This is a Rust implementation of the DDSketch quantile sketch algorithm. DDSketch is a fully-mergeable quantile sketch with relative-error guarantees.
- Implements the DDSketch algorithm with configurable relative error guarantees
- Optimized for high-throughput data collection scenarios
- Designed for distributed systems with efficient sketch merging
use ddsketchy::{DDSketch, DDSketchError};
fn main() -> Result<(), DDSketchError> {
// Create a new sketch with 1% relative error
let mut sketch = DDSketch::new(0.01)?;
// Add some values
sketch.add(1.0);
sketch.add(2.0);
sketch.add(3.0);
sketch.add(4.0);
sketch.add(5.0);
// Get the 50th percentile (median)
let median = sketch.quantile(0.5)?;
println!("Median: {}", median);
// Get the 90th percentile
let p90 = sketch.quantile(0.9)?;
println!("90th percentile: {}", p90);
Ok(())
}ddsketchy supports optional serialization via serde. Serialization is disabled by default to keep the library dependency-free. To enable it, add the serde feature
use ddsketchy::DDSketch;
fn main() -> Result<(), Box<dyn std::error::Error>> {
let mut sketch = DDSketch::new(0.01)?;
sketch.add(1.0);
sketch.add(2.0);
// Serialize to JSON
let json = serde_json::to_string(&sketch)?;
println!("Serialized: {}", json);
// Deserialize from JSON
let restored: DDSketch = serde_json::from_str(&json)?;
// Verify the sketch works correctly
assert_eq!(sketch.count(), restored.count());
assert_eq!(sketch.quantile(0.5)?, restored.quantile(0.5)?);
Ok(())
}The serialization handles all internal state including infinity values for min/max bounds in empty sketches. Empty sketches serialize min/max as null values, while sketches with data serialize them as numbers.
- DDSketch: A Fast and Fully-Mergeable Quantile Sketch with Relative-Error Guarantees - The original paper describing the DDSketch algorithm