Files
LokiStack/backend/services/collector.go
T
hermes-agent 7e3356e32b feat(collectors): 主机采集器容器优先托管 + 采集器日志查看 + 推送地址纠正
现象
1. 主机采集器只有 systemd 一条路,跨发行版差异大(这轮已踩过 param 被删、日志路径、
   时间戳格式三处);进程存活/开机自启也没有统一保证。
2. 「采集器日志」在主机模式下无法从平台查看,排「起没起来/在报什么错」必须 SSH。
3. 实测发现:Loki 部署把 listenIP 填成 0.0.0.0 时,部署记录 endpoint 是
   http://0.0.0.0:3100,采集器照抄去推 -> connect: connection refused,
   而容器/systemd 都是 running、平台 status=active —— 日志静默丢失。

根因
1. 托管方式写死在代码里(promtail 二进制 + systemd 单元),没有能力探测。
2. 没有对外暴露采集器自身日志的接口。
3. 渲染 promtail 配置时直接用 deployment.Endpoint,没有校验它是不是「可连接的推送地址」;
   另外容器用 bridge 网络,127.0.0.1 在容器里指向容器自身,同机 Loki 也会推失败。

修法
- services/collector_host_runtime.go(新增)
  * detectHostCapabilities:一次 SSH 往返探测 架构/init/systemd/docker可用/开机自启/SELinux/journald
  * chooseHostRunMode:docker 优先 -> systemd 兜底 -> 两者都没有则明确报错(不裸跑进程)
  * installHostDocker:grafana/promtail:3.4.0 容器,--restart=always --memory=256m --cpus=1
    --log-opt max-size=10m,max-file=3,挂 /etc/promtail(:ro) /var/lib/promtail /var/log(:ro)
    (+journald 相关),SELinux Enforcing 时 --security-opt label=disable;
    先 docker run --rm -check-syntax 校验配置,再替换正式配置
  * waitContainerRunning:看 Status + RestartCount,起来就崩立即报错并带回 docker logs
  * hostPushEndpoint:0.0.0.0/:: 换真实主机 IP;容器模式下 127.0.0.1/localhost 也换;
    换不了明确报错
  * CollectorLogs:主机 docker logs / journalctl,K8s DaemonSet Pod 日志、Sidecar fluent-bit 容器日志
- collector_deployer.go:installHost 探测->选模式->渲染配置->分发;systemd 路径从 docker 迁回时
  清理同名容器;卸载时容器与 systemd 两种方式都清(RunMode 对老记录是空的)
- 拉不到镜像(内网无外网)时降级 systemd 二进制托管,其它错误原样抛出
- models/collector.go:新增 runMode 并纳入 Sanitized;services.Update 白名单同步
- 接口 GET /api/v1/collectors/:id/logs?tail=N;前端卡片加「运行方式」+「日志」弹窗
- build.sh:detect_go 的 INFO 走 stderr(原来被打进 GO_CMD,构建必失败)

验证(172.19.16.8 / pre-cex-ops-jenkins,collector id=2)
- 安装后 runMode=docker,容器 Up,RestartPolicy=always,systemd promtail 自动停用
- 推送地址自动纠正为 http://172.19.16.8:3100,Loki 近 10 分钟查到 4919 条
  job=system/job=app、host=pre-cex-ops-jenkins
- docker 垫片模拟「docker 不可用」-> 自动回退 runMode=systemd、unit active+enabled、
  日志接口 source=systemd;移除垫片重装 -> 切回 docker
- 卸载:容器 0 个、unit 文件不存在、/etc/promtail 不存在、runMode 清空;重装恢复 running/0 重启
- K8s 采集器日志接口 source=kubernetes(顺带暴露 429 限流,见后续日志量治理)
- go vet 干净;go test ./... 全绿(新增 chooseHostRunMode / hostPushEndpoint 等 6 组用例);
  tsc -b 通过
- 镜像 lghost/lokistack:latest = :v0.1 = sha256:906a6d5369d7,Pod imageID 一致
2026-10-28 20:01:38 +00:00

526 lines
16 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"
"strings"
"time"
"lokistack/models"
"lokistack/repository"
"lokistack/utils"
)
type CollectorService struct{}
func NewCollectorService() *CollectorService {
return &CollectorService{}
}
type CreateCollectorRequest struct {
Name string
Type string
LokiDeploymentID uint
HostID *uint
ClusterID *uint
TargetDeployment string
Namespace string
LogPath string
Labels string
ScrapeJobs []models.ScrapeJob
}
// fillAssociations 填充关联的 Loki / Host / Cluster(用于配置生成与展示)
func (s *CollectorService) fillAssociations(c *models.Collector) {
db := repository.GetDB()
db.RLock()
defer db.RUnlock()
for i := range db.Deployments {
if db.Deployments[i].ID == c.LokiDeploymentID {
d := db.Deployments[i]
c.LokiDeployment = &d
break
}
}
if c.HostID != nil {
for i := range db.Hosts {
if db.Hosts[i].ID == *c.HostID {
h := db.Hosts[i]
c.Host = &h
break
}
}
}
if c.ClusterID != nil {
for i := range db.Clusters {
if db.Clusters[i].ID == *c.ClusterID {
cl := db.Clusters[i]
c.Cluster = &cl
break
}
}
}
// 配置快照按当前字段实时渲染:DB 里存的 Config 只是上次安装时的副本,
// 用户改完采集项还没点「部署」时,看配置应该看到「将要生效的 YAML」。
// renderConfig 只读 c 的字段、不取锁,所以在 RLock 内调用是安全的。
if cfg, err := s.renderConfig(c); err == nil {
c.Config = cfg
}
}
// parseScrapeJobs 把接口层传来的 scrapeJobs([]interface{} / JSON 字符串)解析成模型。
// gin 的 map 绑定拿不到强类型,统一过一遍 JSON 转换,避免手写一堆类型断言。
func parseScrapeJobs(v interface{}) ([]models.ScrapeJob, error) {
if v == nil {
return nil, nil
}
raw, err := json.Marshal(v)
if err != nil {
return nil, fmt.Errorf("采集项格式不合法:%w", err)
}
var jobs []models.ScrapeJob
if err := json.Unmarshal(raw, &jobs); err != nil {
return nil, fmt.Errorf("采集项格式不合法:%w", err)
}
return jobs, nil
}
// renderConfig 按类型渲染当前配置(不落库)。
func (s *CollectorService) renderConfig(c *models.Collector) (string, error) {
endpoint := ""
if c.LokiDeployment != nil {
endpoint = c.LokiDeployment.Endpoint
}
clusterName := "host"
if c.Cluster != nil && c.Cluster.Name != "" {
clusterName = c.Cluster.Name
}
switch c.Type {
case "host-promtail":
// 主机名与 syslog 格式:优先用上次安装时探测到的真实值,
// 从未安装过时退回平台登记的主机名/IP(预览会提示这一点)。
hostname := c.ResolvedHostname
if hostname == "" && c.Host != nil {
hostname = c.Host.Name
}
if hostname == "" && c.Host != nil {
hostname = c.Host.HostIP
}
// 预览要与线上一致:这里也走一遍推送地址纠正(0.0.0.0 / 容器内回环)
pushEndpoint := endpoint
hostIP := ""
if c.Host != nil {
hostIP = c.Host.HostIP
}
if v, perr := hostPushEndpoint(endpoint, hostIP, c.RunMode == HostRunModeDocker); perr == nil {
pushEndpoint = v
}
return PromtailConfigForHost(HostPromtailOptions{
Endpoint: pushEndpoint,
Cluster: "host",
LabelsJSON: c.Labels,
Hostname: hostname,
SyslogFormat: c.SyslogFormat,
SystemLogPaths: c.SystemLogPaths,
LogPath: c.LogPath,
Jobs: c.ScrapeJobs,
})
case "k8s-daemonset":
ns := c.Namespace
if ns == "" && c.LokiDeployment != nil {
ns = c.LokiDeployment.Namespace
}
if ns == "" {
ns = "log"
}
return PromtailDaemonsetYAML(ns, endpoint, clusterName, c.Labels, c.TargetNamespaces, c.AppLabelSource, c.AppInclude), nil
case "k8s-fluentbit-sidecar":
host, port := parseLokiURL(endpoint)
app := c.TargetDeployment
ns := c.Namespace
if ns == "" {
ns = "default"
}
return FluentbitConfig(host, port, c.LogPath, app, ns), nil
}
return "", nil
}
// generateConfigSnapshot 按类型生成配置快照(渲染失败返回空串,不阻断保存)
func (s *CollectorService) generateConfigSnapshot(c *models.Collector) string {
cfg, err := s.renderConfig(c)
if err != nil {
return ""
}
return cfg
}
// PreviewConfig 按「当前已保存配置 + 本次提交的改动」渲染 YAML 预览。
// 只渲染不落库:用户在编辑器里点「预览」时用,避免为了看一眼 YAML 就改线上采集。
func (s *CollectorService) PreviewConfig(id uint, updates map[string]interface{}) (map[string]interface{}, error) {
c, err := s.Get(id)
if err != nil {
return nil, err
}
if v, ok := updates["logPath"].(string); ok {
c.LogPath = v
}
if v, ok := updates["labels"].(string); ok {
c.Labels = v
}
if v, ok := updates["scrapeJobs"]; ok {
jobs, perr := parseScrapeJobs(v)
if perr != nil {
return nil, perr
}
c.ScrapeJobs = jobs
}
if c.Type == "host-promtail" {
if err := models.ValidateScrapeJobs(c.ScrapeJobs); err != nil {
return nil, err
}
}
cfg, err := s.renderConfig(c)
if err != nil {
return nil, err
}
warnings := make([]string, 0, 2)
if c.Type == "host-promtail" {
if c.SyslogFormat == "" {
warnings = append(warnings, "尚未成功安装过:系统日志时间戳格式会在安装时按目标机实际格式探测(ISO8601 / 传统 syslog)")
}
if c.ResolvedHostname == "" {
warnings = append(warnings, "尚未成功安装过:host 标签将使用平台登记的主机名,实际安装时会替换为目标机真实 hostname")
}
// 推送地址填写了通配/回环地址时提前提示:这类地址采集器推不上去,
// 表现是「容器/systemd 都 running,但 Loki 里一条日志都没有」。
rawEndpoint := ""
if c.LokiDeployment != nil {
rawEndpoint = c.LokiDeployment.Endpoint
}
hostIP := ""
if c.Host != nil {
hostIP = c.Host.HostIP
}
if fixed, perr := hostPushEndpoint(rawEndpoint, hostIP, c.RunMode == HostRunModeDocker); perr == nil && fixed != rawEndpoint {
warnings = append(warnings, fmt.Sprintf("Loki 推送地址 %s 不可用,部署时会自动改为 %s(建议修正 Loki 部署的监听地址)", rawEndpoint, fixed))
}
}
return map[string]interface{}{"config": cfg, "warnings": warnings, "hostname": c.ResolvedHostname, "syslogFormat": c.SyslogFormat}, nil
}
func (s *CollectorService) Create(req *CreateCollectorRequest) (*models.Collector, error) {
switch req.Type {
case "host-promtail", "k8s-daemonset", "k8s-fluentbit-sidecar":
default:
return nil, fmt.Errorf("不支持的采集器类型: %s", req.Type)
}
if req.LokiDeploymentID == 0 {
return nil, fmt.Errorf("请选择关联的 Loki 部署")
}
if req.Type == "host-promtail" && (req.HostID == nil || *req.HostID == 0) {
return nil, fmt.Errorf("主机采集器需要选择目标主机")
}
if (req.Type == "k8s-daemonset" || req.Type == "k8s-fluentbit-sidecar") && (req.ClusterID == nil || *req.ClusterID == 0) {
return nil, fmt.Errorf("Kubernetes 采集器需要选择目标集群")
}
if req.Type == "k8s-fluentbit-sidecar" && req.TargetDeployment == "" {
return nil, fmt.Errorf("Sidecar 采集器需要填写目标 Deployment 名称")
}
refType := "cluster"
if req.Type == "host-promtail" {
refType = "host"
}
collector := models.Collector{
ID: repository.GetDB().NextID("collectors"),
Name: req.Name,
Type: req.Type,
LokiDeploymentID: req.LokiDeploymentID,
RefType: refType,
HostID: req.HostID,
ClusterID: req.ClusterID,
TargetDeployment: req.TargetDeployment,
Namespace: req.Namespace,
LogPath: req.LogPath,
Labels: req.Labels,
ScrapeJobs: req.ScrapeJobs,
Status: "pending",
CreatedAt: time.Now(),
UpdatedAt: time.Now(),
}
s.fillAssociations(&collector)
if collector.LokiDeployment == nil {
return nil, fmt.Errorf("关联的 Loki 部署不存在")
}
collector.Config = s.generateConfigSnapshot(&collector)
db := repository.GetDB()
db.Lock()
db.Collectors = append(db.Collectors, collector)
db.Unlock()
db.RequestSave()
go s.Install(collector.ID)
return &collector, nil
}
func (s *CollectorService) Get(id uint) (*models.Collector, error) {
db := repository.GetDB()
db.RLock()
var found *models.Collector
for i := range db.Collectors {
if db.Collectors[i].ID == id {
c := db.Collectors[i]
found = &c
break
}
}
db.RUnlock()
if found == nil {
return nil, utils.ErrNotFound
}
s.fillAssociations(found)
return found, nil
}
func (s *CollectorService) List(page, pageSize int) ([]models.Collector, int64, error) {
db := repository.GetDB()
db.RLock()
total := int64(len(db.Collectors))
start := (page - 1) * pageSize
if start < 0 || start >= len(db.Collectors) {
db.RUnlock()
return []models.Collector{}, total, nil
}
end := start + pageSize
if end > len(db.Collectors) {
end = len(db.Collectors)
}
result := make([]models.Collector, end-start)
copy(result, db.Collectors[start:end])
db.RUnlock()
for i := range result {
s.fillAssociations(&result[i])
}
return result, total, nil
}
func (s *CollectorService) Update(id uint, updates map[string]interface{}) error {
db := repository.GetDB()
db.Lock()
defer db.Unlock()
for i := range db.Collectors {
if db.Collectors[i].ID == id {
if v, ok := updates["name"].(string); ok && v != "" {
db.Collectors[i].Name = v
}
if v, ok := updates["status"].(string); ok && v != "" {
db.Collectors[i].Status = v
}
if v, ok := updates["statusMessage"].(string); ok {
db.Collectors[i].StatusMessage = v
}
if v, ok := updates["logPath"].(string); ok {
if v != db.Collectors[i].LogPath {
db.Collectors[i].StatusMessage = "采集路径已保存,点击部署以应用新配置"
}
db.Collectors[i].LogPath = v
}
if v, ok := updates["labels"].(string); ok {
db.Collectors[i].Labels = v
}
// targetNamespaces: 采集哪些命名空间([]string;空=全集群)
if v, ok := updates["targetNamespaces"]; ok {
var nss []string
switch t := v.(type) {
case []interface{}:
for _, item := range t {
if s, ok := item.(string); ok {
if s = strings.TrimSpace(s); s != "" {
nss = append(nss, s)
}
}
}
case []string:
for _, s := range t {
if s = strings.TrimSpace(s); s != "" {
nss = append(nss, s)
}
}
case string:
// 兼容 "astralx,log" 这种逗号分隔写法
for _, s := range strings.Split(t, ",") {
if s = strings.TrimSpace(s); s != "" {
nss = append(nss, s)
}
}
}
if len(nss) != len(db.Collectors[i].TargetNamespaces) {
db.Collectors[i].StatusMessage = "采集命名空间已保存,点击部署以应用新配置"
} else {
for j := range nss {
if nss[j] != db.Collectors[i].TargetNamespaces[j] {
db.Collectors[i].StatusMessage = "采集命名空间已保存,点击部署以应用新配置"
break
}
}
}
db.Collectors[i].TargetNamespaces = nss
}
// appLabelSource:app 标签取值来源(空=默认 controller_name 回退链)
if v, ok := updates["appLabelSource"].(string); ok {
if strings.TrimSpace(v) != db.Collectors[i].AppLabelSource {
db.Collectors[i].StatusMessage = "app 标签来源已保存,点击部署以应用新配置"
}
db.Collectors[i].AppLabelSource = strings.TrimSpace(v)
}
// appInclude:只采集这些 app(白名单,空=不限)
if v, ok := updates["appInclude"]; ok {
var apps []string
switch t := v.(type) {
case []interface{}:
for _, item := range t {
if s, ok := item.(string); ok {
if s = strings.TrimSpace(s); s != "" {
apps = append(apps, s)
}
}
}
case []string:
for _, s := range t {
if s = strings.TrimSpace(s); s != "" {
apps = append(apps, s)
}
}
case string:
for _, s := range strings.Split(t, ",") {
if s = strings.TrimSpace(s); s != "" {
apps = append(apps, s)
}
}
}
if len(apps) != len(db.Collectors[i].AppInclude) {
db.Collectors[i].StatusMessage = "采集 app 白名单已保存,点击部署以应用新配置"
}
db.Collectors[i].AppInclude = apps
}
// scrapeJobs:主机模式的结构化采集项(B 档)
if v, ok := updates["scrapeJobs"]; ok {
jobs, jerr := parseScrapeJobs(v)
if jerr != nil {
return jerr
}
if verr := models.ValidateScrapeJobs(jobs); verr != nil {
return verr
}
if len(jobs) != len(db.Collectors[i].ScrapeJobs) {
db.Collectors[i].StatusMessage = "采集项已保存,点击部署以应用新配置"
}
db.Collectors[i].ScrapeJobs = jobs
}
// 以下三个字段由安装流程写回(探测结果 / 渲染快照),不在前端表单里
if v, ok := updates["resolvedHostname"].(string); ok && v != "" {
db.Collectors[i].ResolvedHostname = v
}
if v, ok := updates["syslogFormat"].(string); ok && v != "" {
db.Collectors[i].SyslogFormat = v
}
// 系统日志路径也是探测结果(Ubuntu 与 RHEL 系不同),
// 不落库的话「配置预览」会和线上实际采集路径不一致。
if v, ok := updates["systemLogPaths"].([]string); ok && len(v) > 0 {
db.Collectors[i].SystemLogPaths = v
}
// runMode:安装时探测出的托管方式(docker | systemd),卸载后清空。
if v, ok := updates["runMode"].(string); ok {
db.Collectors[i].RunMode = v
}
if v, ok := updates["config"].(string); ok && v != "" {
db.Collectors[i].Config = v
}
db.Collectors[i].UpdatedAt = time.Now()
db.RequestSave()
return nil
}
}
return utils.ErrNotFound
}
func (s *CollectorService) Delete(id uint) error {
collector, err := s.Get(id)
if err != nil {
return utils.ErrNotFound
}
db := repository.GetDB()
db.Lock()
for i, c := range db.Collectors {
if c.ID == id {
db.Collectors = append(db.Collectors[:i], db.Collectors[i+1:]...)
break
}
}
db.Unlock()
db.RequestSave()
// 异步卸载(用拷贝的 collector,避免删除后取不到)
go func() {
ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
defer cancel()
NewCollectorDeployer(collector, collector.LokiDeployment, nil).Uninstall(ctx)
}()
return nil
}
// Install 异步部署采集器
func (s *CollectorService) Install(id uint) {
collector, err := s.Get(id)
if err != nil {
return
}
if collector.LokiDeployment == nil {
s.Update(id, map[string]interface{}{"status": "error", "statusMessage": "关联的 Loki 不存在"})
return
}
ctx, cancel := context.WithTimeout(context.Background(), 10*time.Minute)
defer cancel()
if err := NewCollectorDeployer(collector, collector.LokiDeployment, nil).Install(ctx); err != nil {
s.Update(id, map[string]interface{}{"status": "error", "statusMessage": err.Error()})
return
}
// 安装时探测到的真实 hostname / syslog 格式 + 实际生效的配置都要落库,
// 否则「配置预览」与线上真实配置不一致(用户没法核对)。
updates := map[string]interface{}{"status": "active", "statusMessage": "采集器已部署"}
if collector.Config != "" {
updates["config"] = collector.Config
}
if collector.ResolvedHostname != "" {
updates["resolvedHostname"] = collector.ResolvedHostname
}
if collector.SyslogFormat != "" {
updates["syslogFormat"] = collector.SyslogFormat
}
if len(collector.SystemLogPaths) > 0 {
updates["systemLogPaths"] = collector.SystemLogPaths
}
// 托管方式(docker | systemd)由安装流程探测后写回,前端据此展示运行方式
updates["runMode"] = collector.RunMode
s.Update(id, updates)
}
// Uninstall 异步卸载采集器
func (s *CollectorService) Uninstall(id uint) {
collector, err := s.Get(id)
if err != nil {
return
}
if collector.LokiDeployment == nil {
return
}
ctx, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
defer cancel()
NewCollectorDeployer(collector, collector.LokiDeployment, nil).Uninstall(ctx)
s.Update(id, map[string]interface{}{"status": "pending", "statusMessage": "采集器已卸载", "runMode": ""})
}