Snowflake ID

Snowflake IDs are 64-bit identifiers designed by Twitter for distributed systems requiring high-throughput, time-sortable ID generation.

Overview

Format

 1234567890123456789
 |-----------------|
   64-bit integer

Bit Structure (Twitter format)

 0 | 41 bits timestamp | 5 bits DC | 5 bits Worker | 12 bits sequence

Characteristics

High Throughput

• 4096 IDs per millisecond per worker
• 4,096,000 IDs per second per worker
• With 32 workers: ~131 million IDs/second

Time-Sortable

IDs generated later have higher values:

$ idt gen snowflake
1234567890123456789
$ idt gen snowflake
1234567890123456790  # Greater value

Compact

64 bits fits in a single long integer in most languages.

Requires Coordination

Unlike UUIDs, Snowflake IDs require assigning machine/datacenter IDs to avoid collisions.

Generation

# Default Snowflake
idt gen snowflake

# With machine ID
idt gen snowflake --machine-id 1

# With datacenter ID
idt gen snowflake --datacenter-id 2

# Both
idt gen snowflake --machine-id 1 --datacenter-id 2

Custom Epochs

Different systems use different epochs:

# Twitter epoch (default): Nov 4, 2010
idt gen snowflake --epoch twitter

# Discord epoch: Jan 1, 2015
idt gen snowflake --epoch discord

# Custom epoch (milliseconds since Unix epoch)
idt gen snowflake --epoch 1420070400000

Environment Variable

Set default epoch via environment:

export IDT_SNOWFLAKE_EPOCH=discord
idt gen snowflake

Inspection

$ idt inspect 1234567890123456789
SNOWFLAKE
  1234567890123456789

  Time (UTC)          2023-01-15T12:34:56.789Z
  Local Time (+09:00) 2023-01-15T21:34:56.789+09:00
  Datacenter   1
  Worker       2
  Sequence     789

  Hex          112210f47de98115
  Base64       ESIQr0fpgRU=
  Int          1234567890123456789

Common Epochs

Comparison with Other IDs

*UUIDv7 has timestamp and sortability

When to Use Snowflake

Good for:

• High-throughput systems (>100k IDs/second)
• Distributed databases
• Systems needing 64-bit IDs
• Real-time analytics (time-ordered)

Consider alternatives if:

• You can’t coordinate machine IDs (use ULID)
• You need 128-bit IDs (use UUID or ULID)
• You’re generating IDs client-side (use NanoID)

Implementation Notes

Machine ID Assignment

In production, assign machine IDs via:

• Configuration files
• Environment variables
• Service discovery (e.g., ZooKeeper, Consul)
• Database sequences

Clock Skew

Snowflake IDs are sensitive to clock skew:

• Use NTP synchronization
• Some implementations wait for the clock to catch up
• Consider using UUIDv7 if clock skew is a concern

Epoch Selection

Choose an epoch close to your system’s start:

• Maximizes the timestamp range
• Twitter’s epoch gives ~69 years from 2010
• A 2024 epoch would give ~69 years from 2024

Variants

Different systems use slightly different bit layouts:

Specification

• Original announcement: https://blog.twitter.com/engineering/en_us/a/2010/announcing-snowflake
• Wikipedia: https://en.wikipedia.org/wiki/Snowflake_ID