How to Avoid Conflicts in a Distributed System

The idea is that with primary backup, you would only write to one of the copies of the data items. But when you start looking at databases that want to provide higher availability and have to be able to operate when there's a network partition, such as in the primary backup situation, the primary might not be able to propagate to the backup. So this is a challenge about how we think about what kind of session guarantees do I have? And then we have things like strong consistency and linearizability and weaker forms like eventual consistency.