Files
aiops/backend/pkg/siteprobe/worker.go
T
Logan 7e1c8f583c feat: 工具配置增强、站点拨测与日志中心多项改进
- 工具配置:新增卸载 API/UI、Go 工具链、流水线构建自动安装依赖
- 修复 ListTools VerifyTool 递归栈溢出导致获取工具配置失败
- Node.js Alpine 安装改用 unofficial-builds musl 预编译包与 index.json 解析 LTS
- 站点拨测:详情页路由、告警自动恢复、并发拨测与默认可观测频率
- 日志中心:RBAC、查询构建器、ES/Loki 增强与审计 request/response 展示
- 发布配置:构建节点、流水线远程执行、Deploy RBAC 导航拆分
2026-07-02 07:47:33 +08:00

127 lines
3.0 KiB
Go

package siteprobe
import (
"context"
"log"
"sync"
"time"
"ezops/model"
"gorm.io/gorm"
)
// Worker 站点拨测后台采集器 // Logan.Li
type Worker struct {
db *gorm.DB
stopCh chan struct{}
wg sync.WaitGroup
}
// NewWorker 创建拨测 worker
func NewWorker(db *gorm.DB) *Worker {
return &Worker{db: db, stopCh: make(chan struct{})}
}
// Start 启动拨测循环
func (w *Worker) Start(intervalSec int) {
if intervalSec < 5 {
intervalSec = 10
}
w.wg.Add(1)
go func() {
defer w.wg.Done()
ticker := time.NewTicker(time.Duration(intervalSec) * time.Second)
defer ticker.Stop()
log.Printf("[SiteProbe] 拨测引擎启动,扫描间隔 %ds", intervalSec)
w.runOnce()
for {
select {
case <-ticker.C:
w.runOnce()
case <-w.stopCh:
log.Printf("[SiteProbe] 拨测引擎停止")
return
}
}
}()
}
// Stop 停止拨测
func (w *Worker) Stop() {
close(w.stopCh)
w.wg.Wait()
}
// RunTaskNow 立即执行单个任务拨测 // Logan.Li
func (w *Worker) RunTaskNow(taskID uint) error {
var task model.SiteTask
if err := w.db.First(&task, taskID).Error; err != nil {
return err
}
w.probeTask(task)
return nil
}
func (w *Worker) runOnce() {
var tasks []model.SiteTask
w.db.Where("status = ?", model.SiteTaskStatusRunning).Find(&tasks)
now := time.Now()
sem := make(chan struct{}, 8) // 最多 8 路并发拨测,避免串行阻塞 // Logan.Li
var wg sync.WaitGroup
for _, task := range tasks {
task.Frequency = NormalizeProbeFrequency(string(task.TaskType), task.Frequency)
if task.LastCheckAt != nil && now.Sub(*task.LastCheckAt) < time.Duration(task.Frequency)*time.Second {
continue
}
wg.Add(1)
go func(t model.SiteTask) {
defer wg.Done()
sem <- struct{}{}
defer func() { <-sem }()
w.probeTask(t)
}(task)
}
wg.Wait()
}
func (w *Worker) probeTask(task model.SiteTask) {
if task.TaskType == "" {
task.TaskType = model.SiteTaskType(inferTaskType(task))
}
outcome := Execute(context.Background(), task)
now := time.Now()
result := model.SiteResult{
TaskID: task.ID,
Status: outcome.Status,
HTTPStatusCode: outcome.HTTPStatus,
ResponseTime: outcome.ResponseTime,
TotalTime: outcome.TotalTime,
RemoteIP: outcome.RemoteIP,
ErrorMessage: outcome.ErrorMessage,
SSLCertExpiry: outcome.SSLCertExpiry,
SSLCertIssuer: outcome.SSLCertIssuer,
SSLCertSubject: outcome.SSLCertSubject,
SSLDaysRemaining: outcome.SSLDaysRemaining,
CheckedAt: now,
}
w.db.Create(&result)
lastResult := "fail"
if outcome.Status == model.SiteResultSuccess {
lastResult = "success"
}
avg := outcome.ResponseTime
if task.AvgResponse > 0 {
avg = (task.AvgResponse + outcome.ResponseTime) / 2
}
w.db.Model(&model.SiteTask{}).Where("id = ?", task.ID).Updates(map[string]interface{}{
"task_type": task.TaskType,
"last_result": lastResult,
"last_check_at": now,
"avg_response": avg,
})
// 拨测恢复后即时关闭探测类告警,与统计栏 down_count 同步 // Logan.Li
ResolveRecoveredProbeAlerts(w.db, task.ID, lastResult)
}