Contatori

High-performance sharded atomic counters for Rust.

A library providing thread-safe, high-performance counters optimized for highly concurrent workloads. This library implements a sharded counter pattern that dramatically reduces contention compared to traditional single atomic counters.

The Problem

In multi-threaded applications, a naive approach to counting uses a single atomic variable shared across all threads. While this is correct, it creates a severe performance bottleneck: every increment operation causes cache line bouncing between CPU cores, as each core must acquire exclusive access to the cache line containing the counter.

This contention grows worse with more threads and higher update frequencies, turning what should be a simple operation into a major scalability bottleneck.

The Solution: Sharded Counters

This library solves the contention problem by sharding counters across multiple slots (64 by default). Each thread is assigned to a specific slot, so threads updating the counter typically operate on different memory locations, eliminating contention.

Design Principles

1. Per-Thread Sharding: Each thread gets assigned a slot index via thread_local!, ensuring that concurrent updates from different threads don't compete for the same cache line.

2. Cache Line Padding: Each slot is wrapped in CachePadded, which adds padding to ensure each atomic value occupies its own cache line (typically 64 bytes). This prevents false sharing where unrelated data on the same cache line causes unnecessary invalidations.

3. Relaxed Ordering: All atomic operations use Ordering::Relaxed since counters don't need to establish happens-before relationships with other memory operations. This allows maximum optimization by the CPU.

4. Aggregation on Read: The global counter value is computed by summing all slots. This makes reads slightly more expensive but keeps writes extremely fast, which is the right trade-off for counters (many writes, few reads).

Performance Benchmark

Single Counter: Sharded vs AtomicUsize

Benchmarked on Apple M2 (8 cores) with 8 threads, each performing 1,000,000 increments (8 million total operations):

┌─────────────────────────────────────────────────────────────────────────────┐
│                    Counter Performance Comparison                           │
│                   (8 threads × 1,000,000 iterations)                        │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  AtomicUsize (single)   ████████████████████████████████████████  162.53 ms │
│                                                                             │
│  Unsigned (sharded)     █                                           2.27 ms │
│                                                                             │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  Speedup: 71.6x faster                                                      │
│                                                                             │
└─────────────────────────────────────────────────────────────────────────────┘

The sharded counter is ~72x faster than a naive atomic counter under high contention. This difference grows with more threads and higher contention.

Contatori vs OpenTelemetry Counters/Gauges

Benchmarked on Apple M2 (8 cores) with 8 threads, each performing 100,000 increments:

┌─────────────────────────────────────────────────────────────────────────────┐
│              Counter Performance: Contatori vs OpenTelemetry                │
│                        (8 threads × 100,000 iterations)                     │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  Simple counter (no labels):                                                │
│                                                                             │
│  OpenTelemetry Counter  ████████████████████████████████████████   25.81 ms │
│                                                                             │
│  contatori Monotone     █                                          0.33 ms  │
│                                                                             │
│  Speedup: 79x faster                                                        │
│                                                                             │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  Labeled counters (rotating GET/POST/PUT/DELETE):                           │
│                                                                             │
│  OpenTelemetry Counter  ████████████████████████████████████████  356.46 ms │
│                                                                             │
│  cont. labeled_group!   ▏                                          0.21 ms  │
│                                                                             │
│  Speedup: 1665x faster                                                      │
│                                                                             │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  High contention (all threads same label):                                  │
│                                                                             │
│  OpenTelemetry Counter  ████████████████████████████████████████  350.45 ms │
│                                                                             │
│  cont. labeled_group!   ▏                                          0.32 ms  │
│                                                                             │
│  Speedup: 1093x faster                                                      │
│                                                                             │
└─────────────────────────────────────────────────────────────────────────────┘

Contatori is 79x to ~1600x faster than OpenTelemetry counters depending on usage pattern. This massive difference comes from:

• Zero runtime overhead: Labels are resolved at compile time
• Sharded storage: Each sub-counter uses the same sharding strategy
• No dynamic dispatch: Direct field access instead of hash lookups

Available Counter Types

Quick Start

Add to your Cargo.toml:

[dependencies]
contatori = "0.7"

Basic Usage

use contatori::counters::unsigned::Unsigned;
use contatori::counters::Observable;

// Create a counter (can be shared across threads via Arc)
let counter = Unsigned::new().with_name("requests");

// Increment from any thread - extremely fast!
counter.add(1);
counter.add(5);

// Read the total value (aggregates all shards)
println!("Total requests: {}", counter.value());
// value() does NOT reset the counter - it just reads
println!("Still: {}", counter.value()); // Still 6

Resettable Counters

To reset a counter when reading (useful for per-period metrics), wrap it with Resettable:

use contatori::counters::unsigned::Unsigned;
use contatori::counters::Observable;
use contatori::adapters::Resettable;

// Create a resettable counter for per-period metrics
let requests_per_second = Resettable::new(Unsigned::new().with_name("requests_per_second"));

requests_per_second.add(100);

// value() returns the value AND resets the counter
let count = requests_per_second.value();
println!("Requests this period: {}", count); // 100
println!("After reset: {}", requests_per_second.value()); // 0

Multi-threaded Usage

use contatori::counters::unsigned::Unsigned;
use contatori::counters::Observable;
use std::sync::Arc;
use std::thread;

let counter = Arc::new(Unsigned::new());
let mut handles = vec![];

for _ in 0..8 {
    let c = Arc::clone(&counter);
    handles.push(thread::spawn(move || {
        for _ in 0..1_000_000 {
            c.add(1);
        }
    }));
}

for h in handles {
    h.join().unwrap();
}

assert_eq!(counter.value(), contatori::counters::CounterValue::Unsigned(8_000_000));

Tracking Statistics

use contatori::counters::minimum::Minimum;
use contatori::counters::maximum::Maximum;
use contatori::counters::average::Average;
use contatori::counters::Observable;

let min_latency = Minimum::new().with_name("latency_min");
let max_latency = Maximum::new().with_name("latency_max");
let avg_latency = Average::new().with_name("latency_avg");

// Record some latencies
for latency in [150, 85, 200, 120, 95] {
    min_latency.observe(latency);
    max_latency.observe(latency);
    avg_latency.observe(latency);
}

println!("Min: {}", min_latency.value());  // 85
println!("Max: {}", max_latency.value());  // 200
println!("Avg: {}", avg_latency.value());  // 130

Thread Safety

All counter types are Send + Sync and can be safely shared across threads using Arc<Counter>. In addition, counters constructors are const functions, enabling global initialization. The sharding ensures that concurrent updates are efficient.

Memory Usage

Each counter uses approximately 8KB of memory (128 slots × 64 bytes per cache line). This is a trade-off: more memory for dramatically better performance under contention.

Serialization & Observers

The library provides modules for serializing and exporting counter values in various formats. Each module is gated behind a feature flag:

Snapshot Module

The snapshot module provides serializable types that work with any serde-compatible format (JSON, YAML, TOML, bincode, etc.).

[dependencies]
contatori = { version = "0.7", features = ["serde"] }

use contatori::counters::unsigned::Unsigned;
use contatori::counters::Observable;
use contatori::snapshot::{CounterSnapshot, MetricsSnapshot};

let requests = Unsigned::new().with_name("requests");
let errors = Unsigned::new().with_name("errors");

requests.add(1000);
errors.add(5);

let counters: Vec<&dyn Observable> = vec![&requests, &errors];

// Collect snapshots
let snapshot = MetricsSnapshot::collect(counters.into_iter());

// Serialize with any serde-compatible format
let json = serde_json::to_string(&snapshot).unwrap();
let yaml = serde_yaml::to_string(&snapshot).unwrap();
let bytes = bincode::serialize(&snapshot).unwrap();

TableObserver

Renders counters as formatted ASCII tables using the tabled crate.

[dependencies]
contatori = { version = "0.7", features = ["table"] }

use contatori::counters::unsigned::Unsigned;
use contatori::counters::Observable;
use contatori::observers::table::{TableObserver, TableStyle};

let requests = Unsigned::new().with_name("requests");
let errors = Unsigned::new().with_name("errors");

requests.add(1000);
errors.add(5);

let counters: Vec<&dyn Observable> = vec![&requests, &errors];

// Standard format (vertical list)
let observer = TableObserver::new().with_style(TableStyle::Rounded);
println!("{}", observer.render(counters.into_iter()));
// ╭──────────┬───────╮
// │ Name     │ Value │
// ├──────────┼───────┤
// │ requests │ 1000  │
// │ errors   │ 5     │
// ╰──────────┴───────╯

// Compact format (multiple columns)
let observer = TableObserver::new()
    .compact(true)
    .columns(3);
println!("{}", observer.render(counters.into_iter()));
// ╭────────────────┬───────────┬──────────────╮
// │ requests: 1000 │ errors: 5 │ latency: 120 │
// ╰────────────────┴───────────┴──────────────╯

Available styles: Ascii, Rounded, Sharp, Modern, Extended, Markdown, ReStructuredText, Dots, Blank, Double

Compact separators: Colon (:), Equals (=), Arrow (→), Pipe (|), Space

JsonObserver

Serializes counters to JSON format using serde.

[dependencies]
contatori = { version = "0.7", features = ["json"] }

use contatori::counters::unsigned::Unsigned;
use contatori::counters::Observable;
use contatori::observers::json::JsonObserver;

let requests = Unsigned::new().with_name("http_requests");
let errors = Unsigned::new().with_name("http_errors");

requests.add(1000);
errors.add(5);

let counters: Vec<&dyn Observable> = vec![&requests, &errors];

// Simple array output
let json = JsonObserver::new()
    .to_json(counters.into_iter())
    .unwrap();

// Pretty-printed output with timestamp wrapper
let json = JsonObserver::new()
    .pretty(true)
    .wrap_in_snapshot(true)
    .include_timestamp(true)
    .to_json(counters.into_iter())
    .unwrap();

PrometheusObserver

Exports counters in Prometheus exposition format using the official prometheus crate.

[dependencies]
contatori = { version = "0.7", features = ["prometheus"] }

Automatic Metric Type Detection

The observer automatically determines the correct Prometheus metric type based on the counter's metric_kind() method:

This means you don't need to manually specify types for most use cases:

use contatori::counters::monotone::Monotone;
use contatori::counters::signed::Signed;
use contatori::counters::{Observable, MetricKind};
use contatori::observers::prometheus::PrometheusObserver;

// Monotone returns MetricKind::Counter, auto-detected as Prometheus Counter
let requests = Monotone::new().with_name("http_requests_total");
assert_eq!(requests.metric_kind(), MetricKind::Counter);
requests.add(1000);

// Signed returns MetricKind::Gauge, auto-detected as Prometheus Gauge
let connections = Signed::new().with_name("active_connections");
assert_eq!(connections.metric_kind(), MetricKind::Gauge);
connections.add(42);

let counters: Vec<&dyn Observable> = vec![&requests, &connections];

let observer = PrometheusObserver::new()
    .with_namespace("myapp")
    .with_help("http_requests_total", "Total number of HTTP requests")
    .with_help("active_connections", "Current number of active connections");

let output = observer.render(counters.into_iter()).unwrap();
// Output will have:
// # TYPE myapp_http_requests_total counter
// # TYPE myapp_active_connections gauge

OpenTelemetryObserver

Exports counters to OpenTelemetry using observable instruments (callbacks). When OpenTelemetry collects metrics, it calls the registered callbacks which read values directly from contatori counters.

[dependencies]
contatori = { version = "0.7", features = ["opentelemetry"] }
opentelemetry = "0.27"
opentelemetry_sdk = { version = "0.27", features = ["rt-tokio"] }
opentelemetry-stdout = { version = "0.27", features = ["metrics"] }
tokio = { version = "1", features = ["rt-multi-thread", "macros"] }

use contatori::counters::monotone::Monotone;
use contatori::counters::unsigned::Unsigned;
use contatori::counters::Observable;
use contatori::observers::opentelemetry::OtelObserver;

use opentelemetry_sdk::metrics::{PeriodicReader, SdkMeterProvider};
use opentelemetry_sdk::runtime;
use std::time::Duration;

// Define static counters
static HTTP_REQUESTS: Monotone = Monotone::new().with_name("http_requests_total");
static ACTIVE_CONNECTIONS: Unsigned = Unsigned::new().with_name("active_connections");

#[tokio::main]
async fn main() {
    // Setup OpenTelemetry with stdout exporter
    let exporter = opentelemetry_stdout::MetricExporter::default();
    let reader = PeriodicReader::builder(exporter, runtime::Tokio)
        .with_interval(Duration::from_secs(60))
        .build();
    let provider = SdkMeterProvider::builder().with_reader(reader).build();
    opentelemetry::global::set_meter_provider(provider.clone());

    // Register contatori metrics with OpenTelemetry
    let observer = OtelObserver::new("my_service");
    let counters: &[&'static (dyn Observable + Send + Sync)] = &[
        &HTTP_REQUESTS,
        &ACTIVE_CONNECTIONS,
    ];
    observer.register(counters).unwrap();

    // Update counters
    HTTP_REQUESTS.add(100);
    ACTIVE_CONNECTIONS.add(5);

    // Flush metrics (they will be printed to stdout)
    provider.force_flush().unwrap();
    provider.shutdown().unwrap();
}

Automatic Metric Type Detection

The observer automatically determines the correct OpenTelemetry instrument type based on the counter's metric_kind() method:

OtelObserver Configuration

Labeled Groups Support

Labeled groups are automatically exported with OpenTelemetry attributes:

use contatori::labeled_group;
use contatori::counters::unsigned::Unsigned;

labeled_group!(
    HttpByMethod,
    "http_requests_by_method",
    "method",
    value: Unsigned,
    get: "GET": Unsigned,
    post: "POST": Unsigned,
);

static HTTP_METHODS: HttpByMethod = HttpByMethod::new();

// Each counter becomes a data point with the "method" attribute
HTTP_METHODS.value.add(150); // Base metric (no label)
HTTP_METHODS.get.add(100);   // method="GET"
HTTP_METHODS.post.add(50);   // method="POST"

Note: Counters must be 'static and implement Send + Sync to be registered with OpenTelemetry, as the callbacks are invoked asynchronously.

Adapters

The library provides adapter types that add additional behavior to counters while maintaining compatibility with the Observable trait.

Resettable

Wraps a counter to reset it when value() is called. Useful for evaluating metrics over observation periods (e.g., requests per second, errors per minute).

use contatori::counters::unsigned::Unsigned;
use contatori::counters::Observable;
use contatori::adapters::Resettable;

let requests = Resettable::new(Unsigned::new().with_name("requests_per_period"));
requests.add(100);

// value() returns the value AND resets the counter
assert_eq!(requests.value().as_u64(), 100);
assert_eq!(requests.value().as_u64(), 0); // Reset to 0!

requests.add(50);
assert_eq!(requests.value().as_u64(), 50); // Just this period

Regular counters (without Resettable) keep their value across reads:

use contatori::counters::unsigned::Unsigned;
use contatori::counters::Observable;

let total = Unsigned::new().with_name("total_requests");
total.add(100);

// value() just reads, does NOT reset
assert_eq!(total.value().as_u64(), 100);
assert_eq!(total.value().as_u64(), 100); // Still 100!

Rate Counter

The Rate counter calculates the rate of change (units per second) over time. It's useful for tracking throughput, request rates, or any metric where you need to know "how fast" something is happening.

use contatori::counters::rate::Rate;
use contatori::counters::Observable;
use std::thread;
use std::time::Duration;

// Can be used as a static
static REQUESTS: Rate = Rate::new().with_name("requests_per_sec");

// Increment like a normal counter
REQUESTS.add(1);
REQUESTS.add(5);

// Get the absolute count
println!("Total: {}", REQUESTS.total_value()); // 6

// Get the rate (units per second)
// First call returns 0.0 and establishes baseline
let rate1 = REQUESTS.rate(); // 0.0

// Add more and wait
REQUESTS.add(1000);
thread::sleep(Duration::from_secs(1));

// Now rate() returns actual rate
let rate2 = REQUESTS.rate(); // ~1000.0 per second

The Rate counter:

• Uses sharded storage like all other counters (high performance)
• Can be initialized in const context (static RATE: Rate = Rate::new())
• Returns MetricKind::Gauge (rates can go up or down)
• Exports as float values in Prometheus

When to Use Sharded Counters

Sharded counters are ideal when:

• Multiple threads frequently update the same counter
• Write performance is more important than read performance
• You're tracking metrics, statistics, or telemetry data

For single-threaded scenarios or rarely-updated counters, a simple AtomicUsize may be more appropriate due to lower memory overhead.

Running Benchmarks

cargo bench

Running Tests

cargo test

License

MIT