gitlab-com-seaweedfs/seaweedfs
1
0
Fork 0
mirror of https://github.com/seaweedfs/seaweedfs.git synced 2024-01-19 02:48:24 +00:00
Code Issues Projects Releases Wiki Activity

fix: avoid deleting one replica without sync (#4875)

Browse source 
* fix: avoid deleting one replica without sync
https://github.com/seaweedfs/seaweedfs/issues/4647

* Update weed/shell/command_volume_fix_replication.go

Co-authored-by: Chris Lu <chrislusf@users.noreply.github.com>

* fix: revert this existing do option to positive

---------

Co-authored-by: Konstantin Lebedev <9497591+kmlebedev@users.noreply.github.co>
Co-authored-by: Chris Lu <chrislusf@users.noreply.github.com>
This commit is contained in:
Konstantin Lebedev 2023-09-28 11:12:10 +05:00 committed by GitHub
parent 44906f1f3b
commit dd580190b4
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 63 additions and 30 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

45
weed/shell/command_volume_check_disk.go
View file

@ -22,8 +22,7 @@ func init() {
} }
type commandVolumeCheckDisk struct { type commandVolumeCheckDisk struct {
env *CommandEnv env *CommandEnv
syncDeletions *bool
} }
func (c *commandVolumeCheckDisk) Name() string { func (c *commandVolumeCheckDisk) Name() string {
@ -51,7 +50,7 @@ func (c *commandVolumeCheckDisk) Do(args []string, commandEnv *CommandEnv, write
verbose := fsckCommand.Bool("v", false, "verbose mode") verbose := fsckCommand.Bool("v", false, "verbose mode")
volumeId := fsckCommand.Uint("volumeId", 0, "the volume id") volumeId := fsckCommand.Uint("volumeId", 0, "the volume id")
applyChanges := fsckCommand.Bool("force", false, "apply the fix") applyChanges := fsckCommand.Bool("force", false, "apply the fix")
c.syncDeletions = fsckCommand.Bool("syncDeleted", false, "sync of deletions the fix") syncDeletions := fsckCommand.Bool("syncDeleted", false, "sync of deletions the fix")
nonRepairThreshold := fsckCommand.Float64("nonRepairThreshold", 0.3, "repair when missing keys is not more than this limit") nonRepairThreshold := fsckCommand.Float64("nonRepairThreshold", 0.3, "repair when missing keys is not more than this limit")
if err = fsckCommand.Parse(args); err != nil { if err = fsckCommand.Parse(args); err != nil {
return nil return nil
@ -97,7 +96,7 @@ func (c *commandVolumeCheckDisk) Do(args []string, commandEnv *CommandEnv, write
continue continue
} }
if err := c.syncTwoReplicas(a, b, *applyChanges, *nonRepairThreshold, *verbose, writer); err != nil { if err := c.syncTwoReplicas(a, b, *applyChanges, *syncDeletions, *nonRepairThreshold, *verbose, writer); err != nil {
fmt.Fprintf(writer, "sync volume %d on %s and %s: %v\n", a.info.Id, a.location.dataNode.Id, b.location.dataNode.Id, err) fmt.Fprintf(writer, "sync volume %d on %s and %s: %v\n", a.info.Id, a.location.dataNode.Id, b.location.dataNode.Id, err)
} }
replicas = replicas[1:] replicas = replicas[1:]
@ -107,17 +106,17 @@ func (c *commandVolumeCheckDisk) Do(args []string, commandEnv *CommandEnv, write
return nil return nil
} }
func (c *commandVolumeCheckDisk) syncTwoReplicas(a *VolumeReplica, b *VolumeReplica, applyChanges bool, nonRepairThreshold float64, verbose bool, writer io.Writer) (err error) { func (c *commandVolumeCheckDisk) syncTwoReplicas(a *VolumeReplica, b *VolumeReplica, applyChanges bool, doSyncDeletions bool, nonRepairThreshold float64, verbose bool, writer io.Writer) (err error) {
aHasChanges, bHasChanges := true, true aHasChanges, bHasChanges := true, true
for aHasChanges || bHasChanges { for aHasChanges || bHasChanges {
if aHasChanges, bHasChanges, err = c.checkBoth(a, b, applyChanges, nonRepairThreshold, verbose, writer); err != nil { if aHasChanges, bHasChanges, err = c.checkBoth(a, b, applyChanges, doSyncDeletions, nonRepairThreshold, verbose, writer); err != nil {
return err return err
} }
} }
return nil return nil
} }
func (c *commandVolumeCheckDisk) checkBoth(a *VolumeReplica, b *VolumeReplica, applyChanges bool, nonRepairThreshold float64, verbose bool, writer io.Writer) (aHasChanges bool, bHasChanges bool, err error) { func (c *commandVolumeCheckDisk) checkBoth(a *VolumeReplica, b *VolumeReplica, applyChanges bool, doSyncDeletions bool, nonRepairThreshold float64, verbose bool, writer io.Writer) (aHasChanges bool, bHasChanges bool, err error) {
aDB, bDB := needle_map.NewMemDb(), needle_map.NewMemDb() aDB, bDB := needle_map.NewMemDb(), needle_map.NewMemDb()
defer func() { defer func() {
aDB.Close() aDB.Close()
@ -126,24 +125,24 @@ func (c *commandVolumeCheckDisk) checkBoth(a *VolumeReplica, b *VolumeReplica, a
// read index db // read index db
readIndexDbCutoffFrom := uint64(time.Now().UnixNano()) readIndexDbCutoffFrom := uint64(time.Now().UnixNano())
if err = c.readIndexDatabase(aDB, a.info.Collection, a.info.Id, pb.NewServerAddressFromDataNode(a.location.dataNode), verbose, writer); err != nil { if err = readIndexDatabase(aDB, a.info.Collection, a.info.Id, pb.NewServerAddressFromDataNode(a.location.dataNode), verbose, writer, c.env.option.GrpcDialOption); err != nil {
return true, true, fmt.Errorf("readIndexDatabase %s volume %d: %v", a.location.dataNode, a.info.Id, err) return true, true, fmt.Errorf("readIndexDatabase %s volume %d: %v", a.location.dataNode, a.info.Id, err)
} }
if err := c.readIndexDatabase(bDB, b.info.Collection, b.info.Id, pb.NewServerAddressFromDataNode(b.location.dataNode), verbose, writer); err != nil { if err := readIndexDatabase(bDB, b.info.Collection, b.info.Id, pb.NewServerAddressFromDataNode(b.location.dataNode), verbose, writer, c.env.option.GrpcDialOption); err != nil {
return true, true, fmt.Errorf("readIndexDatabase %s volume %d: %v", b.location.dataNode, b.info.Id, err) return true, true, fmt.Errorf("readIndexDatabase %s volume %d: %v", b.location.dataNode, b.info.Id, err)
} }
// find and make up the differences // find and make up the differences
if aHasChanges, err = c.doVolumeCheckDisk(bDB, aDB, b, a, verbose, writer, applyChanges, nonRepairThreshold, readIndexDbCutoffFrom); err != nil { if aHasChanges, err = doVolumeCheckDisk(bDB, aDB, b, a, verbose, writer, applyChanges, doSyncDeletions, nonRepairThreshold, readIndexDbCutoffFrom, c.env.option.GrpcDialOption); err != nil {
return true, true, fmt.Errorf("doVolumeCheckDisk source:%s target:%s volume %d: %v", b.location.dataNode.Id, a.location.dataNode.Id, b.info.Id, err) return true, true, fmt.Errorf("doVolumeCheckDisk source:%s target:%s volume %d: %v", b.location.dataNode.Id, a.location.dataNode.Id, b.info.Id, err)
} }
if bHasChanges, err = c.doVolumeCheckDisk(aDB, bDB, a, b, verbose, writer, applyChanges, nonRepairThreshold, readIndexDbCutoffFrom); err != nil { if bHasChanges, err = doVolumeCheckDisk(aDB, bDB, a, b, verbose, writer, applyChanges, doSyncDeletions, nonRepairThreshold, readIndexDbCutoffFrom, c.env.option.GrpcDialOption); err != nil {
return true, true, fmt.Errorf("doVolumeCheckDisk source:%s target:%s volume %d: %v", a.location.dataNode.Id, b.location.dataNode.Id, a.info.Id, err) return true, true, fmt.Errorf("doVolumeCheckDisk source:%s target:%s volume %d: %v", a.location.dataNode.Id, b.location.dataNode.Id, a.info.Id, err)
} }
return return
} }
func (c *commandVolumeCheckDisk) doVolumeCheckDisk(minuend, subtrahend *needle_map.MemDb, source, target *VolumeReplica, verbose bool, writer io.Writer, applyChanges bool, nonRepairThreshold float64, cutoffFromAtNs uint64) (hasChanges bool, err error) { func doVolumeCheckDisk(minuend, subtrahend *needle_map.MemDb, source, target *VolumeReplica, verbose bool, writer io.Writer, applyChanges bool, doSyncDeletions bool, nonRepairThreshold float64, cutoffFromAtNs uint64, grpcDialOption grpc.DialOption) (hasChanges bool, err error) {
// find missing keys // find missing keys
// hash join, can be more efficient // hash join, can be more efficient
@ -158,7 +157,7 @@ func (c *commandVolumeCheckDisk) doVolumeCheckDisk(minuend, subtrahend *needle_m
return nil return nil
} }
if doCutoffOfLastNeedle { if doCutoffOfLastNeedle {
if needleMeta, err := readNeedleMeta(c.env.option.GrpcDialOption, pb.NewServerAddressFromDataNode(source.location.dataNode), source.info.Id, minuendValue); err == nil { if needleMeta, err := readNeedleMeta(grpcDialOption, pb.NewServerAddressFromDataNode(source.location.dataNode), source.info.Id, minuendValue); err == nil {
// needles older than the cutoff time are not missing yet // needles older than the cutoff time are not missing yet
if needleMeta.AppendAtNs > cutoffFromAtNs { if needleMeta.AppendAtNs > cutoffFromAtNs {
return nil return nil
@ -193,7 +192,7 @@ func (c *commandVolumeCheckDisk) doVolumeCheckDisk(minuend, subtrahend *needle_m
} }
for _, needleValue := range missingNeedles { for _, needleValue := range missingNeedles {
needleBlob, err := readSourceNeedleBlob(c.env.option.GrpcDialOption, pb.NewServerAddressFromDataNode(source.location.dataNode), source.info.Id, needleValue) needleBlob, err := readSourceNeedleBlob(grpcDialOption, pb.NewServerAddressFromDataNode(source.location.dataNode), source.info.Id, needleValue)
if err != nil { if err != nil {
return hasChanges, err return hasChanges, err
} }
@ -208,13 +207,13 @@ func (c *commandVolumeCheckDisk) doVolumeCheckDisk(minuend, subtrahend *needle_m
hasChanges = true hasChanges = true
if err = c.writeNeedleBlobToTarget(pb.NewServerAddressFromDataNode(target.location.dataNode), source.info.Id, needleValue, needleBlob); err != nil { if err = writeNeedleBlobToTarget(grpcDialOption, pb.NewServerAddressFromDataNode(target.location.dataNode), source.info.Id, needleValue, needleBlob); err != nil {
return hasChanges, err return hasChanges, err
} }
} }
if *c.syncDeletions && len(partiallyDeletedNeedles) > 0 { if doSyncDeletions && len(partiallyDeletedNeedles) > 0 {
var fidList []string var fidList []string
for _, needleValue := range partiallyDeletedNeedles { for _, needleValue := range partiallyDeletedNeedles {
fidList = append(fidList, needleValue.Key.FileId(source.info.Id)) fidList = append(fidList, needleValue.Key.FileId(source.info.Id))
@ -224,7 +223,7 @@ func (c *commandVolumeCheckDisk) doVolumeCheckDisk(minuend, subtrahend *needle_m
} }
deleteResults, deleteErr := operation.DeleteFilesAtOneVolumeServer( deleteResults, deleteErr := operation.DeleteFilesAtOneVolumeServer(
pb.NewServerAddressFromDataNode(target.location.dataNode), pb.NewServerAddressFromDataNode(target.location.dataNode),
c.env.option.GrpcDialOption, fidList, false) grpcDialOption, fidList, false)
if deleteErr != nil { if deleteErr != nil {
return hasChanges, deleteErr return hasChanges, deleteErr
} }
@ -255,9 +254,9 @@ func readSourceNeedleBlob(grpcDialOption grpc.DialOption, sourceVolumeServer pb.
return return
} }
func (c *commandVolumeCheckDisk) writeNeedleBlobToTarget(targetVolumeServer pb.ServerAddress, volumeId uint32, needleValue needle_map.NeedleValue, needleBlob []byte) error { func writeNeedleBlobToTarget(grpcDialOption grpc.DialOption, targetVolumeServer pb.ServerAddress, volumeId uint32, needleValue needle_map.NeedleValue, needleBlob []byte) error {
return operation.WithVolumeServerClient(false, targetVolumeServer, c.env.option.GrpcDialOption, func(client volume_server_pb.VolumeServerClient) error { return operation.WithVolumeServerClient(false, targetVolumeServer, grpcDialOption, func(client volume_server_pb.VolumeServerClient) error {
_, err := client.WriteNeedleBlob(context.Background(), &volume_server_pb.WriteNeedleBlobRequest{ _, err := client.WriteNeedleBlob(context.Background(), &volume_server_pb.WriteNeedleBlobRequest{
VolumeId: volumeId, VolumeId: volumeId,
NeedleId: uint64(needleValue.Key), NeedleId: uint64(needleValue.Key),
@ -269,10 +268,10 @@ func (c *commandVolumeCheckDisk) writeNeedleBlobToTarget(targetVolumeServer pb.S
} }
func (c *commandVolumeCheckDisk) readIndexDatabase(db *needle_map.MemDb, collection string, volumeId uint32, volumeServer pb.ServerAddress, verbose bool, writer io.Writer) error { func readIndexDatabase(db *needle_map.MemDb, collection string, volumeId uint32, volumeServer pb.ServerAddress, verbose bool, writer io.Writer, grpcDialOption grpc.DialOption) error {
var buf bytes.Buffer var buf bytes.Buffer
if err := c.copyVolumeIndexFile(collection, volumeId, volumeServer, &buf, verbose, writer); err != nil { if err := copyVolumeIndexFile(collection, volumeId, volumeServer, &buf, verbose, writer, grpcDialOption); err != nil {
return err return err
} }
@ -282,9 +281,9 @@ func (c *commandVolumeCheckDisk) readIndexDatabase(db *needle_map.MemDb, collect
return db.LoadFilterFromReaderAt(bytes.NewReader(buf.Bytes()), true, false) return db.LoadFilterFromReaderAt(bytes.NewReader(buf.Bytes()), true, false)
} }
func (c *commandVolumeCheckDisk) copyVolumeIndexFile(collection string, volumeId uint32, volumeServer pb.ServerAddress, buf *bytes.Buffer, verbose bool, writer io.Writer) error { func copyVolumeIndexFile(collection string, volumeId uint32, volumeServer pb.ServerAddress, buf *bytes.Buffer, verbose bool, writer io.Writer, grpcDialOption grpc.DialOption) error {
return operation.WithVolumeServerClient(true, volumeServer, c.env.option.GrpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error { return operation.WithVolumeServerClient(true, volumeServer, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
ext := ".idx" ext := ".idx"

48
weed/shell/command_volume_fix_replication.go
View file

@ -6,8 +6,10 @@ import (
"fmt" "fmt"
"github.com/seaweedfs/seaweedfs/weed/pb" "github.com/seaweedfs/seaweedfs/weed/pb"
"github.com/seaweedfs/seaweedfs/weed/storage/needle" "github.com/seaweedfs/seaweedfs/weed/storage/needle"
"github.com/seaweedfs/seaweedfs/weed/storage/needle_map"
"github.com/seaweedfs/seaweedfs/weed/storage/types" "github.com/seaweedfs/seaweedfs/weed/storage/types"
"golang.org/x/exp/slices" "golang.org/x/exp/slices"
"google.golang.org/grpc"
"io" "io"
"path/filepath" "path/filepath"
"strconv" "strconv"
@ -57,7 +59,8 @@ func (c *commandVolumeFixReplication) Do(args []string, commandEnv *CommandEnv,
volFixReplicationCommand := flag.NewFlagSet(c.Name(), flag.ContinueOnError) volFixReplicationCommand := flag.NewFlagSet(c.Name(), flag.ContinueOnError)
c.collectionPattern = volFixReplicationCommand.String("collectionPattern", "", "match with wildcard characters '*' and '?'") c.collectionPattern = volFixReplicationCommand.String("collectionPattern", "", "match with wildcard characters '*' and '?'")
skipChange := volFixReplicationCommand.Bool("n", false, "skip the changes") skipChange := volFixReplicationCommand.Bool("n", false, "skip the changes")
noDelete := volFixReplicationCommand.Bool("noDelete", false, "Do not delete over-replicated volumes, only fix under-replication") doDelete := volFixReplicationCommand.Bool("doDelete", true, "Also delete over-replicated volumes besides fixing under-replication")
doCheck := volFixReplicationCommand.Bool("doCheck", true, "Also check synchronization before deleting")
retryCount := volFixReplicationCommand.Int("retry", 5, "how many times to retry") retryCount := volFixReplicationCommand.Int("retry", 5, "how many times to retry")
volumesPerStep := volFixReplicationCommand.Int("volumesPerStep", 0, "how many volumes to fix in one cycle") volumesPerStep := volFixReplicationCommand.Int("volumesPerStep", 0, "how many volumes to fix in one cycle")
@ -70,7 +73,6 @@ func (c *commandVolumeFixReplication) Do(args []string, commandEnv *CommandEnv,
} }
takeAction := !*skipChange takeAction := !*skipChange
doDeletes := !*noDelete
underReplicatedVolumeIdsCount := 1 underReplicatedVolumeIdsCount := 1
for underReplicatedVolumeIdsCount > 0 { for underReplicatedVolumeIdsCount > 0 {
@ -111,14 +113,14 @@ func (c *commandVolumeFixReplication) Do(args []string, commandEnv *CommandEnv,
return fmt.Errorf("lock is lost") return fmt.Errorf("lock is lost")
} }
if len(overReplicatedVolumeIds) > 0 && doDeletes { if len(overReplicatedVolumeIds) > 0 && *doDelete {
if err := c.deleteOneVolume(commandEnv, writer, takeAction, overReplicatedVolumeIds, volumeReplicas, allLocations, pickOneReplicaToDelete); err != nil { if err := c.deleteOneVolume(commandEnv, writer, takeAction, *doCheck, overReplicatedVolumeIds, volumeReplicas, allLocations, pickOneReplicaToDelete); err != nil {
return err return err
} }
} }
if len(misplacedVolumeIds) > 0 && doDeletes { if len(misplacedVolumeIds) > 0 && *doDelete {
if err := c.deleteOneVolume(commandEnv, writer, takeAction, misplacedVolumeIds, volumeReplicas, allLocations, pickOneMisplacedVolume); err != nil { if err := c.deleteOneVolume(commandEnv, writer, takeAction, *doCheck, misplacedVolumeIds, volumeReplicas, allLocations, pickOneMisplacedVolume); err != nil {
return err return err
} }
} }
@ -189,7 +191,28 @@ func collectVolumeReplicaLocations(topologyInfo *master_pb.TopologyInfo) (map[ui
type SelectOneVolumeFunc func(replicas []*VolumeReplica, replicaPlacement *super_block.ReplicaPlacement) *VolumeReplica type SelectOneVolumeFunc func(replicas []*VolumeReplica, replicaPlacement *super_block.ReplicaPlacement) *VolumeReplica
func (c *commandVolumeFixReplication) deleteOneVolume(commandEnv *CommandEnv, writer io.Writer, takeAction bool, overReplicatedVolumeIds []uint32, volumeReplicas map[uint32][]*VolumeReplica, allLocations []location, selectOneVolumeFn SelectOneVolumeFunc) error { func checkOneVolume(a *VolumeReplica, b *VolumeReplica, writer io.Writer, grpcDialOption grpc.DialOption) (err error) {
aDB, bDB := needle_map.NewMemDb(), needle_map.NewMemDb()
defer func() {
aDB.Close()
bDB.Close()
}()
// read index db
readIndexDbCutoffFrom := uint64(time.Now().UnixNano())
if err = readIndexDatabase(aDB, a.info.Collection, a.info.Id, pb.NewServerAddressFromDataNode(a.location.dataNode), false, writer, grpcDialOption); err != nil {
return fmt.Errorf("readIndexDatabase %s volume %d: %v", a.location.dataNode, a.info.Id, err)
}
if err := readIndexDatabase(bDB, b.info.Collection, b.info.Id, pb.NewServerAddressFromDataNode(b.location.dataNode), false, writer, grpcDialOption); err != nil {
return fmt.Errorf("readIndexDatabase %s volume %d: %v", b.location.dataNode, b.info.Id, err)
}
if _, err = doVolumeCheckDisk(aDB, bDB, a, b, false, writer, true, false, float64(1), readIndexDbCutoffFrom, grpcDialOption); err != nil {
return fmt.Errorf("doVolumeCheckDisk source:%s target:%s volume %d: %v", a.location.dataNode.Id, b.location.dataNode.Id, a.info.Id, err)
}
return
}
func (c *commandVolumeFixReplication) deleteOneVolume(commandEnv *CommandEnv, writer io.Writer, takeAction bool, doCheck bool, overReplicatedVolumeIds []uint32, volumeReplicas map[uint32][]*VolumeReplica, allLocations []location, selectOneVolumeFn SelectOneVolumeFunc) error {
for _, vid := range overReplicatedVolumeIds { for _, vid := range overReplicatedVolumeIds {
replicas := volumeReplicas[vid] replicas := volumeReplicas[vid]
replicaPlacement, _ := super_block.NewReplicaPlacementFromByte(byte(replicas[0].info.ReplicaPlacement)) replicaPlacement, _ := super_block.NewReplicaPlacementFromByte(byte(replicas[0].info.ReplicaPlacement))
@ -224,6 +247,17 @@ func (c *commandVolumeFixReplication) deleteOneVolume(commandEnv *CommandEnv, wr
break break
} }
if doCheck {
for _, replicaB := range replicas {
if replicaB.location.dataNode == replica.location.dataNode {
continue
}
if err := checkOneVolume(replica, replicaB, writer, commandEnv.option.GrpcDialOption); err != nil {
return fmt.Errorf("sync volume %d on %s and %s: %v\n", replica.info.Id, replica.location.dataNode.Id, replicaB.location.dataNode.Id, err)
}
}
}
if err := deleteVolume(commandEnv.option.GrpcDialOption, needle.VolumeId(replica.info.Id), if err := deleteVolume(commandEnv.option.GrpcDialOption, needle.VolumeId(replica.info.Id),
pb.NewServerAddressFromDataNode(replica.location.dataNode), false); err != nil { pb.NewServerAddressFromDataNode(replica.location.dataNode), false); err != nil {
return fmt.Errorf("deleting volume %d from %s : %v", replica.info.Id, replica.location.dataNode.Id, err) return fmt.Errorf("deleting volume %d from %s : %v", replica.info.Id, replica.location.dataNode.Id, err)