EventStores

EventStores are data stores designed to persist events produced by event-driven architectures, especially those implementing event sourcing. In event sourcing, system state is derived from a sequence of events rather than a single current snapshot. An event store records events in streams, often scoped by aggregate or entity, using append-only writes to preserve order and immutability. Each event typically carries a type, a data payload, and metadata such as a timestamp, sequence number, version, and correlation identifiers.

Reading data from an event store generally involves replaying the events for a given stream to reconstruct state, or using read models and projections that transform the event stream into queryable views. Event stores commonly support features such as per-stream ordering, multiple streams, event versioning, snapshotting to speed up state reconstruction, retention policies, and durability guarantees. They may also provide built-in support for projections, allowing developers to create derived read models in real time.

Event stores are used to enable auditability, traceability, and time-based queries, as well as to power CQRS architectures where write models emit events and read models consume them. They differ from general-purpose relational or document stores, which maintain current-state records and support ad-hoc queries; event stores emphasize long-term history, append-only semantics, and replayability.

Common implementations include EventStoreDB, which emphasizes event streams and projections; and distributed logs like Apache Kafka, which are often used as the backbone for event-sourced systems, though Kafka is a log rather than a full event store. Trade-offs include complexity of querying, need for read-models, and challenges with schema evolution and retention.

See also: event sourcing, CQRS, event projection, append-only storage.