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
This commit is contained in:
1 parent
cde7bbe540
commit
7e1c8f583c
84 files changed
+7729
-1507
No files matched your search
@@ -192,6 +192,7 @@ func (e *AlertEngine) evalSiteRule(rule model.MonitorAlertRule, task model.SiteT
|
||||
}
|
||||
|
||||
if !triggered {
|
||||
e.resolveActiveSiteAlerts(task.ID, rule.Metric)
|
||||
return
|
||||
}
|
||||
|
||||
@@ -294,22 +295,79 @@ func extractHostMetricValue(report model.AgentReport, metric string) (float64, b
|
||||
}
|
||||
}
|
||||
|
||||
// autoResolveSiteSslAlerts 证书续期后自动恢复 SSL 过期告警 // Logan.Li
|
||||
func (e *AlertEngine) autoResolveSiteSslAlerts() {
|
||||
// resolveActiveSiteAlerts 指标恢复正常时关闭对应活跃告警 // Logan.Li
|
||||
func (e *AlertEngine) resolveActiveSiteAlerts(taskID uint, triggerType string) {
|
||||
now := time.Now()
|
||||
e.db.Model(&model.SiteAlert{}).
|
||||
Where("task_id = ? AND status = ? AND trigger_type = ?", taskID, model.SiteAlertStatusActive, triggerType).
|
||||
Updates(map[string]interface{}{
|
||||
"status": model.SiteAlertStatusResolved, "resolved_at": now,
|
||||
})
|
||||
}
|
||||
|
||||
// siteRuleStillTriggered 判断任务是否仍满足告警规则 // Logan.Li
|
||||
func (e *AlertEngine) siteRuleStillTriggered(rule model.MonitorAlertRule, task model.SiteTask) bool {
|
||||
switch rule.Metric {
|
||||
case "probe_fail":
|
||||
return task.LastResult == "fail" || task.LastResult == "timeout"
|
||||
case "probe_timeout":
|
||||
return task.LastResult == "timeout"
|
||||
case "http_status_error":
|
||||
return task.LastResult == "fail"
|
||||
case "response_time":
|
||||
return e.compareOperator(rule.Operator, task.AvgResponse, rule.Threshold)
|
||||
case "availability":
|
||||
return e.compareOperator(rule.Operator, task.SuccessRate, rule.Threshold)
|
||||
case "ssl_expiry_days":
|
||||
if !siteprobe.IsSSLProbeTask(task) {
|
||||
return false
|
||||
}
|
||||
var result model.SiteResult
|
||||
e.db.Where("task_id = ? AND ssl_cert_expiry IS NOT NULL", task.ID).
|
||||
Order("checked_at DESC").First(&result)
|
||||
if result.ID == 0 {
|
||||
return false
|
||||
}
|
||||
return e.compareOperator(rule.Operator, float64(result.SSLDaysRemaining), rule.Threshold)
|
||||
default:
|
||||
return true
|
||||
}
|
||||
}
|
||||
|
||||
// autoResolveSiteAlerts 扫描活跃拨测告警,指标已恢复则自动关闭(含 SSL)// Logan.Li
|
||||
func (e *AlertEngine) autoResolveSiteAlerts() {
|
||||
var alerts []model.SiteAlert
|
||||
e.db.Where("status = ? AND trigger_type = ?", model.SiteAlertStatusActive, "ssl_expiry_days").Find(&alerts)
|
||||
e.db.Where("status = ?", model.SiteAlertStatusActive).Find(&alerts)
|
||||
if len(alerts) == 0 {
|
||||
return
|
||||
}
|
||||
taskCache := map[uint]model.SiteTask{}
|
||||
ruleCache := map[string]model.MonitorAlertRule{}
|
||||
now := time.Now()
|
||||
for _, alert := range alerts {
|
||||
var result model.SiteResult
|
||||
e.db.Where("task_id = ? AND ssl_cert_expiry IS NOT NULL", alert.TaskID).
|
||||
Order("checked_at DESC").First(&result)
|
||||
if result.ID == 0 {
|
||||
continue
|
||||
task, ok := taskCache[alert.TaskID]
|
||||
if !ok {
|
||||
var t model.SiteTask
|
||||
if e.db.First(&t, alert.TaskID).Error != nil {
|
||||
continue
|
||||
}
|
||||
task = t
|
||||
taskCache[alert.TaskID] = task
|
||||
}
|
||||
if float64(result.SSLDaysRemaining) >= 30 {
|
||||
rule, ok := ruleCache[alert.Title]
|
||||
if !ok {
|
||||
var r model.MonitorAlertRule
|
||||
e.db.Where("category = ? AND name = ?", model.MonitorAlertSite, alert.Title).First(&r)
|
||||
if r.ID == 0 {
|
||||
r.Metric = alert.TriggerType
|
||||
}
|
||||
rule = r
|
||||
ruleCache[alert.Title] = rule
|
||||
}
|
||||
if rule.Metric == "" {
|
||||
rule.Metric = alert.TriggerType
|
||||
}
|
||||
if !e.siteRuleStillTriggered(rule, task) {
|
||||
e.db.Model(&alert).Updates(map[string]interface{}{
|
||||
"status": model.SiteAlertStatusResolved, "resolved_at": now,
|
||||
})
|
||||
|
||||
Reference in new issue
Block a user