Grafana Beyla 是一款 eBPF 采集器，适合用来体验无侵入采集 HTTP 服务 RED 指标的工作方式。本文保留原始命令和指标样例，重点说明如何准备测试服务、启动 Beyla、查看 Prometheus 指标，以及后续如何接入采集器和 Grafana Dashboard。
核心要点
- Beyla 的价值在于通过 eBPF 获取服务请求信息，不需要在示例 Go 服务中手工埋点。
- 本文使用 RED（Rate、Errors、Duration）指标作为验证目标，重点观察请求量、错误状态码和延迟分布。
- Beyla 通过
BEYLA_PROMETHEUS_PORT暴露 Prometheus 格式指标，通过OPEN_PORT指定要观测的服务端口。 - 指标可以被 vmagent、Categraf、Prometheus 等采集器抓取，再进入后端存储并通过 Grafana 分析。
- 原文示例基于早期 Beyla 版本，生产使用前需要结合当前版本文档复核参数和兼容性。
eBPF 基础介绍可以参考我之前的文章《eBPF Hello world》。理论上，eBPF 可以拿到服务收到的请求信息，比如QPS、延迟、成功率等，这些数据对于应用级监控至关重要，Grafana Beyla 就是为此而生的。
准备一个可观测的 HTTP 服务
要测试使用 Beyla 采集服务的 RED（Rate-Errors-Duration） 指标，那首先得有个服务，这里我用的是 answer（ https://answer.flashcat.cloud ） 论坛，你也可以自己搞一个简单的 http 服务，比如：
package main
import (
"net/http"
"strconv"
"time"
)
func handleRequest(rw http.ResponseWriter, req *http.Request) {
status := 200
for k, v := range req.URL.Query() {
if len(v) == 0 {
continue
}
switch k {
case "status":
if s, err := strconv.Atoi(v[0]); err == nil {
status = s
}
case "delay":
if d, err := time.ParseDuration(v[0]); err == nil {
time.Sleep(d)
}
}
}
rw.WriteHeader(status)
}
func main() {
http.ListenAndServe(":8080",
http.HandlerFunc(handleRequest))
}
上面这个代码，保存成 server.go，然后用 go run server.go 即可运行，当然，前提是你机器上有 go 开发环境。这个小服务，可以接收两个参数，一个是 status，用来指定返回的 http 状态码，另一个是 delay，用来指定延迟多久返回，比如：
curl -v "http://localhost:8080/foo?status=404"
上面的命令，会返回 404 状态码，如果想延迟 1 秒返回，可以这样：
curl -v "http://localhost:8080/foo?delay=1s"
接下来，我们就可以使用 Beyla 采集这个服务的 RED 指标了。这个示例服务的好处是输入、状态码和延迟都可控，方便验证 Beyla 是否真的采集到了请求维度的数据。
下载 Beyla
我的机器上有 go 开发环境，所以我直接使用 go install 安装了，你也可以去 Beyla 的 release 页面下载二进制包，然后解压缩使用。
go install github.com/grafana/beyla/cmd/beyla@latest
运行 Beyla
使用下面的命令运行 Beyla：
$ BEYLA_PROMETHEUS_PORT=8999 PRINT_TRACES=true OPEN_PORT=8080 sudo -E beyla
或者直接使用 root 账号运行，比如我是这么跑的：
$ BEYLA_PROMETHEUS_PORT=8999 PRINT_TRACES=true OPEN_PORT=8080 beyla
解释一下这几个参数：
- BEYLA_PROMETHEUS_PORT: Beyla 要监听的端口，通过这个端口暴露 metrics 指标数据
- PRINT_TRACES: 是否打印 trace 日志
- OPEN_PORT: Beyla 采集的目标服务监听的端口，这里是 8080，上面给出的那段 go server 的代码就是监听在 8080，我的机器上 answer 论坛程序也是监听在 8080，你要监控的程序如果不是监听在 8080，可以在换成你自己的端口
| 环境变量 | 作用 | 示例值 |
|---|---|---|
BEYLA_PROMETHEUS_PORT |
Beyla 暴露 Prometheus 指标的端口 | 8999 |
PRINT_TRACES |
控制是否在终端打印 trace 日志 | true |
OPEN_PORT |
被 Beyla 观测的目标服务端口 | 8080 |
查看 Prometheus 指标
运行之后，可以通过 curl 查看指标：
curl http://localhost:8999/metrics
返回的内容如下：
# HELP http_client_duration_seconds duration of HTTP service calls from the client side, in seconds
# TYPE http_client_duration_seconds histogram
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.005"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.01"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.025"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.05"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.075"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.1"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.25"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.5"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.75"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="1"} 0
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="2.5"} 1
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="5"} 1
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="7.5"} 1
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="10"} 1
http_client_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="+Inf"} 1
http_client_duration_seconds_sum{http_method="GET",http_status_code="200",service_name="answer"} 1.668771575
http_client_duration_seconds_count{http_method="GET",http_status_code="200",service_name="answer"} 1
# HELP http_client_request_size_bytes size, in bytes, of the HTTP request body as sent from the client side
# TYPE http_client_request_size_bytes histogram
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="32"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="64"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="128"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="256"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="512"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="1024"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="2048"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="4096"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="8192"} 1
http_client_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="+Inf"} 1
http_client_request_size_bytes_sum{http_method="GET",http_status_code="200",service_name="answer"} 0
http_client_request_size_bytes_count{http_method="GET",http_status_code="200",service_name="answer"} 1
# HELP http_server_duration_seconds duration of HTTP service calls from the server side, in seconds
# TYPE http_server_duration_seconds histogram
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0"} 0
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.005"} 201
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.01"} 789
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.025"} 799
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.05"} 799
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.075"} 799
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.1"} 799
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.25"} 799
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.5"} 799
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0.75"} 799
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="1"} 799
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="2.5"} 800
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="5"} 800
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="7.5"} 800
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="10"} 800
http_server_duration_seconds_bucket{http_method="GET",http_status_code="200",service_name="answer",le="+Inf"} 800
http_server_duration_seconds_sum{http_method="GET",http_status_code="200",service_name="answer"} 5.752096697000003
http_server_duration_seconds_count{http_method="GET",http_status_code="200",service_name="answer"} 800
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0"} 0
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0.005"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0.01"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0.025"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0.05"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0.075"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0.1"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0.25"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0.5"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0.75"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="1"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="2.5"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="5"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="7.5"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="10"} 1
http_server_duration_seconds_bucket{http_method="GET",http_status_code="302",service_name="answer",le="+Inf"} 1
http_server_duration_seconds_sum{http_method="GET",http_status_code="302",service_name="answer"} 0.001523002
http_server_duration_seconds_count{http_method="GET",http_status_code="302",service_name="answer"} 1
# HELP http_server_request_size_bytes size, in bytes, of the HTTP request body as received at the server side
# TYPE http_server_request_size_bytes histogram
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="0"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="32"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="64"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="128"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="256"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="512"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="1024"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="2048"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="4096"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="8192"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="200",service_name="answer",le="+Inf"} 800
http_server_request_size_bytes_sum{http_method="GET",http_status_code="200",service_name="answer"} 0
http_server_request_size_bytes_count{http_method="GET",http_status_code="200",service_name="answer"} 800
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="0"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="32"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="64"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="128"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="256"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="512"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="1024"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="2048"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="4096"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="8192"} 1
http_server_request_size_bytes_bucket{http_method="GET",http_status_code="302",service_name="answer",le="+Inf"} 1
http_server_request_size_bytes_sum{http_method="GET",http_status_code="302",service_name="answer"} 0
http_server_request_size_bytes_count{http_method="GET",http_status_code="302",service_name="answer"} 1
# HELP promhttp_metric_handler_errors_total Total number of internal errors encountered by the promhttp metric handler.
# TYPE promhttp_metric_handler_errors_total counter
promhttp_metric_handler_errors_total{cause="encoding"} 0
promhttp_metric_handler_errors_total{cause="gathering"} 0
这段输出里已经能看到 http_server_duration_seconds、http_client_duration_seconds、请求大小、状态码、服务名等信息。对于应用监控来说，最先关注的通常是三类指标：
| 观测目标 | 对应指标样例 | 排障价值 |
|---|---|---|
| 请求延迟 | http_server_duration_seconds_bucket、http_server_duration_seconds_sum |
判断服务响应是否变慢，并支持按桶计算延迟分布 |
| 状态码 | http_status_code 标签 |
区分 2xx、3xx、4xx、5xx 请求，定位错误趋势 |
| 请求数量 | http_server_duration_seconds_count |
估算请求量，结合错误和延迟判断服务健康度 |
这些指标就可以用采集器来抓了，比如 vmagent、categraf、prometheus 等，完事之后入库，使用 Grafana 展示分析即可，经常关注本公众号的读者对于这些知识应该比较熟悉了，这里不再赘述。Beyla 默认提供了一个 Grafana Dashboard，可以导入测试：https://github.com/grafana/beyla/tree/main/grafana。
常见问题
Beyla 适合替代应用内埋点吗？
从本文示例看，Beyla 更适合先解决“无需改业务代码也能采集请求 RED 指标”的问题。它可以降低接入门槛，但是否替代应用内埋点，要看你是否还需要业务语义、代码级 span、用户维度等更细粒度信息。
为什么示例里需要指定 OPEN_PORT？
OPEN_PORT 用来告诉 Beyla 观测哪个本地服务端口。本文的 Go 服务和 answer 示例都监听 8080，所以命令中使用 OPEN_PORT=8080。
采集到指标后下一步做什么？
下一步是让 vmagent、Categraf 或 Prometheus 抓取 http://localhost:8999/metrics，写入时序库，再导入 Beyla Dashboard 或自建 Grafana 面板分析延迟、错误和请求量。
结语
Beyla 的核心价值是把 eBPF 的无侵入能力带到应用级可观测场景中。本文演示的是最小闭环：准备 HTTP 服务、启动 Beyla、查看 Prometheus 指标、再交给采集器和 Grafana 使用。对于想理解 eBPF 应用监控路径的同学，这已经足够作为一次可复现的尝鲜实验。
需要注意的是，本文原始实践基于 Beyla 早期阶段，生产落地前应复核当前版本的参数、权限要求和运行限制。
