Summary¶
After exploring the different families of databases, we can now summarize what they bring to the table – and where they fall short when applied to event sourcing. This section highlights the key takeaways from the comparisons and outlines why specialized event stores usually provide the best long-term fit.
Relational Databases¶
Relational systems are mature, stable, and well-understood. They provide strong ACID guarantees, powerful query languages, and a rich ecosystem. However, they were never designed for append-only workloads. Enforcing immutability, replaying large event streams, or scaling horizontally requires significant effort and custom infrastructure. They are suitable for prototypes or small-scale systems but rarely sustainable for event sourcing at scale.
NoSQL Databases¶
NoSQL solutions offer flexibility and scalability. They can store vast amounts of data across clusters and handle high write throughput. Yet, they trade off strict consistency and often lack mechanisms for immutability, replay, and efficient querying. Using NoSQL for event sourcing typically means building many essential features yourself, which adds operational complexity.
Event Streaming Systems¶
Event streaming systems excel at real-time distribution and integration. They are perfect for connecting services, processing events on the fly, and supporting high-throughput messaging. But they are not designed for permanent storage. Retention windows, partitioning constraints, and limited querying capabilities make them unsuitable as the sole persistence layer for event sourcing. They are complementary to an event store, not a replacement for one.
Specialized Event Stores¶
Specialized event stores are built for event sourcing from the ground up. They enforce append-only semantics, provide immutability guarantees, support replays and snapshots, and offer subscriptions for consuming events in real time. Many also include query capabilities, scalability features, and compliance mechanisms. Their narrow focus means they are not general-purpose databases, but for event sourcing they align directly with the core requirements.
The Bigger Picture¶
Looking across all families, the pattern is clear:
- General-purpose databases can be adapted to store events, but with compromises and complexity.
- Streaming systems are excellent companions but cannot replace permanent storage.
- Specialized event stores provide the most natural and reliable foundation for event sourcing.
This does not mean that relational or NoSQL databases are "wrong" – they may be adequate in limited contexts. But for organizations aiming to adopt event sourcing as a first-class architectural principle, specialized solutions are the most sustainable choice.
Conclusion¶
Choosing the right kind of database is less about individual products and more about recognizing which family of databases truly matches the requirements of event sourcing. The evidence points toward specialized event stores as the option that delivers on those requirements without forcing additional infrastructure or fragile workarounds.
Next up: Event Store Landscape – a closer look at the variety of specialized event stores and how they compare.