Design a unique ID generator in HLD - System Design Exercise

Design: Unique ID Generator

Design covers the ID generation service, its architecture, data flow, and scaling. Does not cover client-side ID usage or storage.

Functional Requirements

FR1: Generate unique IDs that are globally unique across distributed systems

FR2: Support high request throughput of up to 100,000 ID requests per second

FR3: IDs should be sortable by creation time

FR4: IDs must be generated with low latency (p99 < 10ms)

FR5: System should be highly available with 99.9% uptime

FR6: Support multiple clients generating IDs concurrently

Non-Functional Requirements

NFR1: No single point of failure

NFR2: IDs must be collision-free even under network partitions

NFR3: System should scale horizontally

NFR4: Generated IDs should be compact (e.g., 64-bit or 128-bit)

NFR5: Avoid reliance on centralized databases for ID generation

Think Before You Design

Questions to Ask

❓ Question 1

❓ Question 2

❓ Question 3

❓ Question 4

❓ Question 5

❓ Question 6

Key Components

Timestamp provider

Node or machine identifier

Sequence number generator

API gateway or load balancer

Distributed coordination or consensus service (optional)

Cache or in-memory store for sequence state

Design Patterns

Snowflake ID generation pattern

UUID version 1 or 4

Database auto-increment with sharding

Timestamp + node ID + sequence number

Leader election for coordination

Reference Architecture

Load Balancer

↓

ID Generator Nodes

↓

Clients receive unique IDs

↓

Timestamp

Components

Load Balancer

Nginx or Cloud Load Balancer

Distributes incoming ID generation requests evenly across ID generator nodes

ID Generator Nodes

Custom service using Snowflake algorithm

Generate unique IDs using timestamp, node ID, and sequence number

Coordination Service

Apache ZooKeeper or etcd

Assigns unique node IDs and manages node membership to avoid collisions

Request Flow

1. Client sends request for unique ID to Load Balancer

2. Load Balancer forwards request to one of the ID Generator Nodes

3. ID Generator Node obtains current timestamp

4. Node uses its unique node ID assigned by Coordination Service

5. Node increments sequence number for IDs generated within the same millisecond

6. Node combines timestamp, node ID, and sequence number to form unique ID

7. Node returns generated unique ID to client

Database Schema

No persistent database required for ID generation. Coordination Service stores ephemeral node IDs and membership info.

Scaling Discussion

Bottlenecks

Sequence number overflow if too many IDs requested in one millisecond per node

Coordination service becoming a single point of failure

Load balancer overload under very high request rates

Clock synchronization issues causing duplicate or out-of-order IDs

Solutions

Increase number of ID generator nodes to distribute load

Use high-availability setup for coordination service with leader election

Implement client-side retries and backoff for load balancer failures

Use NTP or GPS for clock synchronization and monitor clock drift

If sequence overflows, wait for next millisecond or reject requests temporarily

Interview Tips

Time: Spend 10 minutes clarifying requirements and constraints, 20 minutes designing architecture and data flow, 10 minutes discussing scaling and failure handling, 5 minutes summarizing.

Explain why global uniqueness and ordering are important

Describe how timestamp, node ID, and sequence number combine to form unique IDs

Discuss trade-offs between centralized and decentralized approaches

Highlight how coordination service prevents node ID collisions

Address how system handles clock issues and high throughput

Mention scalability and fault tolerance strategies