main.go 5.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163
  1. package main
  2. import (
  3. commonConfig "cicv-data-closedloop/aarch64/pjibot_guide/common/config"
  4. commonService "cicv-data-closedloop/aarch64/pjibot_guide/common/service"
  5. "cicv-data-closedloop/aarch64/pjibot_guide/common/variable"
  6. "cicv-data-closedloop/aarch64/pjibot_guide/control/pkg"
  7. "cicv-data-closedloop/common/config/c_log"
  8. "cicv-data-closedloop/common/entity"
  9. "cicv-data-closedloop/common/util"
  10. "fmt"
  11. "net/rpc"
  12. "os"
  13. "runtime"
  14. "time"
  15. )
  16. var (
  17. applicationName = "pji-control"
  18. localStatus = "idle"
  19. cloudStatus = "NONE"
  20. lastLocalStatus = "idle"
  21. lastCloudStatus = "NONE"
  22. localTurnLength = 1 // s,本地状态刷新时间
  23. cloudTurnLength = 60 // s,云端状态刷新时间
  24. renewTurnLength = 3 // s,续约状态刷新时间
  25. waitStopLength = 1 // min,停止master前等待时间
  26. launchedFlag = false
  27. renewedFlag = false
  28. renewTimer *time.Timer // 续约定时器
  29. RenewDur = 5 // min, 续约时间
  30. )
  31. func init() {
  32. runtime.GOMAXPROCS(1)
  33. // 初始化日志配置
  34. c_log.InitLog(variable.LogDir, applicationName)
  35. // 初始化本地配置文件(第1处配置,在本地文件)
  36. commonConfig.InitLocalConfig(variable.LocalConfigPath)
  37. // 初始化Oss连接信息
  38. commonConfig.InitOssConfig()
  39. // 初始化业务逻辑配置信息,配置文件在oss上(第2处配置,在oss文件)
  40. commonConfig.InitCloudConfig()
  41. // 初始化rpc客户端,用于杀死旧的采集程序
  42. // 初始化websocket配置
  43. commonConfig.InitWebsocketConfig()
  44. }
  45. func renew() {
  46. for {
  47. time.Sleep(time.Duration(renewTurnLength) * time.Second)
  48. if localStatus == "running" && !renewedFlag { // 设备处于运行状态,且尚未续约
  49. c_log.GlobalLogger.Info("设备仍处于运行状态,续约 - 开始。")
  50. if renewTimer != nil {
  51. renewTimer.Stop()
  52. }
  53. renewedFlag = true
  54. renewTimer = time.AfterFunc(time.Duration(RenewDur)*time.Minute, func() {
  55. renewedFlag = false
  56. })
  57. c_log.GlobalLogger.Infof("续约时间【%v】分钟 - 成功。", RenewDur)
  58. }
  59. }
  60. }
  61. func startMasterNode() {
  62. c_log.GlobalLogger.Info("获取数据闭环平台最新配置。")
  63. fmt.Println("1")
  64. commonConfig.InitPlatformConfig()
  65. if _, err := util.ExecuteWithPath(commonConfig.LocalConfig.RestartCmd.Dir, commonConfig.LocalConfig.RestartCmd.Name, commonConfig.LocalConfig.RestartCmd.Args...); err != nil {
  66. c_log.GlobalLogger.Info("启动新程序失败,【path】=", commonConfig.LocalConfig.RestartCmd.Dir, "【cmd】=", commonConfig.LocalConfig.RestartCmd.Name, commonConfig.LocalConfig.RestartCmd.Args, ":", err)
  67. os.Exit(-1)
  68. }
  69. c_log.GlobalLogger.Info("启动任务,本地执行启动命令:【path】=", commonConfig.LocalConfig.RestartCmd.Dir, "【cmd】=", commonConfig.LocalConfig.RestartCmd.Name, commonConfig.LocalConfig.RestartCmd.Args)
  70. launchedFlag = true
  71. c_log.GlobalLogger.Info("数采程序启动 - 成功。")
  72. }
  73. func stopMasterNode() {
  74. // 发送rpc信号杀死采集程序
  75. var killArgs commonService.KillSignal
  76. killArgs = commonService.KillSignal{NodeName: "master", DropUploadData: commonConfig.PlatformConfig.DropUploadData, Restart: false}
  77. c_log.GlobalLogger.Info("杀死任务,发送rpc结束信号:", killArgs)
  78. KillRpcClient, err := rpc.Dial("tcp", commonConfig.LocalConfig.Node.Ip+":"+commonConfig.CloudConfig.RpcPort)
  79. if err != nil {
  80. // 此处如果连接失败说明采集程序已经停止了
  81. lastCloudStatus = "NONE"
  82. c_log.GlobalLogger.Error("采集程序已经停止:", err)
  83. return
  84. }
  85. reply := 0
  86. if err = KillRpcClient.Call("KillService.Kill", killArgs, &reply); err != nil {
  87. c_log.GlobalLogger.Error("发送 rpc 请求到 master 报错:", err)
  88. }
  89. c_log.GlobalLogger.Info("结束任务后,将数据闭环平台配置置空。")
  90. commonConfig.PlatformConfig = commonConfig.PlatformConfigStruct{}
  91. if err = KillRpcClient.Close(); err != nil {
  92. // 不做处理
  93. }
  94. launchedFlag = false
  95. c_log.GlobalLogger.Info("数采程序关闭 - 成功。")
  96. }
  97. func main() {
  98. // 更新本地任务状态
  99. go pkg.GetLocalStatus(&localStatus, &lastLocalStatus, localTurnLength)
  100. // 更新云端任务状态
  101. go pkg.GetCloudStatus(&cloudStatus, &lastCloudStatus, cloudTurnLength)
  102. for {
  103. if launchedFlag { // 当前已启动master节点
  104. time.Sleep(time.Duration(cloudTurnLength) * time.Second)
  105. } else {
  106. time.Sleep(time.Duration(localTurnLength) * time.Second)
  107. }
  108. fmt.Println("localStatus: ", localStatus, "lastLocalStatus: ", lastLocalStatus)
  109. fmt.Println("cloudStatus: ", cloudStatus, "lastCloudStatus: ", lastCloudStatus)
  110. // 综合判断 cloudStatus 和 localStatus
  111. // cloudStatus
  112. // UN_CHANGE 没有新的任务,无需更改
  113. // CHANGE 有新的任务,需要杀死旧的数采任务并重启
  114. // NONE 设备没有配置任务,需要杀死旧的数采任务
  115. // localStatus
  116. // idle 空闲状态,此状态下不启动数采任务
  117. // running 繁忙状态,此状态需要启动数采任务
  118. // error 错误状态,此状态下不启动数采任务
  119. // 本地任务状态负责启停master
  120. if localStatus == "running" {
  121. if !launchedFlag { // 目前未启动数采程序
  122. c_log.GlobalLogger.Info("数采程序启动 - 开始。")
  123. startMasterNode()
  124. }
  125. } else if localStatus == "idle" {
  126. if !renewedFlag && launchedFlag && len(entity.TimeWindowConsumerQueue) == 0 {
  127. time.Sleep(time.Duration(waitStopLength) * time.Minute)
  128. c_log.GlobalLogger.Info("设备不在运行状态且没有待处理的数据,数采程序关闭 - 开始。")
  129. stopMasterNode()
  130. }
  131. } else if localStatus == "error" {
  132. c_log.GlobalLogger.Error("设备运行状态出错,停止数采程序。")
  133. stopMasterNode()
  134. } else {
  135. c_log.GlobalLogger.Error("未知的设备运行状态。【status】=", localStatus)
  136. }
  137. // 定期检查本地任务状态,执行续约,避免短时间内多次启停
  138. go renew()
  139. // 云端任务状态负责更新配置
  140. go pkg.GetCloudConfig(cloudStatus, lastCloudStatus, cloudTurnLength)
  141. }
  142. }