Back to System Design HubCategory: distributed-systems

intermediate12 minute deep dive

Design a Distributed Unique ID Generator (Twitter Snowflake)

Design a lightweight, highly available distributed unique ID generation service that outputs 64-bit numerically sortable IDs without cross-network lock contention.

✓ Functional Requirements

IDs must be globally unique across distributed datacenters.
IDs must be 64-bit integers (fits inside standard 64-bit SQL BIGINT).
IDs must be roughly time-sortable chronologically.
System should generate over 10,000 unique IDs per millisecond per node.

⚡ Non-Functional Scale Goals

High Availability: 99.999% uptime (ID generation must never block application servers).
Ultra Low Latency: ID generation overhead < 0.2ms.
No central database auto-increment lock bottlenecks.

1. Key System Components

client

Application Services

Microservices requesting unique primary keys before performing database inserts.

Tech: gRPC / HTTP Client

service

Snowflake ID Generator Instances

Stateless worker nodes generating 64-bit integer IDs using local millisecond time & sequence counter.

Tech: Go / C++ / Rust Service

service

Apache ZooKeeper Coordinator

Registers worker instances and assigns unique 10-bit Worker IDs (0 to 1023) at startup.

Tech: Apache ZooKeeper / Consul

2. End-to-End Execution Flow

App ServerID Generator Node

Calls local or embedded Snowflake worker.

Request Next IDgRPC

ID Generator NodeID Generator Node

Assemble 64-bit ID: (timestamp << 22) | (worker_id << 12) | sequence

Bitwise Bitshift OperationInternal Bus

ID Generator NodeApp Server

ID returned in < 0.1ms without network locks.

Return 64-bit BIGINT (e.g. 1728394028341982208)gRPC

Back to System Design HubCategory: distributed-systems

intermediate12 minute deep dive

Design a Distributed Unique ID Generator (Twitter Snowflake)

Design a lightweight, highly available distributed unique ID generation service that outputs 64-bit numerically sortable IDs without cross-network lock contention.

✓ Functional Requirements

IDs must be globally unique across distributed datacenters.
IDs must be 64-bit integers (fits inside standard 64-bit SQL BIGINT).
IDs must be roughly time-sortable chronologically.
System should generate over 10,000 unique IDs per millisecond per node.

⚡ Non-Functional Scale Goals

High Availability: 99.999% uptime (ID generation must never block application servers).
Ultra Low Latency: ID generation overhead < 0.2ms.
No central database auto-increment lock bottlenecks.

1. Key System Components

client

Application Services

Microservices requesting unique primary keys before performing database inserts.

Tech: gRPC / HTTP Client

service

Snowflake ID Generator Instances

Stateless worker nodes generating 64-bit integer IDs using local millisecond time & sequence counter.

Tech: Go / C++ / Rust Service

service

Apache ZooKeeper Coordinator

Registers worker instances and assigns unique 10-bit Worker IDs (0 to 1023) at startup.

Tech: Apache ZooKeeper / Consul

2. End-to-End Execution Flow

App ServerID Generator Node

Calls local or embedded Snowflake worker.

Request Next IDgRPC

ID Generator NodeID Generator Node

Assemble 64-bit ID: (timestamp << 22) | (worker_id << 12) | sequence

Bitwise Bitshift OperationInternal Bus

ID Generator NodeApp Server

ID returned in < 0.1ms without network locks.

Return 64-bit BIGINT (e.g. 1728394028341982208)gRPC