KJZH001/nezha-2023
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

⚡️ v0.7.0 优化 Dashboard 内存占用

Browse Source
This commit is contained in:
naiba 2021-05-10 17:30:18 +08:00
parent fb686b2f57
commit 842233976b
5 changed files with 93 additions and 81 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
README.md
View File

@ -1,7 +1,7 @@
<div align="center" style="background-color: white">
<img width="500" style="max-width:100%" src="https://raw.githubusercontent.com/naiba/nezha/master/resource/static/brand.png" title="哪吒监控">
<br><br>
<img src="https://img.shields.io/github/workflow/status/naiba/nezha/Dashboard%20image?label=Dash%20v0.6.7&logo=github&style=for-the-badge">&nbsp;<img src="https://img.shields.io/github/v/release/naiba/nezha?color=brightgreen&label=Agent&style=for-the-badge&logo=github">&nbsp;<img src="https://img.shields.io/github/workflow/status/naiba/nezha/Agent%20release?label=Agent%20CI&logo=github&style=for-the-badge">&nbsp;<img src="https://img.shields.io/badge/Installer-v0.5.0-brightgreen?style=for-the-badge&logo=linux">
<img src="https://img.shields.io/github/workflow/status/naiba/nezha/Dashboard%20image?label=Dash%20v0.7.0&logo=github&style=for-the-badge">&nbsp;<img src="https://img.shields.io/github/v/release/naiba/nezha?color=brightgreen&label=Agent&style=for-the-badge&logo=github">&nbsp;<img src="https://img.shields.io/github/workflow/status/naiba/nezha/Agent%20release?label=Agent%20CI&logo=github&style=for-the-badge">&nbsp;<img src="https://img.shields.io/badge/Installer-v0.5.0-brightgreen?style=for-the-badge&logo=linux">
<br>
<p>:trollface: 哪吒监控 一站式轻监控轻运维系统。支持系统状态、HTTP(SSL 证书变更、即将到期、到期)、TCP、Ping 监控报警，命令批量执行和计划任务。</p>
</div>
@ -18,7 +18,7 @@
## 安装脚本
**推荐配置：** 安装前解析 _两个域名_ 到面板服务器，一个作为 _公开访问_ ，可以 **接入 CDN**，比如 (status.nai.ba)；另外一个作为安装 Agent 时连接 Dashboard 使用，**不能接入 CDN** 直接暴露面板主机 IP，比如（randomdashboard.nai.ba）。
**推荐配置：** 安装前准备 _两个域名_，一个可以 **接入 CDN** 作为 _公开访问_，比如 (status.nai.ba)；另外一个解析到面板服务器作为 Agent 连接 Dashboard 使用，**不能接入 CDN** 直接暴露面板主机 IP，比如（randomdashboard.nai.ba）。
```shell
curl -L https://raw.githubusercontent.com/naiba/nezha/master/script/install.sh -o nezha.sh && chmod +x nezha.sh
@ -59,11 +59,13 @@ URL 里面也可放置占位符，请求时会进行简单的字符串替换。
1. 添加通知方式
- server 酱示例
- 名称：server 酱
- URL：https://sc.ftqq.com/SCUrandomkeys.send?text=#NEZHA#
- 请求方式: GET
- 请求类型: 默认
- Body: 空
- wxpusher 示例，需要关注你的应用
- 名称: wxpusher
@ -73,6 +75,7 @@ URL 里面也可放置占位符，请求时会进行简单的字符串替换。
- Body: `{"appToken":"你的appToken","topicIds":[],"content":"#NEZHA#","contentType":"1","uids":["你的uid"]}`
- telegram 示例 [@haitau](https://github.com/haitau) 贡献
- 名称：telegram 机器人消息通知
- URL：https://api.telegram.org/botXXXXXX/sendMessage?chat_id=YYYYYY&text=#NEZHA#
- 请求方式: GET
@ -122,7 +125,7 @@ URL 里面也可放置占位符，请求时会进行简单的字符串替换。
</style>
```
- DayNight 主题更改进度条颜色示例（来自 [@hyt-allen-xu](https://github.com/hyt-allen-xu)，欢迎 PR）
- DayNight 主题更改进度条颜色示例（来自 [@hyt-allen-xu](https://github.com/hyt-allen-xu)）
```
<style>
@ -132,7 +135,7 @@ URL 里面也可放置占位符，请求时会进行简单的字符串替换。
</style>
```
- 默认主题修改 LOGO、移除版权示例（来自 [@iLay1678](https://github.com/iLay1678)，欢迎 PR）
- 默认主题修改 LOGO、移除版权示例（来自 [@iLay1678](https://github.com/iLay1678)）
```
<style>
@ -157,7 +160,8 @@ URL 里面也可放置占位符，请求时会进行简单的字符串替换。
}
</script>
```
- DayNight 移除版权示例（来自 [@hyt-allen-xu](https://github.com/hyt-allen-xu)，欢迎 PR）
- DayNight 移除版权示例（来自 [@hyt-allen-xu](https://github.com/hyt-allen-xu)）
```
<script>

2
cmd/agent/main.go
View File

@ -80,7 +80,7 @@ func main() {
var debug bool
flag.String("i", "", "unused 旧Agent配置兼容")
flag.BoolVar(&debug, "d", false, "允许不安全连接")
flag.BoolVar(&debug, "d", false, "开启调试信息")
flag.StringVar(&server, "s", "localhost:5555", "管理面板RPC端口")
flag.StringVar(&clientSecret, "p", "", "Agent连接Secret")
flag.Parse()

2
script/nezha-agent.service
View File

@ -15,7 +15,7 @@ Type=simple
User=root
Group=root
WorkingDirectory=/opt/nezha/agent/
ExecStart=/opt/nezha/agent/nezha-agent -d -s nz_grpc_host:nz_grpc_port -p nz_client_secret
ExecStart=/opt/nezha/agent/nezha-agent -s nz_grpc_host:nz_grpc_port -p nz_client_secret
Restart=always
#Environment=DEBUG=true

2
service/dao/dao.go
View File

@ -13,7 +13,7 @@ import (
pb "github.com/naiba/nezha/proto"
)
var Version = "v0.6.7" // ！！记得修改 README 中的 badge 版本！！
var Version = "v0.7.0" // ！！记得修改 README 中的 badge 版本！！
const (
SnapshotDelay = 3

150
service/dao/servicesentinel.go
View File

@ -3,6 +3,7 @@ package dao
import (
"fmt"
"log"
"sort"
"strings"
"sync"
"time"
@ -15,20 +16,29 @@ const _CurrentStatusSize = 30 // 统计 5 分钟内的数据为当前状态
var ServiceSentinelShared *ServiceSentinel
type ReportData struct {
Data *pb.TaskResult
Reporter uint64
}
type _TodayStatsOfMonitor struct {
Up int
Down int
Delay float32
}
func NewServiceSentinel() {
ServiceSentinelShared = &ServiceSentinel{
serviceResponseChannel: make(chan ReportData, 200),
serviceResponseDataStoreTodaySavedIndex: make(map[uint64]int),
serviceCurrentStatusIndex: make(map[uint64]int),
serviceCurrentStatusData: make(map[uint64][]model.MonitorHistory),
serviceResponseDataStoreTodayLastSave: make(map[uint64]time.Time),
latestDate: make(map[uint64]string),
lastStatus: make(map[uint64]string),
serviceResponseDataStoreCurrentUp: make(map[uint64]uint64),
serviceResponseDataStoreCurrentDown: make(map[uint64]uint64),
monitors: make(map[uint64]model.Monitor),
serviceResponseDataStoreToday: make(map[uint64][]model.MonitorHistory),
sslCertCache: make(map[uint64]string),
serviceReportChannel: make(chan ReportData, 200),
serviceStatusToday: make(map[uint64]*_TodayStatsOfMonitor),
serviceCurrentStatusIndex: make(map[uint64]int),
serviceCurrentStatusData: make(map[uint64][]model.MonitorHistory),
latestDate: make(map[uint64]string),
lastStatus: make(map[uint64]string),
serviceResponseDataStoreCurrentUp: make(map[uint64]uint64),
serviceResponseDataStoreCurrentDown: make(map[uint64]uint64),
monitors: make(map[uint64]model.Monitor),
sslCertCache: make(map[uint64]string),
}
ServiceSentinelShared.OnMonitorUpdate()
@ -38,48 +48,48 @@ func NewServiceSentinel() {
DB.Where("created_at >= ?", today).Find(&mhs)
// 加载当日记录
totalDelay := make(map[uint64]float32)
for i := 0; i < len(mhs); i++ {
ServiceSentinelShared.serviceResponseDataStoreToday[mhs[i].MonitorID] =
append(ServiceSentinelShared.serviceResponseDataStoreToday[mhs[i].MonitorID], mhs[i])
if mhs[i].Successful {
ServiceSentinelShared.serviceStatusToday[mhs[i].MonitorID].Up++
totalDelay[mhs[i].MonitorID] += mhs[i].Delay
} else {
ServiceSentinelShared.serviceStatusToday[mhs[i].MonitorID].Down++
}
}
for id, delay := range totalDelay {
ServiceSentinelShared.serviceStatusToday[id].Delay = delay / float32(ServiceSentinelShared.serviceStatusToday[id].Up)
}
// 更新入库时间及当日数据入库游标
for k := range ServiceSentinelShared.monitors {
ServiceSentinelShared.latestDate[k] = time.Now().Format("02-Jan-06")
ServiceSentinelShared.serviceResponseDataStoreTodaySavedIndex[k] = len(ServiceSentinelShared.serviceResponseDataStoreToday[k])
}
go ServiceSentinelShared.worker()
}
type ReportData struct {
Data *pb.TaskResult
Reporter uint64
}
/*
使用缓存 channel，处理上报的 Service 请求结果，然后判断是否需要报警
需要记录上一次的状态信息
*/
type ServiceSentinel struct {
serviceResponseDataStoreLock sync.RWMutex
monitorsLock sync.RWMutex
serviceResponseChannel chan ReportData
serviceResponseDataStoreTodaySavedIndex map[uint64]int
serviceCurrentStatusIndex map[uint64]int
serviceCurrentStatusData map[uint64][]model.MonitorHistory
serviceResponseDataStoreTodayLastSave map[uint64]time.Time
latestDate map[uint64]string
lastStatus map[uint64]string
serviceResponseDataStoreCurrentUp map[uint64]uint64
serviceResponseDataStoreCurrentDown map[uint64]uint64
monitors map[uint64]model.Monitor
serviceResponseDataStoreToday map[uint64][]model.MonitorHistory
sslCertCache map[uint64]string
serviceResponseDataStoreLock sync.RWMutex
monitorsLock sync.RWMutex
serviceReportChannel chan ReportData
serviceStatusToday map[uint64]*_TodayStatsOfMonitor
serviceCurrentStatusIndex map[uint64]int
serviceCurrentStatusData map[uint64][]model.MonitorHistory
latestDate map[uint64]string
lastStatus map[uint64]string
serviceResponseDataStoreCurrentUp map[uint64]uint64
serviceResponseDataStoreCurrentDown map[uint64]uint64
monitors map[uint64]model.Monitor
sslCertCache map[uint64]string
}
func (ss *ServiceSentinel) Dispatch(r ReportData) {
ss.serviceResponseChannel <- r
ss.serviceReportChannel <- r
}
func (ss *ServiceSentinel) Monitors() []model.Monitor {
@ -89,6 +99,9 @@ func (ss *ServiceSentinel) Monitors() []model.Monitor {
for _, v := range ss.monitors {
monitors = append(monitors, v)
}
sort.SliceStable(monitors, func(i, j int) bool {
return monitors[i].ID < monitors[j].ID
})
return monitors
}
@ -103,21 +116,21 @@ func (ss *ServiceSentinel) OnMonitorUpdate() {
if len(ss.serviceCurrentStatusData[monitors[i].ID]) == 0 {
ss.serviceCurrentStatusData[monitors[i].ID] = make([]model.MonitorHistory, _CurrentStatusSize)
}
if ss.serviceStatusToday[monitors[i].ID] == nil {
ss.serviceStatusToday[monitors[i].ID] = &_TodayStatsOfMonitor{}
}
}
}
func (ss *ServiceSentinel) OnMonitorDelete(id uint64) {
ss.serviceResponseDataStoreLock.Lock()
defer ss.serviceResponseDataStoreLock.Unlock()
delete(ss.serviceResponseDataStoreTodaySavedIndex, id)
delete(ss.serviceCurrentStatusIndex, id)
delete(ss.serviceCurrentStatusData, id)
delete(ss.serviceResponseDataStoreTodayLastSave, id)
delete(ss.latestDate, id)
delete(ss.lastStatus, id)
delete(ss.serviceResponseDataStoreCurrentUp, id)
delete(ss.serviceResponseDataStoreCurrentDown, id)
delete(ss.serviceResponseDataStoreToday, id)
delete(ss.sslCertCache, id)
ss.monitorsLock.Lock()
defer ss.monitorsLock.Unlock()
@ -176,17 +189,12 @@ func (ss *ServiceSentinel) LoadStats() map[uint64]*model.ServiceItemResponse {
}
}
msm[k].Monitor = ss.monitors[k]
v := ss.serviceResponseDataStoreToday[k]
for i := 0; i < len(v); i++ {
if v[i].Successful {
msm[k].Up[29]++
msm[k].TotalUp++
} else {
msm[k].Down[29]++
msm[k].TotalDown++
}
msm[k].Delay[29] = (msm[k].Delay[29]*float32(msm[k].Up[29]) + v[i].Delay) / float32(msm[k].Up[29]+1)
}
v := ss.serviceStatusToday[k]
msm[k].Up[29] = v.Up
msm[k].Down[29] = v.Down
msm[k].TotalUp += uint64(v.Up)
msm[k].TotalDown += uint64(v.Down)
msm[k].Delay[29] = v.Delay
}
// 最后 5 分钟的状态 与 monitor 对象填充
for k, v := range ss.serviceResponseDataStoreCurrentDown {
@ -212,7 +220,7 @@ func getStateStr(percent uint64) string {
}
func (ss *ServiceSentinel) worker() {
for r := range ss.serviceResponseChannel {
for r := range ss.serviceReportChannel {
if ss.monitors[r.Data.GetId()].ID == 0 {
continue
}
@ -221,33 +229,33 @@ func (ss *ServiceSentinel) worker() {
// 先查看是否到下一天
nowDate := time.Now().Format("02-Jan-06")
if nowDate != ss.latestDate[mh.MonitorID] {
// 清理前一天数据
ss.latestDate[mh.MonitorID] = nowDate
dataToSave := ss.serviceResponseDataStoreToday[mh.MonitorID][ss.serviceResponseDataStoreTodaySavedIndex[mh.MonitorID]:]
if err := DB.Create(&dataToSave).Error; err != nil {
log.Println(err)
}
ss.serviceResponseDataStoreTodaySavedIndex[mh.MonitorID] = 0
ss.serviceResponseDataStoreToday[mh.MonitorID] = []model.MonitorHistory{}
ss.serviceResponseDataStoreCurrentDown[mh.MonitorID] = 0
ss.serviceResponseDataStoreCurrentUp[mh.MonitorID] = 0
ss.serviceResponseDataStoreTodayLastSave[mh.MonitorID] = time.Now()
ss.serviceResponseDataStoreCurrentDown[mh.MonitorID] = 0
ss.serviceStatusToday[mh.MonitorID].Delay = 0
ss.serviceStatusToday[mh.MonitorID].Up = 0
ss.serviceStatusToday[mh.MonitorID].Down = 0
}
// 储存至当日数据
ss.serviceResponseDataStoreToday[mh.MonitorID] = append(ss.serviceResponseDataStoreToday[mh.MonitorID], mh)
// 每20分钟入库一次
if time.Now().After(ss.serviceResponseDataStoreTodayLastSave[mh.MonitorID].Add(time.Minute * 20)) {
ss.serviceResponseDataStoreTodayLastSave[mh.MonitorID] = time.Now()
dataToSave := ss.serviceResponseDataStoreToday[mh.MonitorID][ss.serviceResponseDataStoreTodaySavedIndex[mh.MonitorID]:]
if err := DB.Create(&dataToSave).Error; err != nil {
log.Println(err)
}
ss.serviceResponseDataStoreTodaySavedIndex[mh.MonitorID] = len(ss.serviceResponseDataStoreToday[mh.MonitorID])
// 写入当天状态
if mh.Successful {
ss.serviceStatusToday[mh.MonitorID].Delay = (ss.serviceStatusToday[mh.
MonitorID].Delay*float32(ss.serviceStatusToday[mh.MonitorID].Up) +
mh.Delay) / float32(ss.serviceStatusToday[mh.MonitorID].Up+1)
ss.serviceStatusToday[mh.MonitorID].Up++
} else {
ss.serviceStatusToday[mh.MonitorID].Down++
}
// 写入当前数据
ss.serviceCurrentStatusData[mh.MonitorID][ss.serviceCurrentStatusIndex[mh.MonitorID]] = mh
ss.serviceCurrentStatusIndex[mh.MonitorID]++
// 数据持久化
if ss.serviceCurrentStatusIndex[mh.MonitorID] == _CurrentStatusSize {
ss.serviceCurrentStatusIndex[mh.MonitorID] = 0
dataToSave := ss.serviceCurrentStatusData[mh.MonitorID]
if err := DB.Create(&dataToSave).Error; err != nil {
log.Println(err)
}
}
// 更新当前状态
ss.serviceResponseDataStoreCurrentUp[mh.MonitorID] = 0
@ -267,13 +275,13 @@ func (ss *ServiceSentinel) worker() {
}
stateStr := getStateStr(upPercent)
if Conf.Debug {
log.Println(ss.monitors[mh.MonitorID].Target, stateStr, "Reporter:", r.Reporter, "Successful:", mh.Successful, "Data:", mh.Data)
log.Println(ss.monitors[mh.MonitorID].Target, stateStr, "Agent:", r.Reporter, "Successful:", mh.Successful, "Response:", mh.Data)
}
if stateStr == "故障" || stateStr != ss.lastStatus[mh.MonitorID] {
ss.monitorsLock.RLock()
isSendNotification := (ss.lastStatus[mh.MonitorID] != "" || stateStr == "故障") && ss.monitors[mh.MonitorID].Notify
isNeedSendNotification := (ss.lastStatus[mh.MonitorID] != "" || stateStr == "故障") && ss.monitors[mh.MonitorID].Notify
ss.lastStatus[mh.MonitorID] = stateStr
if isSendNotification {
if isNeedSendNotification {
go SendNotification(fmt.Sprintf("服务监控：%s 服务状态：%s", ss.monitors[mh.MonitorID].Name, stateStr), true)
}
ss.monitorsLock.RUnlock()