JarCastingtime-id
Generates 27-character, time-ordered, k-sortable, URL-safe, globally unique identifiers.
How do I add it to my project?
<dependency>
<groupId>com.codahale</groupId>
<artifactId>time-id</artifactId>
<version>0.4.5</version>
</dependency>
|
Note
|
The module name for Java 9+ is com.codahale.timeid. |
How do I use it?
import com.codahale.timeid.IdGenerator;
class Example {
public static void main(String[] args) throws InterruptedException {
final IdGenerator generator = new IdGenerator();
for (int i = 0; i < 5; i++) {
System.out.println(generator.generate());
Thread.sleep(1000);
}
}
}
1KoN0_AUFKtrzGS_vay5hlOyCFF 1KoN1$8dtcqssctVzGFD1O900o$ 1KoN1Q7uJBNHyQuEo8y$eND_2dk 1KoN1VejHzyMi1MJLNM4sqtmNJ3 1KoN1vqpjk0nUfE25pUjZXCFLjN
|
Tip
|
IdGenerator is thread-safe but synchronized. Highly concurrent applications may wish to use thread-local instances. |
How does it work?
Each ID consists of a 32-bit, big-endian timestamp (the number of seconds since 1.4e9 seconds after the Unix epoch), plus 128 bits of random data, for a total of 160 bits of information.
The random data is produced using ChaCha20 in a fast-key-erasure construction, with a per-ID iteration of the ChaCha20 block transform. The first 256 bits of the result are used as the key for the next ID; the next 128 bits are used in the ID; the remaining state is discarded. This construction is an order of magnitude faster than the fastest java.util.SecureRandom implementation, is nonblocking, has a very small memory footprint, operates in constant time, offers forward secrecy, requires no hardware support, and has performance characteristics independent of JVM configuration.
The timestamp and the random data are encoded with Radix-64 using an alphabet which is both URL-safe and which preserves lexical ordering. The result is a 27-character, URL-safe string which can be used in systems which are unaware of its internal structure (e.g., databases, file systems) to store time-ordered data with unique IDs.
Is it fast?
Benchmark Mode Cnt Score Error Units Benchmarks.generate sample 7041460 270.418 ± 2.743 ns/op Benchmarks.generate:generate·p0.00 sample 211.000 ns/op Benchmarks.generate:generate·p0.50 sample 246.000 ns/op Benchmarks.generate:generate·p0.90 sample 259.000 ns/op Benchmarks.generate:generate·p0.95 sample 264.000 ns/op Benchmarks.generate:generate·p0.99 sample 405.000 ns/op Benchmarks.generate:generate·p0.999 sample 3516.000 ns/op Benchmarks.generate:generate·p0.9999 sample 25472.000 ns/op Benchmarks.generate:generate·p1.00 sample 2912256.000 ns/op Benchmarks.generate:·gc.alloc.rate sample 20 293.922 ± 4.292 MB/sec Benchmarks.generate:·gc.alloc.rate.norm sample 20 72.007 ± 0.001 B/op Benchmarks.generate:·gc.churn.G1_Eden_Space sample 20 293.985 ± 8.721 MB/sec Benchmarks.generate:·gc.churn.G1_Eden_Space.norm sample 20 72.017 ± 1.684 B/op Benchmarks.generate:·gc.churn.G1_Old_Gen sample 20 0.001 ± 0.001 MB/sec Benchmarks.generate:·gc.churn.G1_Old_Gen.norm sample 20 ≈ 10⁻⁴ B/op Benchmarks.generate:·gc.count sample 20 334.000 counts Benchmarks.generate:·gc.time sample 20 208.000 ms
It’s pretty fast. Has a nicely flat latency profile. Doesn’t generate a lot of garbage, either. Mostly just the String allocation. All the internal state is either reused or stack-allocated.
License
Copyright © 2019 Coda Hale
Distributed under the Apache License 2.0.