Single node architecture models are inefficient as they cause a single point of failure, hence there is the use of distributed replicas (ref: Distributed Consensus DB)

Storing data in multiple nodes makes it more reliable, but data can be inconsistent.
Suppose, we want to lock operations until it becomes consistent, then it can be unavailable for that time.

This is the Consistency Trade off

Then there's Master - Slave Distributed System, as seen in Distributed Consensus DB

Similar to the TCP-IP protocol

To ensure consistency on saves, we use a 2-way ACK,

1. First a "prepare" req is send from master -> slave
2. If failure at any point -> rollback the system
3. Once prepare is done, ask to commit, after commit -> master updates itself as it knows it is consistent to save data now.