OpenTelemetry instrumentation for Tokio runtime metrics.

This crate provides observability for Tokio runtimes by exposing runtime metrics through OpenTelemetry.

Installation

[dependencies]
opentelemetry-instrumentation-tokio = "0.1"
opentelemetry = "0.31"
opentelemetry_sdk = "0.31"
tokio = { version = "1", features = ["rt-multi-thread"] }

Quick Start

use opentelemetry_sdk::metrics::SdkMeterProvider;

#[tokio::main]
async fn main() {
    // Setup your meter provider
    let provider = SdkMeterProvider::builder().build();
    opentelemetry::global::set_meter_provider(provider);
    
    // Instrument the current runtime
    opentelemetry_instrumentation_tokio::observe_current_runtime();
    
    // Your application code
}

Configuration

Explicit Runtime Handle

let handle = tokio::runtime::Handle::current();
opentelemetry_instrumentation_tokio::observe_runtime(&handle);

Multiple Runtimes

Use custom labels to distinguish metrics from different runtimes. Labels are merged with the automatically added tokio.runtime.id (when available) so you can disambiguate runtimes without manual guards or deduplication.

use opentelemetry_instrumentation_tokio::Config;

Config::new()
    .with_label("runtime.name", "api")
    .observe_current_runtime();

let runtime = tokio::runtime::Builder::new_multi_thread()
    .enable_all()
    .build()
    .unwrap();

runtime.block_on(async {
    Config::new()
        .with_label("runtime.name", "worker")
        .observe_current_runtime();
});

use opentelemetry::KeyValue;

Config::new()
    .with_labels([
        KeyValue::new("service", "api"),
        KeyValue::new("env", "production"),
    ])
    .observe_current_runtime();

Metrics

Always Available (4 metrics)

These metrics work without any special configuration:

• tokio.workers - Number of worker threads
• tokio.global_queue_depth - Tasks in global queue
• tokio.worker.park_count - Worker park operations (per-worker)
• tokio.worker.busy_duration - Worker busy time in ms (per-worker)

Requires tokio_unstable (19 additional metrics)

Most metrics require building with the tokio_unstable cfg flag:

RUSTFLAGS="--cfg tokio_unstable" cargo build

See the Tokio documentation for more details.

Available metrics with tokio_unstable:

Runtime-level metrics:

• tokio.blocking_threads - Blocking thread pool size
• tokio.active_tasks - Number of alive tasks
• tokio.idle_blocking_threads - Idle blocking threads
• tokio.remote_schedules - Remote task schedules
• tokio.budget_forced_yields - Budget-forced yields
• tokio.spawned_tasks_count - Total spawned tasks
• tokio.blocking_queue_depth - Blocking queue depth

I/O driver metrics:

• tokio.io_driver.fd_registrations - FD registrations
• tokio.io_driver.fd_deregistrations - FD deregistrations
• tokio.io_driver.fd_readies - Ready events processed

Per-worker metrics (all with tokio.worker.index attribute):

• tokio.worker.noops - No-op wake-ups
• tokio.worker.task_steals - Tasks stolen
• tokio.worker.steal_operations - Steal operations
• tokio.worker.polls - Task polls
• tokio.worker.local_schedules - Local task schedules
• tokio.worker.overflows - Local queue overflows
• tokio.worker.local_queue_depth - Local queue depth
• tokio.worker.mean_poll_time - Mean poll duration (ns)
• tokio.worker.poll_time_bucket - Poll time histogram (requires config + runtime support)

License

Licensed under the Apache License, Version 2.0.