Minitrace

A high-performance, ergonomic timeline tracing library for Rust.

Concepts

Span

A Span represents an individual unit of work done. It contains:

• An operation name
• A start timestamp and finish timestamp
• A set of key-value properties
• A reference to a causally-related Span

To record such a span, a struct of the same name, Span, is provided. Its lifetime indicates the begin and the end of the span.

A new Span can be created via Span::root(), Span::from_parent() and Span::from_parents(), and the last two methods refer existing related Spans to the new Span.

Spans are thread-safe and can be sent and accessed across threads.

Collector

A Collector associated to a root Span is used to collect all spans related to a request.

Local Span & Local Span Guard

Cooperate with LocalCollector, LocalSpan provides an extremely low-cost tracing mechanism for performance-sensitive apps.

A LocalSpan is recorded by a LocalSpanGuard which is thread-local and created via method LocalSpan::enter(). A LocalSpanGuard's creation means a LocalSpan's begin and its destruction means a LocalSpan's end. A completed LocalSpan will be finally stored in thread local.

The causally-relation between LocalSpans is built implicitly. It doesn't need to explicitly pass any contexts to connect two LocalSpans. Even within a deeply nested function calls, the inner LocalSpanGuard can automatically figure out its parent.

Local Collector

LocalCollector is provided to retrieve LocalSpans from thread local. Such collected LocalSpans can be mounted to a normal Span.

At most time, we should use the thread-local tracing mechanism (i.e LocalSpan and LocalCollector) to achieve the high performance goal. In other situations where execution is crossing thread bound, and a Span has to be passed from one thread to another thread, we just need to simply mount LocalSpans, which is collected by different LocalCollectors, to the Span.

Usage

[dependencies]
minitrace = { git = "https://github.com/tikv/minitrace-rust.git" }

Record a Span

To record a common span:

use minitrace::*;

let _span_guard = LocalSpan::enter("my event");

To add properties:

use minitrace::*;

// add a property for a span
let _span_guard = LocalSpan::enter("my event").with_property(|| ("key", String::from("value")));

// or add multiple properties for a span
let _span_guard = LocalSpan::enter("my event").with_properties(|| {
    vec![
        ("key1", String::from("value1")),
        ("key2", String::from("value2")),
    ]
});

Synchronous Example

A common pattern to trace synchronous code:

• Create a root Span and a Collector via Span::root(), then attach the Span to the current thread.
• Add LocalSpan::enter()s somewhere, e.g. at the beginning of a code scope, at the beginning of a function, to record spans.
• Make sure the root Span and all guards are dropped, then call Collector's collect to get all Spans.

use minitrace::*;

let collector = {
    let (root_span, collector) = Span::root("root");
    let _span_guard = root_span.enter();

    let _local_span_guard = LocalSpan::enter("child");

    // do something ...

    collector
};

let spans: Vec<span::Span> = collector.collect();

Asynchronous Example

To trace asynchronous code, we usually transmit Span from one thread to another thread.

Threads

use minitrace::*;

let collector = {
    let (root_span, collector) = Span::root("task1");
    let _span_guard = root_span.enter();

    let _local_span_guard = LocalSpan::enter("span of task1");
    
    // To trace a child task
    let span = Span::from_local_parent("task2");
    std::thread::spawn(move || {
        let _span_guard = span.enter();

        let _loal_span_guard = Span::enter("span of also task2");
    });

    collector
};

let spans: Vec<span::Span> = collector.collect();

Futures

We provide two Future adaptors:

• in_local_span: call LocalSpan::enter at every poll
• in_span: wrap the Future with a Span, then call Span::try_enter at every poll

The in_span adaptor is commonly used on a Future submitting to a runtime.

use minitrace::*;

let collector = {
    let (root_span, collector) = Span::root("root");
    let _span_guard = root_span.enter();

    // To trace another task
    runtime::spawn(async {
        let _ = async {
            // some works
        }.in_local_span("");
    }.in_span(Span::from_local_parent("new task")));

    collector
};

let spans: Vec<span::Span> = collector.collect();

Macros

We provide two macros to help reduce boilerplate code:

• trace
• trace_async

For normal functions, you can change:

use minitrace::*;

fn amazing_func() {
    let _span_guard = LocalSpan::enter("wow");

    // some works
}

to

use minitrace::*;
use minitrace_macro::trace;

#[trace("wow")]
fn amazing_func() {
    // some works
}

For async functions, you can change:

use minitrace::*;

async fn amazing_async_func() {
    async {
        // some works
    }
    .in_local_span("wow")
    .await
}

to

use minitrace::*;
use minitrace_macro::trace_async;

#[trace_async("wow")]
async fn amazing_async_func() {
    // some works
}

To access these macros, a dependency should be added as:

[dependencies]
minitrace-macro = { git = "https://github.com/tikv/minitrace-rust.git" }

User Interface

We support visualization provided by an amazing tracing platform Jaeger.

To experience, a dependency should be added as:

[dependencies]
minitrace-jaeger = { git = "https://github.com/tikv/minitrace-rust.git" }

Report to Jaeger

use minitrace_jaeger::Reporter;

let spans = /* collect from a collector */;

let socket = SocketAddr::new("127.0.0.1".parse().unwrap(), 6831);

const TRACE_ID: u64 = 42;
const SPAN_ID_PREFIX: u32 = 42;
const ROOT_PARENT_SPAN_ID: u64 = 0;
let bytes = Reporter::encode(
    String::from("service name"),
    TRACE_ID,
    ROOT_PARENT_SPAN_ID,
    SPAN_ID_PREFIX,
    &spans,
)
.expect("encode error");
Reporter::report(socket, &bytes).expect("report error");

Setup Jaeger

docker run --rm -d -p6831:6831/udp -p16686:16686 --name jaeger jaegertracing/all-in-one:latest

Run Synchronous Example

cargo run --example synchronous

Open http://localhost:16686 to see the results.

Run Asynchronous Example

cargo run --example asynchronous

Open http://localhost:16686 to see the results.