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

391 lines
15 KiB

5 years ago
6 years ago
6 years ago
6 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
12 years ago
5 years ago
4 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
4 years ago
5 years ago
5 years ago
4 years ago
4 years ago
5 years ago
4 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
5 years ago
  1. package command
  2. import (
  3. "fmt"
  4. "net/http"
  5. httppprof "net/http/pprof"
  6. "os"
  7. "runtime/pprof"
  8. "strconv"
  9. "strings"
  10. "time"
  11. "github.com/seaweedfs/seaweedfs/weed/storage/types"
  12. "github.com/spf13/viper"
  13. "google.golang.org/grpc"
  14. "github.com/seaweedfs/seaweedfs/weed/util/grace"
  15. "github.com/seaweedfs/seaweedfs/weed/pb"
  16. "github.com/seaweedfs/seaweedfs/weed/security"
  17. "github.com/seaweedfs/seaweedfs/weed/util/httpdown"
  18. "google.golang.org/grpc/reflection"
  19. "github.com/seaweedfs/seaweedfs/weed/glog"
  20. "github.com/seaweedfs/seaweedfs/weed/pb/volume_server_pb"
  21. weed_server "github.com/seaweedfs/seaweedfs/weed/server"
  22. stats_collect "github.com/seaweedfs/seaweedfs/weed/stats"
  23. "github.com/seaweedfs/seaweedfs/weed/storage"
  24. "github.com/seaweedfs/seaweedfs/weed/util"
  25. )
  26. var (
  27. v VolumeServerOptions
  28. )
  29. type VolumeServerOptions struct {
  30. port *int
  31. portGrpc *int
  32. publicPort *int
  33. folders []string
  34. folderMaxLimits []int32
  35. idxFolder *string
  36. ip *string
  37. publicUrl *string
  38. bindIp *string
  39. mastersString *string
  40. masters []pb.ServerAddress
  41. idleConnectionTimeout *int
  42. dataCenter *string
  43. rack *string
  44. whiteList []string
  45. indexType *string
  46. diskType *string
  47. fixJpgOrientation *bool
  48. readMode *string
  49. cpuProfile *string
  50. memProfile *string
  51. compactionMBPerSecond *int
  52. fileSizeLimitMB *int
  53. concurrentUploadLimitMB *int
  54. concurrentDownloadLimitMB *int
  55. pprof *bool
  56. preStopSeconds *int
  57. metricsHttpPort *int
  58. // pulseSeconds *int
  59. inflightUploadDataTimeout *time.Duration
  60. hasSlowRead *bool
  61. readBufferSizeMB *int
  62. }
  63. func init() {
  64. cmdVolume.Run = runVolume // break init cycle
  65. v.port = cmdVolume.Flag.Int("port", 8080, "http listen port")
  66. v.portGrpc = cmdVolume.Flag.Int("port.grpc", 0, "grpc listen port")
  67. v.publicPort = cmdVolume.Flag.Int("port.public", 0, "port opened to public")
  68. v.ip = cmdVolume.Flag.String("ip", util.DetectedHostAddress(), "ip or server name, also used as identifier")
  69. v.publicUrl = cmdVolume.Flag.String("publicUrl", "", "Publicly accessible address")
  70. v.bindIp = cmdVolume.Flag.String("ip.bind", "", "ip address to bind to. If empty, default to same as -ip option.")
  71. v.mastersString = cmdVolume.Flag.String("mserver", "localhost:9333", "comma-separated master servers")
  72. v.preStopSeconds = cmdVolume.Flag.Int("preStopSeconds", 10, "number of seconds between stop send heartbeats and stop volume server")
  73. // v.pulseSeconds = cmdVolume.Flag.Int("pulseSeconds", 5, "number of seconds between heartbeats, must be smaller than or equal to the master's setting")
  74. v.idleConnectionTimeout = cmdVolume.Flag.Int("idleTimeout", 30, "connection idle seconds")
  75. v.dataCenter = cmdVolume.Flag.String("dataCenter", "", "current volume server's data center name")
  76. v.rack = cmdVolume.Flag.String("rack", "", "current volume server's rack name")
  77. v.indexType = cmdVolume.Flag.String("index", "memory", "Choose [memory|leveldb|leveldbMedium|leveldbLarge] mode for memory~performance balance.")
  78. v.diskType = cmdVolume.Flag.String("disk", "", "[hdd|ssd|<tag>] hard drive or solid state drive or any tag")
  79. v.fixJpgOrientation = cmdVolume.Flag.Bool("images.fix.orientation", false, "Adjust jpg orientation when uploading.")
  80. v.readMode = cmdVolume.Flag.String("readMode", "proxy", "[local|proxy|redirect] how to deal with non-local volume: 'not found|proxy to remote node|redirect volume location'.")
  81. v.cpuProfile = cmdVolume.Flag.String("cpuprofile", "", "cpu profile output file")
  82. v.memProfile = cmdVolume.Flag.String("memprofile", "", "memory profile output file")
  83. v.compactionMBPerSecond = cmdVolume.Flag.Int("compactionMBps", 0, "limit background compaction or copying speed in mega bytes per second")
  84. v.fileSizeLimitMB = cmdVolume.Flag.Int("fileSizeLimitMB", 256, "limit file size to avoid out of memory")
  85. v.concurrentUploadLimitMB = cmdVolume.Flag.Int("concurrentUploadLimitMB", 256, "limit total concurrent upload size")
  86. v.concurrentDownloadLimitMB = cmdVolume.Flag.Int("concurrentDownloadLimitMB", 256, "limit total concurrent download size")
  87. v.pprof = cmdVolume.Flag.Bool("pprof", false, "enable pprof http handlers. precludes --memprofile and --cpuprofile")
  88. v.metricsHttpPort = cmdVolume.Flag.Int("metricsPort", 0, "Prometheus metrics listen port")
  89. v.idxFolder = cmdVolume.Flag.String("dir.idx", "", "directory to store .idx files")
  90. v.inflightUploadDataTimeout = cmdVolume.Flag.Duration("inflightUploadDataTimeout", 60*time.Second, "inflight upload data wait timeout of volume servers")
  91. v.hasSlowRead = cmdVolume.Flag.Bool("hasSlowRead", false, "<experimental> if true, this prevents slow reads from blocking other requests, but large file read P99 latency will increase.")
  92. v.readBufferSizeMB = cmdVolume.Flag.Int("readBufferSizeMB", 4, "<experimental> larger values can optimize query performance but will increase some memory usage,Use with hasSlowRead normally.")
  93. }
  94. var cmdVolume = &Command{
  95. UsageLine: "volume -port=8080 -dir=/tmp -max=5 -ip=server_name -mserver=localhost:9333",
  96. Short: "start a volume server",
  97. Long: `start a volume server to provide storage spaces
  98. `,
  99. }
  100. var (
  101. volumeFolders = cmdVolume.Flag.String("dir", os.TempDir(), "directories to store data files. dir[,dir]...")
  102. maxVolumeCounts = cmdVolume.Flag.String("max", "8", "maximum numbers of volumes, count[,count]... If set to zero, the limit will be auto configured as free disk space divided by volume size.")
  103. volumeWhiteListOption = cmdVolume.Flag.String("whiteList", "", "comma separated Ip addresses having write permission. No limit if empty.")
  104. minFreeSpacePercent = cmdVolume.Flag.String("minFreeSpacePercent", "1", "minimum free disk space (default to 1%). Low disk space will mark all volumes as ReadOnly (deprecated, use minFreeSpace instead).")
  105. minFreeSpace = cmdVolume.Flag.String("minFreeSpace", "", "min free disk space (value<=100 as percentage like 1, other as human readable bytes, like 10GiB). Low disk space will mark all volumes as ReadOnly.")
  106. )
  107. func runVolume(cmd *Command, args []string) bool {
  108. util.LoadConfiguration("security", false)
  109. // If --pprof is set we assume the caller wants to be able to collect
  110. // cpu and memory profiles via go tool pprof
  111. if !*v.pprof {
  112. grace.SetupProfiling(*v.cpuProfile, *v.memProfile)
  113. }
  114. go stats_collect.StartMetricsServer(*v.metricsHttpPort)
  115. minFreeSpaces := util.MustParseMinFreeSpace(*minFreeSpace, *minFreeSpacePercent)
  116. v.masters = pb.ServerAddresses(*v.mastersString).ToAddresses()
  117. v.startVolumeServer(*volumeFolders, *maxVolumeCounts, *volumeWhiteListOption, minFreeSpaces)
  118. return true
  119. }
  120. func (v VolumeServerOptions) startVolumeServer(volumeFolders, maxVolumeCounts, volumeWhiteListOption string, minFreeSpaces []util.MinFreeSpace) {
  121. // Set multiple folders and each folder's max volume count limit'
  122. v.folders = strings.Split(volumeFolders, ",")
  123. for _, folder := range v.folders {
  124. if err := util.TestFolderWritable(util.ResolvePath(folder)); err != nil {
  125. glog.Fatalf("Check Data Folder(-dir) Writable %s : %s", folder, err)
  126. }
  127. }
  128. // set max
  129. maxCountStrings := strings.Split(maxVolumeCounts, ",")
  130. for _, maxString := range maxCountStrings {
  131. if max, e := strconv.ParseInt(maxString, 10, 64); e == nil {
  132. v.folderMaxLimits = append(v.folderMaxLimits, int32(max))
  133. } else {
  134. glog.Fatalf("The max specified in -max not a valid number %s", maxString)
  135. }
  136. }
  137. if len(v.folderMaxLimits) == 1 && len(v.folders) > 1 {
  138. for i := 0; i < len(v.folders)-1; i++ {
  139. v.folderMaxLimits = append(v.folderMaxLimits, v.folderMaxLimits[0])
  140. }
  141. }
  142. if len(v.folders) != len(v.folderMaxLimits) {
  143. glog.Fatalf("%d directories by -dir, but only %d max is set by -max", len(v.folders), len(v.folderMaxLimits))
  144. }
  145. if len(minFreeSpaces) == 1 && len(v.folders) > 1 {
  146. for i := 0; i < len(v.folders)-1; i++ {
  147. minFreeSpaces = append(minFreeSpaces, minFreeSpaces[0])
  148. }
  149. }
  150. if len(v.folders) != len(minFreeSpaces) {
  151. glog.Fatalf("%d directories by -dir, but only %d minFreeSpacePercent is set by -minFreeSpacePercent", len(v.folders), len(minFreeSpaces))
  152. }
  153. // set disk types
  154. var diskTypes []types.DiskType
  155. diskTypeStrings := strings.Split(*v.diskType, ",")
  156. for _, diskTypeString := range diskTypeStrings {
  157. diskTypes = append(diskTypes, types.ToDiskType(diskTypeString))
  158. }
  159. if len(diskTypes) == 1 && len(v.folders) > 1 {
  160. for i := 0; i < len(v.folders)-1; i++ {
  161. diskTypes = append(diskTypes, diskTypes[0])
  162. }
  163. }
  164. if len(v.folders) != len(diskTypes) {
  165. glog.Fatalf("%d directories by -dir, but only %d disk types is set by -disk", len(v.folders), len(diskTypes))
  166. }
  167. // security related white list configuration
  168. v.whiteList = util.StringSplit(volumeWhiteListOption, ",")
  169. if *v.ip == "" {
  170. *v.ip = util.DetectedHostAddress()
  171. glog.V(0).Infof("detected volume server ip address: %v", *v.ip)
  172. }
  173. if *v.bindIp == "" {
  174. *v.bindIp = *v.ip
  175. }
  176. if *v.publicPort == 0 {
  177. *v.publicPort = *v.port
  178. }
  179. if *v.portGrpc == 0 {
  180. *v.portGrpc = 10000 + *v.port
  181. }
  182. if *v.publicUrl == "" {
  183. *v.publicUrl = util.JoinHostPort(*v.ip, *v.publicPort)
  184. }
  185. volumeMux := http.NewServeMux()
  186. publicVolumeMux := volumeMux
  187. if v.isSeparatedPublicPort() {
  188. publicVolumeMux = http.NewServeMux()
  189. }
  190. if *v.pprof {
  191. volumeMux.HandleFunc("/debug/pprof/", httppprof.Index)
  192. volumeMux.HandleFunc("/debug/pprof/cmdline", httppprof.Cmdline)
  193. volumeMux.HandleFunc("/debug/pprof/profile", httppprof.Profile)
  194. volumeMux.HandleFunc("/debug/pprof/symbol", httppprof.Symbol)
  195. volumeMux.HandleFunc("/debug/pprof/trace", httppprof.Trace)
  196. }
  197. volumeNeedleMapKind := storage.NeedleMapInMemory
  198. switch *v.indexType {
  199. case "leveldb":
  200. volumeNeedleMapKind = storage.NeedleMapLevelDb
  201. case "leveldbMedium":
  202. volumeNeedleMapKind = storage.NeedleMapLevelDbMedium
  203. case "leveldbLarge":
  204. volumeNeedleMapKind = storage.NeedleMapLevelDbLarge
  205. }
  206. volumeServer := weed_server.NewVolumeServer(volumeMux, publicVolumeMux,
  207. *v.ip, *v.port, *v.portGrpc, *v.publicUrl,
  208. v.folders, v.folderMaxLimits, minFreeSpaces, diskTypes,
  209. *v.idxFolder,
  210. volumeNeedleMapKind,
  211. v.masters, 5, *v.dataCenter, *v.rack,
  212. v.whiteList,
  213. *v.fixJpgOrientation, *v.readMode,
  214. *v.compactionMBPerSecond,
  215. *v.fileSizeLimitMB,
  216. int64(*v.concurrentUploadLimitMB)*1024*1024,
  217. int64(*v.concurrentDownloadLimitMB)*1024*1024,
  218. *v.inflightUploadDataTimeout,
  219. *v.hasSlowRead,
  220. *v.readBufferSizeMB,
  221. )
  222. // starting grpc server
  223. grpcS := v.startGrpcService(volumeServer)
  224. // starting public http server
  225. var publicHttpDown httpdown.Server
  226. if v.isSeparatedPublicPort() {
  227. publicHttpDown = v.startPublicHttpService(publicVolumeMux)
  228. if nil == publicHttpDown {
  229. glog.Fatalf("start public http service failed")
  230. }
  231. }
  232. // starting the cluster http server
  233. clusterHttpServer := v.startClusterHttpService(volumeMux)
  234. grace.OnReload(volumeServer.LoadNewVolumes)
  235. stopChan := make(chan bool)
  236. grace.OnInterrupt(func() {
  237. fmt.Println("volume server has been killed")
  238. // Stop heartbeats
  239. if !volumeServer.StopHeartbeat() {
  240. volumeServer.SetStopping()
  241. glog.V(0).Infof("stop send heartbeat and wait %d seconds until shutdown ...", *v.preStopSeconds)
  242. time.Sleep(time.Duration(*v.preStopSeconds) * time.Second)
  243. }
  244. shutdown(publicHttpDown, clusterHttpServer, grpcS, volumeServer)
  245. stopChan <- true
  246. })
  247. select {
  248. case <-stopChan:
  249. }
  250. }
  251. func shutdown(publicHttpDown httpdown.Server, clusterHttpServer httpdown.Server, grpcS *grpc.Server, volumeServer *weed_server.VolumeServer) {
  252. // firstly, stop the public http service to prevent from receiving new user request
  253. if nil != publicHttpDown {
  254. glog.V(0).Infof("stop public http server ... ")
  255. if err := publicHttpDown.Stop(); err != nil {
  256. glog.Warningf("stop the public http server failed, %v", err)
  257. }
  258. }
  259. glog.V(0).Infof("graceful stop cluster http server ... ")
  260. if err := clusterHttpServer.Stop(); err != nil {
  261. glog.Warningf("stop the cluster http server failed, %v", err)
  262. }
  263. glog.V(0).Infof("graceful stop gRPC ...")
  264. grpcS.GracefulStop()
  265. volumeServer.Shutdown()
  266. pprof.StopCPUProfile()
  267. }
  268. // check whether configure the public port
  269. func (v VolumeServerOptions) isSeparatedPublicPort() bool {
  270. return *v.publicPort != *v.port
  271. }
  272. func (v VolumeServerOptions) startGrpcService(vs volume_server_pb.VolumeServerServer) *grpc.Server {
  273. grpcPort := *v.portGrpc
  274. grpcL, err := util.NewListener(util.JoinHostPort(*v.bindIp, grpcPort), 0)
  275. if err != nil {
  276. glog.Fatalf("failed to listen on grpc port %d: %v", grpcPort, err)
  277. }
  278. grpcS := pb.NewGrpcServer(security.LoadServerTLS(util.GetViper(), "grpc.volume"))
  279. volume_server_pb.RegisterVolumeServerServer(grpcS, vs)
  280. reflection.Register(grpcS)
  281. go func() {
  282. if err := grpcS.Serve(grpcL); err != nil {
  283. glog.Fatalf("start gRPC service failed, %s", err)
  284. }
  285. }()
  286. return grpcS
  287. }
  288. func (v VolumeServerOptions) startPublicHttpService(handler http.Handler) httpdown.Server {
  289. publicListeningAddress := util.JoinHostPort(*v.bindIp, *v.publicPort)
  290. glog.V(0).Infoln("Start Seaweed volume server", util.Version(), "public at", publicListeningAddress)
  291. publicListener, e := util.NewListener(publicListeningAddress, time.Duration(*v.idleConnectionTimeout)*time.Second)
  292. if e != nil {
  293. glog.Fatalf("Volume server listener error:%v", e)
  294. }
  295. pubHttp := httpdown.HTTP{StopTimeout: 5 * time.Minute, KillTimeout: 5 * time.Minute}
  296. publicHttpDown := pubHttp.Serve(&http.Server{Handler: handler}, publicListener)
  297. go func() {
  298. if err := publicHttpDown.Wait(); err != nil {
  299. glog.Errorf("public http down wait failed, %v", err)
  300. }
  301. }()
  302. return publicHttpDown
  303. }
  304. func (v VolumeServerOptions) startClusterHttpService(handler http.Handler) httpdown.Server {
  305. var (
  306. certFile, keyFile string
  307. )
  308. if viper.GetString("https.volume.key") != "" {
  309. certFile = viper.GetString("https.volume.cert")
  310. keyFile = viper.GetString("https.volume.key")
  311. }
  312. listeningAddress := util.JoinHostPort(*v.bindIp, *v.port)
  313. glog.V(0).Infof("Start Seaweed volume server %s at %s", util.Version(), listeningAddress)
  314. listener, e := util.NewListener(listeningAddress, time.Duration(*v.idleConnectionTimeout)*time.Second)
  315. if e != nil {
  316. glog.Fatalf("Volume server listener error:%v", e)
  317. }
  318. httpDown := httpdown.HTTP{
  319. KillTimeout: time.Minute,
  320. StopTimeout: 30 * time.Second,
  321. CertFile: certFile,
  322. KeyFile: keyFile}
  323. httpS := &http.Server{Handler: handler}
  324. if viper.GetString("https.volume.ca") != "" {
  325. clientCertFile := viper.GetString("https.volume.ca")
  326. httpS.TLSConfig = security.LoadClientTLSHTTP(clientCertFile)
  327. }
  328. clusterHttpServer := httpDown.Serve(httpS, listener)
  329. go func() {
  330. if e := clusterHttpServer.Wait(); e != nil {
  331. glog.Fatalf("Volume server fail to serve: %v", e)
  332. }
  333. }()
  334. return clusterHttpServer
  335. }