You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

serve.go 9.5 kB

10 months ago
10 months ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362
  1. package cmd
  2. import (
  3. "context"
  4. "fmt"
  5. "net"
  6. "os"
  7. "time"
  8. "github.com/go-co-op/gocron/v2"
  9. "gitlink.org.cn/cloudream/storage/agent/internal/http"
  10. "gitlink.org.cn/cloudream/storage/agent/internal/tickevent"
  11. "gitlink.org.cn/cloudream/common/pkgs/ioswitch/exec"
  12. "gitlink.org.cn/cloudream/common/pkgs/logger"
  13. "gitlink.org.cn/cloudream/common/pkgs/mq"
  14. cdssdk "gitlink.org.cn/cloudream/common/sdks/storage"
  15. "gitlink.org.cn/cloudream/storage/agent/internal/config"
  16. "gitlink.org.cn/cloudream/storage/agent/internal/task"
  17. stgglb "gitlink.org.cn/cloudream/storage/common/globals"
  18. stgmod "gitlink.org.cn/cloudream/storage/common/models"
  19. "gitlink.org.cn/cloudream/storage/common/pkgs/accessstat"
  20. "gitlink.org.cn/cloudream/storage/common/pkgs/connectivity"
  21. "gitlink.org.cn/cloudream/storage/common/pkgs/distlock"
  22. "gitlink.org.cn/cloudream/storage/common/pkgs/downloader"
  23. "gitlink.org.cn/cloudream/storage/common/pkgs/downloader/strategy"
  24. agtrpc "gitlink.org.cn/cloudream/storage/common/pkgs/grpc/agent"
  25. "gitlink.org.cn/cloudream/storage/common/pkgs/metacache"
  26. "gitlink.org.cn/cloudream/storage/common/pkgs/storage/agtpool"
  27. "gitlink.org.cn/cloudream/storage/common/pkgs/sysevent"
  28. "gitlink.org.cn/cloudream/storage/common/pkgs/uploader"
  29. "google.golang.org/grpc"
  30. agtmq "gitlink.org.cn/cloudream/storage/common/pkgs/mq/agent"
  31. coormq "gitlink.org.cn/cloudream/storage/common/pkgs/mq/coordinator"
  32. grpcsvc "gitlink.org.cn/cloudream/storage/agent/internal/grpc"
  33. cmdsvc "gitlink.org.cn/cloudream/storage/agent/internal/mq"
  34. )
  35. func serve(configPath string) {
  36. err := config.Init(configPath)
  37. if err != nil {
  38. fmt.Printf("init config failed, err: %s", err.Error())
  39. os.Exit(1)
  40. }
  41. err = logger.Init(&config.Cfg().Logger)
  42. if err != nil {
  43. fmt.Printf("init logger failed, err: %s", err.Error())
  44. os.Exit(1)
  45. }
  46. stgglb.InitLocal(&config.Cfg().Local)
  47. stgglb.InitMQPool(config.Cfg().RabbitMQ)
  48. stgglb.InitAgentRPCPool(&agtrpc.PoolConfig{})
  49. // 获取Hub配置
  50. hubCfg := downloadHubConfig()
  51. // 初始化存储服务管理器
  52. stgAgts := agtpool.NewPool()
  53. for _, stg := range hubCfg.Storages {
  54. err := stgAgts.SetupAgent(stg)
  55. if err != nil {
  56. fmt.Printf("init storage %v: %v", stg.Storage.String(), err)
  57. os.Exit(1)
  58. }
  59. }
  60. // 初始化执行器
  61. worker := exec.NewWorker()
  62. // 初始化HTTP服务
  63. httpSvr, err := http.NewServer(config.Cfg().ListenAddr, http.NewService(&worker, stgAgts))
  64. if err != nil {
  65. logger.Fatalf("new http server failed, err: %s", err.Error())
  66. }
  67. go serveHTTP(httpSvr)
  68. // 启动网络连通性检测,并就地检测一次
  69. conCol := connectivity.NewCollector(&config.Cfg().Connectivity, func(collector *connectivity.Collector) {
  70. log := logger.WithField("Connectivity", "")
  71. coorCli, err := stgglb.CoordinatorMQPool.Acquire()
  72. if err != nil {
  73. log.Warnf("acquire coordinator mq failed, err: %s", err.Error())
  74. return
  75. }
  76. defer stgglb.CoordinatorMQPool.Release(coorCli)
  77. cons := collector.GetAll()
  78. hubCons := make([]cdssdk.HubConnectivity, 0, len(cons))
  79. for _, con := range cons {
  80. var delay *float32
  81. if con.Latency != nil {
  82. v := float32(con.Latency.Microseconds()) / 1000
  83. delay = &v
  84. }
  85. hubCons = append(hubCons, cdssdk.HubConnectivity{
  86. FromHubID: *stgglb.Local.HubID,
  87. ToHubID: con.ToHubID,
  88. Latency: delay,
  89. TestTime: con.TestTime,
  90. })
  91. }
  92. _, err = coorCli.UpdateHubConnectivities(coormq.ReqUpdateHubConnectivities(hubCons))
  93. if err != nil {
  94. log.Warnf("update hub connectivities: %v", err)
  95. }
  96. })
  97. conCol.CollectInPlace()
  98. // 初始化元数据缓存服务
  99. metacacheHost := metacache.NewHost()
  100. go metacacheHost.Serve()
  101. stgMeta := metacacheHost.AddStorageMeta()
  102. hubMeta := metacacheHost.AddHubMeta()
  103. conMeta := metacacheHost.AddConnectivity()
  104. // 启动访问统计服务
  105. acStat := accessstat.NewAccessStat(accessstat.Config{
  106. // TODO 考虑放到配置里
  107. ReportInterval: time.Second * 10,
  108. })
  109. go serveAccessStat(acStat)
  110. // 初始化分布式锁服务
  111. distlock, err := distlock.NewService(&config.Cfg().DistLock)
  112. if err != nil {
  113. logger.Fatalf("new ipfs failed, err: %s", err.Error())
  114. }
  115. // 初始化下载策略选择器
  116. strgSel := strategy.NewSelector(config.Cfg().DownloadStrategy, stgMeta, hubMeta, conMeta)
  117. // 初始化下载器
  118. dlder := downloader.NewDownloader(config.Cfg().Downloader, &conCol, stgAgts, strgSel)
  119. // 初始化上传器
  120. uploader := uploader.NewUploader(distlock, &conCol, stgAgts, stgMeta)
  121. // 初始化任务管理器
  122. taskMgr := task.NewManager(distlock, &conCol, &dlder, acStat, stgAgts, uploader)
  123. // 初始化系统事件发布器
  124. evtPub, err := sysevent.NewPublisher(sysevent.ConfigFromMQConfig(config.Cfg().RabbitMQ), &stgmod.SourceHub{
  125. HubID: hubCfg.Hub.HubID,
  126. HubName: hubCfg.Hub.Name,
  127. })
  128. if err != nil {
  129. logger.Errorf("new sysevent publisher: %v", err)
  130. os.Exit(1)
  131. }
  132. go servePublisher(evtPub)
  133. // 初始化定时任务执行器
  134. sch := setupTickTask(stgAgts, evtPub)
  135. defer sch.Shutdown()
  136. // 启动命令服务器
  137. // TODO 需要设计AgentID持久化机制
  138. agtSvr, err := agtmq.NewServer(cmdsvc.NewService(&taskMgr, stgAgts), config.Cfg().ID, config.Cfg().RabbitMQ)
  139. if err != nil {
  140. logger.Fatalf("new agent server failed, err: %s", err.Error())
  141. }
  142. agtSvr.OnError(func(err error) {
  143. logger.Warnf("agent server err: %s", err.Error())
  144. })
  145. go serveAgentServer(agtSvr)
  146. // 启动GRPC服务
  147. listenAddr := config.Cfg().GRPC.MakeListenAddress()
  148. lis, err := net.Listen("tcp", listenAddr)
  149. if err != nil {
  150. logger.Fatalf("listen on %s failed, err: %s", listenAddr, err.Error())
  151. }
  152. s := grpc.NewServer()
  153. agtrpc.RegisterAgentServer(s, grpcsvc.NewService(&worker, stgAgts))
  154. go serveGRPC(s, lis)
  155. go serveDistLock(distlock)
  156. foever := make(chan struct{})
  157. <-foever
  158. }
  159. func downloadHubConfig() coormq.GetHubConfigResp {
  160. coorCli, err := stgglb.CoordinatorMQPool.Acquire()
  161. if err != nil {
  162. logger.Errorf("new coordinator client: %v", err)
  163. os.Exit(1)
  164. }
  165. defer stgglb.CoordinatorMQPool.Release(coorCli)
  166. cfgResp, err := coorCli.GetHubConfig(coormq.ReqGetHubConfig(cdssdk.HubID(config.Cfg().ID)))
  167. if err != nil {
  168. logger.Errorf("getting hub config: %v", err)
  169. os.Exit(1)
  170. }
  171. return *cfgResp
  172. }
  173. func servePublisher(evtPub *sysevent.Publisher) {
  174. logger.Info("start serving sysevent publisher")
  175. ch := evtPub.Start()
  176. loop:
  177. for {
  178. val, err := ch.Receive().Wait(context.Background())
  179. if err != nil {
  180. logger.Errorf("sysevent publisher stopped with error: %s", err.Error())
  181. break
  182. }
  183. switch val := val.(type) {
  184. case sysevent.PublishError:
  185. logger.Errorf("publishing event: %v", val)
  186. case sysevent.PublisherExited:
  187. if val.Err != nil {
  188. logger.Errorf("publisher exited with error: %v", val.Err)
  189. } else {
  190. logger.Info("publisher exited")
  191. }
  192. break loop
  193. case sysevent.OtherError:
  194. logger.Errorf("sysevent: %v", val)
  195. }
  196. }
  197. logger.Info("sysevent publisher stopped")
  198. // TODO 仅简单结束了程序
  199. os.Exit(1)
  200. }
  201. func setupTickTask(agtPool *agtpool.AgentPool, evtPub *sysevent.Publisher) gocron.Scheduler {
  202. sch, err := gocron.NewScheduler()
  203. if err != nil {
  204. logger.Errorf("new cron scheduler: %s", err.Error())
  205. os.Exit(1)
  206. }
  207. sch.NewJob(gocron.DailyJob(1, gocron.NewAtTimes(
  208. gocron.NewAtTime(0, 0, 0),
  209. )), gocron.NewTask(tickevent.ReportStorageStats, agtPool, evtPub))
  210. sch.NewJob(gocron.DailyJob(1, gocron.NewAtTimes(
  211. gocron.NewAtTime(0, 0, 1),
  212. )), gocron.NewTask(tickevent.ReportHubTransferStats, evtPub))
  213. sch.NewJob(gocron.DailyJob(1, gocron.NewAtTimes(
  214. gocron.NewAtTime(0, 0, 2),
  215. )), gocron.NewTask(tickevent.ReportHubStorageTransferStats, evtPub))
  216. return sch
  217. }
  218. func serveAgentServer(server *agtmq.Server) {
  219. logger.Info("start serving command server")
  220. ch := server.Start()
  221. loop:
  222. for {
  223. val, err := ch.Receive()
  224. if err != nil {
  225. logger.Errorf("command server stopped with error: %s", err.Error())
  226. break
  227. }
  228. switch val := val.(type) {
  229. case error:
  230. logger.Errorf("rabbitmq connect with error: %v", val)
  231. case mq.ServerExit:
  232. if val.Error != nil {
  233. logger.Errorf("rabbitmq server exit with error: %v", val.Error)
  234. } else {
  235. logger.Info("rabbitmq server exit")
  236. }
  237. break loop
  238. }
  239. }
  240. logger.Info("command server stopped")
  241. // TODO 仅简单结束了程序
  242. os.Exit(1)
  243. }
  244. func serveGRPC(s *grpc.Server, lis net.Listener) {
  245. logger.Info("start serving grpc")
  246. err := s.Serve(lis)
  247. if err != nil {
  248. logger.Errorf("grpc stopped with error: %s", err.Error())
  249. }
  250. logger.Info("grpc stopped")
  251. // TODO 仅简单结束了程序
  252. os.Exit(1)
  253. }
  254. func serveHTTP(server *http.Server) {
  255. logger.Info("start serving http")
  256. err := server.Serve()
  257. if err != nil {
  258. logger.Errorf("http stopped with error: %s", err.Error())
  259. }
  260. logger.Info("http stopped")
  261. // TODO 仅简单结束了程序
  262. os.Exit(1)
  263. }
  264. func serveDistLock(svc *distlock.Service) {
  265. logger.Info("start serving distlock")
  266. err := svc.Serve()
  267. if err != nil {
  268. logger.Errorf("distlock stopped with error: %s", err.Error())
  269. }
  270. logger.Info("distlock stopped")
  271. // TODO 仅简单结束了程序
  272. os.Exit(1)
  273. }
  274. func serveAccessStat(svc *accessstat.AccessStat) {
  275. logger.Info("start serving access stat")
  276. ch := svc.Start()
  277. loop:
  278. for {
  279. val, err := ch.Receive()
  280. if err != nil {
  281. logger.Errorf("access stat stopped with error: %v", err)
  282. break
  283. }
  284. switch val := val.(type) {
  285. case error:
  286. logger.Errorf("access stat stopped with error: %v", val)
  287. break loop
  288. }
  289. }
  290. logger.Info("access stat stopped")
  291. // TODO 仅简单结束了程序
  292. os.Exit(1)
  293. }

本项目旨在将云际存储公共基础设施化,使个人及企业可低门槛使用高效的云际存储服务(安装开箱即用云际存储客户端即可,无需关注其他组件的部署),同时支持用户灵活便捷定制云际存储的功能细节。