feat(v2.0.0): tsdb (#1162)

* feat: tsdb * fix(ci): remove --parseGoList=false from swag init to fix dependency resolution * fix(ci): fix swag init directory and temporary remove s390x support due to cgo issues * fix(ci): fix swag init output directory to cmd/dashboard/docs * fix(ci): set GOTOOLCHAIN=auto for gosec * feat: add system storage maintenance for SQLite and TSDB * shit * feat: add s390x support and improve service monitoring * ci: upgrade goreleaser-cross image to v1.25 * ci: add libzstd-dev:s390x for cross-compilation * ci: build libzstd for s390x from source * ci: add libzstd_linux_s390x.go for gozstd linking * ci: use vendor mode for s390x gozstd build * ci: clone zstd source for s390x build * refactor(tsdb): rename MaxDiskUsageGB to MinFreeDiskSpaceGB and optimize queries - Rename config to accurately reflect VictoriaMetrics behavior: minimum free disk space threshold - Add QueryServiceHistoryByServerID for batch query optimization - Fix hasStatus to avoid false status counting when only delay data exists - Fix service aggregation boundary: use successCount*2 >= count - Fix serviceID parsing with strconv.ParseUint error handling - Add TagFiltersCacheSize for better query performance * feat(api): add server metrics endpoint and simplify service history response - Add /server/:id/metrics API for querying TSDB server metrics - Simplify getServiceHistory by removing redundant data conversion - Change AvgDelay type from float32 to float64 - Remove generated swagger docs (to be regenerated) - Update TSDB query, writer and tests * chore: 临时禁用不支持前端 * ci: cache zstd build for s390x to speed up CI * fix(tsdb): fix race conditions, data correctness and optimize performance - Fix TOCTOU race between IsClosed() and write/query by holding RLock - Fix delay=0 excluded from stats by using hasDelay flag instead of value > 0 - Fix fmt.Sscanf -> strconv.ParseUint for server_id parsing with error logging - Fix buffer unbounded growth by flushing inside lock when over maxSize - Split makeMetricRow into makeServerMetricRow/makeServiceMetricRow - Extract InitGlobalSettings() from Open() for VictoriaMetrics globals - Remove redundant instance/GetInstance/SetInstance singleton - Add error logging for silently skipped block decode errors - Optimize WriteBatch* to build all rows in single write call - Optimize downsample to use linear scan instead of map for sorted data - Optimize query slice reuse across block iterations * 服务添加DisplayIndex (#1166) * 服务添加DisplayIndex * 根据ai建议修改 --------- Co-authored-by: huYang <306061454@qq.com> * fix(tsdb): restore SQLite fallback and monthly status reload on restart - Restore ServiceHistory model and SQLite write fallback when TSDB is disabled - Reload monthlyStatus (30-day) and serviceStatusToday from TSDB/SQLite on startup - Add SQLite fallback query for /service/:id/history and /server/:id/service - Remove breaking GET /service/:id endpoint, keep /service/:id/history only - Add QueryServiceDailyStats to TSDB for per-day aggregation - Add tests for monthly status and today stats loading from both TSDB and SQLite - Migrate ServiceHistory table only when TSDB is disabled * ci: exclude false-positive gosec rules G117, G703, G704 * feat(api): expose tsdb_enabled in setting response * ci: restore G115 exclusion accidentally dropped in previous commit * fix: update version numbers for OfficialAdmin and Official templates * chore: upgrade frontend * chore: upgrade frontend --------- Co-authored-by: 胡说丷刂 <34758853+laosan-xx@users.noreply.github.com> Co-authored-by: huYang <306061454@qq.com>
2026-05-06 13:48:52 +00:00 · 2026-02-15 13:13:33 +08:00
parent 4c4758207d
commit e61772e858
28 changed files with 3054 additions and 221 deletions
@@ -12,6 +12,7 @@ import (
 	"github.com/jinzhu/copier"
 	geoipx "github.com/nezhahq/nezha/pkg/geoip"
 	"github.com/nezhahq/nezha/pkg/grpcx"
+	"github.com/nezhahq/nezha/pkg/tsdb"

 	"github.com/nezhahq/nezha/model"
 	pb "github.com/nezhahq/nezha/proto"
@@ -114,6 +115,44 @@ func (s *NezhaHandler) ReportSystemState(stream pb.NezhaService_ReportSystemStat
 		server.LastActive = time.Now()
 		server.State = &innerState

+		if singleton.TSDBEnabled() {
+			maxTemp := 0.0
+			for _, t := range innerState.Temperatures {
+				if t.Temperature > maxTemp {
+					maxTemp = t.Temperature
+				}
+			}
+			maxGPU := 0.0
+			for _, g := range innerState.GPU {
+				if g > maxGPU {
+					maxGPU = g
+				}
+			}
+			if err := singleton.TSDBShared.WriteServerMetrics(&tsdb.ServerMetrics{
+				ServerID:       clientID,
+				Timestamp:      time.Now(),
+				CPU:            innerState.CPU,
+				MemUsed:        innerState.MemUsed,
+				SwapUsed:       innerState.SwapUsed,
+				DiskUsed:       innerState.DiskUsed,
+				NetInSpeed:     innerState.NetInSpeed,
+				NetOutSpeed:    innerState.NetOutSpeed,
+				NetInTransfer:  innerState.NetInTransfer,
+				NetOutTransfer: innerState.NetOutTransfer,
+				Load1:          innerState.Load1,
+				Load5:          innerState.Load5,
+				Load15:         innerState.Load15,
+				TCPConnCount:   innerState.TcpConnCount,
+				UDPConnCount:   innerState.UdpConnCount,
+				ProcessCount:   innerState.ProcessCount,
+				Temperature:    maxTemp,
+				Uptime:         innerState.Uptime,
+				GPU:            maxGPU,
+			}); err != nil {
+				log.Printf("NEZHA>> Failed to write server metrics to TSDB: %v", err)
+			}
+		}
+
 		// 应对 dashboard / agent 重启的情况，如果从未记录过，先打点，等到小时时间点时入库
 		if server.PrevTransferInSnapshot == 0 || server.PrevTransferOutSnapshot == 0 {
 			server.PrevTransferInSnapshot = state.NetInTransfer
@@ -2,14 +2,14 @@
  name: "OfficialAdmin"
  repository: "https://github.com/nezhahq/admin-frontend"
  author: "nezhahq"
-  version: "v1.14.7"
+  version: "v2.0.3"
  is_admin: true
  is_official: true
 - path: "user-dist"
  name: "Official"
  repository: "https://github.com/hamster1963/nezha-dash-v1"
  author: "hamster1963"
-  version: "v1.33.0"
+  version: "v2.0.0"
  is_official: true
 - path: "nazhua-dist"
  name: "Nazhua"
@@ -16,6 +16,7 @@ import (
 	"golang.org/x/exp/constraints"

 	"github.com/nezhahq/nezha/model"
+	"github.com/nezhahq/nezha/pkg/tsdb"
 	"github.com/nezhahq/nezha/pkg/utils"
 	pb "github.com/nezhahq/nezha/proto"
 )
@@ -39,7 +40,7 @@ type ReportData struct {
 type _TodayStatsOfService struct {
 	Up    uint64  // 今日在线计数
 	Down  uint64  // 今日离线计数
-	Delay float32 // 今日平均延迟
+	Delay float64 // 今日平均延迟
 }

 type serviceResponseData = _TodayStatsOfService
@@ -51,8 +52,9 @@ type serviceTaskStatus struct {
 }

 type pingStore struct {
-	count int
-	ping  float32
+	count        int
+	ping         float64
+	successCount int
 }

 /*
@@ -108,23 +110,7 @@ func NewServiceSentinel(serviceSentinelDispatchBus chan<- *model.Service) (*Serv

 	year, month, day := time.Now().Date()
 	today := time.Date(year, month, day, 0, 0, 0, 0, Loc)
-
-	var mhs []model.ServiceHistory
-	// 加载当日记录
-	DB.Where("created_at >= ? AND server_id = 0", today).Find(&mhs)
-	totalDelay := make(map[uint64]float32)
-	totalDelayCount := make(map[uint64]float32)
-	for _, mh := range mhs {
-		totalDelay[mh.ServiceID] += mh.AvgDelay
-		totalDelayCount[mh.ServiceID]++
-		ss.serviceStatusToday[mh.ServiceID].Up += mh.Up
-		ss.monthlyStatus[mh.ServiceID].TotalUp += mh.Up
-		ss.serviceStatusToday[mh.ServiceID].Down += mh.Down
-		ss.monthlyStatus[mh.ServiceID].TotalDown += mh.Down
-	}
-	for id, delay := range totalDelay {
-		ss.serviceStatusToday[id].Delay = delay / float32(totalDelayCount[id])
-	}
+	ss.loadTodayStats(today)

 	// 启动服务监控器
 	go ss.worker()
@@ -135,6 +121,12 @@ func NewServiceSentinel(serviceSentinelDispatchBus chan<- *model.Service) (*Serv
 		return nil, err
 	}

+	// 每周日凌晨 4:00 执行系统存储维护
+	_, err = CronShared.AddFunc("0 0 4 * * 0", PerformMaintenance)
+	if err != nil {
+		log.Printf("NEZHA>> Warning: failed to schedule maintenance task: %v", err)
+	}
+
 	return ss, nil
 }

@@ -171,6 +163,16 @@ func (ss *ServiceSentinel) Dispatch(r ReportData) {
 	ss.serviceReportChannel <- r
 }

+// sortServices 按 DisplayIndex 降序、ID 升序排列服务列表
+func sortServices(services []*model.Service) {
+	slices.SortFunc(services, func(a, b *model.Service) int {
+		if a.DisplayIndex != b.DisplayIndex {
+			return cmp.Compare(b.DisplayIndex, a.DisplayIndex)
+		}
+		return cmp.Compare(a.ID, b.ID)
+	})
+}
+
 func (ss *ServiceSentinel) UpdateServiceList() {
 	ss.servicesLock.RLock()
 	defer ss.servicesLock.RUnlock()
@@ -179,9 +181,7 @@ func (ss *ServiceSentinel) UpdateServiceList() {
 	defer ss.serviceListLock.Unlock()

 	ss.serviceList = utils.MapValuesToSlice(ss.services)
-	slices.SortFunc(ss.serviceList, func(a, b *model.Service) int {
-		return cmp.Compare(a.ID, b.ID)
-	})
+	sortServices(ss.serviceList)
 }

 // loadServiceHistory 加载服务监控器的历史状态信息
@@ -207,6 +207,7 @@ func (ss *ServiceSentinel) loadServiceHistory() error {
 		ss.serviceStatusToday[service.ID] = &_TodayStatsOfService{}
 	}
 	ss.serviceList = services
+	sortServices(ss.serviceList)

 	year, month, day := time.Now().Date()
 	today := time.Date(year, month, day, 0, 0, 0, 0, Loc)
@@ -215,33 +216,111 @@ func (ss *ServiceSentinel) loadServiceHistory() error {
 		ss.monthlyStatus[service.ID] = &serviceResponseItem{
 			service: service,
 			ServiceResponseItem: model.ServiceResponseItem{
-				Delay: &[30]float32{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
-				Up:    &[30]uint64{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
-				Down:  &[30]uint64{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+				Delay: &[30]float64{},
+				Up:    &[30]uint64{},
+				Down:  &[30]uint64{},
 			},
 		}
 	}

-	// 加载服务监控历史记录
-	var mhs []model.ServiceHistory
-	DB.Where("created_at > ? AND created_at < ? AND server_id = 0", today.AddDate(0, 0, -29), today).Find(&mhs)
-	var delayCount = make(map[int]int)
-	for _, mh := range mhs {
-		dayIndex := 28 - (int(today.Sub(mh.CreatedAt).Hours()) / 24)
-		if dayIndex < 0 {
-			continue
-		}
-		ss.monthlyStatus[mh.ServiceID].Delay[dayIndex] = (ss.monthlyStatus[mh.ServiceID].Delay[dayIndex]*float32(delayCount[dayIndex]) + mh.AvgDelay) / float32(delayCount[dayIndex]+1)
-		delayCount[dayIndex]++
-		ss.monthlyStatus[mh.ServiceID].Up[dayIndex] += mh.Up
-		ss.monthlyStatus[mh.ServiceID].TotalUp += mh.Up
-		ss.monthlyStatus[mh.ServiceID].Down[dayIndex] += mh.Down
-		ss.monthlyStatus[mh.ServiceID].TotalDown += mh.Down
+	if TSDBEnabled() {
+		ss.loadMonthlyStatusFromTSDB(services, today)
+	} else {
+		ss.loadMonthlyStatusFromDB(today)
 	}

 	return nil
 }

+func (ss *ServiceSentinel) loadMonthlyStatusFromTSDB(services []*model.Service, today time.Time) {
+	for _, service := range services {
+		dailyStats, err := TSDBShared.QueryServiceDailyStats(service.ID, today, 30)
+		if err != nil {
+			log.Printf("NEZHA>> Failed to load TSDB history for service %d: %v", service.ID, err)
+			continue
+		}
+		ms := ss.monthlyStatus[service.ID]
+		for i := 0; i < 29; i++ {
+			ms.Up[i] = dailyStats[i].Up
+			ms.TotalUp += dailyStats[i].Up
+			ms.Down[i] = dailyStats[i].Down
+			ms.TotalDown += dailyStats[i].Down
+			ms.Delay[i] = dailyStats[i].Delay
+		}
+	}
+}
+
+func (ss *ServiceSentinel) loadMonthlyStatusFromDB(today time.Time) {
+	var mhs []model.ServiceHistory
+	DB.Where("created_at > ? AND created_at < ? AND server_id = 0", today.AddDate(0, 0, -29), today).Find(&mhs)
+	delayCount := make(map[uint64]map[int]int)
+	for _, mh := range mhs {
+		dayIndex := 28 - int(today.Sub(mh.CreatedAt).Hours())/24
+		if dayIndex < 0 {
+			continue
+		}
+		ms := ss.monthlyStatus[mh.ServiceID]
+		if ms == nil {
+			continue
+		}
+		if delayCount[mh.ServiceID] == nil {
+			delayCount[mh.ServiceID] = make(map[int]int)
+		}
+		ms.Delay[dayIndex] = (ms.Delay[dayIndex]*float64(delayCount[mh.ServiceID][dayIndex]) + mh.AvgDelay) / float64(delayCount[mh.ServiceID][dayIndex]+1)
+		delayCount[mh.ServiceID][dayIndex]++
+		ms.Up[dayIndex] += mh.Up
+		ms.TotalUp += mh.Up
+		ms.Down[dayIndex] += mh.Down
+		ms.TotalDown += mh.Down
+	}
+}
+
+func (ss *ServiceSentinel) loadTodayStats(today time.Time) {
+	if TSDBEnabled() {
+		for serviceID, ms := range ss.monthlyStatus {
+			result, err := TSDBShared.QueryServiceHistory(serviceID, tsdb.Period1Day)
+			if err != nil {
+				log.Printf("NEZHA>> Failed to load TSDB today stats for service %d: %v", serviceID, err)
+				continue
+			}
+			var totalUp, totalDown uint64
+			var totalDelay float64
+			var delayCount int
+			for _, serverStats := range result.Servers {
+				totalUp += serverStats.Stats.TotalUp
+				totalDown += serverStats.Stats.TotalDown
+				if serverStats.Stats.AvgDelay > 0 {
+					totalDelay += serverStats.Stats.AvgDelay
+					delayCount++
+				}
+			}
+			ss.serviceStatusToday[serviceID].Up = totalUp
+			ss.serviceStatusToday[serviceID].Down = totalDown
+			if delayCount > 0 {
+				ss.serviceStatusToday[serviceID].Delay = totalDelay / float64(delayCount)
+			}
+			ms.TotalUp += totalUp
+			ms.TotalDown += totalDown
+		}
+	} else {
+		var mhs []model.ServiceHistory
+		DB.Where("created_at >= ? AND server_id = 0", today).Find(&mhs)
+		totalDelay := make(map[uint64]float64)
+		totalDelayCount := make(map[uint64]int)
+		for _, mh := range mhs {
+			ss.serviceStatusToday[mh.ServiceID].Up += mh.Up
+			ss.monthlyStatus[mh.ServiceID].TotalUp += mh.Up
+			ss.serviceStatusToday[mh.ServiceID].Down += mh.Down
+			ss.monthlyStatus[mh.ServiceID].TotalDown += mh.Down
+			totalDelay[mh.ServiceID] += mh.AvgDelay
+			totalDelayCount[mh.ServiceID]++
+		}
+		for id, delay := range totalDelay {
+			ss.serviceStatusToday[id].Delay = delay / float64(totalDelayCount[id])
+		}
+	}
+}
+
 func (ss *ServiceSentinel) Update(m *model.Service) error {
 	ss.serviceResponseDataStoreLock.Lock()
 	defer ss.serviceResponseDataStoreLock.Unlock()
@@ -266,9 +345,9 @@ func (ss *ServiceSentinel) Update(m *model.Service) error {
 		ss.monthlyStatus[m.ID] = &serviceResponseItem{
 			service: m,
 			ServiceResponseItem: model.ServiceResponseItem{
-				Delay: &[30]float32{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
-				Up:    &[30]uint64{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
-				Down:  &[30]uint64{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
+				Delay: &[30]float64{},
+				Up:    &[30]uint64{},
+				Down:  &[30]uint64{},
 			},
 		}
 		if ss.serviceCurrentStatusData[m.ID] == nil {
@@ -406,6 +485,7 @@ func (ss *ServiceSentinel) worker() {

 		mh := r.Data
 		if mh.Type == model.TaskTypeTCPPing || mh.Type == model.TaskTypeICMPPing {
+			// TCP/ICMP Ping 使用平均值计算后再写入
 			serviceTcpMap, ok := ss.serviceResponsePing[mh.GetId()]
 			if !ok {
 				serviceTcpMap = make(map[uint64]*pingStore)
@@ -416,28 +496,56 @@ func (ss *ServiceSentinel) worker() {
 				ts = &pingStore{}
 			}
 			ts.count++
-			ts.ping = (ts.ping*float32(ts.count-1) + mh.Delay) / float32(ts.count)
+			ts.ping = (ts.ping*float64(ts.count-1) + float64(mh.Delay)) / float64(ts.count)
+			if mh.Successful {
+				ts.successCount++
+			}
 			if ts.count == Conf.AvgPingCount {
-				if err := DB.Create(&model.ServiceHistory{
-					ServiceID: mh.GetId(),
-					AvgDelay:  ts.ping,
-					Data:      mh.Data,
-					ServerID:  r.Reporter,
-				}).Error; err != nil {
-					log.Printf("NEZHA>> Failed to save service monitor metrics: %v", err)
+				if TSDBEnabled() {
+					if err := TSDBShared.WriteServiceMetrics(&tsdb.ServiceMetrics{
+						ServiceID:  mh.GetId(),
+						ServerID:   r.Reporter,
+						Timestamp:  time.Now(),
+						Delay:      ts.ping,
+						Successful: ts.successCount*2 >= ts.count,
+					}); err != nil {
+						log.Printf("NEZHA>> Failed to save service monitor metrics to TSDB: %v", err)
+					}
+				} else {
+					if err := DB.Create(&model.ServiceHistory{
+						ServiceID: mh.GetId(),
+						AvgDelay:  ts.ping,
+						Data:      mh.Data,
+						ServerID:  r.Reporter,
+					}).Error; err != nil {
+						log.Printf("NEZHA>> Failed to save service monitor metrics: %v", err)
+					}
 				}
 				ts.count = 0
-				ts.ping = mh.Delay
+				ts.ping = 0
+				ts.successCount = 0
 			}
 			serviceTcpMap[r.Reporter] = ts
+		} else {
+			if TSDBEnabled() {
+				if err := TSDBShared.WriteServiceMetrics(&tsdb.ServiceMetrics{
+					ServiceID:  mh.GetId(),
+					ServerID:   r.Reporter,
+					Timestamp:  time.Now(),
+					Delay:      float64(mh.Delay),
+					Successful: mh.Successful,
+				}); err != nil {
+					log.Printf("NEZHA>> Failed to save service monitor metrics to TSDB: %v", err)
+				}
+			}
 		}

 		ss.serviceResponseDataStoreLock.Lock()
 		// 写入当天状态
 		if mh.Successful {
 			ss.serviceStatusToday[mh.GetId()].Delay = (ss.serviceStatusToday[mh.
-				GetId()].Delay*float32(ss.serviceStatusToday[mh.GetId()].Up) +
-				mh.Delay) / float32(ss.serviceStatusToday[mh.GetId()].Up+1)
+				GetId()].Delay*float64(ss.serviceStatusToday[mh.GetId()].Up) +
+				float64(mh.Delay)) / float64(ss.serviceStatusToday[mh.GetId()].Up+1)
 			ss.serviceStatusToday[mh.GetId()].Up++
 		} else {
 			ss.serviceStatusToday[mh.GetId()].Down++
@@ -463,7 +571,7 @@ func (ss *ServiceSentinel) worker() {
 				rd := ss.serviceResponseDataStore[mh.GetId()]
 				if cs.Successful {
 					rd.Up++
-					rd.Delay = (rd.Delay*float32(rd.Up-1) + cs.Delay) / float32(rd.Up)
+					rd.Delay = (rd.Delay*float64(rd.Up-1) + float64(cs.Delay)) / float64(rd.Up)
 				} else {
 					rd.Down++
 				}
@@ -482,20 +590,20 @@ func (ss *ServiceSentinel) worker() {
 			stateCode = GetStatusCode(upPercent)
 		}

-		// 数据持久化
 		if len(ss.serviceCurrentStatusData[mh.GetId()].result) == _CurrentStatusSize {
 			ss.serviceCurrentStatusData[mh.GetId()].t = currentTime
-			rd := ss.serviceResponseDataStore[mh.GetId()]
-			if err := DB.Create(&model.ServiceHistory{
-				ServiceID: mh.GetId(),
-				AvgDelay:  rd.Delay,
-				Data:      mh.Data,
-				Up:        rd.Up,
-				Down:      rd.Down,
-			}).Error; err != nil {
-				log.Printf("NEZHA>> Failed to save service monitor metrics: %v", err)
+			if !TSDBEnabled() {
+				rd := ss.serviceResponseDataStore[mh.GetId()]
+				if err := DB.Create(&model.ServiceHistory{
+					ServiceID: mh.GetId(),
+					AvgDelay:  rd.Delay,
+					Data:      mh.Data,
+					Up:        rd.Up,
+					Down:      rd.Down,
+				}).Error; err != nil {
+					log.Printf("NEZHA>> Failed to save service monitor metrics: %v", err)
+				}
 			}
-
 			ss.serviceCurrentStatusData[mh.GetId()].result = ss.serviceCurrentStatusData[mh.GetId()].result[:0]
 		}

@@ -0,0 +1,336 @@
+package singleton
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"gorm.io/driver/sqlite"
+	"gorm.io/gorm"
+
+	"github.com/nezhahq/nezha/model"
+	"github.com/nezhahq/nezha/pkg/tsdb"
+)
+
+func newTestSentinel(serviceIDs []uint64) *ServiceSentinel {
+	ss := &ServiceSentinel{
+		serviceStatusToday: make(map[uint64]*_TodayStatsOfService),
+		monthlyStatus:      make(map[uint64]*serviceResponseItem),
+	}
+	for _, id := range serviceIDs {
+		ss.serviceStatusToday[id] = &_TodayStatsOfService{}
+		ss.monthlyStatus[id] = &serviceResponseItem{
+			service: &model.Service{Common: model.Common{ID: id}},
+			ServiceResponseItem: model.ServiceResponseItem{
+				Delay: &[30]float64{},
+				Up:    &[30]uint64{},
+				Down:  &[30]uint64{},
+			},
+		}
+	}
+	return ss
+}
+
+func setupTestDB(t *testing.T) func() {
+	t.Helper()
+	var err error
+	DB, err = gorm.Open(sqlite.Open(":memory:"), &gorm.Config{})
+	require.NoError(t, err)
+	require.NoError(t, DB.AutoMigrate(model.ServiceHistory{}))
+	return func() { DB = nil }
+}
+
+func setupTestTSDB(t *testing.T) (*tsdb.TSDB, func()) {
+	t.Helper()
+	tempDir, err := os.MkdirTemp("", "tsdb_sentinel_test")
+	require.NoError(t, err)
+	config := &tsdb.Config{
+		DataPath:           filepath.Join(tempDir, "tsdb"),
+		RetentionDays:      30,
+		MinFreeDiskSpaceGB: 1,
+		DedupInterval:      time.Second,
+	}
+	db, err := tsdb.Open(config)
+	require.NoError(t, err)
+	TSDBShared = db
+	return db, func() {
+		db.Close()
+		TSDBShared = nil
+		os.RemoveAll(tempDir)
+	}
+}
+
+func TestLoadMonthlyStatusFromDB(t *testing.T) {
+	cleanup := setupTestDB(t)
+	defer cleanup()
+
+	year, month, day := time.Now().Date()
+	today := time.Date(year, month, day, 0, 0, 0, 0, time.UTC)
+
+	serviceID := uint64(1)
+	ss := newTestSentinel([]uint64{serviceID})
+
+	DB.Create(&model.ServiceHistory{
+		ServiceID: serviceID,
+		ServerID:  0,
+		AvgDelay:  10.0,
+		Up:        5,
+		Down:      1,
+		CreatedAt: today.Add(-25 * time.Hour),
+	})
+	DB.Create(&model.ServiceHistory{
+		ServiceID: serviceID,
+		ServerID:  0,
+		AvgDelay:  20.0,
+		Up:        3,
+		Down:      2,
+		CreatedAt: today.Add(-25 * time.Hour),
+	})
+	DB.Create(&model.ServiceHistory{
+		ServiceID: serviceID,
+		ServerID:  0,
+		AvgDelay:  30.0,
+		Up:        10,
+		Down:      0,
+		CreatedAt: today.Add(-49 * time.Hour),
+	})
+
+	ss.loadMonthlyStatusFromDB(today)
+
+	ms := ss.monthlyStatus[serviceID]
+
+	// day -1: index 27, two records with AvgDelay 10 and 20
+	assert.InDelta(t, 15.0, ms.Delay[27], 0.01)
+	assert.Equal(t, uint64(8), ms.Up[27])
+	assert.Equal(t, uint64(3), ms.Down[27])
+
+	// day -2: index 26
+	assert.InDelta(t, 30.0, ms.Delay[26], 0.01)
+	assert.Equal(t, uint64(10), ms.Up[26])
+	assert.Equal(t, uint64(0), ms.Down[26])
+
+	// totals
+	assert.Equal(t, uint64(18), ms.TotalUp)
+	assert.Equal(t, uint64(3), ms.TotalDown)
+
+	// today (index 29) should be untouched
+	assert.Equal(t, float64(0), ms.Delay[29])
+	assert.Equal(t, uint64(0), ms.Up[29])
+}
+
+func TestLoadMonthlyStatusFromDB_IgnoresToday(t *testing.T) {
+	cleanup := setupTestDB(t)
+	defer cleanup()
+
+	year, month, day := time.Now().Date()
+	today := time.Date(year, month, day, 0, 0, 0, 0, time.UTC)
+
+	serviceID := uint64(1)
+	ss := newTestSentinel([]uint64{serviceID})
+
+	DB.Create(&model.ServiceHistory{
+		ServiceID: serviceID,
+		ServerID:  0,
+		AvgDelay:  50.0,
+		Up:        100,
+		Down:      5,
+		CreatedAt: today.Add(2 * time.Hour),
+	})
+
+	ss.loadMonthlyStatusFromDB(today)
+
+	ms := ss.monthlyStatus[serviceID]
+	assert.Equal(t, uint64(0), ms.TotalUp)
+	assert.Equal(t, uint64(0), ms.TotalDown)
+}
+
+func TestLoadMonthlyStatusFromDB_UnknownServiceIgnored(t *testing.T) {
+	cleanup := setupTestDB(t)
+	defer cleanup()
+
+	year, month, day := time.Now().Date()
+	today := time.Date(year, month, day, 0, 0, 0, 0, time.UTC)
+
+	ss := newTestSentinel([]uint64{1})
+
+	DB.Create(&model.ServiceHistory{
+		ServiceID: 999,
+		ServerID:  0,
+		AvgDelay:  10.0,
+		Up:        5,
+		Down:      1,
+		CreatedAt: today.Add(-25 * time.Hour),
+	})
+
+	ss.loadMonthlyStatusFromDB(today)
+
+	ms := ss.monthlyStatus[uint64(1)]
+	assert.Equal(t, uint64(0), ms.TotalUp)
+}
+
+func TestLoadTodayStatsFromDB(t *testing.T) {
+	cleanup := setupTestDB(t)
+	defer cleanup()
+
+	year, month, day := time.Now().Date()
+	today := time.Date(year, month, day, 0, 0, 0, 0, time.UTC)
+
+	serviceID := uint64(1)
+	ss := newTestSentinel([]uint64{serviceID})
+
+	DB.Create(&model.ServiceHistory{
+		ServiceID: serviceID,
+		ServerID:  0,
+		AvgDelay:  10.0,
+		Up:        5,
+		Down:      1,
+		CreatedAt: today.Add(1 * time.Hour),
+	})
+	DB.Create(&model.ServiceHistory{
+		ServiceID: serviceID,
+		ServerID:  0,
+		AvgDelay:  30.0,
+		Up:        3,
+		Down:      2,
+		CreatedAt: today.Add(2 * time.Hour),
+	})
+
+	ss.loadTodayStats(today)
+
+	st := ss.serviceStatusToday[serviceID]
+	assert.Equal(t, uint64(8), st.Up)
+	assert.Equal(t, uint64(3), st.Down)
+	assert.InDelta(t, 20.0, st.Delay, 0.01)
+
+	ms := ss.monthlyStatus[serviceID]
+	assert.Equal(t, uint64(8), ms.TotalUp)
+	assert.Equal(t, uint64(3), ms.TotalDown)
+}
+
+func TestLoadMonthlyStatusFromTSDB(t *testing.T) {
+	db, cleanup := setupTestTSDB(t)
+	defer cleanup()
+
+	year, month, day := time.Now().Date()
+	today := time.Date(year, month, day, 0, 0, 0, 0, time.UTC)
+
+	serviceID := uint64(1)
+	services := []*model.Service{{Common: model.Common{ID: serviceID}}}
+	ss := newTestSentinel([]uint64{serviceID})
+
+	yesterday := today.Add(-25 * time.Hour)
+	for i := 0; i < 5; i++ {
+		ts := yesterday.Add(time.Duration(i) * time.Minute)
+		require.NoError(t, db.WriteServiceMetrics(&tsdb.ServiceMetrics{
+			ServiceID:  serviceID,
+			ServerID:   1,
+			Timestamp:  ts,
+			Delay:      float64(10 + i),
+			Successful: true,
+		}))
+	}
+	for i := 0; i < 3; i++ {
+		require.NoError(t, db.WriteServiceMetrics(&tsdb.ServiceMetrics{
+			ServiceID:  serviceID,
+			ServerID:   1,
+			Timestamp:  yesterday.Add(time.Duration(i+10) * time.Minute),
+			Delay:      float64(20 + i),
+			Successful: false,
+		}))
+	}
+
+	db.Flush()
+
+	ss.loadMonthlyStatusFromTSDB(services, today)
+
+	ms := ss.monthlyStatus[serviceID]
+	// day -1: dayIndex 28
+	assert.Equal(t, uint64(5), ms.Up[28])
+	assert.Equal(t, uint64(3), ms.Down[28])
+	assert.Equal(t, uint64(5), ms.TotalUp)
+	assert.Equal(t, uint64(3), ms.TotalDown)
+	assert.Greater(t, ms.Delay[28], float64(0))
+
+	// today (index 29) should be untouched
+	assert.Equal(t, uint64(0), ms.Up[29])
+}
+
+func TestLoadTodayStatsFromTSDB(t *testing.T) {
+	db, cleanup := setupTestTSDB(t)
+	defer cleanup()
+
+	year, month, day := time.Now().Date()
+	today := time.Date(year, month, day, 0, 0, 0, 0, time.UTC)
+
+	serviceID := uint64(1)
+	ss := newTestSentinel([]uint64{serviceID})
+
+	now := time.Now()
+	for i := 0; i < 4; i++ {
+		ts := now.Add(-time.Duration(i) * time.Minute)
+		require.NoError(t, db.WriteServiceMetrics(&tsdb.ServiceMetrics{
+			ServiceID:  serviceID,
+			ServerID:   1,
+			Timestamp:  ts,
+			Delay:      float64(10 + i),
+			Successful: true,
+		}))
+	}
+	for i := 0; i < 2; i++ {
+		ts := now.Add(-time.Duration(i+10) * time.Minute)
+		require.NoError(t, db.WriteServiceMetrics(&tsdb.ServiceMetrics{
+			ServiceID:  serviceID,
+			ServerID:   1,
+			Timestamp:  ts,
+			Delay:      0,
+			Successful: false,
+		}))
+	}
+
+	db.Flush()
+
+	ss.loadTodayStats(today)
+
+	st := ss.serviceStatusToday[serviceID]
+	assert.Greater(t, st.Up, uint64(0))
+	assert.Greater(t, st.Down, uint64(0))
+
+	ms := ss.monthlyStatus[serviceID]
+	assert.Equal(t, st.Up, ms.TotalUp)
+	assert.Equal(t, st.Down, ms.TotalDown)
+}
+
+func TestLoadMonthlyStatusFromTSDB_NoDoubleCountToday(t *testing.T) {
+	db, cleanup := setupTestTSDB(t)
+	defer cleanup()
+
+	year, month, day := time.Now().Date()
+	today := time.Date(year, month, day, 0, 0, 0, 0, time.UTC)
+
+	serviceID := uint64(1)
+	services := []*model.Service{{Common: model.Common{ID: serviceID}}}
+	ss := newTestSentinel([]uint64{serviceID})
+
+	now := time.Now()
+	for i := 0; i < 5; i++ {
+		require.NoError(t, db.WriteServiceMetrics(&tsdb.ServiceMetrics{
+			ServiceID:  serviceID,
+			ServerID:   1,
+			Timestamp:  now.Add(-time.Duration(i) * time.Minute),
+			Delay:      10.0,
+			Successful: true,
+		}))
+	}
+	db.Flush()
+
+	ss.loadMonthlyStatusFromTSDB(services, today)
+	totalAfterMonthly := ss.monthlyStatus[serviceID].TotalUp
+
+	ss.loadTodayStats(today)
+	totalAfterToday := ss.monthlyStatus[serviceID].TotalUp
+
+	assert.Equal(t, totalAfterMonthly+ss.serviceStatusToday[serviceID].Up, totalAfterToday)
+}
@@ -87,12 +87,13 @@ func InitDBFromPath(path string) error {
 	}
 	err = DB.AutoMigrate(model.Server{}, model.User{}, model.ServerGroup{}, model.NotificationGroup{},
 		model.Notification{}, model.AlertRule{}, model.Service{}, model.NotificationGroupNotification{},
-		model.ServiceHistory{}, model.Cron{}, model.Transfer{}, model.ServerGroupServer{},
+		model.Cron{}, model.Transfer{}, model.ServerGroupServer{},
 		model.NAT{}, model.DDNSProfile{}, model.NotificationGroupNotification{},
 		model.WAF{}, model.Oauth2Bind{})
 	if err != nil {
 		return err
 	}
+
 	return nil
 }

@@ -130,14 +131,9 @@ func RecordTransferHourlyUsage(servers ...*model.Server) {
 	log.Printf("NEZHA>> Saved traffic metrics to database. Affected %d row(s), Error: %v", len(txs), DB.Create(txs).Error)
 }

-// CleanServiceHistory 清理无效或过时的 监控记录 和 流量记录
-func CleanServiceHistory() {
-	// 清理已被删除的服务器的监控记录与流量记录
-	DB.Unscoped().Delete(&model.ServiceHistory{}, "created_at < ? OR service_id NOT IN (SELECT `id` FROM services)", time.Now().AddDate(0, 0, -30))
-	// 由于网络监控记录的数据较多，并且前端仅使用了 1 天的数据
-	// 考虑到 sqlite 数据量问题，仅保留一天数据，
-	// server_id = 0 的数据会用于/service页面的可用性展示
-	DB.Unscoped().Delete(&model.ServiceHistory{}, "(created_at < ? AND server_id != 0) OR service_id NOT IN (SELECT `id` FROM services)", time.Now().AddDate(0, 0, -1))
+// CleanMonitorHistory 清理流量记录（TSDB 有自己的保留策略）
+func CleanMonitorHistory() {
+	// 清理已被删除的服务器的流量记录
 	DB.Unscoped().Delete(&model.Transfer{}, "server_id NOT IN (SELECT `id` FROM servers)")
 	// 计算可清理流量记录的时长
 	var allServerKeep time.Time
@@ -179,6 +175,28 @@ func CleanServiceHistory() {
 	}
 }

+// PerformMaintenance 执行系统维护（SQLite VACUUM 和 TSDB 维护）
+func PerformMaintenance() {
+	log.Println("NEZHA>> Starting system maintenance...")
+
+	// 1. SQLite 维护
+	if DB != nil {
+		log.Println("NEZHA>> SQLite: Starting VACUUM...")
+		if err := DB.Exec("VACUUM").Error; err != nil {
+			log.Printf("NEZHA>> SQLite: VACUUM failed: %v", err)
+		} else {
+			log.Println("NEZHA>> SQLite: VACUUM completed")
+		}
+	}
+
+	// 2. TSDB 维护
+	if TSDBEnabled() {
+		TSDBShared.Maintenance()
+	}
+
+	log.Println("NEZHA>> System maintenance completed")
+}
+
 // IPDesensitize 根据设置选择是否对IP进行打码处理 返回处理后的IP(关闭打码则返回原IP)
 func IPDesensitize(ip string) string {
 	if Conf.EnablePlainIPInNotification {
@@ -0,0 +1,73 @@
+package singleton
+
+import (
+	"log"
+	"time"
+
+	"github.com/nezhahq/nezha/model"
+	"github.com/nezhahq/nezha/pkg/tsdb"
+)
+
+var TSDBShared *tsdb.TSDB
+
+func InitTSDB() error {
+	config := &tsdb.Config{
+		RetentionDays:      30,
+		MinFreeDiskSpaceGB: 1,
+		MaxMemoryMB:        256,
+	}
+
+	if Conf.TSDB.DataPath != "" {
+		config.DataPath = Conf.TSDB.DataPath
+	}
+	if Conf.TSDB.RetentionDays > 0 {
+		config.RetentionDays = Conf.TSDB.RetentionDays
+	}
+	if Conf.TSDB.MinFreeDiskSpaceGB > 0 {
+		config.MinFreeDiskSpaceGB = Conf.TSDB.MinFreeDiskSpaceGB
+	}
+	if Conf.TSDB.MaxMemoryMB > 0 {
+		config.MaxMemoryMB = Conf.TSDB.MaxMemoryMB
+	}
+	if Conf.TSDB.WriteBufferSize > 0 {
+		config.WriteBufferSize = Conf.TSDB.WriteBufferSize
+	}
+	if Conf.TSDB.WriteBufferFlushInterval > 0 {
+		config.WriteBufferFlushInterval = time.Duration(Conf.TSDB.WriteBufferFlushInterval) * time.Second
+	}
+
+	if !config.Enabled() {
+		log.Println("NEZHA>> TSDB is disabled (tsdb.data_path not configured)")
+		if DB != nil {
+			return DB.AutoMigrate(model.ServiceHistory{})
+		}
+		return nil
+	}
+
+	var err error
+	TSDBShared, err = tsdb.Open(config)
+	if err != nil {
+		return err
+	}
+
+	log.Println("NEZHA>> TSDB initialized successfully")
+
+	if DB != nil && DB.Migrator().HasTable("service_histories") {
+		log.Println("NEZHA>> Dropping legacy service_histories table (TSDB is now enabled). Historical data will NOT be migrated.")
+		if err := DB.Migrator().DropTable("service_histories"); err != nil {
+			log.Printf("NEZHA>> Warning: failed to drop service_histories table: %v", err)
+		}
+	}
+
+	return nil
+}
+
+func TSDBEnabled() bool {
+	return TSDBShared != nil && !TSDBShared.IsClosed()
+}
+
+func CloseTSDB() {
+	if TSDBShared != nil {
+		TSDBShared.Close()
+	}
+}