the most common way of generating unique ideas is to use some auto_increment function of the SQL of choice. Unfortunately, in distributed systems this won't cut it.. but why?

On s single server app without much traffic, this may work, as there won’t be any concurrent ID generation, like there will be in a multi-server application. i.e. this feature can result in conflicts when implemented in distributed environments

The Requirements On Our System

The IDs need to be unique, but should they have more features? → Yes, as well as miminimizing the duplicated to basically 0, we should be able to sort IDs by time

How big can a single ID be? → 64 bits

How many IDs are we expected to generate? → Potentially billions

Should we support unique ID generation in distributed environment? → Yes, we will be working in distributed environment as we expect our application to scale horizontally with time

What Are Our Approaches to Solve This?

Multimaster Approach

• here we are simply incrementing the IDs, but with a twist. We are taking the number of nodes/servers and we are incrementing each ID by this number
• for example if we have two nodes - n1 and n2, the ID generation will be as:

n1 → 0,2,4,6

n2 → 1,3,5,7

The limitations of this approach are that we won’t be able to sort the IDs by time as required and adding/removing nodes/servers to our system will bring a great deal of complexity and potential duplication in the ID generation process

The Ticket Server Approach

• here, we basically have a centralized node that deals with the ID generation and all other nodes get the generated ID from this central node