Buriburizaemon/nezha_domains
1
0
Fork 0
mirror of https://github.com/Buriburizaem0n/nezha_domains.git synced 2026-02-05 13:10:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

🔖 dashboard v0.10.0

Browse Source
This commit is contained in:
naiba
2021-09-27 21:18:09 +08:00
parent ac3cfa162c
commit 4249fa82d7
11 changed files with 81 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
service/dao/alertsentinel.go
View File

@@ -54,7 +54,7 @@ func AlertSentinelStart() {
checkCount++
if lastPrint.Before(startedAt.Add(-1 * time.Hour)) {
if Conf.Debug {
log.Println("报警规则检测每小时", checkCount, "次", startedAt, time.Now())
log.Println("NEZHA>> 报警规则检测每小时", checkCount, "次", startedAt, time.Now())
}
checkCount = 0
lastPrint = startedAt
@@ -114,11 +114,11 @@ func checkStatus() {
max, passed := alert.Check(alertsStore[alert.ID][server.ID])
if !passed {
alertsPrevState[alert.ID][server.ID] = _RuleCheckFail
message := fmt.Sprintf("报警规则：%s，服务器：%s(%s)，逮到咯，快去看看！", alert.Name, server.Name, utils.IPDesensitize(server.Host.IP))
message := fmt.Sprintf("[主机故障] %s(%s) 规则：%s，", server.Name, utils.IPDesensitize(server.Host.IP), alert.Name)
go SendNotification(message, true)
} else {
if alertsPrevState[alert.ID][server.ID] == _RuleCheckFail {
message := fmt.Sprintf("报警规则：%s，服务器：%s(%s)，已恢复正常", alert.Name, server.Name, utils.IPDesensitize(server.Host.IP))
message := fmt.Sprintf("[主机恢复] %s(%s) 规则：%s", server.Name, utils.IPDesensitize(server.Host.IP), alert.Name)
go SendNotification(message, true)
}
alertsPrevState[alert.ID][server.ID] = _RuleCheckPass

6
service/dao/dao.go
View File

@@ -13,7 +13,7 @@ import (
pb "github.com/naiba/nezha/proto"
)
var Version = "v0.9.34" // ！！记得修改 README 中的 badge 版本！！
var Version = "v0.10.0" // ！！记得修改 README 中的 badge 版本！！
var (
Conf *model.Config
@@ -64,7 +64,7 @@ func ManualTrigger(c *model.Cron) {
Type: model.TaskTypeCommand,
})
} else {
SendNotification(fmt.Sprintf("计划任务：%s，服务器：%s 离线，无法执行。", c.Name, ServerList[c.Servers[j]].Name), false)
SendNotification(fmt.Sprintf("[任务失败] %s，服务器 %s 离线，无法执行。", c.Name, ServerList[c.Servers[j]].Name), false)
}
}
}
@@ -91,7 +91,7 @@ func CronTrigger(cr model.Cron) func() {
Type: model.TaskTypeCommand,
})
} else {
SendNotification(fmt.Sprintf("计划任务：%s，服务器：%s 离线，无法执行。", cr.Name, s.Name), false)
SendNotification(fmt.Sprintf("[任务失败] %s，服务器 %s 离线，无法执行。", cr.Name, s.Name), false)
}
}
}

4
service/dao/notification.go
View File

@@ -71,7 +71,7 @@ func SendNotification(desc string, muteable bool) {
if !flag {
if Conf.Debug {
log.Println("静音的重复通知：", desc, muteable)
log.Println("NEZHA>> 静音的重复通知：", desc, muteable)
}
return
}
@@ -81,7 +81,7 @@ func SendNotification(desc string, muteable bool) {
defer notificationsLock.RUnlock()
for i := 0; i < len(notifications); i++ {
if err := notifications[i].Send(desc); err != nil {
log.Println("发送通知失败：", err)
log.Println("NEZHA>> 发送通知失败：", err)
}
}
}

12
service/dao/servicesentinel.go
View File

@@ -150,6 +150,7 @@ func (ss *ServiceSentinel) loadMonitorHistory() {
monitors[i].CronJobID, err = ss.dispatchCron.AddFunc(task.CronSpec(), func() {
ss.dispatchBus <- task
})
log.Println("NEZHA>> 服务监控任务", monitors[i].ID, monitors[i].Name, monitors[i].CronJobID)
if err != nil {
panic(err)
}
@@ -282,7 +283,8 @@ func getStateStr(percent uint64) string {
func (ss *ServiceSentinel) worker() {
for r := range ss.serviceReportChannel {
if ss.monitors[r.Data.GetId()].ID == 0 {
if ss.monitors[r.Data.GetId()] == nil || ss.monitors[r.Data.GetId()].ID == 0 {
log.Printf("NEZAH>> 错误的服务监控上报 %+v", r)
continue
}
mh := model.PB2MonitorHistory(r.Data)
@@ -315,7 +317,7 @@ func (ss *ServiceSentinel) worker() {
ss.serviceCurrentStatusIndex[mh.MonitorID] = 0
dataToSave := ss.serviceCurrentStatusData[mh.MonitorID]
if err := DB.Create(&dataToSave).Error; err != nil {
log.Println("服务监控数据持久化失败：", err)
log.Println("NEZHA>> 服务监控数据持久化失败：", err)
}
}
// 更新当前状态
@@ -337,7 +339,7 @@ func (ss *ServiceSentinel) worker() {
stateStr := getStateStr(upPercent)
if !mh.Successful {
ServerLock.RLock()
log.Println("服务故障上报：", ss.monitors[mh.MonitorID].Target, stateStr, "上报者：", ServerList[r.Reporter].Name, "请求输出：", mh.Data)
log.Println("NEZHA>> 服务故障上报：", ss.monitors[mh.MonitorID].Target, stateStr, "上报者：", ServerList[r.Reporter].Name, "请求输出：", mh.Data)
ServerLock.RUnlock()
}
if stateStr == "故障" || stateStr != ss.lastStatus[mh.MonitorID] {
@@ -345,7 +347,7 @@ func (ss *ServiceSentinel) worker() {
isNeedSendNotification := (ss.lastStatus[mh.MonitorID] != "" || stateStr == "故障") && ss.monitors[mh.MonitorID].Notify
ss.lastStatus[mh.MonitorID] = stateStr
if isNeedSendNotification {
go SendNotification(fmt.Sprintf("服务监控：%s 服务状态：%s", ss.monitors[mh.MonitorID].Name, stateStr), true)
go SendNotification(fmt.Sprintf("[服务%s] %s", stateStr, ss.monitors[mh.MonitorID].Name), true)
}
ss.monitorsLock.RUnlock()
}
@@ -389,7 +391,7 @@ func (ss *ServiceSentinel) worker() {
if errMsg != "" {
ss.monitorsLock.RLock()
if ss.monitors[mh.MonitorID].Notify {
go SendNotification(fmt.Sprintf("服务监控：%s %s", ss.monitors[mh.MonitorID].Name, errMsg), true)
go SendNotification(fmt.Sprintf("[SSL] %s %s", ss.monitors[mh.MonitorID].Name, errMsg), true)
}
ss.monitorsLock.RUnlock()
}

18
service/rpc/nezha.go
View File

@@ -21,12 +21,7 @@ func (s *NezhaHandler) ReportTask(c context.Context, r *pb.TaskResult) (*pb.Rece
if clientID, err = s.Auth.Check(c); err != nil {
return nil, err
}
if r.GetType() != model.TaskTypeCommand {
dao.ServiceSentinelShared.Dispatch(dao.ReportData{
Data: r,
Reporter: clientID,
})
} else {
if r.GetType() == model.TaskTypeCommand {
// 处理上报的计划任务
dao.CronLock.RLock()
defer dao.CronLock.RUnlock()
@@ -35,16 +30,21 @@ func (s *NezhaHandler) ReportTask(c context.Context, r *pb.TaskResult) (*pb.Rece
dao.ServerLock.RLock()
defer dao.ServerLock.RUnlock()
if cr.PushSuccessful && r.GetSuccessful() {
dao.SendNotification(fmt.Sprintf("成功计划任务：%s ，服务器：%s，日志：\n%s", cr.Name, dao.ServerList[clientID].Name, r.GetData()), false)
dao.SendNotification(fmt.Sprintf("[任务成功] %s ，服务器：%s，日志：\n%s", cr.Name, dao.ServerList[clientID].Name, r.GetData()), false)
}
if !r.GetSuccessful() {
dao.SendNotification(fmt.Sprintf("失败计划任务：%s ，服务器：%s，日志：\n%s", cr.Name, dao.ServerList[clientID].Name, r.GetData()), false)
dao.SendNotification(fmt.Sprintf("[任务失败] %s ，服务器：%s，日志：\n%s", cr.Name, dao.ServerList[clientID].Name, r.GetData()), false)
}
dao.DB.Model(cr).Updates(model.Cron{
LastExecutedAt: time.Now().Add(time.Second * -1 * time.Duration(r.GetDelay())),
LastResult: r.GetSuccessful(),
})
}
} else if model.IsServiceSentinelNeeded(r.GetType()) {
dao.ServiceSentinelShared.Dispatch(dao.ReportData{
Data: r,
Reporter: clientID,
})
}
return &pb.Receipt{Proced: true}, nil
}
@@ -101,7 +101,7 @@ func (s *NezhaHandler) ReportSystemInfo(c context.Context, r *pb.Host) (*pb.Rece
host.IP != "" &&
dao.ServerList[clientID].Host.IP != host.IP {
dao.SendNotification(fmt.Sprintf(
"IP变更提醒 服务器：%s ，旧IP：%s，新IP：%s。",
"[IP变更] %s ，旧IP：%s，新IP：%s。",
dao.ServerList[clientID].Name, utils.IPDesensitize(dao.ServerList[clientID].Host.IP), utils.IPDesensitize(host.IP)), true)
}