Google Cloud Platform for Developers

Entity groups and consistency

Depending on how data is modeled, entities in Datastore may be strongly or weakly consistent. This is due to the inherent trade off of distributed systems between consistency, availability, and partition tolerance, commonly referred to as CAP theorem. By default, entities are eventually consistent. This means that changes to data take time to propagate throughout the system. During this time, two observers of the data may see different results. For many cases, this is a totally acceptable trade off for performance and scalability.

In some cases, eventual consistency is simply not an option. For example, a trading platform that is eventually consistent may result in serious issues such as double-spending. To address ...