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

♻️ 重构部分代码，打印调试信息

Browse Source
This commit is contained in:
naiba 2021-04-20 19:30:34 +08:00
parent d3c3e55c88
commit 6286e34af0
7 changed files with 70 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
cmd/agent/main.go
View File

@ -35,11 +35,10 @@ var (
)
var (
reporting bool
client pb.NezhaServiceClient
ctx = context.Background()
delayWhenError = time.Second * 10 // Agent 重连间隔
updateCh = make(chan struct{}, 0) // Agent 自动更新间隔
delayWhenError = time.Second * 10 // Agent 重连间隔
updateCh = make(chan struct{}) // Agent 自动更新间隔
httpClient = &http.Client{
Transport: &http.Transport{
TLSClientConfig: &tls.Config{InsecureSkipVerify: true},
@ -179,12 +178,14 @@ func doTask(task *pb.Task) {
start := time.Now()
resp, err := httpClient.Get(task.GetData())
if err == nil {
result.Delay = float32(time.Now().Sub(start).Microseconds()) / 1000.0
// 检查 HTTP Response 状态
result.Delay = float32(time.Since(start).Microseconds()) / 1000.0
if resp.StatusCode > 399 || resp.StatusCode < 200 {
err = errors.New("\n应用错误：" + resp.Status)
}
}
if err == nil {
// 检查 SSL 证书信息
if strings.HasPrefix(task.GetData(), "https://") {
c := cert.NewCert(task.GetData()[8:])
if c.Error != "" {
@ -197,6 +198,7 @@ func doTask(task *pb.Task) {
result.Successful = true
}
} else {
// HTTP 请求失败
result.Data = err.Error()
}
case model.TaskTypeICMPPing:
@ -219,7 +221,7 @@ func doTask(task *pb.Task) {
if err == nil {
conn.Write([]byte("ping\n"))
conn.Close()
result.Delay = float32(time.Now().Sub(start).Microseconds()) / 1000.0
result.Delay = float32(time.Since(start).Microseconds()) / 1000.0
result.Successful = true
} else {
result.Data = err.Error()
@ -260,7 +262,7 @@ func doTask(task *pb.Task) {
result.Data = string(output)
result.Successful = true
}
result.Delay = float32(time.Now().Sub(startedAt).Seconds())
result.Delay = float32(time.Since(startedAt).Seconds())
default:
log.Printf("Unknown action: %v", task)
}

2
cmd/dashboard/controller/oauth2.go
View File

@ -4,7 +4,6 @@ import (
"context"
"errors"
"fmt"
"log"
"net/http"
"strings"
@ -71,7 +70,6 @@ func (oa *oauth2controller) callback(c *gin.Context) {
if err == nil {
gu, _, err = client.Users.Get(ctx, "")
}
log.Printf("%+v", gu)
if err != nil {
mygin.ShowErrorPage(c, mygin.ErrInfo{
Code: http.StatusBadRequest,

2
cmd/playground/main.go
View File

@ -34,7 +34,7 @@ func tcpping() {
}
conn.Write([]byte("ping\n"))
conn.Close()
fmt.Println(time.Now().Sub(start).Microseconds(), float32(time.Now().Sub(start).Microseconds())/1000.0)
fmt.Println(time.Since(start).Microseconds(), float32(time.Since(start).Microseconds())/1000.0)
}
func sysinfo() {

4
service/dao/alertsentinel.go
View File

@ -43,7 +43,9 @@ func AlertSentinelStart() {
checkStatus()
checkCount++
if lastPrint.Before(startedAt.Add(-1 * time.Hour)) {
log.Println("报警规则检测每小时", checkCount, "次", startedAt, time.Now())
if Conf.Debug {
log.Println("报警规则检测每小时", checkCount, "次", startedAt, time.Now())
}
checkCount = 0
lastPrint = startedAt
}

2
service/dao/dao.go
View File

@ -46,7 +46,7 @@ func ReSortServer() {
sort.SliceStable(SortedServerList, func(i, j int) bool {
if SortedServerList[i].DisplayIndex == SortedServerList[j].DisplayIndex {
return SortedServerList[i].ID < SortedServerList[i].ID
return SortedServerList[i].ID < SortedServerList[j].ID
}
return SortedServerList[i].DisplayIndex > SortedServerList[j].DisplayIndex
})

79
service/dao/servicesentinel.go
View File

@ -15,7 +15,7 @@ var ServiceSentinelShared *ServiceSentinel
func NewServiceSentinel() {
ServiceSentinelShared = &ServiceSentinel{
serviceResponseChannel: make(chan *pb.TaskResult, 200),
serviceResponseChannel: make(chan ReportData, 200),
serviceResponseDataStoreTodaySavedIndex: make(map[uint64]int),
serviceCurrentStatusIndex: make(map[uint64]int),
serviceCurrentStatusData: make(map[uint64][]model.MonitorHistory),
@ -26,14 +26,35 @@ func NewServiceSentinel() {
serviceResponseDataStoreCurrentDown: make(map[uint64]uint64),
monitors: make(map[uint64]model.Monitor),
serviceResponseDataStoreToday: make(map[uint64][]model.MonitorHistory),
sslCertCache: make(map[uint64]string),
}
ServiceSentinelShared.OnMonitorUpdate()
year, month, day := time.Now().Date()
today := time.Date(year, month, day, 0, 0, 0, 0, time.Local)
var mhs []model.MonitorHistory
DB.Where("created_at >= ?", today).Find(&mhs)
// 加载当日记录
for i := 0; i < len(mhs); i++ {
ServiceSentinelShared.serviceResponseDataStoreToday[mhs[i].MonitorID] =
append(ServiceSentinelShared.serviceResponseDataStoreToday[mhs[i].MonitorID], mhs[i])
}
// 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
@ -41,7 +62,7 @@ func NewServiceSentinel() {
type ServiceSentinel struct {
serviceResponseDataStoreLock sync.RWMutex
monitorsLock sync.RWMutex
serviceResponseChannel chan *pb.TaskResult
serviceResponseChannel chan ReportData
serviceResponseDataStoreTodaySavedIndex map[uint64]int
serviceCurrentStatusIndex map[uint64]int
serviceCurrentStatusData map[uint64][]model.MonitorHistory
@ -52,9 +73,10 @@ type ServiceSentinel struct {
serviceResponseDataStoreCurrentDown map[uint64]uint64
monitors map[uint64]model.Monitor
serviceResponseDataStoreToday map[uint64][]model.MonitorHistory
sslCertCache map[uint64]string
}
func (ss *ServiceSentinel) Dispatch(r *pb.TaskResult) {
func (ss *ServiceSentinel) Dispatch(r ReportData) {
ss.serviceResponseChannel <- r
}
@ -94,6 +116,7 @@ func (ss *ServiceSentinel) OnMonitorDelete(id uint64) {
delete(ss.serviceResponseDataStoreCurrentUp, id)
delete(ss.serviceResponseDataStoreCurrentDown, id)
delete(ss.serviceResponseDataStoreToday, id)
delete(ss.sslCertCache, id)
ss.monitorsLock.Lock()
defer ss.monitorsLock.Unlock()
delete(ss.monitors, id)
@ -186,10 +209,10 @@ func getStateStr(percent uint64) string {
func (ss *ServiceSentinel) worker() {
for r := range ss.serviceResponseChannel {
if ss.monitors[r.GetId()].ID == 0 {
if ss.monitors[r.Data.GetId()].ID == 0 {
continue
}
mh := model.PB2MonitorHistory(r)
mh := model.PB2MonitorHistory(r.Data)
ss.serviceResponseDataStoreLock.Lock()
// 先查看是否到下一天
nowDate := time.Now().Format("02-Jan-06")
@ -239,30 +262,34 @@ func (ss *ServiceSentinel) worker() {
upPercent = ss.serviceResponseDataStoreCurrentUp[mh.MonitorID] * 100 / (ss.serviceResponseDataStoreCurrentDown[mh.MonitorID] + ss.serviceResponseDataStoreCurrentUp[mh.MonitorID])
}
stateStr := getStateStr(upPercent)
log.Println(ss.monitors[mh.MonitorID].Target, stateStr)
if Conf.Debug {
log.Println(ss.monitors[mh.MonitorID].Target, stateStr, "Reporter:", r.Reporter, "Successful:", mh.Successful, "Data:", 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
ss.lastStatus[mh.MonitorID] = stateStr
if isSendNotification {
SendNotification(fmt.Sprintf("服务监控：%s 服务状态：%s", ss.monitors[mh.MonitorID].Name, stateStr), true)
go SendNotification(fmt.Sprintf("服务监控：%s 服务状态：%s", ss.monitors[mh.MonitorID].Name, stateStr), true)
}
ss.monitorsLock.RUnlock()
}
ss.serviceResponseDataStoreLock.Unlock()
// SSL 证书报警
var errMsg string
if strings.HasPrefix(r.GetData(), "SSL证书错误：") {
if strings.HasPrefix(mh.Data, "SSL证书错误：") {
// 排除 i/o timeont、connection timeout、EOF 错误
if !strings.HasSuffix(r.GetData(), "timeout") &&
!strings.HasSuffix(r.GetData(), "EOF") &&
!strings.HasSuffix(r.GetData(), "timed out") {
errMsg = r.GetData()
if !strings.HasSuffix(mh.Data, "timeout") &&
!strings.HasSuffix(mh.Data, "EOF") &&
!strings.HasSuffix(mh.Data, "timed out") {
errMsg = mh.Data
}
} else {
var last model.MonitorHistory
var newCert = strings.Split(r.GetData(), "|")
var newCert = strings.Split(mh.Data, "|")
if len(newCert) > 1 {
if ss.sslCertCache[mh.MonitorID] == "" {
ss.sslCertCache[mh.MonitorID] = mh.Data
}
expiresNew, _ := time.Parse("2006-01-02 15:04:05 -0700 MST", newCert[1])
// 证书过期提醒
if expiresNew.Before(time.Now().AddDate(0, 0, 7)) {
@ -271,23 +298,23 @@ func (ss *ServiceSentinel) worker() {
expiresNew.Format("2006-01-02 15:04:05"))
}
// 证书变更提醒
if err := DB.Where("monitor_id = ? AND data LIKE ?", r.GetId(), "%|%").Order("id DESC").First(&last).Error; err == nil {
var oldCert = strings.Split(last.Data, "|")
var expiresOld time.Time
if len(oldCert) > 1 {
expiresOld, _ = time.Parse("2006-01-02 15:04:05 -0700 MST", oldCert[1])
}
if last.Data != "" && oldCert[0] != newCert[0] && !expiresNew.Equal(expiresOld) {
errMsg = fmt.Sprintf(
"SSL证书变更，旧：%s, %s 过期；新：%s, %s 过期。",
oldCert[0], expiresOld.Format("2006-01-02 15:04:05"), newCert[0], expiresNew.Format("2006-01-02 15:04:05"))
}
var oldCert = strings.Split(ss.sslCertCache[mh.MonitorID], "|")
var expiresOld time.Time
if len(oldCert) > 1 {
expiresOld, _ = time.Parse("2006-01-02 15:04:05 -0700 MST", oldCert[1])
}
if oldCert[0] != newCert[0] && !expiresNew.Equal(expiresOld) {
errMsg = fmt.Sprintf(
"SSL证书变更，旧：%s, %s 过期；新：%s, %s 过期。",
oldCert[0], expiresOld.Format("2006-01-02 15:04:05"), newCert[0], expiresNew.Format("2006-01-02 15:04:05"))
}
}
}
if errMsg != "" {
ss.monitorsLock.RLock()
SendNotification(fmt.Sprintf("服务监控：%s %s", ss.monitors[mh.MonitorID].Name, errMsg), true)
if ss.monitors[mh.MonitorID].Notify {
go SendNotification(fmt.Sprintf("服务监控：%s %s", ss.monitors[mh.MonitorID].Name, errMsg), true)
}
ss.monitorsLock.RUnlock()
}
}

5
service/rpc/nezha.go
View File

@ -21,7 +21,10 @@ func (s *NezhaHandler) ReportTask(c context.Context, r *pb.TaskResult) (*pb.Rece
return nil, err
}
if r.GetType() != model.TaskTypeCommand {
dao.ServiceSentinelShared.Dispatch(r)
dao.ServiceSentinelShared.Dispatch(dao.ReportData{
Data: r,
Reporter: clientID,
})
} else {
// 处理上报的计划任务
dao.CronLock.RLock()