rust使用opel将trace数据导入tempo


 TL;DR
修改Cargo.toml,将其内容修改如下:
难度 中等

新建项目

cargo new otlp_test

修改项目依赖

修改Cargo.toml,将其内容修改如下:

[package]
name = "otlp_test"
version = "0.1.0"
edition = "2024"

[dependencies]
opentelemetry = "0.23"
opentelemetry-otlp = "0.16"
tokio = { version = "1", features = ["full"] }
opentelemetry_sdk = { version = "0.23", features = ["rt-tokio"] }

添加导入逻辑

将main.rs替换为如下内容:

use opentelemetry::{global, Context, KeyValue};
use opentelemetry::trace::{Span, SpanKind, TraceContextExt, Tracer};
use opentelemetry_sdk::trace::{self, Sampler};
use opentelemetry_sdk::Resource;
use opentelemetry_otlp::WithExportConfig;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // ===== 构建 OTLP SpanExporter(注意:build!)=====
    let exporter = opentelemetry_otlp::new_exporter()
        .tonic()
        .with_endpoint("http://192.168.3.99:4317")
        .build_span_exporter()?;

    // ===== TracerProvider =====
    let tracer_provider = trace::TracerProvider::builder()
        .with_config(
            trace::Config::default()
                .with_sampler(Sampler::AlwaysOn)
                .with_resource(Resource::new(vec![
                    KeyValue::new("service.name", "rust-tempo-demo"),
                    KeyValue::new("service.version", "0.1.0"),
                ])),
        )
        .with_batch_exporter(exporter, opentelemetry_sdk::runtime::Tokio)
        .build();

    global::set_tracer_provider(tracer_provider);

    let tracer = global::tracer("demo");

    // ===== Root span =====
    // let cx = opentelemetry::Context::current();
    // let mut root = tracer.start_with_context("http_request", &cx);
    let mut root = tracer.start("http_request");
    root.set_attribute(KeyValue::new("http.method", "GET"));
    root.set_attribute(KeyValue::new("http.route", "/demo"));
    let cx = Context::current_with_span(root);

    // ===== Child span =====
    {
        let mut child = tracer
            .span_builder("db_query")
            .with_kind(SpanKind::Internal)
            .start_with_context(&tracer, &cx);

        child.set_attribute(KeyValue::new("db.system", "postgresql"));
        child.set_attribute(KeyValue::new(
            "db.statement",
            "SELECT * FROM test",
        ));

        std::thread::sleep(std::time::Duration::from_millis(120));
        child.end();
    }

    cx.span().end();

    // ===== 非常重要:flush =====
    global::shutdown_tracer_provider();

    Ok(())
}

基本概念详解

  • exporter:出口
  • provider:入口与生命周期管理
  • pipeline:处理流水线
Instrumentation (Span / Metric)
        ↓
   Provider (TracerProvider / MeterProvider)
        ↓
      Pipeline
  (Sampler / Processor / Aggregation)
        ↓
     Exporter
        ↓
 Tempo / Jaeger / Prometheus / OTLP Collector

trace/span/metric是底层数据模型:

span有如下特点:

  • 一次操作
  • 开始/结束时间
  • 有属性
  • 有事件
  • 有父子关系
HTTP request
 └─ SQL query
     └─ index scan

trace是一棵span树,使用traceID标识。span是最小单位,trace是span的集合。

metric是数值型时间序列,主要是统计信息。

provider

provider是入口+全局管理者。

  1. TraceProvider
let provider = TracerProvider::builder()
    .with_span_processor(...)
    .build();
  • 创建Tracer
  • 管理生命周期
  • 持有pipeline
  • flush/shutdown
  1. pipeline

trace pipeline典型组成:

  • Sampler:采样器
  • SpanProcessor:处理器
  • Exporter:发送到后端

常见的trace exporter:

  • opentelemetry_otlp OTLP gRPC / HTTP
  • opentelemetry_jaeger Jaeger
  • opentelemetry_zipkin Zipkin
  • stdout 打印
let exporter = opentelemetry_otlp::new_exporter().tonic();

trace的完整生命周期如下:

你调用 tracer.start()
   ↓
Tracer(来自 TracerProvider)
   ↓
Sampler 决定是否记录
   ↓
Span 生命周期交给 SpanProcessor
   ↓
Span 结束
   ↓
Exporter 发出去

对应sdk结构关系:

TracerProvider
 ├── Resource
 ├── Sampler
 └── SpanProcessor
      └── Exporter
  • trace / metric “记录什么数据”
  • provider “谁来创建 & 管理”
  • pipeline “数据如何处理”
  • exporter “数据发到哪”

文章作者: growdu
版权声明: 本博客所有文章除特別声明外,均采用 CC BY 4.0 许可协议。转载请注明来源 growdu !
这篇还发到了
今日头条 drafted
  目录