main.go 6.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166
  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. maxRetryCount = 10 // 查询配置最大重试次数
  31. )
  32. func init() {
  33. runtime.GOMAXPROCS(1)
  34. // 初始化日志配置
  35. c_log.InitLog(variable.LogDir, applicationName)
  36. // 初始化本地配置文件(第1处配置,在本地文件)
  37. commonConfig.InitLocalConfig(variable.LocalConfigPath)
  38. // 初始化Oss连接信息
  39. commonConfig.InitOssConfig()
  40. // 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)
  41. commonConfig.InitCloudConfig()
  42. // 初始化rpc客户端,用于杀死旧的采集程序
  43. // 初始化websocket配置
  44. commonConfig.InitWebsocketConfig()
  45. }
  46. func renew() {
  47. for {
  48. time.Sleep(time.Duration(renewTurnLength) * time.Second)
  49. if localStatus == "running" && !renewedFlag { // 设备处于运行状态,且尚未续约
  50. c_log.GlobalLogger.Info("设备仍处于运行状态,续约 - 开始。")
  51. if renewTimer != nil {
  52. renewTimer.Stop()
  53. }
  54. renewedFlag = true
  55. renewTimer = time.AfterFunc(time.Duration(RenewDur)*time.Minute, func() {
  56. renewedFlag = false
  57. })
  58. c_log.GlobalLogger.Infof("续约时间【%v】分钟 - 成功。", RenewDur)
  59. }
  60. }
  61. }
  62. func startMasterNode() {
  63. c_log.GlobalLogger.Info("获取数据闭环平台最新配置。")
  64. if commonConfig.CheckPlatformConfigStatus(maxRetryCount) {
  65. c_log.GlobalLogger.Info("查询到数据闭环平台有配置任务。")
  66. commonConfig.InitPlatformConfig()
  67. }
  68. if _, err := util.ExecuteWithPath(commonConfig.LocalConfig.RestartCmd.Dir, commonConfig.LocalConfig.RestartCmd.Name, commonConfig.LocalConfig.RestartCmd.Args...); err != nil {
  69. 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)
  70. os.Exit(-1)
  71. }
  72. 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)
  73. launchedFlag = true
  74. c_log.GlobalLogger.Info("数采程序启动 - 成功。")
  75. }
  76. func stopMasterNode() {
  77. // 发送rpc信号杀死采集程序
  78. var killArgs commonService.KillSignal
  79. killArgs = commonService.KillSignal{NodeName: "master", DropUploadData: commonConfig.PlatformConfig.DropUploadData, Restart: false}
  80. c_log.GlobalLogger.Info("杀死任务,发送rpc结束信号:", killArgs)
  81. KillRpcClient, err := rpc.Dial("tcp", commonConfig.LocalConfig.Node.Ip+":"+commonConfig.CloudConfig.RpcPort)
  82. if err != nil {
  83. // 此处如果连接失败说明采集程序已经停止了
  84. lastCloudStatus = "NONE"
  85. c_log.GlobalLogger.Error("采集程序已经停止:", err)
  86. return
  87. }
  88. reply := 0
  89. if err = KillRpcClient.Call("KillService.Kill", killArgs, &reply); err != nil {
  90. c_log.GlobalLogger.Error("发送 rpc 请求到 master 报错:", err)
  91. }
  92. c_log.GlobalLogger.Info("结束任务后,将数据闭环平台配置置空。")
  93. commonConfig.PlatformConfig = commonConfig.PlatformConfigStruct{}
  94. if err = KillRpcClient.Close(); err != nil {
  95. // 不做处理
  96. }
  97. launchedFlag = false
  98. c_log.GlobalLogger.Info("数采程序关闭 - 成功。")
  99. }
  100. func main() {
  101. // 更新本地任务状态
  102. go pkg.GetLocalStatus(&localStatus, &lastLocalStatus, localTurnLength)
  103. // 更新云端任务状态
  104. go pkg.GetCloudStatus(&cloudStatus, &lastCloudStatus, cloudTurnLength)
  105. // 定期检查本地任务状态,执行续约,避免短时间内多次启停
  106. go renew()
  107. // 云端任务状态负责更新配置
  108. go pkg.GetCloudConfig(&cloudStatus, &lastCloudStatus, cloudTurnLength)
  109. for {
  110. if launchedFlag { // 当前已启动master节点
  111. time.Sleep(time.Duration(cloudTurnLength) * time.Second)
  112. } else {
  113. time.Sleep(time.Duration(localTurnLength) * time.Second)
  114. }
  115. fmt.Println("localStatus: ", localStatus, "lastLocalStatus: ", lastLocalStatus)
  116. fmt.Println("cloudStatus: ", cloudStatus, "lastCloudStatus: ", lastCloudStatus)
  117. // 综合判断 cloudStatus 和 localStatus
  118. // cloudStatus
  119. // UN_CHANGE 没有新的任务,无需更改
  120. // CHANGE 有新的任务,需要杀死旧的数采任务并重启
  121. // NONE 设备没有配置任务,需要杀死旧的数采任务
  122. // localStatus
  123. // idle 空闲状态,此状态下不启动数采任务
  124. // running 繁忙状态,此状态需要启动数采任务
  125. // error 错误状态,此状态下不启动数采任务
  126. // 本地任务状态负责启停master
  127. if localStatus == "running" {
  128. if !launchedFlag { // 目前未启动数采程序
  129. c_log.GlobalLogger.Info("数采程序启动 - 开始。")
  130. startMasterNode()
  131. }
  132. } else if localStatus == "idle" {
  133. if !renewedFlag && launchedFlag && len(entity.TimeWindowConsumerQueue) == 0 {
  134. time.Sleep(time.Duration(waitStopLength) * time.Minute)
  135. 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。")
  136. stopMasterNode()
  137. }
  138. } else if localStatus == "error" {
  139. c_log.GlobalLogger.Error("设备运行状态出错,停止数采程序。")
  140. stopMasterNode()
  141. } else {
  142. c_log.GlobalLogger.Error("未知的设备运行状态。【status】=", localStatus)
  143. }
  144. }
  145. }