Crate ferroid

Expand description

§ferroid

ferroid is a Rust crate for generating and parsing Snowflake and ULID identifiers.

§Features

📌 Bit-level compatibility with major Snowflake and ULID formats
🧩 Pluggable clocks and RNGs via TimeSource and RandSource
🧵 Lock-free, lock-based, and single-threaded generators
📐 Custom layouts via define_snowflake_id! and define_ulid! macros
🔢 Crockford base32 support with base32 feature flag

§📦 Supported Layouts

§Snowflake

Platform	Timestamp Bits	Machine ID Bits	Sequence Bits	Epoch
Twitter	41	10	12	2010-11-04 01:42:54.657
Discord	42	10	12	2015-01-01 00:00:00.000
Instagram	41	13	10	2011-01-01 00:00:00.000
Mastodon	48	0	16	1970-01-01 00:00:00.000

§Ulid

Platform	Timestamp Bits	Random Bits	Epoch
ULID	48	80	1970-01-01 00:00:00.000

§🔧 Generator Comparison

Snowflake Generator	Monotonic	Thread-Safe	Lock-Free	Throughput	Use Case
`BasicSnowflakeGenerator`	✅	❌	❌	Highest	Single-threaded or generator per thread
`LockSnowflakeGenerator`	✅	✅	❌	Medium	Fair multithreaded access
`AtomicSnowflakeGenerator`	✅	✅	✅	High	Fast concurrent generation (less fair)

Ulid Generator	Monotonic	Thread-Safe	Lock-Free	Throughput	Use Case
`BasicUlidGenerator`	❌	✅	❌	Slow	Thread-safe, always random, but slow
`BasicMonoUlidGenerator`	✅	❌	❌	Highest	Single-threaded or generator per thread
`LockMonoUlidGenerator`	✅	✅	❌	High	Fair multithreaded access

§🚀 Usage

§Thread Locals

The simplest way to generate a ULID is via Ulid, which provides a thread-local generator that can produce both non-monotonic and monotonic ULIDs:

#[cfg(all(feature = "ulid", feature = "thread_local"))]
{
    use ferroid::{ULID, Ulid};

    // A ULID (slower, always random within the same millisecond)
    let id: ULID = Ulid::new_ulid();

    // A monotonic ULID (faster, increments within the same millisecond)
    let id: ULID = Ulid::new_mono_ulid();
}

Thread-local generators are not currently available for SnowflakeId-style IDs because they rely on a valid machine_id to avoid collisions. Mapping unique machine_ids across threads requires coordination beyond what thread_local! alone can guarantee.

§Crockford Base32

Enable the base32 feature to support Crockford Base32 encoding and decoding of IDs. This is useful when you need fixed-width, URL-safe, and lexicographically sortable strings (e.g. for databases, logs, or URLs).

With base32 enabled, each ID type automatically implements fmt::Display, which internally uses .encode(). IDs also implement TryFrom<&str> and FromStr, both of which decode via .decode().

For explicit, allocation-free formatting, use .encode() to get a lightweight formatter. This avoids committing to a specific string type and lets the consumer control how and when to render the result. The formatter uses a stack-allocated buffer and avoids heap allocation by default. To enable .to_string() and other owned string functionality, enable the alloc feature.

#[cfg(all(feature = "base32", feature = "snowflake"))]
{
    use ferroid::{Base32SnowExt, Base32SnowFormatter, SnowflakeId, SnowflakeTwitterId};
    use core::str::FromStr;

    let id = SnowflakeTwitterId::from(123_456, 0, 42);

    assert_eq!(format!("{id}"), "00000F280001A");
    assert_eq!(id.encode(), "00000F280001A");
    assert_eq!(SnowflakeTwitterId::decode("00000F280001A").unwrap(), id);
    assert_eq!(SnowflakeTwitterId::try_from("00000F280001A").unwrap(), id);
    assert_eq!(SnowflakeTwitterId::from_str("00000F280001A").unwrap(), id);
}

#[cfg(all(feature = "base32", feature = "ulid"))]
{
    use ferroid::{Base32UlidExt, Base32UlidFormatter, UlidId, ULID};
    use core::str::FromStr;

    let id = ULID::from(123_456, 42);

    assert_eq!(format!("{id}"), "0000003RJ0000000000000001A");
    assert_eq!(id.encode(), "0000003RJ0000000000000001A");
    assert_eq!(ULID::decode("0000003RJ0000000000000001A").unwrap(), id);
    assert_eq!(ULID::try_from("0000003RJ0000000000000001A").unwrap(), id);
    assert_eq!(ULID::from_str("0000003RJ0000000000000001A").unwrap(), id);
}

⚠️ Decoding and Overflow: ULID Spec vs. Ferroid

Base32 encodes in 5-bit chunks. That means:

A u32 (32 bits) maps to 7 Base32 characters (7 × 5 = 35 bits)
A u64 (64 bits) maps to 13 Base32 characters (13 × 5 = 65 bits)
A u128 (128 bits) maps to 26 Base32 characters (26 × 5 = 130 bits)

This creates an invariant: an encoded string may contain more bits than the target type can hold.

The ULID specification is strict:

Technically, a 26-character Base32 encoded string can contain 130 bits of information, whereas a ULID must only contain 128 bits. Therefore, the largest valid ULID encoded in Base32 is 7ZZZZZZZZZZZZZZZZZZZZZZZZZ, which corresponds to an epoch time of 281474976710655 or 2 ^ 48 - 1.

Any attempt to decode or encode a ULID larger than this should be rejected by all implementations, to prevent overflow bugs.

Ferroid takes a more flexible stance:

Strings like "ZZZZZZZZZZZZZZZZZZZZZZZZZZ" (which technically overflow) are accepted and decoded without error.
However, if any of the overflowed bits fall into reserved regions, which must remain zero, decoding will fail with Base32Error::DecodeOverflow.

This allows any 13-character Base32 string to decode into a u64, or any 26-character string into a u128, as long as reserved layout constraints aren’t violated. If the layout defines no reserved bits, decoding is always considered valid.

For example:

A ULID has no reserved bits, so decoding will never fail due to overflow.
A SnowflakeTwitterId reserves the highest bit, so decoding must ensure that bit remains unset.

If reserved bits are set during decoding, Ferroid returns a Base32Error::DecodeOverflow { id } containing the full (invalid) ID. You can recover by calling .into_valid() to mask off reserved bits-allowing either explicit error handling or silent correction.

§Generate an ID

§Clocks

In std environments, you can use the default MonotonicClock implementation. It is thread-safe, lightweight to clone, and intended to be shared across the application. If you’re using multiple generators, clone and reuse the same clock instance.

By default, MonotonicClock::default() sets the offset to UNIX_EPOCH. You should override this depending on the ID specification. For example, Twitter IDs use TWITTER_EPOCH, which begins at Thursday, November 4, 2010, 01:42:54.657 UTC (millisecond zero).

#[cfg(all(feature = "std", feature = "alloc"))]
{
    use ferroid::{MonotonicClock, UNIX_EPOCH};

    // Same as MonotonicClock::default();
    let clock = MonotonicClock::with_epoch(UNIX_EPOCH);

    // let generator0 = BasicSnowflakeGenerator::new(0, clock.clone());
    // let generator1 = BasicSnowflakeGenerator::new(1, clock.clone());
}

§Synchronous Generators

Calling next_id() may yield Pending if the current sequence is exhausted. Please note that while this behavior is exposed to provide maximum flexibility, you must be generating enough IDs per millisecond to draw out the Pending path. You may spin, yield, or sleep depending on your environment:

#[cfg(all(feature = "std", feature = "alloc", feature = "snowflake"))]
{
    use ferroid::{MonotonicClock, IdGenStatus, TWITTER_EPOCH, BasicSnowflakeGenerator, SnowflakeTwitterId};

    let generator = BasicSnowflakeGenerator::new(0, MonotonicClock::with_epoch(TWITTER_EPOCH));

    let id: SnowflakeTwitterId = loop {
        match generator.next_id() {
            IdGenStatus::Ready { id } => break id,
            IdGenStatus::Pending { yield_for } => {
                println!("Exhausted; wait for: {}ms", yield_for);
                core::hint::spin_loop(); // Blocking spin: burns CPU, but yields the lowest latency.
                // std::thread::yield_now(); // Optional: yields to OS, still busy-waits.
                // std::thread::sleep(Duration::from_millis(yield_for.to_u64())); // Lowest CPU use, but imprecise and may oversleep.
                //
                // For non-blocking ID generation, use the async API (see below).
            }
        }
    };
}

#[cfg(all(feature = "std", feature = "alloc", feature = "ulid"))]
{
    use ferroid::{MonotonicClock, IdGenStatus, ThreadRandom, BasicUlidGenerator, ULID};

    let generator = BasicUlidGenerator::new(MonotonicClock::default(), ThreadRandom::default());

    let id: ULID = loop {
        match generator.next_id() {
            IdGenStatus::Ready { id } => break id,
            IdGenStatus::Pending { yield_for } => {
                println!("Exhausted; wait for: {}ms", yield_for);
                core::hint::spin_loop(); // Blocking spin: burns CPU, but yields the lowest latency.
                // std::thread::yield_now(); // Optional: yields to OS, still busy-waits.
                // std::thread::sleep(Duration::from_millis(yield_for.to_u64())); // Lowest CPU use, but imprecise and may oversleep.
                //
                // For non-blocking ID generation, use the async API (see below).
            }
        }
    };
}

§Asynchronous Generators

If you’re in an async context (e.g., using Tokio or Smol), enable one of the following features to avoid blocking behavior:

aysnc-tokio
async-smol

These features extend the generator to yield cooperatively when it returns Pending, causing the current task to sleep for the specified yield_for duration (typically ~1ms). While this is fully non-blocking, it may oversleep slightly due to OS or executor timing precision, potentially reducing peak throughput.

#[cfg(feature = "async-tokio")]
{
    use ferroid::{Result, MonotonicClock, MASTODON_EPOCH, UNIX_EPOCH};

    #[tokio::main]
    async fn main() -> Result<()> {
        #[cfg(feature = "snowflake")]
        {
            use ferroid::{
                BasicSnowflakeGenerator, SnowflakeMastodonId,
                SnowflakeGeneratorAsyncTokioExt
            };

            let generator = BasicSnowflakeGenerator::new(0, MonotonicClock::with_epoch(MASTODON_EPOCH));

            let id: SnowflakeMastodonId = generator.try_next_id_async().await?;
            println!("Generated ID: {}", id);
        }

        #[cfg(feature = "ulid")]
        {
            use ferroid::{ThreadRandom, UlidGeneratorAsyncTokioExt, BasicUlidGenerator, ULID};

            let generator = BasicUlidGenerator::new(MonotonicClock::with_epoch(UNIX_EPOCH), ThreadRandom::default());

            let id: ULID = generator.try_next_id_async().await?;
            println!("Generated ID: {}", id);
        }
        Ok(())
    }
    main().expect("failed to run")
}

#[cfg(feature = "async-smol")]
{
    use ferroid::{Result, MonotonicClock};

    fn main() -> Result<()> {
        smol::block_on(async {
            #[cfg(feature = "snowflake")]
            {
                use ferroid::{
                    BasicSnowflakeGenerator, SnowflakeMastodonId,
                    SnowflakeGeneratorAsyncSmolExt, MASTODON_EPOCH
                };

                let generator = BasicSnowflakeGenerator::new(0, MonotonicClock::with_epoch(MASTODON_EPOCH));

                let id: SnowflakeMastodonId = generator.try_next_id_async().await?;
                println!("Generated ID: {}", id);
            }

            #[cfg(feature = "ulid")]
            {
                use ferroid::{ThreadRandom, UlidGeneratorAsyncSmolExt, BasicUlidGenerator, ULID, UNIX_EPOCH};

                let generator = BasicUlidGenerator::new(MonotonicClock::with_epoch(UNIX_EPOCH), ThreadRandom::default());

                let id: ULID = generator.try_next_id_async().await?;
                println!("Generated ID: {}", id);
            }

            Ok(())
        })
    }
    main().expect("failed to run")
}

§Custom Layouts

To gain more control or optimize for different performance characteristics, you can define a custom layout.

Use the define_* macros below to create a new struct with your chosen name. The resulting type behaves just like built-in types such as SnowflakeTwitterId or ULID, with no extra setup required and full compatibility with the existing API.

#[cfg(feature = "snowflake")]
{
    use ferroid::{define_snowflake_id};

    // Example: a 64-bit Twitter-like ID layout
    //
    //  Bit Index:  63           63 62            22 21             12 11             0
    //              +--------------+----------------+-----------------+---------------+
    //  Field:      | reserved (1) | timestamp (41) | machine ID (10) | sequence (12) |
    //              +--------------+----------------+-----------------+---------------+
    //              |<----------- MSB ---------- 64 bits ----------- LSB ------------>|
    define_snowflake_id!(
        MyCustomId, u64,
        reserved: 1,
        timestamp: 41,
        machine_id: 10,
        sequence: 12
    );
}

#[cfg(feature = "ulid")]
{
    use ferroid::define_ulid;

    // Example: a 128-bit ULID using the Ulid layout
    //
    // - 0 bits reserved
    // - 48 bits timestamp
    // - 80 bits random
    //
    //  Bit Index:  127            80 79           0
    //              +----------------+-------------+
    //  Field:      | timestamp (48) | random (80) |
    //              +----------------+-------------+
    //              |<-- MSB -- 128 bits -- LSB -->|
    define_ulid!(
        MyULID, u128,
        reserved: 0,
        timestamp: 48,
        random: 80
    );
}

⚠️ Note: When using the snowflake macro, you must specify all four sections (in order): reserved, timestamp, machine_id, and sequence-even if a section uses 0 bits.

The reserved bits are always set to zero and can be reserved for future use.

Similarly, the ulid macro requires all three fields: reserved, timestamp, and random.

§Behavior

§Snowflake

If the clock advances: reset sequence to 0 → IdGenStatus::Ready
If the clock is unchanged: increment sequence → IdGenStatus::Ready
If the clock goes backward: return IdGenStatus::Pending
If the sequence increment overflows: return IdGenStatus::Pending

§Ulid

This implementation respects monotonicity within the same millisecond in a single generator by incrementing the random portion of the ID and guarding against overflow.

If the clock advances: generate new random → IdGenStatus::Ready
If the clock is unchanged: increment random → IdGenStatus::Ready
If the clock goes backward: return IdGenStatus::Pending
If the random increment overflows: return IdGenStatus::Pending

§Probability of ID Collisions

When generating time-sortable IDs that use random bits, it’s important to estimate the probability of collisions (i.e., two IDs being the same within the same millisecond), given your ID layout and system throughput.

§Monotonic IDs with Multiple ULID Generators

If you have $g$ generators (e.g., distributed nodes), and each generator produces $k$ sequential (monotonic) IDs per millisecond by incrementing from a random starting point, the probability that any two generators produce overlapping IDs in the same millisecond is approximately:

$$P_\text{collision} \approx \frac{g(g-1)(2k-1)}{2 \cdot 2^r}$$

Where:

$g$ = number of generators
$k$ = number of monotonic IDs per generator per millisecond
$r$ = number of random bits per ID
$P_\text{collision}$ = probability of at least one collision

Note: The formula above uses the approximate (birthday bound) model, which assumes that:

$k \ll 2^r$ and $g \ll 2^r$

Each generator’s range of $k$ IDs starts at a uniformly random position within the $r$-bit space

§Estimating Time Until a Collision Occurs

While collisions only happen within a single millisecond, we often want to know how long it takes before any collision happens, given continuous generation over time.

The expected time in milliseconds to reach a 50% chance of collision is:

$T_{\text{50%}} \approx \frac{\ln 2}{P_\text{collision}} = \frac{0.6931 \cdot 2 \cdot 2^r}{g(g - 1)(2k - 1)}$

This is derived from the cumulative probability formula:

$P_\text{collision}(T) = 1 - (1 - P_\text{collision})^T$

Solving for $T$ when $P_\text{collision}(T) = 0.5$:

$(1 - P_\text{collision})^T = 0.5$

$\Rightarrow T \approx \frac{\ln(0.5)}{\ln(1 - P_\text{collision})}$

Using the approximation $\ln(1 - x) \approx -x$ for small $x$, this simplifies to:

$\Rightarrow T \approx \frac{\ln 2}{P_\text{collision}}$

The $\ln 2$ term arises because $\ln(0.5) = -\ln 2$. After $T_\text{50%}$ milliseconds, there’s a 50% chance that at least one collision has occurred.

Generators ($g$)	IDs per generator per ms ($k$)	$P_\text{collision}$	Estimated Time to 50% Collision ($T_{\text{50%}}$)
1	1	$0$ (single generator; no collision possible)	∞ (no collision possible)
1	65,536	$0$ (single generator; no collision possible)	∞ (no collision possible)
2	1	$\displaystyle \frac{2 \times 1 \times 1}{2 \cdot 2^{80}} \approx 8.27 \times 10^{-25}$	$\approx 8.38 \times 10^{23} \text{ ms}$
2	65,536	$\displaystyle \frac{2 \times 1 \times 131{,}071}{2 \cdot 2^{80}} \approx 1.08 \times 10^{-19}$	$\approx 6.41 \times 10^{18} \text{ ms}$
1,000	1	$\displaystyle \frac{1{,}000 \times 999 \times 1}{2 \cdot 2^{80}} \approx 4.13 \times 10^{-19}$	$\approx 1.68 \times 10^{18} \text{ ms}$
1,000	65,536	$\displaystyle \frac{1{,}000 \times 999 \times 131{,}071}{2 \cdot 2^{80}} \approx 5.42 \times 10^{-14}$	$\approx 1.28 \times 10^{13} \text{ ms} \approx 406\ years$

§📈 Benchmarks

Snowflake ID generation is theoretically capped by:

max IDs/sec = 2^sequence_bits × 1000ms

For example, Twitter-style IDs (12 sequence bits) allow:

4096 IDs/ms × 1000 ms/sec = ~4M IDs/sec

To benchmark this, we generate IDs in chunks of 4096, which aligns with the sequence limit per millisecond in Snowflake layouts. For ULIDs, we use the same chunk size for consistency, but this number does not represent a hard throughput cap - ULID generation is probabilistic: monotonicity within the same millisecond increments the random bit value. Chunking here primarily serves to keep the benchmark code consistent.

Async benchmarks are tricky because a single generator’s performance is affected by task scheduling, which is not predictable and whose scheduler typically has a resolution of 1 millisecond. By the time a task is scheduled to execute (i.e., generate an ID), a millisecond may have already passed, potentially resetting any sequence counter or monotonic increment - thus, never truly testing the hot path. To mitigate this, async tests measure maximum throughput: each task generates a batch of IDs and may await on any of them. This approach offsets idle time on one generator with active work on another, yielding more representative throughput numbers.

§Snowflake:

Sync: Benchmarks the hot path without yielding to the clock.
Async: Also uses 4096-ID batches, but may yield (sequence exhaustion/CAS failure) or await due to task scheduling, reducing throughput.

§ULID:

Sync & Async: Uses the same 4096-ID batches. Due to random number generation, monotonic increments may overflow randomly, reflecting real-world behavior. In general, it is rare for ULIDs to overflow.

Tests were ran on an M1 Macbook Pro 14“, 32GB, 10 cores (8 performance, 2 efficiency).

§Synchronous Generators

Generator	Time per ID	Throughput
`BasicSnowflakeGenerator`	~2.8 ns	~353M IDs/sec
`LockSnowflakeGenerator`	~8.9 ns	~111M IDs/sec
`AtomicSnowflakeGenerator`	~3.1 ns	~320M IDs/sec
`BasicUlidGenerator`	~20.4 ns	~44M IDs/sec
`BasicMonoUlidGenerator`	~3.4 ns	~288M IDs/sec
`LockMonoUlidGenerator`	~9.2 ns	~109M IDs/sec

§Thread Local Generators

Generator	Time per ID	Throughput
`Ulid::new_ulid`	~24 ns	~41.7M IDs/sec
`Ulid::new_mono_ulid`	~5.6 ns	~178M IDs/sec

§Async (Tokio Runtime) - Peak throughput

Generator	Generators	Time per ID	Throughput
`LockSnowflakeGenerator`	1024	~1.46 ns	~687M IDs/sec
`AtomicSnowflakeGenerator`	1024	~0.86 ns	~1.17B IDs/sec
`LockMonoUlidGenerator`	1024	~1.57 ns	~635M IDs/sec

§Async (Smol Runtime) - Peak throughput

Generator	Generators	Time per ID	Throughput
`LockSnowflakeGenerator`	1024	~1.40 ns	~710M IDs/sec
`AtomicSnowflakeGenerator`	1024	~0.62 ns	~1.61B IDs/sec
`LockMonoUlidGenerator`	1024	~1.32 ns	~756M IDs/sec

To run all benchmarks:

cargo criterion --all-features

§🧪 Testing

Run all tests with:

cargo test --all-features

§📄 License

Licensed under either of:

at your option.

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Macros§

cfg_base32: Helper to implement fmt::Display when the base32 feature is enabled.
define_snowflake_id: A macro for defining a bit layout for a custom Snowflake ID using four required components: reserved, timestamp, machine_id, and sequence.
define_ulid: A macro for defining a bit layout for a custom Ulid using three required components: reserved, timestamp, and random.

Structs§

AtomicSnowflakeGenerator: A lock-free Snowflake ID generator suitable for multi-threaded environments.
Base32SnowFormatter: A reusable builder that owns the Base32 buffer and formats an ID.
Base32SnowFormatterRef: A builder that borrows a user-supplied buffer for Base32 formatting.
Base32UlidFormatter: A reusable builder that owns the Base32 buffer and formats an ID.
Base32UlidFormatterRef: A builder that borrows a user-supplied buffer for Base32 formatting.
BasicMonoUlidGenerator: A monotonic ULID-style ID generator suitable for single-threaded environments.
BasicSnowflakeGenerator: A non-concurrent Snowflake ID generator suitable for single-threaded environments.
BasicUlidGenerator: A non-monotonic ULID-style ID generator suitable for multi-threaded environments.
LockMonoUlidGenerator: A monotonic ULID-style ID generator suitable for multi-threaded environments.
LockSnowflakeGenerator: A lock-based Snowflake ID generator suitable for multi-threaded environments.
MonotonicClock: A monotonic time source that returns elapsed time since process start, offset from a user-defined epoch.
SmolSleep: An implementation of SleepProvider using Smol’s timer.
SmolSleepFuture: Internal future returned by SmolSleep::sleep_for.
SmolYield: An implementation of SleepProvider using Smol’s yield.
SnowflakeDiscordId: A 64-bit Snowflake ID using the Discord layout
SnowflakeGeneratorFuture: A future that polls a SnowflakeGenerator until it is ready to produce an ID.
SnowflakeInstagramId: A 64-bit Snowflake ID using the Instagram layout
SnowflakeMastodonId: A 64-bit Snowflake ID using the Mastodon layout
SnowflakeTwitterId: A 64-bit Snowflake ID using the Twitter layout
ThreadRandom: A RandSource that uses the thread-local RNG (rand::thread_local()).
TokioSleep: An implementation of SleepProvider using Tokio’s timer.
TokioYield: An implementation of SleepProvider using Tokio’s yield.
ULID: A 128-bit ULID
Ulid: A thread-local ULID generator with monotonic and non-monotonic modes.
UlidGeneratorFuture: A future that polls a UlidGenerator until it is ready to produce an ID.

Enums§

Backoff: Backoff strategies for handling monotonic ULID overflow.
Base32Error: Errors that can occur while decoding Crockford Base32 strings.
Error: All possible errors that ferroid can produce.
IdGenStatus: Represents the result of attempting to generate a new Snowflake ID.

Constants§

DISCORD_EPOCH: Discord epoch: Thursday, January 1, 2015 00:00:00 UTC
INSTAGRAM_EPOCH: Instagram epoch: Saturday, January 1, 2011 00:00:00 UTC
MASTODON_EPOCH: Mastodon epoch: Thursday, January 1, 1970 00:00:00 UTC
TWITTER_EPOCH: Twitter epoch: Thursday, November 4, 2010 1:42:54.657 UTC
UNIX_EPOCH: Unix epoch: Thursday, January 1, 1970 00:00:00 UTC

Traits§

Base32SnowExt: Extension trait for Crockford Base32 encoding and decoding of ID types.
Base32UlidExt: Extension trait for Crockford Base32 encoding and decoding of ID types.
BeBytes: A trait for types that can be encoded to and decoded from big-endian bytes.
Id: A trait for types that wrap a primitive scalar identifier.
RandSource: A trait for random sources that return a random byte integers.
SleepProvider: A trait that abstracts over how to sleep for a given Duration in async contexts.
SnowflakeGenerator: A minimal interface for generating Snowflake IDs
SnowflakeGeneratorAsyncExt: Extension trait for asynchronously generating Snowflake IDs.
SnowflakeGeneratorAsyncSmolExt: Extension trait for asynchronously generating Snowflake IDs using the smol async runtime.
SnowflakeGeneratorAsyncTokioExt: Extension trait for asynchronously generating Snowflake IDs using the tokio async runtime.
SnowflakeId: A trait representing a layout-compatible Snowflake ID generator.
TimeSource: A trait for time sources that return a monotonic or wall-clock timestamp.
ToU64: Trait for converting numeric-like values into a u64.
UlidGenerator: A minimal interface for generating Ulid IDs
UlidGeneratorAsyncExt: Extension trait for asynchronously generating ULIDs.
UlidGeneratorAsyncSmolExt: Extension trait for asynchronously generating ULIDs using the smol async runtime.
UlidGeneratorAsyncTokioExt: Extension trait for asynchronously generating ULIDs using the tokio async runtime.
UlidId: Trait for layout-compatible ULID-style identifiers.

Type Aliases§

Result: A result type that is infallible by default.

Crate ferroidCopy item path

§ferroid

§Features

§📦 Supported Layouts

§Snowflake

§Ulid

§🔧 Generator Comparison

§🚀 Usage

§Thread Locals

§Crockford Base32

§Generate an ID

§Clocks

§Synchronous Generators

§Asynchronous Generators

§Custom Layouts

§Behavior

§Snowflake

§Ulid

§Probability of ID Collisions

§Monotonic IDs with Multiple ULID Generators

§Estimating Time Until a Collision Occurs

§📈 Benchmarks

§Snowflake:

§ULID:

§Synchronous Generators

§Thread Local Generators

§Async (Tokio Runtime) - Peak throughput

§Async (Smol Runtime) - Peak throughput

§🧪 Testing

§📄 License

Macros§

Structs§

Enums§

Constants§

Traits§

Type Aliases§

Crate ferroid