probabilistic-rs

Probabilistic data structures in Rust with Python bindings and HTTP API.

Features

• Bloom Filter — fast membership testing with bulk ops and optional Fjall persistence
• Expiring Bloom Filter — auto-expires elements via sliding time windows
• HTTP API — REST server with Swagger UI, managing multiple named filters
• Python Bindings — native wheels via PyO3/maturin
• CLI + TUI — interactive terminal interface

Installation

Rust:

cargo add probabilistic-rs

Python:

pip install probabilistic-rs

Quick Start

Rust — Bloom Filter

use probabilistic_rs::bloom::{BloomFilter, BloomFilterConfigBuilder, BloomFilterOps};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let config = BloomFilterConfigBuilder::default()
        .capacity(10000)
        .false_positive_rate(0.01)
        .build()?;

    let filter = BloomFilter::create(config).await?;
    filter.insert(b"item1")?;
    assert!(filter.contains(b"item1")?);
    Ok(())
}

Rust — Expiring Bloom Filter

use probabilistic_rs::ebloom::{
    ExpiringBloomFilter, ExpiringFilterConfigBuilder, ExpiringBloomFilterOps,
};
use std::time::Duration;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let config = ExpiringFilterConfigBuilder::default()
        .capacity(1000)
        .false_positive_rate(0.01)
        .level_duration(Duration::from_secs(60))
        .max_levels(3)
        .build()?;

    let mut filter = ExpiringBloomFilter::new(config)?;
    filter.insert(b"test_item")?;
    assert!(filter.query(b"test_item")?);
    Ok(())
}

Python

from probabilistic_rs import BloomFilter, ExpiringBloomFilter

bf = BloomFilter(capacity=10000, false_positive_rate=0.01)
bf.insert(b"item1")
assert bf.contains(b"item1")

ebf = ExpiringBloomFilter(capacity=1000, false_positive_rate=0.01, ttl_seconds=60)
ebf.insert(b"temp_item")
assert ebf.query(b"temp_item")

HTTP API

# Start server (default: localhost:3000)
probabilistic-server

# Create a filter, insert, query
curl -X POST http://localhost:3000/api/v1/bloom/create \
  -H "Content-Type: application/json" \
  -d '{"name":"my-filter","capacity":10000,"false_positive_rate":0.01}'

curl -X POST http://localhost:3000/api/v1/bloom/insert \
  -H "Content-Type: application/json" \
  -d '{"name":"my-filter","item":"hello"}'

curl -X POST http://localhost:3000/api/v1/bloom/contains \
  -H "Content-Type: application/json" \
  -d '{"name":"my-filter","item":"hello"}'

Endpoints: create, delete, insert, contains, bulk_insert, bulk_contains, clear, stats, list — available for both /api/v1/bloom and /api/v1/ebloom. Swagger UI at /swagger-ui.

CLI

# Create filter
expblf create --db-path myfilter.fjall --capacity 10000 --fpr 0.01

# Operations
expblf load --db-path myfilter.fjall insert --element "key"
expblf load --db-path myfilter.fjall check --element "key"

# Interactive TUI
expblf tui --db-path myfilter.fjall

Benchmarks

Measured on Apple M-series via cargo bench (criterion, 100 samples). Times are total for N operations.

Bloom Filter (in-memory)

Operation	1K elements	100K elements	1M elements
Insert	60.2 µs	6.15 ms	64.1 ms
Query	61.3 µs	6.17 ms	63.1 ms

Expiring Bloom Filter (in-memory, 3 levels — 5 levels nearly identical)

Operation	1K elements	100K elements	1M elements
Insert	63.8 µs	6.61 ms	68.0 ms
Query	63.4 µs	6.52 ms	67.0 ms
Bulk insert	59.7 µs	6.16 ms	63.8 ms
Bulk query	62.3 µs	6.41 ms	65.8 ms
Level rotate	224 µs	255 µs	—

Both filters sustain ~15–17M ops/s (60–65 ns/op) across all dataset sizes. The expiring filter adds ~5–10% overhead over plain bloom due to multi-level bookkeeping. Bulk operations match or slightly outperform single-item ops. Level rotation (TTL expiry) takes ~250 µs regardless of filter size.

Configuration

Parameter	Description	Default
capacity	Max elements	1,000,000
false_positive_rate	Desired FPR	0.01
level_duration	TTL per level (expiring)	60s
max_levels	Filter levels (expiring)	3

License

MIT