A Swift micro-framework for generating compact identifiers that are time ordered in distributed systems without the need for synchronization
Switch branches/tags
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.



CI Status Version License Platform

A Swift micro-framework for generating compact identifiers that are time ordered without the need for synchronization.


  • Generate identifiers for data objects.
  • Fits as positive values in 64-bit integer data type — easy to process and store.
  • Roughly time-ordered identifiers.

Intended uses

  • Great for use in Core Data as a way to generate permanent identifiers without needing to save the objects beforehand.
  • Also works well in iCloud scenarios — single-user, multiple devices, and no server-based logic.
  • Should also function pretty good in collaborative multi-device and multi-user scenarios with less than four million concurrent users.
  • Works great as server-side ID generators, even distributed systems environments.


  • Xcode 9.2
  • Swift 4.0
  • macOS 10.12+
  • iOS 10.3+
  • tvOS 10.2+
  • watchOS 3.2+


MiniFlake is available through CocoaPods. To install it, simply add the following line to your Podfile:

pod 'MiniFlake'


First import the library in the top of your source file:

import MiniFlake

Core Data

In your NSManagedObject subclass, override awakeFromInsert and use MiniFlake's NSManagedObjectContext extension method nextFlakeID to pre-populate your object's identifier. You would need to have an indexed attribute of type Int64 as this identifier.

override func awakeFromInsert() {
    self.primitiveIdentifierValue = self.managedObjectContext!.nextFlakeID()
    // ... continue initialization ...

Other Entity Types

The nextFlakeID method is also available as an extension to Thread. You can use this to generate identifiers for your custom persistence objects.

let generatedID = Thread.current.nextFlakeID()

Always use Thread.current and do not call the method for another thread object since ID generation is not thread-safe.

Advanced Uses

The primary classes are FlakeMaker and InProcessFlakeMaker. The former requires you to provide a unique instance number whereas the latter will manage these instance numbers for you. Note that instance numbers are 10-bit values, hence there cannot be more than 1024 generator instances at any given moment or the identifiers may overlap. Call method nextValue() on objects of both classes to generate ID values.

let generator = InProcessFlakeMaker()
let runCount = 100
var results = Set<Int64>(minimumCapacity:runCount)
for _ in 0..<runCount {


The nextFlakeID extension methods are available to call from Objective-C sources, which should cover most use cases. However the FlakeMaker and InProcessFlakeMaker classes are Swift-only.

Importing into Objective-C source file:

#import <MiniFlake/MiniFlake-Swift.h>

Setup of Core Data object:

-(void) awakeFromInsert {
    [super awakeFromInsert];
    self.primitiveIdentifierValue = [self.managedObjectContext nextFlakeID];
    // ... continue initialization ...

Generic identifier creation:

int64_t generatedID = [NSThread.currentThread nextFlakeID];


Open MiniFlake.xcworkspace from the Example folder and run the MiniFlake_Example.app target using Xcode. This is a Mac app that generates an amount of identifiers into a text view, ready for copy-pasting. The main functionality that calls the library is inside the class ViewController, method startGenerate().

How it works

The identifier is inspired by Twitter’s Snowflake ID generator, in which each identifier value is composed by the following components.

  1. 41 bits timestamp, millisecond precision with a custom epoch.
  2. 12 bits sequence number.
  3. 10 bits generator instance identifier.

The timestamp occupies the more significant bits. With 41 bits it would be good for over 69 years before overflowing. The custom epoch is 1 March 2018, hence the identifier won’t wrap around until the year 2087. Even then it would still work but generates negative values instead, hence easily detectable and fixable.

With 12 bits available for the sequence number, each worker can generate over four thousand unique values per millisecond. Because it can count up to 4095 per millisecond, there’s much less likely to have an ID clash as opposed to using a plain Unix timestamp as an ID generator. Moreover an overflow in this field would get passed on to the timestamp field, hence the generator would still create valid values.

The instance identifier helps mitigate the possibility of duplicates. In server-based environments, you would associate this with the service worker number — the instance number of the microservice which hands out these ID values. In on-device scenarios, I’ve created category methods that associates the ID generator with either the thread or the Core Data context.


Copyright © Sasmito Adibowo


MIT license, a.k.a. Simplified BSD. See the LICENSE.markdown file for more information.