489 lines
12 KiB
Go
489 lines
12 KiB
Go
package services
|
||
|
||
import (
|
||
"log"
|
||
"sync"
|
||
"time"
|
||
|
||
"mengyaping-backend/config"
|
||
"mengyaping-backend/models"
|
||
"mengyaping-backend/storage"
|
||
"mengyaping-backend/utils"
|
||
)
|
||
|
||
// MonitorService 监控服务
|
||
type MonitorService struct {
|
||
httpClient *utils.HTTPClient
|
||
storage *storage.Storage
|
||
stopCh chan struct{}
|
||
rescheduleCh chan struct{}
|
||
running bool
|
||
mu sync.Mutex
|
||
}
|
||
|
||
var (
|
||
monitorService *MonitorService
|
||
monitorOnce sync.Once
|
||
)
|
||
|
||
// GetMonitorService 获取监控服务单例
|
||
func GetMonitorService() *MonitorService {
|
||
monitorOnce.Do(func() {
|
||
st := storage.GetStorage()
|
||
cfg := config.GetConfig()
|
||
monitorService = &MonitorService{
|
||
httpClient: utils.NewHTTPClient(cfg.Monitor.Timeout),
|
||
storage: st,
|
||
stopCh: make(chan struct{}),
|
||
rescheduleCh: make(chan struct{}, 1),
|
||
}
|
||
})
|
||
return monitorService
|
||
}
|
||
|
||
// Start 启动监控服务
|
||
func (s *MonitorService) Start() {
|
||
s.mu.Lock()
|
||
if s.running {
|
||
s.mu.Unlock()
|
||
return
|
||
}
|
||
s.running = true
|
||
s.mu.Unlock()
|
||
|
||
log.Println("监控服务已启动")
|
||
go s.runLoop()
|
||
}
|
||
|
||
// Reschedule 在检测间隔配置变更后唤醒主循环以使用新周期（非阻塞）
|
||
func (s *MonitorService) Reschedule() {
|
||
select {
|
||
case s.rescheduleCh <- struct{}{}:
|
||
default:
|
||
}
|
||
}
|
||
|
||
func (s *MonitorService) runLoop() {
|
||
go s.checkAll()
|
||
|
||
d := config.GetConfig().Monitor.Interval
|
||
if d <= 0 {
|
||
d = 30 * time.Minute
|
||
}
|
||
timer := time.NewTimer(d)
|
||
defer timer.Stop()
|
||
|
||
for {
|
||
select {
|
||
case <-s.stopCh:
|
||
if !timer.Stop() {
|
||
select {
|
||
case <-timer.C:
|
||
default:
|
||
}
|
||
}
|
||
log.Println("监控服务已停止")
|
||
return
|
||
case <-s.rescheduleCh:
|
||
if !timer.Stop() {
|
||
select {
|
||
case <-timer.C:
|
||
default:
|
||
}
|
||
}
|
||
d = config.GetConfig().Monitor.Interval
|
||
if d <= 0 {
|
||
d = 30 * time.Minute
|
||
}
|
||
timer.Reset(d)
|
||
case <-timer.C:
|
||
go s.checkAll()
|
||
d = config.GetConfig().Monitor.Interval
|
||
if d <= 0 {
|
||
d = 30 * time.Minute
|
||
}
|
||
timer.Reset(d)
|
||
}
|
||
}
|
||
}
|
||
|
||
// Stop 停止监控服务
|
||
func (s *MonitorService) Stop() {
|
||
s.mu.Lock()
|
||
defer s.mu.Unlock()
|
||
|
||
if s.running {
|
||
close(s.stopCh)
|
||
s.running = false
|
||
}
|
||
}
|
||
|
||
// checkAll 检查所有网站（错峰执行，避免并发暴涨）
|
||
func (s *MonitorService) checkAll() {
|
||
websites := s.storage.GetWebsites()
|
||
semaphore := make(chan struct{}, 3) // 最多 3 个并发检测
|
||
|
||
for i, website := range websites {
|
||
// 每个网站之间间隔 1 秒，把检测分散开
|
||
if i > 0 {
|
||
time.Sleep(1 * time.Second)
|
||
}
|
||
|
||
var wg sync.WaitGroup
|
||
for _, urlInfo := range website.URLs {
|
||
wg.Add(1)
|
||
go func(w models.Website, u models.URLInfo) {
|
||
defer wg.Done()
|
||
semaphore <- struct{}{}
|
||
defer func() { <-semaphore }()
|
||
s.checkURL(w, u)
|
||
}(website, urlInfo)
|
||
}
|
||
wg.Wait()
|
||
}
|
||
|
||
// 检测完毕后，逐个解析 DNS
|
||
s.resolveAllWebsiteIPs(websites)
|
||
|
||
// 保存记录
|
||
s.storage.SaveAll()
|
||
log.Printf("本轮检测完成，共 %d 个网站", len(websites))
|
||
}
|
||
|
||
// resolveAllWebsiteIPs 逐个解析所有网站域名 IP（每次都刷新）
|
||
func (s *MonitorService) resolveAllWebsiteIPs(websites []models.Website) {
|
||
for i, website := range websites {
|
||
if len(website.URLs) == 0 {
|
||
continue
|
||
}
|
||
if i > 0 {
|
||
time.Sleep(500 * time.Millisecond)
|
||
}
|
||
s.resolveWebsiteIP(website)
|
||
}
|
||
}
|
||
|
||
// resolveWebsiteIP 解析单个网站的域名 IP
|
||
func (s *MonitorService) resolveWebsiteIP(website models.Website) {
|
||
if len(website.URLs) == 0 {
|
||
return
|
||
}
|
||
|
||
if utils.IsIntranetURL(website.URLs[0].URL) {
|
||
w := s.storage.GetWebsite(website.ID)
|
||
if w != nil && len(w.IPAddresses) > 0 {
|
||
w.IPAddresses = nil
|
||
w.UpdatedAt = time.Now()
|
||
s.storage.UpdateWebsite(*w)
|
||
}
|
||
return
|
||
}
|
||
|
||
ips, err := utils.ResolveDomainIPs(website.URLs[0].URL)
|
||
if err != nil {
|
||
log.Printf("DNS解析失败 [%s]: %v", website.Name, err)
|
||
return
|
||
}
|
||
|
||
if len(ips) == 0 {
|
||
return
|
||
}
|
||
|
||
w := s.storage.GetWebsite(website.ID)
|
||
if w == nil {
|
||
return
|
||
}
|
||
|
||
w.IPAddresses = ips
|
||
w.UpdatedAt = time.Now()
|
||
s.storage.UpdateWebsite(*w)
|
||
log.Printf("DNS解析 [%s] → %v", website.Name, ips)
|
||
}
|
||
|
||
// checkURL 检查单个URL（带重试）
|
||
func (s *MonitorService) checkURL(website models.Website, urlInfo models.URLInfo) {
|
||
cfg := config.GetConfig()
|
||
maxRetries := cfg.Monitor.RetryCount
|
||
|
||
var result utils.CheckResult
|
||
|
||
for attempt := 0; attempt <= maxRetries; attempt++ {
|
||
if attempt > 0 {
|
||
time.Sleep(time.Duration(attempt) * 2 * time.Second)
|
||
log.Printf("重试 [%s] %s - 第 %d 次重试", website.Name, urlInfo.URL, attempt)
|
||
}
|
||
|
||
result = s.httpClient.CheckWebsiteStatus(urlInfo.URL)
|
||
if result.Error == nil && utils.IsSuccessStatus(result.StatusCode) {
|
||
break
|
||
}
|
||
}
|
||
|
||
record := models.MonitorRecord{
|
||
WebsiteID: website.ID,
|
||
URLID: urlInfo.ID,
|
||
URL: urlInfo.URL,
|
||
StatusCode: result.StatusCode,
|
||
Latency: result.Latency.Milliseconds(),
|
||
IsUp: result.Error == nil && utils.IsSuccessStatus(result.StatusCode),
|
||
CheckedAt: time.Now(),
|
||
}
|
||
|
||
if result.Error != nil {
|
||
record.Error = result.Error.Error()
|
||
}
|
||
|
||
s.storage.AddRecord(record)
|
||
|
||
// 仅当网站无标题时才做完整检测来获取元数据
|
||
if record.IsUp && website.Title == "" {
|
||
fullResult := s.httpClient.CheckWebsite(urlInfo.URL)
|
||
if fullResult.Title != "" {
|
||
w := s.storage.GetWebsite(website.ID)
|
||
if w != nil {
|
||
w.Title = fullResult.Title
|
||
w.UpdatedAt = time.Now()
|
||
s.storage.UpdateWebsite(*w)
|
||
}
|
||
}
|
||
}
|
||
|
||
log.Printf("检测 [%s] %s - 状态码: %d, 延迟: %dms, 可用: %v",
|
||
website.Name, urlInfo.URL, result.StatusCode, result.Latency.Milliseconds(), record.IsUp)
|
||
}
|
||
|
||
// CheckWebsiteNow 立即检查指定网站（状态 + DNS，等待完成后保存）
|
||
func (s *MonitorService) CheckWebsiteNow(websiteID string) {
|
||
website := s.storage.GetWebsite(websiteID)
|
||
if website == nil {
|
||
return
|
||
}
|
||
|
||
// 逐个检测该网站的所有 URL
|
||
for _, urlInfo := range website.URLs {
|
||
s.checkURL(*website, urlInfo)
|
||
}
|
||
|
||
// 刷新 DNS
|
||
s.resolveWebsiteIP(*website)
|
||
|
||
s.storage.SaveAll()
|
||
}
|
||
|
||
// GetWebsiteStatus 获取网站状态
|
||
func (s *MonitorService) GetWebsiteStatus(websiteID string) *models.WebsiteStatus {
|
||
website := s.storage.GetWebsite(websiteID)
|
||
if website == nil {
|
||
return nil
|
||
}
|
||
|
||
isIntranet := len(website.URLs) > 0 && utils.IsIntranetURL(website.URLs[0].URL)
|
||
status := &models.WebsiteStatus{
|
||
Website: *website,
|
||
IsIntranet: isIntranet,
|
||
URLStatuses: []models.URLStatus{},
|
||
}
|
||
|
||
now := time.Now()
|
||
since24h := now.Add(-24 * time.Hour)
|
||
since7d := now.Add(-7 * 24 * time.Hour)
|
||
since30d := now.Add(-30 * 24 * time.Hour)
|
||
|
||
var totalUptime24h, totalUptime7d float64
|
||
var urlCount int
|
||
|
||
for _, urlInfo := range website.URLs {
|
||
urlStatus := s.getURLStatus(website.ID, urlInfo, since24h, since7d)
|
||
status.URLStatuses = append(status.URLStatuses, urlStatus)
|
||
|
||
totalUptime24h += urlStatus.Uptime24h
|
||
totalUptime7d += urlStatus.Uptime7d
|
||
urlCount++
|
||
}
|
||
|
||
if urlCount > 0 {
|
||
status.Uptime24h = totalUptime24h / float64(urlCount)
|
||
status.Uptime7d = totalUptime7d / float64(urlCount)
|
||
}
|
||
|
||
status.DailyHistory = s.storage.GetWebsiteDailyAggregates(website.ID, since30d)
|
||
if pt, ut := s.storage.GetWebsiteDayProbeTotals(website.ID, since30d); pt > 0 {
|
||
status.Uptime30d = float64(ut) / float64(pt) * 100
|
||
}
|
||
|
||
// 获取最后检测时间
|
||
for _, urlStatus := range status.URLStatuses {
|
||
if urlStatus.CurrentState.CheckedAt.After(status.LastChecked) {
|
||
status.LastChecked = urlStatus.CurrentState.CheckedAt
|
||
}
|
||
}
|
||
|
||
return status
|
||
}
|
||
|
||
// getURLStatus 获取URL状态：每个 URL 仅 1 次读小时表 + 1 次读 latest（在内存拆 24h/7d）
|
||
func (s *MonitorService) getURLStatus(websiteID string, urlInfo models.URLInfo, since24h, since7d time.Time) models.URLStatus {
|
||
var latest *models.MonitorRecord
|
||
if rec := s.storage.GetLatestRecord(websiteID, urlInfo.ID); rec != nil {
|
||
latest = rec
|
||
}
|
||
hRows := s.storage.GetProbeHourRowsForURL(websiteID, urlInfo.ID, since7d)
|
||
return s.buildURLStatusFromHours(websiteID, urlInfo, since24h, since7d, latest, hRows)
|
||
}
|
||
|
||
func hourlyStatFromProbeRow(h storage.MonitorProbeHour) models.HourlyStats {
|
||
st := models.HourlyStats{
|
||
Hour: h.HourAt,
|
||
TotalCount: h.ProbeCount,
|
||
UpCount: h.UpCount,
|
||
}
|
||
if h.ProbeCount > 0 {
|
||
st.AvgLatency = h.LatencySum / int64(h.ProbeCount)
|
||
st.Uptime = float64(h.UpCount) / float64(h.ProbeCount) * 100
|
||
}
|
||
return st
|
||
}
|
||
|
||
func hourWindowsFromRows(rows []storage.MonitorProbeHour, since24h, since7d time.Time) (
|
||
stats24h []models.HourlyStats,
|
||
stats7d []models.HourlyStats,
|
||
t24, u24, l24, t7, u7 int64,
|
||
) {
|
||
for _, h := range rows {
|
||
if !h.HourAt.Before(since7d) {
|
||
t7 += int64(h.ProbeCount)
|
||
u7 += int64(h.UpCount)
|
||
stats7d = append(stats7d, hourlyStatFromProbeRow(h))
|
||
}
|
||
if !h.HourAt.Before(since24h) {
|
||
t24 += int64(h.ProbeCount)
|
||
u24 += int64(h.UpCount)
|
||
l24 += h.LatencySum
|
||
stats24h = append(stats24h, hourlyStatFromProbeRow(h))
|
||
}
|
||
}
|
||
return stats24h, stats7d, t24, u24, l24, t7, u7
|
||
}
|
||
|
||
func (s *MonitorService) buildURLStatusFromHours(
|
||
websiteID string,
|
||
urlInfo models.URLInfo,
|
||
since24h, since7d time.Time,
|
||
latest *models.MonitorRecord,
|
||
hRows []storage.MonitorProbeHour,
|
||
) models.URLStatus {
|
||
urlStatus := models.URLStatus{URLInfo: urlInfo}
|
||
if latest != nil {
|
||
urlStatus.CurrentState = *latest
|
||
}
|
||
s24, s7, t24, u24, l24, t7, u7 := hourWindowsFromRows(hRows, since24h, since7d)
|
||
urlStatus.History24h = hourlyStatsToMonitorRecords(websiteID, urlInfo.ID, s24)
|
||
if t24 > 0 {
|
||
urlStatus.Uptime24h = float64(u24) / float64(t24) * 100
|
||
urlStatus.AvgLatency = l24 / t24
|
||
}
|
||
urlStatus.History7d = s7
|
||
if t7 > 0 {
|
||
urlStatus.Uptime7d = float64(u7) / float64(t7) * 100
|
||
}
|
||
return urlStatus
|
||
}
|
||
|
||
func hourlyStatsToMonitorRecords(websiteID, urlID string, stats []models.HourlyStats) []models.MonitorRecord {
|
||
out := make([]models.MonitorRecord, 0, len(stats))
|
||
for _, h := range stats {
|
||
isUp := h.TotalCount > 0 && h.UpCount*2 >= h.TotalCount
|
||
out = append(out, models.MonitorRecord{
|
||
WebsiteID: websiteID,
|
||
URLID: urlID,
|
||
Latency: h.AvgLatency,
|
||
IsUp: isUp,
|
||
CheckedAt: h.Hour,
|
||
})
|
||
}
|
||
return out
|
||
}
|
||
|
||
// GetAllWebsiteStatuses 获取所有网站状态（批量查 probe 表，避免每个站点/URL 多次 SQL）
|
||
func (s *MonitorService) GetAllWebsiteStatuses() []models.WebsiteStatus {
|
||
websites := s.storage.GetWebsites()
|
||
if len(websites) == 0 {
|
||
return nil
|
||
}
|
||
|
||
now := time.Now()
|
||
since24h := now.Add(-24 * time.Hour)
|
||
since7d := now.Add(-7 * 24 * time.Hour)
|
||
since30d := now.Add(-30 * 24 * time.Hour)
|
||
since30dDay := time.Date(since30d.Year(), since30d.Month(), since30d.Day(), 0, 0, 0, 0, since30d.Location())
|
||
|
||
latestMap := s.storage.GetProbeLatestKeyMap()
|
||
hourMap := s.storage.GroupProbeHoursSince(since7d)
|
||
dailyBySite, uptime30dBySite := s.storage.LoadAllWebsiteDayRollups(since30dDay)
|
||
|
||
statuses := make([]models.WebsiteStatus, 0, len(websites))
|
||
for _, w := range websites {
|
||
daily := dailyBySite[w.ID]
|
||
u30 := uptime30dBySite[w.ID]
|
||
st := s.buildWebsiteStatusBulk(w, since24h, since7d, latestMap, hourMap, daily, u30)
|
||
if st != nil {
|
||
statuses = append(statuses, *st)
|
||
}
|
||
}
|
||
return statuses
|
||
}
|
||
|
||
func (s *MonitorService) buildWebsiteStatusBulk(
|
||
w models.Website,
|
||
since24h, since7d time.Time,
|
||
latestMap map[string]models.MonitorRecord,
|
||
hourMap map[string]map[string][]storage.MonitorProbeHour,
|
||
daily []models.DailyStats,
|
||
uptime30d float64,
|
||
) *models.WebsiteStatus {
|
||
isIntranet := len(w.URLs) > 0 && utils.IsIntranetURL(w.URLs[0].URL)
|
||
status := &models.WebsiteStatus{
|
||
Website: w,
|
||
IsIntranet: isIntranet,
|
||
URLStatuses: []models.URLStatus{},
|
||
}
|
||
|
||
siteHours := hourMap[w.ID]
|
||
if siteHours == nil {
|
||
siteHours = map[string][]storage.MonitorProbeHour{}
|
||
}
|
||
|
||
var totalUptime24h, totalUptime7d float64
|
||
urlCount := 0
|
||
for _, urlInfo := range w.URLs {
|
||
key := w.ID + "\x00" + urlInfo.ID
|
||
var latest *models.MonitorRecord
|
||
if rec, ok := latestMap[key]; ok {
|
||
r := rec
|
||
latest = &r
|
||
}
|
||
hRows := siteHours[urlInfo.ID]
|
||
urlSt := s.buildURLStatusFromHours(w.ID, urlInfo, since24h, since7d, latest, hRows)
|
||
status.URLStatuses = append(status.URLStatuses, urlSt)
|
||
totalUptime24h += urlSt.Uptime24h
|
||
totalUptime7d += urlSt.Uptime7d
|
||
urlCount++
|
||
}
|
||
if urlCount > 0 {
|
||
status.Uptime24h = totalUptime24h / float64(urlCount)
|
||
status.Uptime7d = totalUptime7d / float64(urlCount)
|
||
}
|
||
if daily != nil {
|
||
status.DailyHistory = daily
|
||
} else {
|
||
status.DailyHistory = []models.DailyStats{}
|
||
}
|
||
status.Uptime30d = uptime30d
|
||
for _, urlSt := range status.URLStatuses {
|
||
if urlSt.CurrentState.CheckedAt.After(status.LastChecked) {
|
||
status.LastChecked = urlSt.CurrentState.CheckedAt
|
||
}
|
||
}
|
||
return status
|
||
}
|