HY2XS_flamy/apps/service/metrics_collector.go

package service

import (
	"regexp"
	"strings"

	"github.com/sirupsen/logrus"
	"hy2xs-admin/dao"
	"hy2xs-admin/model/entity"
	"hy2xs-admin/model/vo"
	"hy2xs-admin/util"
	"sync"
	"time"
)

const hysteriaVersionCacheTTL = 15 * time.Minute
const statsRetention = 7 * 24 * time.Hour

var ansiRe = regexp.MustCompile(`\x1b\[[0-9;]*[A-Za-z]`)
var hysteriaVersionRe = regexp.MustCompile(`(?i)v?\d+\.\d+\.\d+(?:[-+][0-9A-Za-z.-]+)?`)

func normalizeHysteriaVersion(raw string) string {
	cleaned := strings.TrimSpace(raw)
	cleaned = ansiRe.ReplaceAllString(cleaned, "")
	match := hysteriaVersionRe.FindString(cleaned)
	if match == "" {
		return "-"
	}
	return strings.TrimSpace(match)
}

type metricsSnapshot struct {
	CollectedAt    int64
	System         vo.DashboardSystemVo
	Hysteria       vo.DashboardHysteriaVo
	OnlinePeers    int64
	OnlineDevices  int64
	CollectorState vo.DataHealthVo
	HysteriaState  vo.DataHealthVo
}

var metricsStore = struct {
	sync.RWMutex
	snapshot      metricsSnapshot
	lastSuccessAt int64
	version       string
	versionAt     time.Time
}{}

func getCachedHysteriaVersion(now time.Time) string {
	metricsStore.RLock()
	if metricsStore.version != "" && now.Sub(metricsStore.versionAt) < hysteriaVersionCacheTTL {
		v := metricsStore.version
		metricsStore.RUnlock()
		return v
	}
	metricsStore.RUnlock()

	content, err := util.Exec(util.GetHysteria2BinPath() + " version")
	if err != nil {
		metricsStore.RLock()
		v := metricsStore.version
		metricsStore.RUnlock()
		if v != "" {
			return v
		}
		return "-"
	}
	normalized := normalizeHysteriaVersion(content)
	if normalized == "-" {
		logrus.Debugf("hysteria version parse failed, raw output: %q", strings.TrimSpace(content))
	}

	metricsStore.Lock()
	metricsStore.version = normalized
	metricsStore.versionAt = now
	metricsStore.Unlock()
	return normalized
}

func CollectMetricsSnapshot() {
	nowMs := time.Now().UnixMilli()
	s := metricsSnapshot{
		CollectedAt:    nowMs,
		CollectorState: vo.DataHealthVo{Status: "ok", LastSuccessAt: nowMs},
		HysteriaState:  vo.DataHealthVo{Status: "ok", LastSuccessAt: nowMs},
	}

	if cpuPercent, err := util.GetCpuPercent(); err == nil {
		s.System.CpuPercent = cpuPercent
	}
	if memPercent, err := util.GetMemPercent(); err == nil {
		s.System.MemPercent = memPercent
	}
	if memInfo, err := util.GetMemInfo(); err == nil {
		s.System.MemUsedBytes = memInfo.Used
		s.System.MemTotalBytes = memInfo.Total
	}
	if diskPercent, err := util.GetDiskPercent(); err == nil {
		s.System.DiskPercent = diskPercent
	}
	if diskInfo, err := util.GetDiskInfo(); err == nil {
		s.System.DiskUsedBytes = diskInfo.Used
		s.System.DiskTotalBytes = diskInfo.Total
	}

	// Два независимых источника, два независимых вопроса.
	//
	// Раньше второй выводился из первого: общий Hysteria2Online отдавал пустую
	// карту БЕЗ ошибки, когда systemctl говорил «служба неактивна», и сюда
	// приходило `err == nil` — то есть `ApiReachable = true` выставлялся, ни
	// разу не сходив в Traffic Stats API. Дашборд утверждал «Hysteria
	// остановлена» и «API доступен, онлайн 0» одновременно.
	serviceState := Hysteria2ServiceState()
	s.Hysteria.ServiceState = string(serviceState)
	s.Hysteria.Running = serviceState == HysteriaServiceActive
	s.Hysteria.Version = getCachedHysteriaVersion(time.Now())

	if onlineMap, err := Hysteria2Online(); err == nil {
		s.Hysteria.ApiReachable = true
		s.OnlinePeers = int64(len(onlineMap))
		for _, c := range onlineMap {
			s.OnlineDevices += c
		}
	} else {
		s.Hysteria.ApiReachable = false
		s.Hysteria.LastApiError = "dashboard.error.trafficApiUnavailable"
		s.HysteriaState.Status = "error"
		s.HysteriaState.MessageKey = "dashboard.error.trafficApiUnavailable"
	}

	metricsStore.Lock()
	metricsStore.snapshot = s
	metricsStore.lastSuccessAt = nowMs
	metricsStore.Unlock()

	running := int64(0)
	if s.Hysteria.Running {
		running = 1
	}
	diskPath := "/"
	if configured := util.GetEnvDiskPath(); configured != "" {
		diskPath = configured
	}
	metric := entity.MetricSample{
		SampledAt:       &nowMs,
		CpuPercent:      &s.System.CpuPercent,
		Load1:           func() *float64 { v := float64(0); return &v }(),
		MemUsedBytes:    func() *int64 { v := int64(s.System.MemUsedBytes); return &v }(),
		MemTotalBytes:   func() *int64 { v := int64(s.System.MemTotalBytes); return &v }(),
		MemPercent:      &s.System.MemPercent,
		DiskPath:        &diskPath,
		DiskUsedBytes:   func() *int64 { v := int64(s.System.DiskUsedBytes); return &v }(),
		DiskTotalBytes:  func() *int64 { v := int64(s.System.DiskTotalBytes); return &v }(),
		DiskPercent:     &s.System.DiskPercent,
		HysteriaRunning: &running,
		OnlinePeers:     &s.OnlinePeers,
		OnlineDevices:   &s.OnlineDevices,
	}
	_ = dao.SaveMetricSample(metric)
}

func CleanupStatsRetention() {
	now := time.Now()
	cutoffMs := now.Add(-statsRetention).UnixMilli()

	hourMs := int64(time.Hour / time.Millisecond)
	dayMs := int64(24 * time.Hour / time.Millisecond)
	cutoffHour := cutoffMs - (cutoffMs % hourMs)
	cutoffDay := cutoffMs - (cutoffMs % dayMs)

	_ = dao.CleanupMetricSample(cutoffMs)
	_ = dao.CleanupTrafficSample(cutoffMs)
	_ = dao.CleanupTrafficAggregateHourly(cutoffHour)
	_ = dao.CleanupTrafficAggregateDaily(cutoffDay)
}

func DashboardSnapshot() metricsSnapshot {
	metricsStore.RLock()
	defer metricsStore.RUnlock()
	return metricsStore.snapshot
}