Files
LokiStack/backend/services/collector_deployer.go
T
opsx 02d1bd87a7 fix(promtail): 修 Sprintf 缺参导致清单非法 + Ingress 模板固定命名空间 + 资源名去 opsxloki
=== 一、致命 bug:Promtail 清单第 10 个 %s 缺参数(导致日志采集中断)===
go vet 报:"format %s reads arg #10, but call has 9 args"。
模板里 DaemonSet 的 `namespace: %s`(第 9 个)和 `image: %s`(第 10 个)
是两个独立占位符,但 Sprintf 只传了 9 个参数,第 10 个变成 %!s(MISSING),
整个多文档 YAML 解析失败:
  "解析 Kubernetes YAML: yaml: line 23: found character that cannot start any token"

而 installDaemonset 是「先删 daemonset/configmap/serviceaccount 再 apply」,
所以每次失败都会把上一份 Promtail 删掉却不重建 → 日志采集中断。

修复:
1) Sprintf 补上 DaemonSet 命名空间参数(第 9 位)
2) 新增 validateManifest():在删旧资源之前先本地解析校验清单,
   校验不过直接中止、不碰任何现有资源 —— 让模板错误永远不会再打断采集
3) 新增回归测试 promtail_manifest_test.go:用平台同一套
   YAMLOrJSONDecoder 解析 3 种参数组合,断言各解析出 5 个对象且无 TAB

=== 二、快捷 LogQL:Ingress 访问日志不再跟随所选命名空间 ===
原实现:next = applyLabelFilter(next,'namespace', namespace.trim() || 'ingress-nginx')
→ 选了 monitor 就生成 {namespace="monitor"} |= "HTTP/",查不到任何日志
  (ingress controller 的 Pod 只在 ingress-nginx 命名空间里)

修复:ingress 模板独立处理
  - 固定用 ingress controller 命名空间(ingress-nginx,缺失则回退 ingress)
  - 不叠加当前选中的 app / pod 过滤(那些属于业务服务)
  - 结果是 {namespace="ingress-nginx"} |= "HTTP/"

=== 三、K8s 资源名去掉 opsxloki ===
ClusterRole / ClusterRoleBinding:opsxloki-promtail → loki-promtail
- PromtailDaemonsetYAML 里 3 处 name 全部改名
- uninstallDaemonset 同步改名,并保留对旧名 opsxloki-promtail 的兼容清理,
  避免升级后留下孤儿 ClusterRole
- 集群侧已实测:新建 loki-promtail,旧的 opsxloki-promtail 已删除

保留的 opsxloki(有兼容性约束,改动会丢数据/掉登录):
- Go module 导入路径 opsxloki/xxx(149 处,纯内部包名)
- 数据文件 opsxloki.json(改了读不到历史数据)
- 前端 localStorage key opsxloki.auth.* / opsxloki.theme.* 等(改了在线用户掉登录)
- Dockerfile 里的 opsxloki 系统用户/二进制名

镜像:lghost/lokistack:latest + v0.1 = sha256:4a5590580cbb
2026-10-27 04:16:55 +00:00

414 lines
13 KiB
Go
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
package services
import (
"context"
"encoding/json"
"fmt"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"strings"
"time"
"opsxloki/models"
)
// CollectorDeployer 采集器部署器,按 type 分发到三种安装/卸载流程
type CollectorDeployer struct {
collector *models.Collector
loki *models.LokiDeployment
logFunc func(level, message, detail string)
}
func NewCollectorDeployer(c *models.Collector, l *models.LokiDeployment, logFunc func(level, message, detail string)) *CollectorDeployer {
return &CollectorDeployer{collector: c, loki: l, logFunc: logFunc}
}
func (d *CollectorDeployer) log(level, message, detail string) {
if d.logFunc != nil {
d.logFunc(level, message, detail)
}
}
// Install 按采集器类型分发安装
func (d *CollectorDeployer) Install(ctx context.Context) error {
endpoint := d.loki.Endpoint
if endpoint == "" {
return fmt.Errorf("关联的 Loki 尚未部署,缺少推送地址(endpoint),请先成功部署该 Loki")
}
d.log("info", fmt.Sprintf("开始部署采集器 %s(类型 %s)", d.collector.Name, d.collector.Type), "")
var err error
switch d.collector.Type {
case "host-promtail":
err = d.installHost(ctx, endpoint)
case "k8s-daemonset":
err = d.installDaemonset(ctx, endpoint)
case "k8s-fluentbit-sidecar":
err = d.installSidecar(ctx, endpoint)
default:
return fmt.Errorf("不支持的采集器类型: %s", d.collector.Type)
}
if err != nil {
d.log("error", fmt.Sprintf("采集器部署失败: %v", err), "")
return err
}
d.log("success", "采集器部署完成", "")
return nil
}
// Uninstall 按采集器类型分发卸载
func (d *CollectorDeployer) Uninstall(ctx context.Context) error {
d.log("info", fmt.Sprintf("开始卸载采集器 %s(类型 %s)", d.collector.Name, d.collector.Type), "")
switch d.collector.Type {
case "host-promtail":
return d.uninstallHost(ctx)
case "k8s-daemonset":
return d.uninstallDaemonset(ctx)
case "k8s-fluentbit-sidecar":
return d.uninstallSidecar(ctx)
}
return nil
}
// ===== host-promtail:SSH 安装 Promtail 二进制 + systemd =====
func (d *CollectorDeployer) installHost(ctx context.Context, endpoint string) error {
host := d.collector.Host
if host == nil {
return fmt.Errorf("主机采集器缺少目标主机")
}
client, err := NewSSHClientWithAuth(HostSSHAuth(host))
if err != nil {
return err
}
defer client.Close()
// 检测架构
archOut, _, _ := client.Execute(ctx, "uname -m")
arch := strings.TrimSpace(archOut)
switch arch {
case "x86_64":
arch = "amd64"
case "aarch64", "arm64":
arch = "arm64"
default:
return fmt.Errorf("不支持的主机架构: %s", arch)
}
// 确定 Promtail 版本(优先使用关联的 Loki 版本)
version := "3.0.0"
if d.loki != nil && d.loki.Version != "" {
version = d.loki.Version
}
// 创建必要目录
d.log("info", "准备目录 /opt/promtail /etc/promtail /var/lib/promtail", "")
if _, _, err := client.Execute(ctx, "mkdir -p /opt/promtail /etc/promtail /var/lib/promtail"); err != nil {
return fmt.Errorf("创建 Promtail 目录失败: %w", err)
}
// 下载 Promtail
d.log("info", fmt.Sprintf("下载 Promtail %s(%s)", version, arch), "")
url := fmt.Sprintf("https://github.com/grafana/loki/releases/download/v%s/promtail-linux-%s.zip", version, arch)
dl := fmt.Sprintf("cd /tmp && curl -fsSL -o p.zip '%s' && unzip -o p.zip -d /opt/promtail/ && chmod +x /opt/promtail/promtail-linux-%s", url, arch)
if _, _, err := client.Execute(ctx, dl); err != nil {
return fmt.Errorf("下载 Promtail 失败: %w", err)
}
client.Execute(ctx, "ln -sf /opt/promtail/promtail-linux-* /usr/local/bin/promtail 2>/dev/null || true")
// 生成并上传配置
logPath := d.collector.LogPath
if logPath == "" {
logPath = "/var/log/*.log"
}
conf := PromtailConfig(endpoint, logPath, "host", d.collector.Labels)
d.log("info", "上传 Promtail 配置", "")
if _, _, err := client.Execute(ctx, fmt.Sprintf("cat > /etc/promtail/promtail.yaml << 'EOF'\n%s\nEOF", conf)); err != nil {
return fmt.Errorf("上传配置失败: %w", err)
}
// systemd 服务(生产级)
service := `[Unit]
Description=Promtail Log Collector
Wants=network-online.target
After=network-online.target
[Service]
Type=simple
User=root
Group=root
ExecStart=/usr/local/bin/promtail -config.file=/etc/promtail/promtail.yaml -positions.enabled=true
ExecReload=/bin/kill -HUP $MAINPID
Restart=always
RestartSec=10
LimitNOFILE=65536
StandardOutput=append:/var/log/promtail.log
StandardError=append:/var/log/promtail.log
[Install]
WantedBy=multi-user.target
`
d.log("info", "安装 systemd 服务", "")
if _, _, err := client.Execute(ctx, fmt.Sprintf("cat > /etc/systemd/system/promtail.service << 'EOF'\n%s\nEOF", service)); err != nil {
return fmt.Errorf("创建 systemd 服务失败: %w", err)
}
if _, _, err := client.Execute(ctx, "systemctl daemon-reload && systemctl enable promtail && systemctl restart promtail"); err != nil {
return fmt.Errorf("启动 Promtail 失败: %w", err)
}
// 保存配置快照
d.collector.Config = conf
d.log("success", "Promtail 部署完成", "")
return nil
}
func (d *CollectorDeployer) uninstallHost(ctx context.Context) error {
host := d.collector.Host
if host == nil {
return nil
}
client, err := NewSSHClientWithAuth(HostSSHAuth(host))
if err != nil {
return err
}
defer client.Close()
client.Execute(ctx, "systemctl stop promtail 2>/dev/null || true")
client.Execute(ctx, "systemctl disable promtail 2>/dev/null || true")
client.Execute(ctx, "rm -rf /etc/systemd/system/promtail.service /usr/local/bin/promtail /etc/promtail /opt/promtail /var/lib/promtail")
client.Execute(ctx, "systemctl daemon-reload")
return nil
}
// ===== k8s-daemonset:helm 安装 Promtail DaemonSet =====
func (d *CollectorDeployer) installDaemonset(ctx context.Context, endpoint string) error {
cluster := d.collector.Cluster
if cluster == nil {
return fmt.Errorf("集群采集器缺少目标集群")
}
// 默认部署到关联 Loki 的命名空间(统一 log 命名空间策略)
ns := d.collector.Namespace
if ns == "" && d.loki != nil {
ns = d.loki.Namespace
}
if ns == "" {
ns = "log"
}
clusterName := "unknown"
if cluster.Name != "" {
clusterName = cluster.Name
}
manifest := PromtailDaemonsetYAML(ns, endpoint, clusterName, d.collector.Labels,
d.collector.TargetNamespaces, d.collector.AppLabelSource, d.collector.AppInclude)
// 关键顺序:先本地校验清单,再删旧资源。
// 否则模板有误时会出现「旧 DaemonSet 已删、新的建不出来」→ 日志采集直接中断。
if err := validateManifest(manifest); err != nil {
d.log("error", fmt.Sprintf("生成的 Promtail 清单无效,已中止且未改动现有资源: %v", err), "")
return fmt.Errorf("Promtail 清单校验失败(现有采集器未受影响): %w", err)
}
// 删除历史残留并通过 Kubernetes API 提交资源。
for _, kind := range []string{"daemonset", "configmap", "serviceaccount"} {
_ = deleteK8sResource(ctx, cluster, kind, ns, "promtail")
}
if err := d.applyYAML(ctx, nil, manifest, "promtail"); err != nil {
return err
}
d.log("info", fmt.Sprintf("等待 DaemonSet promtail 就绪(命名空间 %s)...", ns), "")
if err := waitDaemonSet(ctx, cluster, ns, "promtail"); err != nil {
return fmt.Errorf("Promtail DaemonSet 未就绪: %w", err)
}
svcURL := endpoint
d.collector.Config = svcURL
d.log("success", fmt.Sprintf("Promtail DaemonSet 已部署到 %s(推送至 %s)", ns, svcURL), "")
return nil
}
// applyYAML 通过 API Server 应用采集器清单。
func (d *CollectorDeployer) applyYAML(ctx context.Context, _ []string, manifest, label string) error {
if d.collector.Cluster == nil {
return fmt.Errorf("采集器缺少集群")
}
if err := applyManifest(ctx, d.collector.Cluster, manifest); err != nil {
d.log("error", fmt.Sprintf("应用 %s 失败: %v", label, err), "")
return err
}
return nil
}
func (d *CollectorDeployer) uninstallDaemonset(ctx context.Context) error {
cluster := d.collector.Cluster
if cluster == nil {
return nil
}
ns := d.collector.Namespace
if ns == "" && d.loki != nil {
ns = d.loki.Namespace
}
if ns == "" {
ns = "log"
}
// clusterrole / clusterrolebinding 用 loki-promtail;
// 旧版本曾用 opsxloki-promtail,这里一并清理以免升级后留下孤儿资源。
for _, item := range [][2]string{
{"daemonset", "promtail"}, {"configmap", "promtail"}, {"serviceaccount", "promtail"},
{"clusterrole", "loki-promtail"}, {"clusterrolebinding", "loki-promtail"},
{"clusterrole", "opsxloki-promtail"}, {"clusterrolebinding", "opsxloki-promtail"},
} {
if err := deleteK8sResource(ctx, cluster, item[0], ns, item[1]); err != nil {
return err
}
}
return nil
}
// ===== k8s-fluentbit-sidecar:kubectl patch 注入 Fluent Bit sidecar =====
func (d *CollectorDeployer) installSidecar(ctx context.Context, endpoint string) error {
cluster := d.collector.Cluster
if cluster == nil {
return fmt.Errorf("Sidecar 采集器缺少目标集群")
}
target := d.collector.TargetDeployment
if target == "" {
return fmt.Errorf("Sidecar 采集器缺少目标 Deployment")
}
client, err := clusterClient(cluster, 30*time.Second)
if err != nil {
return err
}
ns := d.collector.Namespace
if ns == "" {
ns = "default"
}
logPath := d.collector.LogPath
if logPath == "" {
logPath = "/var/log/app/*.log"
}
app := target
// 1. 检查目标 deployment 是否存在
d.log("info", fmt.Sprintf("检查目标 Deployment %s/%s", ns, target), "")
deployment, err := client.AppsV1().Deployments(ns).Get(ctx, target, metav1.GetOptions{})
if err != nil {
return fmt.Errorf("读取目标 Deployment %s/%s: %w", ns, target, err)
}
deploymentJSON, err := json.Marshal(deployment)
if err != nil {
return err
}
mount, err := resolveSidecarLogMount(deploymentJSON, target, logPath, d.collector.ID)
if err != nil {
return err
}
hasSidecar := false
for _, container := range deployment.Spec.Template.Spec.Containers {
if container.Name == "fluent-bit" {
hasSidecar = true
break
}
}
// 3. 部署 ConfigMap
lokiHost, lokiPort := parseLokiURL(endpoint)
conf := FluentbitConfig(lokiHost, lokiPort, logPath, app, ns)
configMapName := fmt.Sprintf("fluent-bit-config-%d", d.collector.ID)
cm := FluentbitConfigMapYAML(conf, ns, configMapName)
d.log("info", "部署 Fluent Bit ConfigMap", "")
if err := applyManifest(ctx, cluster, cm); err != nil {
return fmt.Errorf("ConfigMap 部署失败: %w", err)
}
d.log("info", fmt.Sprintf("为 %s 配置 Fluent Bit Sidecar", target), "")
if err := patchDeployment(ctx, cluster, ns, target, SidecarPatchYAML(configMapName, mount)); err != nil {
return fmt.Errorf("Sidecar 注入失败: %w", err)
}
if hasSidecar {
if err := restartDeployment(ctx, cluster, ns, target); err != nil {
return fmt.Errorf("Sidecar 配置已更新,但重启 Deployment 失败: %w", err)
}
}
d.log("info", "等待 Pod 重建", "")
if err := waitDeployment(ctx, cluster, ns, target); err != nil {
return fmt.Errorf("Sidecar Deployment 未就绪: %w", err)
}
// DNS 自愈:CoreDNS 异常集群上 svc 域名不可解析 → 用 Loki ClusterIP 重建 ConfigMap 并滚动重启
if strings.Contains(endpoint, ".svc") {
host := strings.Split(strings.Split(endpoint, "//")[1], ":")[0]
if podDNSBroken(ctx, cluster, ns, target, "fluent-bit", host) {
d.log("warn", "fluent-bit 无法解析 Loki 服务名,回退 ClusterIP…", "")
if ip, ierr := serviceClusterIP(ctx, cluster, lokiNamespaceOf(d.loki, endpoint), "loki"); ierr == nil && ip != "" {
conf2 := FluentbitConfig(ip, lokiPortOf(endpoint), logPath, target, ns)
cm2 := FluentbitConfigMapYAML(conf2, ns, configMapName)
if err := applyManifest(ctx, cluster, cm2); err == nil {
_ = restartDeployment(ctx, cluster, ns, target)
d.log("info", "fluent-bit 推送地址已回退为 ClusterIP: "+ip, "")
}
}
}
}
return nil
}
func (d *CollectorDeployer) uninstallSidecar(ctx context.Context) error {
cluster := d.collector.Cluster
if cluster == nil || d.collector.TargetDeployment == "" {
return nil
}
client, err := clusterClient(cluster, 30*time.Second)
if err != nil {
return err
}
ns := d.collector.Namespace
if ns == "" {
ns = "default"
}
deployment, err := client.AppsV1().Deployments(ns).Get(ctx, d.collector.TargetDeployment, metav1.GetOptions{})
if err != nil {
return err
}
deploymentJSON, err := json.Marshal(deployment)
if err != nil {
return err
}
removePatch, err := sidecarRemovePatchYAML(deploymentJSON, d.collector.ID)
if err != nil {
return err
}
if err := patchDeployment(ctx, cluster, ns, d.collector.TargetDeployment, removePatch); err != nil {
return fmt.Errorf("移除 Sidecar 失败: %w", err)
}
if err := deleteK8sResource(ctx, cluster, "configmap", ns, fmt.Sprintf("fluent-bit-config-%d", d.collector.ID)); err != nil {
return err
}
return nil
}
// lokiPortOf 从 endpoint 提取端口(默认 3100)
func lokiPortOf(endpoint string) string {
rest := strings.SplitN(endpoint, "//", 2)
if len(rest) != 2 {
return "3100"
}
hostPart := rest[1]
if i := strings.Index(hostPart, ":"); i >= 0 {
port := hostPart[i+1:]
if j := strings.IndexAny(port, "/"); j >= 0 {
port = port[:j]
}
if port != "" {
return port
}
}
return "3100"
}