搜索到
2
篇与
的结果
-
容器化 + 上线到 K8s + 被 Prom 抓取 容器化 + 上线到 K8s + 被 Prom 抓取 目标：Exporter 以容器形式部署到集群，并被 Prometheus 抓取。 任务 构建镜像并 push：make docker IMG=…。 应用 k8s/deployment.yaml + Service；为你的命名空间调好 selector。 如果你集群用 Prometheus Operator，创建 ServiceMonitor 让它自动发现 scrape 目标。 Prometheus target 变为 Up，图表能看到你的新指标。 若用 Operator，ServiceMonitor 生效（label 选择器正确）。一、准备条件你有 Docker、kubectl 权限，能访问集群。 Harbor 私有仓库可用（如果是私有，需要在 K8s 里创建 imagePullSecret，步骤见第 4 步）。 Exporter 监听 0.0.0.0:9090 且有 /metrics（若没有，见“可选：最小 main.go”）。二、（可选）最小 main.go（若你项目还没暴露 /metrics）在你的可执行入口里加入（或新建 main.go）：package main import ( "log" "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/collectors" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { reg := prometheus.NewRegistry() reg.MustRegister(collectors.NewGoCollector(), collectors.NewProcessCollector(collectors.ProcessCollectorOpts{})) // TODO: 在这里注册你自己的指标 reg.MustRegister(yourMetric) mux := http.NewServeMux() mux.Handle("/metrics", promhttp.HandlerFor(reg, promhttp.HandlerOpts{})) //（若你有 /live、/ready 就更好） mux.HandleFunc("/live", func(w http.ResponseWriter, r *http.Request) { w.WriteHeader(200) }) mux.HandleFunc("/ready", func(w http.ResponseWriter, r *http.Request) { w.WriteHeader(200) }) log.Println("listening on :9090") log.Fatal(http.ListenAndServe(":9090", mux)) } 三、Dockerfile（多阶段构建，体积小、非 root 运行） 在项目根目录新建 Dockerfile：root@k8s-03:/woke/go-password-validator-main# cat Dockerfile # ---- build ---- FROM golang:1.23-alpine AS build WORKDIR /src COPY go.mod go.sum ./ RUN go mod download COPY . . # 如果入口在仓库根目录的 main.go： RUN CGO_ENABLED=0 GOOS=linux GOARCH=amd64 \ go build -ldflags="-s -w" -o /out/exporter . # ---- runtime ---- FROM gcr.io/distroless/static:nonroot USER nonroot:nonroot COPY --from=build /out/exporter /exporter EXPOSE 9090 ENTRYPOINT ["/exporter"] 四、Makefile 目标 在项目根目录的 Makefile 增加：root@k8s-03:/woke/go-password-validator-main# cat Makefile # ---------- 现有的目标，保留 ---------- test: go test ./... fmt: go fmt ./... vet: go vet ./... install-lint: # 建议用 go install（Go 1.16+） go install golang.org/x/lint/golint@latest @go env GOPATH >/dev/null lint: go list ./... | grep -v /vendor/ | xargs -L1 golint -set_exit_status install-staticcheck: go install honnef.co/go/tools/cmd/staticcheck@latest staticcheck: staticcheck -f stylish ./... # ---------- Docker 相关，新增 ---------- REGISTRY ?= 192.168.30.180:30003 REPO ?= test/go TAG ?= $(shell date +%Y%m%d-%H%M%S) IMG := $(REGISTRY)/$(REPO):$(TAG) .PHONY: docker docker-build docker-push docker: docker-build docker-push docker-build: docker build -t $(IMG) . docker-push: docker push $(IMG) 使用：# 登录你的 Harbor docker login harbor.example.com # 构建并 push（替换成你自己的 Harbor 项目/镜像名） make docker IMG=harbor.example.com/ops/go-password-exporter:v0.1.0 #如果配置好了Makefile文件 可以直接执行 make docker 也可以像上面这样定制五、Harbor 私有仓库拉取凭据（K8s imagePullSecret）注意：imagePullSecret 是命名空间级的，需要在哪个 ns 拉镜像，就在哪个 ns 创建；要全局通用只能在每个 ns 都创建（或用准入控制器自动注入）。#示例，在 monitoring 命名空间创建： kubectl -n monitoring create secret docker-registry harbor-cred \ --docker-server=192.168.30.180:30003 \ --docker-username='admin' \ --docker-password='Harbor12345' \ --docker-email='ops@example.com' 六、部署到 K8s（Deployment + Service）#deployment.yaml apiVersion: apps/v1 kind: Deployment metadata: name: pass-validator-exporter namespace: monitoring labels: app.kubernetes.io/name: pass-validator-exporter spec: replicas: 1 selector: matchLabels: app.kubernetes.io/name: pass-validator-exporter template: metadata: labels: app.kubernetes.io/name: pass-validator-exporter spec: imagePullSecrets: - name: harbor-cred containers: - name: exporter image: 192.168.30.180:30003/test/go:20250905-154053 imagePullPolicy: IfNotPresent ports: - name: http containerPort: 9090 readinessProbe: httpGet: { path: /ready, port: http } initialDelaySeconds: 3 periodSeconds: 5 livenessProbe: httpGet: { path: /live, port: http } initialDelaySeconds: 10 periodSeconds: 10 resources: requests: { cpu: "50m", memory: "64Mi" } limits: { cpu: "200m", memory: "128Mi" } securityContext: allowPrivilegeEscalation: false readOnlyRootFilesystem: true runAsNonRoot: true runAsUser: 65532#service.yaml apiVersion: v1 kind: Service metadata: name: pass-validator-exporter namespace: monitoring labels: app.kubernetes.io/name: pass-validator-exporter # 若你是“裸 Prometheus + 注解发现”，可打开以下注解： # annotations: # prometheus.io/scrape: "true" # prometheus.io/path: "/metrics" # prometheus.io/port: "9090" spec: type: ClusterIP selector: app.kubernetes.io/name: pass-validator-exporter ports: - name: http port: 9090 targetPort: http kubectl apply -f k8s/deployment.yaml kubectl apply -f k8s/service.yaml # 验证： kubectl -n monitoring get pods -l app.kubernetes.io/name=pass-validator-exporter -w kubectl -n monitoring get svc pass-validator-exporter # 本地看指标（可选）： kubectl -n monitoring port-forward svc/pass-validator-exporter 19090:9090 curl http://127.0.0.1:19090/metrics | head七、让 Prometheus 抓取方案 A：你在用 Prometheus Operator / kube-prometheus-stack（推荐） 关键点：ServiceMonitor 的 selector 要能匹配到 Service 的 label；同时 ServiceMonitor 自己的 label 要匹配到 Prometheus 实例的 serviceMonitorSelector。大多数 Helm 安装用的 label 是 release: kube-prometheus-stack（按你的集群实际为准）。# 看 Prometheus 实例的 selector 要求（取第一个 Prometheus 资源） kubectl -n monitoring get prometheus -o jsonpath='{.items[0].spec.serviceMonitorSelector.matchLabels}{"\n"}' # 一般会看到 {"release":"<你的 Helm release 名>"}#创建 servicemonitor.yaml 如果上面没有查到直接删除 labels: # release: kube-prometheus-stack # ←改成你 Prometheus 实例在选的那个 label apiVersion: monitoring.coreos.com/v1 kind: ServiceMonitor metadata: name: pass-validator-exporter namespace: monitoring labels: release: kube-prometheus-stack # ←改成你 Prometheus 实例在选的那个 label spec: namespaceSelector: matchNames: ["monitoring"] selector: matchLabels: app.kubernetes.io/name: pass-validator-exporter # 要和 Service 的 label 完全一致 endpoints: - port: http # 要和 Service 端口名一致 path: /metrics interval: 15s scrapeTimeout: 10s#应用并验证： kubectl apply -f servicemonitor.yaml # 在 Prometheus UI -> Status -> Targets 里应该出现你的 target，并且是 UP。 #启动pod测试集群内获取是否正常 kubectl -n monitoring run tmp-curl --rm -it --image=curlimages/curl --restart=Never \ -- http://pass-validator-exporter.monitoring.svc:9090/metrics | head #可以在浏览器IP+端口访问 kubectl -n monitoring port-forward --address 0.0.0.0 svc/prometheus-k8s 9090:9090方案 B：裸 Prometheus（没装 Operator）你需要在 Prometheus 的 prometheus.yml 里启用 kubernetes_sd_configs，并用注解过滤（上面 Service 已给注解示例）。典型片段如下（放到你的 scrape_configs 里）：- job_name: 'kubernetes-services' kubernetes_sd_configs: - role: endpoints relabel_configs: - source_labels: [__meta_kubernetes_service_annotation_prometheus_io_scrape] action: keep regex: "true" - source_labels: [__meta_kubernetes_service_annotation_prometheus_io_path] action: replace target: __metrics_path__ regex: "(.+)" - source_labels: [__address__, __meta_kubernetes_service_annotation_prometheus_io_port] action: replace target: __address__ regex: (.+?)(?::\d+)?;(\d+) replacement: $1:$2 #改好配置后重启 Prometheus，去 Targets 看是否 UP。八、迭代上线（改代码后快速发版）# 1) 构建并推送新镜像 make docker IMG=harbor.example.com/ops/go-password-exporter:v0.1.1 # 2) 滚动更新 Deployment（或直接改 yaml） kubectl -n monitoring set image deploy/pass-validator-exporter exporter=harbor.example.com/ops/go-password-exporter:v0.1.1 kubectl -n monitoring rollout status deploy/pass-validator-exporter 九、验收清单Prometheus Targets 里看到 pass-validator-exporter → UP。 在 Prometheus 或 Grafana 中能查询到你的新指标（例如：your_metric_name）。 如果是 Operator：ServiceMonitor 的 labels.release 与 Prometheus 实例选择器 匹配；ServiceMonitor.spec.selector.matchLabels 与 Service 的 labels 匹配；endpoints.port 与 Service 端口名 一致。十、常见坑速查镜像拉不下来：Deployment 没配置 imagePullSecrets，或 secret 在错的命名空间。 Target 一直 DOWN：容器没监听 0.0.0.0；端口/路径不一致；Service 选择器写错；endpoints.port 名字不对。 Operator 抓不到：ServiceMonitor 的 labels 与 Prometheus 的 serviceMonitorSelector 不匹配。 /metrics 超时：指标收集阻塞（锁/IO）；建议拆分收集逻辑并设置 scrapeTimeout。 权限：Exporter 是 HTTP 只读，不需要额外 RBAC。 -
Exporte + 基础指标 一、解压 & 目录说明unzip sys-exporter.zip -d sys-exporter cd sys-exporter tree -L 3 # . # ├── Makefile # ├── README.md # ├── cmd/sys-exporter/main.go # ├── go.mod # └── internal/collectors/ # ├── disk_usage.go # ├── disk_usage_test.go # ├── net_conn.go # ├── net_conn_test.go # ├── proc_rss.go # └── proc_rss_test.go 关键点： 使用自定义 Registry，同时注册 process 和 go collector。 /metrics 使用 promhttp.HandlerFor(reg, promhttp.HandlerOpts{EnableOpenMetrics:true})。 自带 /live、/ready，/ready 会尝试 Gather()，失败返回 503。 已挂好 /debug/pprof/*，方便做 CPU/内存采样。 三个示例指标： sys_mount_usage_ratio{mountpoint="/data"} （附带 sys_mount_used_bytes / sys_mount_total_bytes） sys_network_open_sockets{proto="tcp|udp", family="ipv4|ipv6"} sys_process_resident_memory_bytes（Exporter 进程自身 RSS）二、运行脚手架（本地 make run） 2.1 前置准备（只做一次）#下载go wget https://go.dev/dl/go1.22.5.linux-amd64.tar.gz #解压go sudo tar -C /usr/local -xzf go1.22.5.linux-amd64.tar.gz # 使用 vim： sudo vim ~/.bashrc #在文件的最后，添加以下几行： export PATH=$PATH:/usr/local/go/bin export GOPATH=$HOME/go export PATH=$PATH:$GOPATH/bin #立即生效 source ~/.bashrc #验证安装 go version #创建工作目录 mkdir -p ~/work && cd ~/work2.2一键跑起来（/metrics、/live、/ready）# 可选：拉依赖 make tidy # 启动 如果没有data目录 需要提前创建 make run # 日志应看到：sys-exporter listening on :9090 (mount=/data)#报错 root@k8s-01:~/woke/sys-exporter# make run GO111MODULE=on go run ./cmd/sys-exporter 2025/08/25 07:47:24 sys-exporter listening on :9090 (mount=/data) root@k8s-01:~/woke/sys-exporter# make test GO111MODULE=on go test ./... -race -count=1 # github.com/example/sys-exporter/internal/collectors [github.com/example/sys-exporter/internal/collectors.test] internal/collectors/disk_usage_test.go:22:37: undefined: testutil.NewGaugeFromDesc ? github.com/example/sys-exporter/cmd/sys-exporter [no test files] FAIL github.com/example/sys-exporter/internal/collectors [build failed] FAIL make: *** [Makefile:10: test] Error 1 root@k8s-01:~/woke/sys-exporter 这是单测里用了不存在的函数导致编译失败。prometheus/testutil 并没有 NewGaugeFromDesc 这个 API（我在示例里多写了一句检查 ratio 的代码）。解决方法：把那几行删掉，用 CollectAndCompare 做断言就够了。 #报错 root\@k8s-01:\~/woke/sys-exporter# make test GO111MODULE=on go test ./... -race -count=1 ? github.com/example/sys-exporter/cmd/sys-exporter \[no test files] \--- FAIL: TestProcRSSCollector (0.00s) proc\_rss\_test.go:31: metrics mismatch: +# HELP sys\_process\_resident\_memory\_bytes Resident memory (RSS) of the exporter process in bytes. +# TYPE sys\_process\_resident\_memory\_bytes gauge +sys\_process\_resident\_memory\_bytes 2.097152e+06 FAIL FAIL github.com/example/sys-exporter/internal/collectors 0.020s FAIL make: \*\*\* \[Makefile:10: test] Error 1 root\@k8s-01:~~/woke/sys-exporter# make tidy go mod tidy root\@k8s-01:~~/woke/sys-exporter# make test GO111MODULE=on go test ./... -race -count=1 ? github.com/example/sys-exporter/cmd/sys-exporter \[no test files] \--- FAIL: TestProcRSSCollector (0.00s) proc\_rss\_test.go:31: metrics mismatch: +# HELP sys\_process\_resident\_memory\_bytes Resident memory (RSS) of the exporter process in bytes. +# TYPE sys\_process\_resident\_memory\_bytes gauge +sys\_process\_resident\_memory\_bytes 2.097152e+06 FAIL FAIL github.com/example/sys-exporter/internal/collectors 0.018s FAIL make: \*\*\* \[Makefile:10: test] Error 1 root\@k8s-01:\~/woke/sys-exporter# cat internal/collectors/disk\_usage\_test.go package collectors import ( "strings" "testing" #解决办法 用下面的内容覆盖 internal/collectors/disk_usage_test.go： package collectors import ( "strings" "testing" "github.com/prometheus/client_golang/prometheus/testutil" ) func TestDiskUsageCollector_Metrics(t *testing.T) { mock := func(path string) (total, avail uint64, err error) { // 100 total, 25 available -> used=75, ratio=0.75 return 100, 25, nil } c := NewDiskUsageCollector("/data", mock) expected := # HELP sys_mount_total_bytes Total bytes for a mountpoint. # TYPE sys_mount_total_bytes gauge sys_mount_total_bytes{mountpoint="/data"} 100 # HELP sys_mount_used_bytes Used bytes for a mountpoint. # TYPE sys_mount_used_bytes gauge sys_mount_used_bytes{mountpoint="/data"} 75 # HELP sys_mount_usage_ratio Disk usage ratio (used/total) for a mountpoint. # TYPE sys_mount_usage_ratio gauge sys_mount_usage_ratio{mountpoint="/data"} 0.75 if err := testutil.CollectAndCompare(c, strings.NewReader(expected)); err != nil { t.Fatalf("unexpected metrics diff: %v", err) } } "github.com/prometheus/client_golang/prometheus/testutil" ) func TestDiskUsageCollector\_Metrics(t \*testing.T) { mock := func(path string) (total, avail uint64, err error) { // 100 total, 25 available -> used=75, ratio=0.75 return 100, 25, nil } c := NewDiskUsageCollector("/data", mock) expected := # HELP sys\_mount\_total\_bytes Total bytes for a mountpoint. # TYPE sys\_mount\_total\_bytes gauge sys\_mount\_total\_bytes{mountpoint="/data"} 100 # HELP sys\_mount\_used\_bytes Used bytes for a mountpoint. # TYPE sys\_mount\_used\_bytes gauge sys\_mount\_used\_bytes{mountpoint="/data"} 75 # HELP sys\_mount\_usage\_ratio Disk usage ratio (used/total) for a mountpoint. # TYPE sys\_mount\_usage\_ratio gauge sys\_mount\_usage\_ratio{mountpoint="/data"} 0.75 \ if err := testutil.CollectAndCompare(c, strings.NewReader(expected)); err != nil { t.Fatalf("unexpected metrics diff: %v", err) } } 你这个失败点在于单测期望的数值格式和PID 路径。 Prometheus 的文本输出里，浮点数经常是科学计数法（2.097152e+06），而你测试里写的是 2097152，导致对比失败。 你的测试里把 status 路径硬编码成 /fake/123/status，但实际 os.Getpid() 不一定是 123，最好用真实 PID 组路径。 把 internal/collectors/proc_rss_test.go 改成下面这样（两点都修了）： package collectors import ( "os" "path/filepath" "strings" "testing" "github.com/prometheus/client_golang/prometheus/testutil" ) func TestProcRSSCollector(t *testing.T) { root := "/fake" // 使用真实 PID 组路径，避免硬编码 123 statusPath := filepath.Join(root, strconv.Itoa(os.Getpid()), "status") data := "Name:\tapp\nVmRSS:\t2048 kB\n" // 2048 KB => 2,097,152 bytes // 只要命中 <root>/<pid>/status 就返回我们伪造的数据 reader := func(path string) ([]byte, error) { if path == statusPath { return []byte(data), nil } // 返回空数据，不要报错，Collector 会自己忽略 return []byte(""), nil } c := NewProcRSSCollector(root, reader) // 注意：Prometheus 文本输出常是科学计数法，这里用 e+06 exp := # HELP sys_process_resident_memory_bytes Resident memory (RSS) of the exporter process in bytes. # TYPE sys_process_resident_memory_bytes gauge sys_process_resident_memory_bytes 2.097152e+06 if err := testutil.CollectAndCompare(c, strings.NewReader(exp)); err != nil { t.Fatalf("metrics mismatch: %v", err) } }#报错 root\@k8s-01:~~/woke/sys-exporter# vi internal/collectors/disk\_usage\_test.go root\@k8s-01:~~/woke/sys-exporter# ls cmd go.mod go.sum internal Makefile README.md root\@k8s-01:~~/woke/sys-exporter# go clean -testcache root\@k8s-01:~~/woke/sys-exporter# make test GO111MODULE=on go test ./... -race -count=1 # github.com/example/sys-exporter/internal/collectors \[github.com/example/sys-exporter/internal/collectors.test] internal/collectors/proc\_rss\_test.go:12:6: TestProcRSSCollector redeclared in this block internal/collectors/disk\_usage\_test.go:12:6: other declaration of TestProcRSSCollector internal/collectors/disk\_usage\_test.go:15:36: undefined: strconv ? github.com/example/sys-exporter/cmd/sys-exporter \[no test files] FAIL github.com/example/sys-exporter/internal/collectors \[build failed] FAIL make: \*\*\* \[Makefile:10: test] Error 1 root\@k8s-01:\~/woke/sys-exporter# cat internal/collectors/disk\_usage\_test.go package collectors import ( "os" "path/filepath" "strings" "testing" "github.com/prometheus/client_golang/prometheus/testutil" ) func TestProcRSSCollector(t \*testing.T) { root := "/fake" // 使用真实 PID 组路径，避免硬编码 123 statusPath := filepath.Join(root, strconv.Itoa(os.Getpid()), "status") data := "Name:\tapp\nVmRSS:\t2048 kB\n" // 2048 KB => 2,097,152 bytes // 只要命中 <root>/<pid>/status 就返回我们伪造的数据 reader := func(path string) ([]byte, error) { if path == statusPath { return []byte(data), nil } // 返回空数据，不要报错，Collector 会自己忽略 return []byte(""), nil } c := NewProcRSSCollector(root, reader) // 注意：Prometheus 文本输出常是科学计数法，这里用 e+06 exp := ` # HELP sys\_process\_resident\_memory\_bytes Resident memory (RSS) of the exporter process in bytes. # TYPE sys\_process\_resident\_memory\_bytes gauge sys\_process\_resident\_memory\_bytes 2.097152e+06 \` if err := testutil.CollectAndCompare(c, strings.NewReader(exp)); err != nil { t.Fatalf("metrics mismatch: %v", err) } } 你把 disk_usage_test.go 覆盖成了 proc_rss_test.go 的内容，所以现在： 出现了重复的 TestProcRSSCollector（两个文件里同名测试）。 且缺少 strconv 的 import 还原 disk_usage_test.go 把它恢复成“磁盘使用率”的测试（不要放 RSS 的测试）： // internal/collectors/disk_usage_test.go package collectors import ( "strings" "testing" "github.com/prometheus/client_golang/prometheus/testutil" ) func TestDiskUsageCollector_Metrics(t *testing.T) { mock := func(path string) (total, avail uint64, err error) { // 100 total, 25 available -> used=75, ratio=0.75 return 100, 25, nil } c := NewDiskUsageCollector("/data", mock) expected := ` # HELP sys_mount_total_bytes Total bytes for a mountpoint. # TYPE sys_mount_total_bytes gauge sys_mount_total_bytes{mountpoint="/data"} 100 # HELP sys_mount_used_bytes Used bytes for a mountpoint. # TYPE sys_mount_used_bytes gauge sys_mount_used_bytes{mountpoint="/data"} 75 # HELP sys_mount_usage_ratio Disk usage ratio (used/total) for a mountpoint. # TYPE sys_mount_usage_ratio gauge sys_mount_usage_ratio{mountpoint="/data"} 0.75 ` if err := testutil.CollectAndCompare(c, strings.NewReader(expected)); err != nil { t.Fatalf("unexpected metrics diff: %v", err) } } 修正 proc_rss_test.go 确保它只包含 RSS 的测试，且 加上 strconv 的 import： // internal/collectors/proc_rss_test.go package collectors import ( "os" "path/filepath" "strconv" "strings" "testing" "github.com/prometheus/client_golang/prometheus/testutil" ) func TestProcRSSCollector(t *testing.T) { root := "/fake" statusPath := filepath.Join(root, strconv.Itoa(os.Getpid()), "status") data := "Name:\tapp\nVmRSS:\t2048 kB\n" // 2048 KB => 2,097,152 bytes reader := func(path string) ([]byte, error) { if path == statusPath { return []byte(data), nil } return []byte(""), nil } c := NewProcRSSCollector(root, reader) exp := ` # HELP sys_process_resident_memory_bytes Resident memory (RSS) of the exporter process in bytes. # TYPE sys_process_resident_memory_bytes gauge sys_process_resident_memory_bytes 2.097152e+06 ` if err := testutil.CollectAndCompare(c, strings.NewReader(exp)); err != nil { t.Fatalf("metrics mismatch: %v", err) } } #正常 root@k8s-01:~/woke/sys-exporter# go clean -testcache make test GO111MODULE=on go test ./... -race -count=1 ? github.com/example/sys-exporter/cmd/sys-exporter [no test files] ok github.com/example/sys-exporter/internal/collectors 1.019s#报错 root@k8s-01:~/woke/sys-exporter# make bench GO111MODULE=on go test ./... -run=^$ -bench=. -benchmem -count=1 -cpuprofile cpu.out -memprofile mem.out cannot use -cpuprofile flag with multiple packages make: *** [Makefile:13: bench] Error 1 #cannot use -cpuprofile flag with multiple packages 意思是：你下了 -cpuprofile 参数，但 go test ./... 会一次性跑多个包（cmd/... 和 internal/...），Go 不允许把多个包的基准同时写到一个 cpu.out 里。 解决方法 只 bench 你有基准的那个包（internal/collectors），而不是所有 ./...。 bench: GO111MODULE=on go test ./internal/collectors -run=^$$ -bench=. -benchmem -count=1 -cpuprofile cpu.out -memprofile mem.out @echo "" @echo "CPU profile: cpu.out | Mem profile: mem.out" @echo "Open an interactive view:" @echo " go tool pprof -http=:9999 cpu.out" 执行make bench 能正常生成 cpu.out 和 mem.out#报错 root\@k8s-01:\~/woke/sys-exporter# make bench GO111MODULE=on go test ./internal/collectors -run=^\$ -bench=. -benchmem -count=1 -cpuprofile cpu.out -memprofile mem.out PASS ok github.com/example/sys-exporter/internal/collectors 0.252s CPU profile: cpu.out | Mem profile: mem.out Open an interactive view: go tool pprof -http=:9999 cpu.out root\@k8s-01:~~/woke/sys-exporter# ls cmd collectors.test cpu.out go.mod go.sum internal Makefile mem.out README.md root\@k8s-01:~~/woke/sys-exporter# go tool pprof -http=:9999 cpu.out Serving web UI on [http://localhost:9999](http://localhost:9999) Couldn't find a suitable web browser! Set the BROWSER environment variable to your desired browser. 浏览器[http://192.168.30.180:9999/](http://192.168.30.180:9999/) 打不开 这是正常现象：现在的 pprof Web UI 只监听在本机回环地址（localhost），所以你从别的机器用 http://192.168.30.180:9999 打不开 #解决 go tool pprof -http=0.0.0.0:9999 cpu.out 如果浏览器打开以后显示：Could not execute dot; may need to install graphviz. 那就需要安装 Graphviz sudo apt-get install -y graphviz sudo yum install -y graphviz 重新打开 pprof Web go tool pprof -http=0.0.0.0:9999 cpu.out 这时点击 “View → Graph” 或 “Flame Graph” 就能显示图像了。 root@k8s-01:~/woke/sys-exporter# go tool pprof -http=0.0.0.0:9999 cpu.out Serving web UI on http://0.0.0.0:9999 Couldn't find a suitable web browser! Set the BROWSER environment variable to your desired browser. cd ~/woke/sys-exporter LISTEN_ADDR=":9090" MOUNTPOINT="/data" make run # 看到日志：sys-exporter listening on :9090 (mount=/data)# 抓在线 profile 并用 Web UI 展示 用正确的 URL抓在线 profile 在 exporter 正常监听后再执行（仍然在服务器上）： go tool pprof -http=0.0.0.0:9999 http://localhost:9090/debug/pprof/profile?seconds=15 注意要点： -http=0.0.0.0:9999 只是 Web UI 的监听地址； profile 的 URL 要作为最后一个参数单独给出（别写到 -http= 里）； 如果你把 exporter 绑在了别的端口，比如 :9100，就把 URL 换成： http://localhost:9100/debug/pprof/profile?seconds=15
