as possible, as they pose challenges to testability and maintainability
• We use indexes wisely to improve query performance by creating indexes only on frequently used fields
• We plan for scalability, designing the schema and queries for growth and increased data volume. This includes partitioning data, sharding data, and using caching to reduce database load. For example, a data range partition proves very efficient when dealing with historical data.
• Finally, documenting the data model and schema is vital for understanding and maintaining the database over time. We document field definitions, relationships, constraints, and key design decisions and tradeoffs made during the data modeling process.