Glossary · System coordination, integration and orchestration
Consistency model
Also known as: data consistency model
German: Konsistenzmodell
In distributed systems, a consistency model is a contract that specifies which values a reader may observe after writes to replicated or shared data, ranging from strong consistency, where every read sees the latest write, to weaker models such as causal or eventual consistency.
- System integration
In one sentence
A consistency model states which values readers may see after writes to shared or replicated data, from strong to eventual consistency.
Example
The recipe database is strongly consistent so that no line starts with an outdated recipe, while dashboard data may be eventually consistent.
How it applies
- Architecture: Stronger consistency needs more coordination and reacts worse to network partitions; weaker consistency is faster and more available but allows readers to see old values. Choose per data item, not once for a whole system.
- Engineering: Setpoints, recipes and safety-relevant parameters usually need strong consistency or explicit version checks. Reporting and monitoring data can often tolerate Stale data for a defined time.
- Documentation: Interface and system documentation should state the consistency guarantee of each shared data set and the maximum expected delay, so that consumers don't assume guarantees the system does not give.
Consistency model vs. data validation
A consistency model is about when a change becomes visible to readers. Data validation checks whether a value is plausible and allowed. A value can be perfectly consistent and still wrong.