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.

261 lines
8.5 KiB

5 years ago
3 years ago
3 years ago
4 years ago
4 years ago
  1. package weed_server
  2. import (
  3. "fmt"
  4. "os"
  5. "time"
  6. "github.com/chrislusf/seaweedfs/weed/operation"
  7. "google.golang.org/grpc"
  8. "github.com/chrislusf/seaweedfs/weed/pb"
  9. "github.com/chrislusf/seaweedfs/weed/security"
  10. "github.com/chrislusf/seaweedfs/weed/storage/backend"
  11. "github.com/chrislusf/seaweedfs/weed/storage/erasure_coding"
  12. "golang.org/x/net/context"
  13. "github.com/chrislusf/seaweedfs/weed/glog"
  14. "github.com/chrislusf/seaweedfs/weed/pb/master_pb"
  15. "github.com/chrislusf/seaweedfs/weed/util"
  16. )
  17. func (vs *VolumeServer) GetMaster() pb.ServerAddress {
  18. return vs.currentMaster
  19. }
  20. func (vs *VolumeServer) checkWithMaster() (err error) {
  21. for {
  22. for _, master := range vs.SeedMasterNodes {
  23. err = operation.WithMasterServerClient(false, master, vs.grpcDialOption, func(masterClient master_pb.SeaweedClient) error {
  24. resp, err := masterClient.GetMasterConfiguration(context.Background(), &master_pb.GetMasterConfigurationRequest{})
  25. if err != nil {
  26. return fmt.Errorf("get master %s configuration: %v", master, err)
  27. }
  28. vs.metricsAddress, vs.metricsIntervalSec = resp.MetricsAddress, int(resp.MetricsIntervalSeconds)
  29. backend.LoadFromPbStorageBackends(resp.StorageBackends)
  30. return nil
  31. })
  32. if err == nil {
  33. return
  34. } else {
  35. glog.V(0).Infof("checkWithMaster %s: %v", master, err)
  36. }
  37. }
  38. time.Sleep(1790 * time.Millisecond)
  39. }
  40. }
  41. func (vs *VolumeServer) heartbeat() {
  42. glog.V(0).Infof("Volume server start with seed master nodes: %v", vs.SeedMasterNodes)
  43. vs.store.SetDataCenter(vs.dataCenter)
  44. vs.store.SetRack(vs.rack)
  45. grpcDialOption := security.LoadClientTLS(util.GetViper(), "grpc.volume")
  46. var err error
  47. var newLeader pb.ServerAddress
  48. for vs.isHeartbeating {
  49. for _, master := range vs.SeedMasterNodes {
  50. if newLeader != "" {
  51. // the new leader may actually is the same master
  52. // need to wait a bit before adding itself
  53. time.Sleep(3 * time.Second)
  54. master = newLeader
  55. }
  56. vs.store.MasterAddress = master
  57. newLeader, err = vs.doHeartbeat(master, grpcDialOption, time.Duration(vs.pulseSeconds)*time.Second)
  58. if err != nil {
  59. glog.V(0).Infof("heartbeat error: %v", err)
  60. time.Sleep(time.Duration(vs.pulseSeconds) * time.Second)
  61. newLeader = ""
  62. vs.store.MasterAddress = ""
  63. }
  64. if !vs.isHeartbeating {
  65. break
  66. }
  67. }
  68. }
  69. }
  70. func (vs *VolumeServer) StopHeartbeat() (isAlreadyStopping bool) {
  71. if !vs.isHeartbeating {
  72. return true
  73. }
  74. vs.isHeartbeating = false
  75. close(vs.stopChan)
  76. return false
  77. }
  78. func (vs *VolumeServer) doHeartbeat(masterAddress pb.ServerAddress, grpcDialOption grpc.DialOption, sleepInterval time.Duration) (newLeader pb.ServerAddress, err error) {
  79. ctx, cancel := context.WithCancel(context.Background())
  80. defer cancel()
  81. grpcConection, err := pb.GrpcDial(ctx, masterAddress.ToGrpcAddress(), grpcDialOption)
  82. if err != nil {
  83. return "", fmt.Errorf("fail to dial %s : %v", masterAddress, err)
  84. }
  85. defer grpcConection.Close()
  86. client := master_pb.NewSeaweedClient(grpcConection)
  87. stream, err := client.SendHeartbeat(ctx)
  88. if err != nil {
  89. glog.V(0).Infof("SendHeartbeat to %s: %v", masterAddress, err)
  90. return "", err
  91. }
  92. glog.V(0).Infof("Heartbeat to: %v", masterAddress)
  93. vs.currentMaster = masterAddress
  94. doneChan := make(chan error, 1)
  95. go func() {
  96. for {
  97. in, err := stream.Recv()
  98. if err != nil {
  99. doneChan <- err
  100. return
  101. }
  102. if len(in.DuplicatedUuids) > 0 {
  103. var duplicatedDir []string
  104. foundDuplicate := false
  105. duplicateSet := make(map[string]struct{})
  106. for _, loc := range vs.store.Locations {
  107. directoryUuid := loc.DirectoryUuid
  108. if _, exists := duplicateSet[directoryUuid]; !exists {
  109. duplicateSet[directoryUuid] = struct{}{}
  110. } else {
  111. foundDuplicate = true
  112. }
  113. for _, uuid := range in.DuplicatedUuids {
  114. if uuid == directoryUuid {
  115. duplicatedDir = append(duplicatedDir, loc.Directory)
  116. }
  117. }
  118. }
  119. if foundDuplicate {
  120. glog.Errorf("Shut down Volume Server due to duplicated volume directories: %v", duplicatedDir)
  121. os.Exit(1)
  122. } else {
  123. glog.Warningf("Receive response of duplicated volume directories: %v, ignored(the check found no duplicates)", duplicatedDir)
  124. }
  125. }
  126. if in.GetVolumeSizeLimit() != 0 && vs.store.GetVolumeSizeLimit() != in.GetVolumeSizeLimit() {
  127. vs.store.SetVolumeSizeLimit(in.GetVolumeSizeLimit())
  128. if vs.store.MaybeAdjustVolumeMax() {
  129. if err = stream.Send(vs.store.CollectHeartbeat()); err != nil {
  130. glog.V(0).Infof("Volume Server Failed to talk with master %s: %v", vs.currentMaster, err)
  131. return
  132. }
  133. }
  134. }
  135. if in.GetLeader() != "" && string(vs.currentMaster) != in.GetLeader() {
  136. glog.V(0).Infof("Volume Server found a new master newLeader: %v instead of %v", in.GetLeader(), vs.currentMaster)
  137. newLeader = pb.ServerAddress(in.GetLeader())
  138. doneChan <- nil
  139. return
  140. }
  141. }
  142. }()
  143. if err = stream.Send(vs.store.CollectHeartbeat()); err != nil {
  144. glog.V(0).Infof("Volume Server Failed to talk with master %s: %v", masterAddress, err)
  145. return "", err
  146. }
  147. if err = stream.Send(vs.store.CollectErasureCodingHeartbeat()); err != nil {
  148. glog.V(0).Infof("Volume Server Failed to talk with master %s: %v", masterAddress, err)
  149. return "", err
  150. }
  151. volumeTickChan := time.Tick(sleepInterval)
  152. ecShardTickChan := time.Tick(17 * sleepInterval)
  153. for {
  154. select {
  155. case volumeMessage := <-vs.store.NewVolumesChan:
  156. deltaBeat := &master_pb.Heartbeat{
  157. NewVolumes: []*master_pb.VolumeShortInformationMessage{
  158. &volumeMessage,
  159. },
  160. }
  161. glog.V(0).Infof("volume server %s:%d adds volume %d", vs.store.Ip, vs.store.Port, volumeMessage.Id)
  162. if err = stream.Send(deltaBeat); err != nil {
  163. glog.V(0).Infof("Volume Server Failed to update to master %s: %v", masterAddress, err)
  164. return "", err
  165. }
  166. case ecShardMessage := <-vs.store.NewEcShardsChan:
  167. deltaBeat := &master_pb.Heartbeat{
  168. NewEcShards: []*master_pb.VolumeEcShardInformationMessage{
  169. &ecShardMessage,
  170. },
  171. }
  172. glog.V(0).Infof("volume server %s:%d adds ec shard %d:%d", vs.store.Ip, vs.store.Port, ecShardMessage.Id,
  173. erasure_coding.ShardBits(ecShardMessage.EcIndexBits).ShardIds())
  174. if err = stream.Send(deltaBeat); err != nil {
  175. glog.V(0).Infof("Volume Server Failed to update to master %s: %v", masterAddress, err)
  176. return "", err
  177. }
  178. case volumeMessage := <-vs.store.DeletedVolumesChan:
  179. deltaBeat := &master_pb.Heartbeat{
  180. DeletedVolumes: []*master_pb.VolumeShortInformationMessage{
  181. &volumeMessage,
  182. },
  183. }
  184. glog.V(0).Infof("volume server %s:%d deletes volume %d", vs.store.Ip, vs.store.Port, volumeMessage.Id)
  185. if err = stream.Send(deltaBeat); err != nil {
  186. glog.V(0).Infof("Volume Server Failed to update to master %s: %v", masterAddress, err)
  187. return "", err
  188. }
  189. case ecShardMessage := <-vs.store.DeletedEcShardsChan:
  190. deltaBeat := &master_pb.Heartbeat{
  191. DeletedEcShards: []*master_pb.VolumeEcShardInformationMessage{
  192. &ecShardMessage,
  193. },
  194. }
  195. glog.V(0).Infof("volume server %s:%d deletes ec shard %d:%d", vs.store.Ip, vs.store.Port, ecShardMessage.Id,
  196. erasure_coding.ShardBits(ecShardMessage.EcIndexBits).ShardIds())
  197. if err = stream.Send(deltaBeat); err != nil {
  198. glog.V(0).Infof("Volume Server Failed to update to master %s: %v", masterAddress, err)
  199. return "", err
  200. }
  201. case <-volumeTickChan:
  202. glog.V(4).Infof("volume server %s:%d heartbeat", vs.store.Ip, vs.store.Port)
  203. vs.store.MaybeAdjustVolumeMax()
  204. if err = stream.Send(vs.store.CollectHeartbeat()); err != nil {
  205. glog.V(0).Infof("Volume Server Failed to talk with master %s: %v", masterAddress, err)
  206. return "", err
  207. }
  208. case <-ecShardTickChan:
  209. glog.V(4).Infof("volume server %s:%d ec heartbeat", vs.store.Ip, vs.store.Port)
  210. if err = stream.Send(vs.store.CollectErasureCodingHeartbeat()); err != nil {
  211. glog.V(0).Infof("Volume Server Failed to talk with master %s: %v", masterAddress, err)
  212. return "", err
  213. }
  214. case err = <-doneChan:
  215. return
  216. case <-vs.stopChan:
  217. var volumeMessages []*master_pb.VolumeInformationMessage
  218. emptyBeat := &master_pb.Heartbeat{
  219. Ip: vs.store.Ip,
  220. Port: uint32(vs.store.Port),
  221. PublicUrl: vs.store.PublicUrl,
  222. MaxFileKey: uint64(0),
  223. DataCenter: vs.store.GetDataCenter(),
  224. Rack: vs.store.GetRack(),
  225. Volumes: volumeMessages,
  226. HasNoVolumes: len(volumeMessages) == 0,
  227. }
  228. glog.V(1).Infof("volume server %s:%d stops and deletes all volumes", vs.store.Ip, vs.store.Port)
  229. if err = stream.Send(emptyBeat); err != nil {
  230. glog.V(0).Infof("Volume Server Failed to update to master %s: %v", masterAddress, err)
  231. return "", err
  232. }
  233. return
  234. }
  235. }
  236. }