mrlee
c8264bf280
CI / verify (push) Canceled after 0s
- 三个新事件（默认关闭）：公网服务端掉线、公网 agent 不可达、配置漂移；连续失败阈值与漂移提醒间隔均可配，达标与恢复各报一次 - 避免误报：只在「用户要服务端运行」时判定掉线；ConnectionState=unknown 按健康处理；没配 agent 的服务端不参与可达性判定 - internal/monitor 改 Source 接口（Routes/Servers/Agents/Drift），frp 适配在 internal/app/monitor_source.go；「连续 N 次」抽成共用 failureTransition，通知一律在锁外调用 - 通知图标分类补齐；通知方式标签改「Webhook / IM 机器人」；新增 6 个设置项；补 serverDown/serverLabel/stateOr 测试 - agent 无改动，AgentVersion 保持 0.12.2
387 lines
12 KiB
Go
387 lines
12 KiB
Go
// Package monitor 周期巡检几类必须持续观察才会出现的问题：主机资源占用、隧道连通性、
|
|
// 公网服务端掉线、agent 不可达、配置漂移。
|
|
// 事件型通知（DDNS 失败、证书到期）由各自模块在发生时直接发出，不需要巡检。
|
|
//
|
|
// 判定在这里，通知只走 notify.Service：开关、阈值、冷却都读设置项，
|
|
// 所以本包不需要知道通知通道的任何细节。
|
|
package monitor
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
"log/slog"
|
|
"strings"
|
|
"sync"
|
|
"time"
|
|
|
|
"github.com/fonu/fonu/internal/notify"
|
|
"github.com/fonu/fonu/internal/settings"
|
|
"github.com/fonu/fonu/internal/sysinfo"
|
|
)
|
|
|
|
// RouteStatus 是巡检关心的隧道状态。
|
|
type RouteStatus struct {
|
|
Domain string
|
|
TunnelOK bool
|
|
Detail string
|
|
}
|
|
|
|
// ServerStatus 是巡检关心的公网服务端状态：只有 DesiredState 表示「用户要它运行」时才判定掉线。
|
|
type ServerStatus struct {
|
|
ID int64
|
|
Name string
|
|
DesiredState string
|
|
ProcessState string
|
|
ConnectionState string
|
|
Detail string
|
|
}
|
|
|
|
// AgentStatus 是某台服务端上 fonu-agent 的可达性；没配 agent 的服务端不会出现在这里。
|
|
type AgentStatus struct {
|
|
ID int64
|
|
Name string
|
|
Reachable bool
|
|
Detail string
|
|
}
|
|
|
|
// DriftStatus 是某台服务端的部署漂移摘要，Count 为 0 表示没有漂移。
|
|
type DriftStatus struct {
|
|
ID int64
|
|
Name string
|
|
Count int
|
|
Items []string
|
|
}
|
|
|
|
// Source 提供巡检需要的外部状态；由 app 侧把 frp 服务适配过来，monitor 因此不依赖 frp 包。
|
|
type Source interface {
|
|
Routes(ctx context.Context) ([]RouteStatus, error)
|
|
Servers(ctx context.Context) ([]ServerStatus, error)
|
|
Agents(ctx context.Context) ([]AgentStatus, error)
|
|
Drift(ctx context.Context) ([]DriftStatus, error)
|
|
}
|
|
|
|
// Checker 每轮都重新读设置项：改阈值或开关不需要重启。
|
|
type Checker struct {
|
|
settings *settings.Store
|
|
notify *notify.Service
|
|
logger *slog.Logger
|
|
source Source
|
|
|
|
mu sync.Mutex
|
|
cpuPrev sysinfo.CPUSample
|
|
resourceAlerts map[string]time.Time
|
|
failures map[string]int
|
|
driftAlerted map[int64]bool
|
|
driftNotified map[int64]time.Time
|
|
}
|
|
|
|
func New(settingsStore *settings.Store, notifySvc *notify.Service, logger *slog.Logger, source Source) *Checker {
|
|
if logger == nil {
|
|
logger = slog.Default()
|
|
}
|
|
return &Checker{
|
|
settings: settingsStore,
|
|
notify: notifySvc,
|
|
logger: logger,
|
|
source: source,
|
|
resourceAlerts: map[string]time.Time{},
|
|
failures: map[string]int{},
|
|
driftAlerted: map[int64]bool{},
|
|
driftNotified: map[int64]time.Time{},
|
|
}
|
|
}
|
|
|
|
// Run 跑一轮巡检。scheduler 已对任务包了 recover，这里不重复兜底。
|
|
func (c *Checker) Run(ctx context.Context) {
|
|
if c == nil || c.settings == nil {
|
|
return
|
|
}
|
|
c.checkResources(ctx)
|
|
c.checkRoutes(ctx)
|
|
c.checkServers(ctx)
|
|
c.checkAgents(ctx)
|
|
c.checkDrift(ctx)
|
|
}
|
|
|
|
func (c *Checker) checkResources(ctx context.Context) {
|
|
// 开关关闭时连采集都省掉（关掉的功能不该有后台开销）
|
|
if enabled, _ := c.settings.GetBool(ctx, settings.KeyNotifyOnResourceThreshold); !enabled {
|
|
return
|
|
}
|
|
cpuLimit, _ := c.settings.GetInt(ctx, settings.KeyNotifyCPUThreshold)
|
|
memoryLimit, _ := c.settings.GetInt(ctx, settings.KeyNotifyMemoryThreshold)
|
|
diskLimit, _ := c.settings.GetInt(ctx, settings.KeyNotifyDiskThreshold)
|
|
|
|
c.mu.Lock()
|
|
stats, sample := sysinfo.Read(c.cpuPrev)
|
|
c.cpuPrev = sample
|
|
c.mu.Unlock()
|
|
|
|
limits := []struct {
|
|
metric string
|
|
label string
|
|
limit int
|
|
value float64
|
|
}{
|
|
{"cpu", "CPU", cpuLimit, stats.CPUPercent},
|
|
{"memory", "内存", memoryLimit, percent(stats.MemUsedBytes, stats.MemTotalBytes)},
|
|
{"disk", "磁盘", diskLimit, percent(stats.DiskUsedBytes, stats.DiskTotalBytes)},
|
|
}
|
|
now := time.Now()
|
|
for _, item := range limits {
|
|
// 阈值 <= 0 表示这一项不告警（CPU 首次采样时占用率为 0，也不会误报）
|
|
if item.limit <= 0 {
|
|
continue
|
|
}
|
|
if item.value < float64(item.limit) {
|
|
// 回落到阈值以下就清冷却：再次超标可以立刻提醒
|
|
c.mu.Lock()
|
|
delete(c.resourceAlerts, item.metric)
|
|
c.mu.Unlock()
|
|
continue
|
|
}
|
|
if !c.allowResourceAlert(item.metric, now) {
|
|
continue
|
|
}
|
|
c.notify.Alert(ctx, notify.EventResourceThreshold, "资源占用过高",
|
|
fmt.Sprintf("%s 使用率 %.1f%%，已超过阈值 %d%%", item.label, item.value, item.limit))
|
|
}
|
|
}
|
|
|
|
// allowResourceAlert 按静默期限流：持续高占用每 DefaultResourceAlertCooldown 提醒一次。
|
|
func (c *Checker) allowResourceAlert(metric string, now time.Time) bool {
|
|
cooldown := time.Duration(notify.DefaultResourceAlertCooldown) * time.Second
|
|
c.mu.Lock()
|
|
defer c.mu.Unlock()
|
|
if last, ok := c.resourceAlerts[metric]; ok && now.Sub(last) < cooldown {
|
|
return false
|
|
}
|
|
c.resourceAlerts[metric] = now
|
|
return true
|
|
}
|
|
|
|
func (c *Checker) checkRoutes(ctx context.Context) {
|
|
if c.source == nil {
|
|
return
|
|
}
|
|
if enabled, _ := c.settings.GetBool(ctx, settings.KeyNotifyOnTunnelDown); !enabled {
|
|
return
|
|
}
|
|
limit, _ := c.settings.GetInt(ctx, settings.KeyNotifyTunnelFailThreshold)
|
|
if limit <= 0 {
|
|
return
|
|
}
|
|
|
|
routes, err := c.source.Routes(ctx)
|
|
if err != nil {
|
|
c.logger.Warn("巡检隧道状态失败", "module", "MONITOR", "error", err.Error())
|
|
return
|
|
}
|
|
for _, route := range routes {
|
|
if strings.TrimSpace(route.Domain) == "" {
|
|
continue
|
|
}
|
|
_, event := c.track("route:"+route.Domain, route.TunnelOK, limit)
|
|
switch event {
|
|
case failureEventDown:
|
|
c.notify.Alert(ctx, notify.EventTunnelDown, "隧道连续不通",
|
|
fmt.Sprintf("%s 连续 %d 次巡检未在 frps 上注册：%s", route.Domain, limit, detailOr(route.Detail)))
|
|
case failureEventRecovered:
|
|
c.notify.Alert(ctx, notify.EventTunnelDown, "隧道已恢复",
|
|
fmt.Sprintf("%s 已重新在 frps 上注册", route.Domain))
|
|
}
|
|
}
|
|
}
|
|
|
|
// checkServers 公网服务端掉线：进程没在跑，或隧道客户端连不上 frps。
|
|
// 只对「用户要它运行」的服务端判定，手动停掉的不报。
|
|
func (c *Checker) checkServers(ctx context.Context) {
|
|
if c.source == nil {
|
|
return
|
|
}
|
|
if enabled, _ := c.settings.GetBool(ctx, settings.KeyNotifyOnFrpsDown); !enabled {
|
|
return
|
|
}
|
|
limit, _ := c.settings.GetInt(ctx, settings.KeyNotifyFrpsFailThreshold)
|
|
if limit <= 0 {
|
|
return
|
|
}
|
|
|
|
servers, err := c.source.Servers(ctx)
|
|
if err != nil {
|
|
c.logger.Warn("巡检服务端状态失败", "module", "MONITOR", "error", err.Error())
|
|
return
|
|
}
|
|
for _, server := range servers {
|
|
label := serverLabel(server.ID, server.Name)
|
|
_, event := c.track(fmt.Sprintf("server:%d", server.ID), !serverDown(server), limit)
|
|
switch event {
|
|
case failureEventDown:
|
|
c.notify.Alert(ctx, notify.EventFrpsServerDown, "公网服务端掉线",
|
|
fmt.Sprintf("%s 连续 %d 次巡检未运行或未连接（进程 %s / 连接 %s）：%s",
|
|
label, limit, stateOr(server.ProcessState), stateOr(server.ConnectionState), detailOr(server.Detail)))
|
|
case failureEventRecovered:
|
|
c.notify.Alert(ctx, notify.EventFrpsServerDown, "公网服务端已恢复",
|
|
fmt.Sprintf("%s 已重新运行并连接", label))
|
|
}
|
|
}
|
|
}
|
|
|
|
// checkAgents agent 不可达：VPS 重启后 agent 没起来是最常见的成因。
|
|
func (c *Checker) checkAgents(ctx context.Context) {
|
|
if c.source == nil {
|
|
return
|
|
}
|
|
if enabled, _ := c.settings.GetBool(ctx, settings.KeyNotifyOnAgentUnreachable); !enabled {
|
|
return
|
|
}
|
|
limit, _ := c.settings.GetInt(ctx, settings.KeyNotifyAgentFailThreshold)
|
|
if limit <= 0 {
|
|
return
|
|
}
|
|
|
|
agents, err := c.source.Agents(ctx)
|
|
if err != nil {
|
|
c.logger.Warn("巡检 agent 可达性失败", "module", "MONITOR", "error", err.Error())
|
|
return
|
|
}
|
|
for _, agent := range agents {
|
|
label := serverLabel(agent.ID, agent.Name)
|
|
_, event := c.track(fmt.Sprintf("agent:%d", agent.ID), agent.Reachable, limit)
|
|
switch event {
|
|
case failureEventDown:
|
|
c.notify.Alert(ctx, notify.EventAgentUnreachable, "公网 agent 不可达",
|
|
fmt.Sprintf("%s 的 fonu-agent 连续 %d 次无响应：%s", label, limit, detailOr(agent.Detail)))
|
|
case failureEventRecovered:
|
|
c.notify.Alert(ctx, notify.EventAgentUnreachable, "公网 agent 已恢复",
|
|
fmt.Sprintf("%s 的 fonu-agent 已恢复响应", label))
|
|
}
|
|
}
|
|
}
|
|
|
|
// checkDrift 配置漂移：VPS 上的 vhost 与 Fonu 记录不一致（被手工改过、或 agent 重装后没重部署）。
|
|
// 它不是「连续 N 次」型故障，因此用冷却时间限流：首次发现即报，之后每 cooldown 提醒一次，漂移清空时补一条恢复。
|
|
func (c *Checker) checkDrift(ctx context.Context) {
|
|
if c.source == nil {
|
|
return
|
|
}
|
|
if enabled, _ := c.settings.GetBool(ctx, settings.KeyNotifyOnRouteDrift); !enabled {
|
|
return
|
|
}
|
|
cooldownHours, _ := c.settings.GetInt(ctx, settings.KeyNotifyDriftCooldownHours)
|
|
if cooldownHours <= 0 {
|
|
cooldownHours = notify.DefaultDriftCooldownHours
|
|
}
|
|
cooldown := time.Duration(cooldownHours) * time.Hour
|
|
|
|
reports, err := c.source.Drift(ctx)
|
|
if err != nil {
|
|
c.logger.Warn("巡检配置漂移失败", "module", "MONITOR", "error", err.Error())
|
|
return
|
|
}
|
|
now := time.Now()
|
|
for _, report := range reports {
|
|
c.mu.Lock()
|
|
alerted := c.driftAlerted[report.ID]
|
|
last := c.driftNotified[report.ID]
|
|
if report.Count == 0 {
|
|
delete(c.driftAlerted, report.ID)
|
|
delete(c.driftNotified, report.ID)
|
|
}
|
|
c.mu.Unlock()
|
|
|
|
label := serverLabel(report.ID, report.Name)
|
|
if report.Count == 0 {
|
|
if alerted {
|
|
c.notify.Alert(ctx, notify.EventRouteDrift, "配置漂移已消除",
|
|
fmt.Sprintf("%s 上的 vhost 已与 Fonu 记录一致", label))
|
|
}
|
|
continue
|
|
}
|
|
if alerted && !last.IsZero() && now.Sub(last) < cooldown {
|
|
continue
|
|
}
|
|
c.mu.Lock()
|
|
c.driftAlerted[report.ID] = true
|
|
c.driftNotified[report.ID] = now
|
|
c.mu.Unlock()
|
|
c.notify.Alert(ctx, notify.EventRouteDrift, "公网反代配置漂移",
|
|
fmt.Sprintf("%s 有 %d 项与 Fonu 记录不一致：%s", label, report.Count, strings.Join(report.Items, "、")))
|
|
}
|
|
}
|
|
|
|
// track 推进某个实体的连续失败计数，返回本轮要发的事件。
|
|
// 锁只包住 map 读写：notify 一律在锁外调用。
|
|
func (c *Checker) track(key string, healthy bool, limit int) (int, failureEvent) {
|
|
c.mu.Lock()
|
|
defer c.mu.Unlock()
|
|
failures, event := failureTransition(c.failures[key], healthy, limit)
|
|
c.failures[key] = failures
|
|
return failures, event
|
|
}
|
|
|
|
type failureEvent string
|
|
|
|
const (
|
|
failureEventNone failureEvent = ""
|
|
failureEventDown failureEvent = "down"
|
|
failureEventRecovered failureEvent = "recovered"
|
|
)
|
|
|
|
// failureTransition 推进某个实体的连续失败计数，返回新计数与本轮要发的事件。
|
|
// 只在「刚好达到阈值」与「从失败中恢复」两个时刻发，避免每轮巡检都刷通知。
|
|
func failureTransition(previous int, healthy bool, limit int) (int, failureEvent) {
|
|
if healthy {
|
|
if previous >= limit {
|
|
return 0, failureEventRecovered
|
|
}
|
|
return 0, failureEventNone
|
|
}
|
|
next := previous + 1
|
|
if next == limit {
|
|
return next, failureEventDown
|
|
}
|
|
return next, failureEventNone
|
|
}
|
|
|
|
// serverDown 判定服务端是否处于「用户要它运行、但它没在跑或没连上」。
|
|
// ConnectionState 为 unknown 时按健康处理：状态源（manager+log）偶尔读不到不代表掉线，避免误报。
|
|
func serverDown(status ServerStatus) bool {
|
|
if status.DesiredState != "running" && status.DesiredState != "starting" {
|
|
return false
|
|
}
|
|
if status.ProcessState == "stopped" || status.ProcessState == "error" {
|
|
return true
|
|
}
|
|
return status.ConnectionState == "disconnected" || status.ConnectionState == "error"
|
|
}
|
|
|
|
func serverLabel(id int64, name string) string {
|
|
if trimmed := strings.TrimSpace(name); trimmed != "" {
|
|
return trimmed
|
|
}
|
|
return fmt.Sprintf("服务端 #%d", id)
|
|
}
|
|
|
|
// stateOr 让通知里的状态字段不会因为空串而变成「（进程 / 连接 ）」。
|
|
func stateOr(state string) string {
|
|
if strings.TrimSpace(state) == "" {
|
|
return "未知"
|
|
}
|
|
return state
|
|
}
|
|
|
|
func percent(used, total uint64) float64 {
|
|
if total == 0 {
|
|
return 0
|
|
}
|
|
return float64(used) / float64(total) * 100
|
|
}
|
|
|
|
func detailOr(detail string) string {
|
|
if strings.TrimSpace(detail) == "" {
|
|
return "未返回具体原因"
|
|
}
|
|
return detail
|
|
}
|