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.

263 lines
8.1 KiB

4 years ago
4 years ago
4 years ago
4 years ago
  1. package shell
  2. import (
  3. "bytes"
  4. "context"
  5. "flag"
  6. "fmt"
  7. "github.com/chrislusf/seaweedfs/weed/operation"
  8. "github.com/chrislusf/seaweedfs/weed/pb/volume_server_pb"
  9. "github.com/chrislusf/seaweedfs/weed/storage/needle_map"
  10. "io"
  11. "math"
  12. "sort"
  13. )
  14. func init() {
  15. Commands = append(Commands, &commandVolumeCheckDisk{})
  16. }
  17. type commandVolumeCheckDisk struct {
  18. env *CommandEnv
  19. }
  20. func (c *commandVolumeCheckDisk) Name() string {
  21. return "volume.check.disk"
  22. }
  23. func (c *commandVolumeCheckDisk) Help() string {
  24. return `check all replicated volumes to find and fix inconsistencies
  25. How it works:
  26. find all volumes that are replicated
  27. for each volume id, if there are more than 2 replicas, find one pair with the largest 2 in file count.
  28. for the pair volume A and B
  29. append entries in A and not in B to B
  30. append entries in B and not in A to A
  31. `
  32. }
  33. func (c *commandVolumeCheckDisk) Do(args []string, commandEnv *CommandEnv, writer io.Writer) (err error) {
  34. if err = commandEnv.confirmIsLocked(); err != nil {
  35. return
  36. }
  37. fsckCommand := flag.NewFlagSet(c.Name(), flag.ContinueOnError)
  38. slowMode := fsckCommand.Bool("slow", false, "slow mode checks all replicas even file counts are the same")
  39. verbose := fsckCommand.Bool("v", false, "verbose mode")
  40. applyChanges := fsckCommand.Bool("force", false, "apply the fix")
  41. nonRepairThreshold := fsckCommand.Float64("nonRepairThreshold", 0.3, "repair when missing keys is not more than this limit")
  42. if err = fsckCommand.Parse(args); err != nil {
  43. return nil
  44. }
  45. c.env = commandEnv
  46. // collect topology information
  47. topologyInfo, _, err := collectTopologyInfo(commandEnv)
  48. if err != nil {
  49. return err
  50. }
  51. volumeReplicas, _ := collectVolumeReplicaLocations(topologyInfo)
  52. // pick 1 pairs of volume replica
  53. fileCount := func(replica *VolumeReplica) uint64 {
  54. return replica.info.FileCount - replica.info.DeleteCount
  55. }
  56. aDB, bDB := needle_map.NewMemDb(), needle_map.NewMemDb()
  57. defer aDB.Close()
  58. defer bDB.Close()
  59. for _, replicas := range volumeReplicas {
  60. sort.Slice(replicas, func(i, j int) bool {
  61. return fileCount(replicas[i]) > fileCount(replicas[j])
  62. })
  63. for len(replicas) >= 2 {
  64. a, b := replicas[0], replicas[1]
  65. if !*slowMode {
  66. if fileCount(a) == fileCount(b) {
  67. replicas = replicas[1:]
  68. continue
  69. }
  70. }
  71. if a.info.ReadOnly || b.info.ReadOnly {
  72. fmt.Fprintf(writer, "skipping readonly volume %d on %s and %s\n", a.info.Id, a.location.dataNode.Id, b.location.dataNode.Id)
  73. replicas = replicas[1:]
  74. continue
  75. }
  76. aHasChanges, bHasChanges := true, true
  77. for aHasChanges || bHasChanges {
  78. // reset index db
  79. aDB.Close()
  80. bDB.Close()
  81. aDB, bDB = needle_map.NewMemDb(), needle_map.NewMemDb()
  82. // read index db
  83. if err := c.readIndexDatabase(aDB, a.info.Collection, a.info.Id, a.location.dataNode.Id, *verbose, writer); err != nil {
  84. return err
  85. }
  86. if err := c.readIndexDatabase(bDB, b.info.Collection, b.info.Id, b.location.dataNode.Id, *verbose, writer); err != nil {
  87. return err
  88. }
  89. // find and make up the differences
  90. if aHasChanges, err = c.doVolumeCheckDisk(aDB, bDB, a, b, *verbose, writer, *applyChanges, *nonRepairThreshold); err != nil {
  91. return err
  92. }
  93. if bHasChanges, err = c.doVolumeCheckDisk(bDB, aDB, b, a, *verbose, writer, *applyChanges, *nonRepairThreshold); err != nil {
  94. return err
  95. }
  96. }
  97. replicas = replicas[1:]
  98. }
  99. }
  100. return nil
  101. }
  102. func (c *commandVolumeCheckDisk) doVolumeCheckDisk(subtrahend, minuend *needle_map.MemDb, source, target *VolumeReplica, verbose bool, writer io.Writer, applyChanges bool, nonRepairThreshold float64) (hasChanges bool, err error) {
  103. // find missing keys
  104. // hash join, can be more efficient
  105. var missingNeedles []needle_map.NeedleValue
  106. var counter int
  107. subtrahend.AscendingVisit(func(value needle_map.NeedleValue) error {
  108. counter++
  109. if _, found := minuend.Get(value.Key); !found {
  110. missingNeedles = append(missingNeedles, value)
  111. }
  112. return nil
  113. })
  114. fmt.Fprintf(writer, "volume %d %s has %d entries, %s missed %d entries\n", source.info.Id, source.location.dataNode.Id, counter, target.location.dataNode.Id, len(missingNeedles))
  115. if counter == 0 || len(missingNeedles) == 0 {
  116. return false, nil
  117. }
  118. missingNeedlesFraction := float64(len(missingNeedles)) / float64(counter)
  119. if missingNeedlesFraction > nonRepairThreshold {
  120. return false, fmt.Errorf(
  121. "failed to start repair volume %d, percentage of missing keys is greater than the threshold: %.2f > %.2f",
  122. source.info.Id, missingNeedlesFraction, nonRepairThreshold)
  123. }
  124. for _, needleValue := range missingNeedles {
  125. needleBlob, err := c.readSourceNeedleBlob(source.location.dataNode.Id, source.info.Id, needleValue)
  126. if err != nil {
  127. return hasChanges, err
  128. }
  129. if !applyChanges {
  130. continue
  131. }
  132. if verbose {
  133. fmt.Fprintf(writer, "read %d,%x %s => %s \n", source.info.Id, needleValue.Key, source.location.dataNode.Id, target.location.dataNode.Id)
  134. }
  135. hasChanges = true
  136. if err = c.writeNeedleBlobToTarget(target.location.dataNode.Id, source.info.Id, needleValue, needleBlob); err != nil {
  137. return hasChanges, err
  138. }
  139. }
  140. return
  141. }
  142. func (c *commandVolumeCheckDisk) readSourceNeedleBlob(sourceVolumeServer string, volumeId uint32, needleValue needle_map.NeedleValue) (needleBlob []byte, err error) {
  143. err = operation.WithVolumeServerClient(sourceVolumeServer, c.env.option.GrpcDialOption, func(client volume_server_pb.VolumeServerClient) error {
  144. resp, err := client.ReadNeedleBlob(context.Background(), &volume_server_pb.ReadNeedleBlobRequest{
  145. VolumeId: volumeId,
  146. NeedleId: uint64(needleValue.Key),
  147. Offset: needleValue.Offset.ToActualOffset(),
  148. Size: int32(needleValue.Size),
  149. })
  150. if err != nil {
  151. return err
  152. }
  153. needleBlob = resp.NeedleBlob
  154. return nil
  155. })
  156. return
  157. }
  158. func (c *commandVolumeCheckDisk) writeNeedleBlobToTarget(targetVolumeServer string, volumeId uint32, needleValue needle_map.NeedleValue, needleBlob []byte) error {
  159. return operation.WithVolumeServerClient(targetVolumeServer, c.env.option.GrpcDialOption, func(client volume_server_pb.VolumeServerClient) error {
  160. _, err := client.WriteNeedleBlob(context.Background(), &volume_server_pb.WriteNeedleBlobRequest{
  161. VolumeId: volumeId,
  162. NeedleId: uint64(needleValue.Key),
  163. Size: int32(needleValue.Size),
  164. NeedleBlob: needleBlob,
  165. })
  166. return err
  167. })
  168. }
  169. func (c *commandVolumeCheckDisk) readIndexDatabase(db *needle_map.MemDb, collection string, volumeId uint32, volumeServer string, verbose bool, writer io.Writer) error {
  170. var buf bytes.Buffer
  171. if err := c.copyVolumeIndexFile(collection, volumeId, volumeServer, &buf, verbose, writer); err != nil {
  172. return err
  173. }
  174. if verbose {
  175. fmt.Fprintf(writer, "load collection %s volume %d index size %d from %s ...\n", collection, volumeId, buf.Len(), volumeServer)
  176. }
  177. return db.LoadFromReaderAt(bytes.NewReader(buf.Bytes()))
  178. }
  179. func (c *commandVolumeCheckDisk) copyVolumeIndexFile(collection string, volumeId uint32, volumeServer string, buf *bytes.Buffer, verbose bool, writer io.Writer) error {
  180. return operation.WithVolumeServerClient(volumeServer, c.env.option.GrpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
  181. ext := ".idx"
  182. copyFileClient, err := volumeServerClient.CopyFile(context.Background(), &volume_server_pb.CopyFileRequest{
  183. VolumeId: volumeId,
  184. Ext: ".idx",
  185. CompactionRevision: math.MaxUint32,
  186. StopOffset: math.MaxInt64,
  187. Collection: collection,
  188. IsEcVolume: false,
  189. IgnoreSourceFileNotFound: false,
  190. })
  191. if err != nil {
  192. return fmt.Errorf("failed to start copying volume %d%s: %v", volumeId, ext, err)
  193. }
  194. err = writeToBuffer(copyFileClient, buf)
  195. if err != nil {
  196. return fmt.Errorf("failed to copy %d%s from %s: %v", volumeId, ext, volumeServer, err)
  197. }
  198. return nil
  199. })
  200. }
  201. func writeToBuffer(client volume_server_pb.VolumeServer_CopyFileClient, buf *bytes.Buffer) error {
  202. for {
  203. resp, receiveErr := client.Recv()
  204. if receiveErr == io.EOF {
  205. break
  206. }
  207. if receiveErr != nil {
  208. return fmt.Errorf("receiving: %v", receiveErr)
  209. }
  210. buf.Write(resp.FileContent)
  211. }
  212. return nil
  213. }