seaweedfs/weed/shell/command_ec_common.go

380 lines
12 KiB
Go
Raw Normal View History

2019-06-03 09:26:31 +00:00
package shell
import (
"context"
"fmt"
"github.com/seaweedfs/seaweedfs/weed/glog"
"github.com/seaweedfs/seaweedfs/weed/operation"
"github.com/seaweedfs/seaweedfs/weed/pb"
"github.com/seaweedfs/seaweedfs/weed/pb/master_pb"
"github.com/seaweedfs/seaweedfs/weed/pb/volume_server_pb"
"github.com/seaweedfs/seaweedfs/weed/storage/erasure_coding"
"github.com/seaweedfs/seaweedfs/weed/storage/needle"
"github.com/seaweedfs/seaweedfs/weed/storage/types"
"golang.org/x/exp/slices"
2019-06-03 09:26:31 +00:00
"google.golang.org/grpc"
"math"
2019-06-03 09:26:31 +00:00
)
func moveMountedShardToEcNode(commandEnv *CommandEnv, existingLocation *EcNode, collection string, vid needle.VolumeId, shardId erasure_coding.ShardId, destinationEcNode *EcNode, applyBalancing bool) (err error) {
2019-06-03 09:26:31 +00:00
if !commandEnv.isLocked() {
return fmt.Errorf("lock is lost")
}
2019-06-11 04:32:56 +00:00
copiedShardIds := []uint32{uint32(shardId)}
2019-06-03 09:26:31 +00:00
2019-06-11 04:32:56 +00:00
if applyBalancing {
2019-06-03 09:26:31 +00:00
existingServerAddress := pb.NewServerAddressFromDataNode(existingLocation.info)
2019-06-11 04:32:56 +00:00
// ask destination node to copy shard and the ecx file from source node, and mount it
copiedShardIds, err = oneServerCopyAndMountEcShardsFromSource(commandEnv.option.GrpcDialOption, destinationEcNode, []uint32{uint32(shardId)}, vid, collection, existingServerAddress)
2019-06-11 04:32:56 +00:00
if err != nil {
return err
}
2019-06-03 09:26:31 +00:00
2019-06-11 04:32:56 +00:00
// unmount the to be deleted shards
err = unmountEcShards(commandEnv.option.GrpcDialOption, vid, existingServerAddress, copiedShardIds)
2019-06-11 04:32:56 +00:00
if err != nil {
return err
}
// ask source node to delete the shard, and maybe the ecx file
err = sourceServerDeleteEcShards(commandEnv.option.GrpcDialOption, collection, vid, existingServerAddress, copiedShardIds)
2019-06-11 04:32:56 +00:00
if err != nil {
return err
}
fmt.Printf("moved ec shard %d.%d %s => %s\n", vid, shardId, existingLocation.info.Id, destinationEcNode.info.Id)
2019-06-03 09:26:31 +00:00
2019-06-06 06:20:26 +00:00
}
2019-06-11 04:32:56 +00:00
destinationEcNode.addEcVolumeShards(vid, collection, copiedShardIds)
existingLocation.deleteEcVolumeShards(vid, copiedShardIds)
2019-06-06 06:20:26 +00:00
return nil
2019-06-03 09:26:31 +00:00
}
func oneServerCopyAndMountEcShardsFromSource(grpcDialOption grpc.DialOption,
targetServer *EcNode, shardIdsToCopy []uint32,
volumeId needle.VolumeId, collection string, existingLocation pb.ServerAddress) (copiedShardIds []uint32, err error) {
2019-06-03 09:26:31 +00:00
fmt.Printf("allocate %d.%v %s => %s\n", volumeId, shardIdsToCopy, existingLocation, targetServer.info.Id)
targetAddress := pb.NewServerAddressFromDataNode(targetServer.info)
err = operation.WithVolumeServerClient(false, targetAddress, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
2019-06-03 09:26:31 +00:00
if targetAddress != existingLocation {
2019-06-03 09:26:31 +00:00
fmt.Printf("copy %d.%v %s => %s\n", volumeId, shardIdsToCopy, existingLocation, targetServer.info.Id)
_, copyErr := volumeServerClient.VolumeEcShardsCopy(context.Background(), &volume_server_pb.VolumeEcShardsCopyRequest{
2019-06-03 09:26:31 +00:00
VolumeId: uint32(volumeId),
Collection: collection,
ShardIds: shardIdsToCopy,
CopyEcxFile: true,
2019-12-23 20:48:20 +00:00
CopyEcjFile: true,
2019-12-28 20:44:59 +00:00
CopyVifFile: true,
SourceDataNode: string(existingLocation),
2019-06-03 09:26:31 +00:00
})
if copyErr != nil {
return fmt.Errorf("copy %d.%v %s => %s : %v\n", volumeId, shardIdsToCopy, existingLocation, targetServer.info.Id, copyErr)
2019-06-03 09:26:31 +00:00
}
}
fmt.Printf("mount %d.%v on %s\n", volumeId, shardIdsToCopy, targetServer.info.Id)
_, mountErr := volumeServerClient.VolumeEcShardsMount(context.Background(), &volume_server_pb.VolumeEcShardsMountRequest{
2019-06-03 09:26:31 +00:00
VolumeId: uint32(volumeId),
Collection: collection,
ShardIds: shardIdsToCopy,
})
if mountErr != nil {
return fmt.Errorf("mount %d.%v on %s : %v\n", volumeId, shardIdsToCopy, targetServer.info.Id, mountErr)
2019-06-03 09:26:31 +00:00
}
if targetAddress != existingLocation {
2019-06-03 09:26:31 +00:00
copiedShardIds = shardIdsToCopy
glog.V(0).Infof("%s ec volume %d deletes shards %+v", existingLocation, volumeId, copiedShardIds)
}
return nil
})
if err != nil {
return
}
return
}
2019-06-11 04:32:56 +00:00
func eachDataNode(topo *master_pb.TopologyInfo, fn func(dc string, rack RackId, dn *master_pb.DataNodeInfo)) {
2019-06-03 09:26:31 +00:00
for _, dc := range topo.DataCenterInfos {
for _, rack := range dc.RackInfos {
for _, dn := range rack.DataNodeInfos {
2019-06-11 04:32:56 +00:00
fn(dc.Id, RackId(rack.Id), dn)
2019-06-03 09:26:31 +00:00
}
}
}
}
func sortEcNodesByFreeslotsDescending(ecNodes []*EcNode) {
slices.SortFunc(ecNodes, func(a, b *EcNode) int {
return b.freeEcSlot - a.freeEcSlot
2019-06-03 09:26:31 +00:00
})
}
func sortEcNodesByFreeslotsAscending(ecNodes []*EcNode) {
slices.SortFunc(ecNodes, func(a, b *EcNode) int {
return a.freeEcSlot - b.freeEcSlot
})
}
2019-06-11 04:32:56 +00:00
type CandidateEcNode struct {
ecNode *EcNode
shardCount int
}
// if the index node changed the freeEcSlot, need to keep every EcNode still sorted
func ensureSortedEcNodes(data []*CandidateEcNode, index int, lessThan func(i, j int) bool) {
for i := index - 1; i >= 0; i-- {
if lessThan(i+1, i) {
swap(data, i, i+1)
} else {
break
}
}
for i := index + 1; i < len(data); i++ {
if lessThan(i, i-1) {
swap(data, i, i-1)
} else {
break
}
}
}
func swap(data []*CandidateEcNode, i, j int) {
t := data[i]
data[i] = data[j]
data[j] = t
}
2019-06-03 09:26:31 +00:00
func countShards(ecShardInfos []*master_pb.VolumeEcShardInformationMessage) (count int) {
for _, ecShardInfo := range ecShardInfos {
shardBits := erasure_coding.ShardBits(ecShardInfo.EcIndexBits)
count += shardBits.ShardIdCount()
}
return
}
2021-02-16 10:47:02 +00:00
func countFreeShardSlots(dn *master_pb.DataNodeInfo, diskType types.DiskType) (count int) {
if dn.DiskInfos == nil {
return 0
}
diskInfo := dn.DiskInfos[string(diskType)]
if diskInfo == nil {
return 0
}
return int(diskInfo.MaxVolumeCount-diskInfo.VolumeCount)*erasure_coding.DataShardsCount - countShards(diskInfo.EcShardInfos)
2019-06-03 09:26:31 +00:00
}
2019-06-11 04:32:56 +00:00
type RackId string
type EcNodeId string
2019-06-03 09:26:31 +00:00
type EcNode struct {
info *master_pb.DataNodeInfo
2019-06-07 07:25:01 +00:00
dc string
2019-06-11 04:32:56 +00:00
rack RackId
2019-06-03 09:26:31 +00:00
freeEcSlot int
}
func (ecNode *EcNode) localShardIdCount(vid uint32) int {
2021-02-16 10:47:02 +00:00
for _, diskInfo := range ecNode.info.DiskInfos {
for _, ecShardInfo := range diskInfo.EcShardInfos {
if vid == ecShardInfo.Id {
shardBits := erasure_coding.ShardBits(ecShardInfo.EcIndexBits)
return shardBits.ShardIdCount()
}
}
}
return 0
}
2019-06-11 04:32:56 +00:00
type EcRack struct {
ecNodes map[EcNodeId]*EcNode
freeEcSlot int
}
func collectEcNodes(commandEnv *CommandEnv, selectedDataCenter string) (ecNodes []*EcNode, totalFreeEcSlots int, err error) {
2019-06-03 09:26:31 +00:00
// list all possible locations
2021-02-22 08:28:42 +00:00
// collect topology information
topologyInfo, _, err := collectTopologyInfo(commandEnv, 0)
2019-06-03 09:26:31 +00:00
if err != nil {
2021-02-22 08:28:42 +00:00
return
2019-06-03 09:26:31 +00:00
}
// find out all volume servers with one slot left.
2021-02-22 08:28:42 +00:00
ecNodes, totalFreeEcSlots = collectEcVolumeServersByDc(topologyInfo, selectedDataCenter)
sortEcNodesByFreeslotsDescending(ecNodes)
return
}
func collectEcVolumeServersByDc(topo *master_pb.TopologyInfo, selectedDataCenter string) (ecNodes []*EcNode, totalFreeEcSlots int) {
eachDataNode(topo, func(dc string, rack RackId, dn *master_pb.DataNodeInfo) {
2019-06-11 04:32:56 +00:00
if selectedDataCenter != "" && selectedDataCenter != dc {
return
}
2021-02-16 10:47:02 +00:00
freeEcSlots := countFreeShardSlots(dn, types.HardDriveType)
ecNodes = append(ecNodes, &EcNode{
info: dn,
dc: dc,
rack: rack,
freeEcSlot: int(freeEcSlots),
})
totalFreeEcSlots += freeEcSlots
2019-06-03 09:26:31 +00:00
})
return
}
func sourceServerDeleteEcShards(grpcDialOption grpc.DialOption, collection string, volumeId needle.VolumeId, sourceLocation pb.ServerAddress, toBeDeletedShardIds []uint32) error {
2019-06-03 09:26:31 +00:00
fmt.Printf("delete %d.%v from %s\n", volumeId, toBeDeletedShardIds, sourceLocation)
return operation.WithVolumeServerClient(false, sourceLocation, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
_, deleteErr := volumeServerClient.VolumeEcShardsDelete(context.Background(), &volume_server_pb.VolumeEcShardsDeleteRequest{
2019-06-03 09:26:31 +00:00
VolumeId: uint32(volumeId),
Collection: collection,
ShardIds: toBeDeletedShardIds,
})
return deleteErr
})
}
func unmountEcShards(grpcDialOption grpc.DialOption, volumeId needle.VolumeId, sourceLocation pb.ServerAddress, toBeUnmountedhardIds []uint32) error {
2019-06-03 09:26:31 +00:00
fmt.Printf("unmount %d.%v from %s\n", volumeId, toBeUnmountedhardIds, sourceLocation)
return operation.WithVolumeServerClient(false, sourceLocation, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
_, deleteErr := volumeServerClient.VolumeEcShardsUnmount(context.Background(), &volume_server_pb.VolumeEcShardsUnmountRequest{
2019-06-03 09:26:31 +00:00
VolumeId: uint32(volumeId),
ShardIds: toBeUnmountedhardIds,
})
return deleteErr
})
}
func mountEcShards(grpcDialOption grpc.DialOption, collection string, volumeId needle.VolumeId, sourceLocation pb.ServerAddress, toBeMountedhardIds []uint32) error {
2019-06-03 09:26:31 +00:00
fmt.Printf("mount %d.%v on %s\n", volumeId, toBeMountedhardIds, sourceLocation)
return operation.WithVolumeServerClient(false, sourceLocation, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
_, mountErr := volumeServerClient.VolumeEcShardsMount(context.Background(), &volume_server_pb.VolumeEcShardsMountRequest{
2019-06-03 09:26:31 +00:00
VolumeId: uint32(volumeId),
Collection: collection,
ShardIds: toBeMountedhardIds,
})
return mountErr
})
}
2019-06-11 04:32:56 +00:00
func ceilDivide(total, n int) int {
return int(math.Ceil(float64(total) / float64(n)))
}
func findEcVolumeShards(ecNode *EcNode, vid needle.VolumeId) erasure_coding.ShardBits {
2021-02-16 13:13:48 +00:00
if diskInfo, found := ecNode.info.DiskInfos[string(types.HardDriveType)]; found {
for _, shardInfo := range diskInfo.EcShardInfos {
if needle.VolumeId(shardInfo.Id) == vid {
return erasure_coding.ShardBits(shardInfo.EcIndexBits)
}
2019-06-11 04:32:56 +00:00
}
}
return 0
}
func (ecNode *EcNode) addEcVolumeShards(vid needle.VolumeId, collection string, shardIds []uint32) *EcNode {
foundVolume := false
2021-02-16 13:33:38 +00:00
diskInfo, found := ecNode.info.DiskInfos[string(types.HardDriveType)]
if found {
2021-02-16 13:13:48 +00:00
for _, shardInfo := range diskInfo.EcShardInfos {
if needle.VolumeId(shardInfo.Id) == vid {
oldShardBits := erasure_coding.ShardBits(shardInfo.EcIndexBits)
newShardBits := oldShardBits
for _, shardId := range shardIds {
newShardBits = newShardBits.AddShardId(erasure_coding.ShardId(shardId))
}
shardInfo.EcIndexBits = uint32(newShardBits)
ecNode.freeEcSlot -= newShardBits.ShardIdCount() - oldShardBits.ShardIdCount()
foundVolume = true
break
2019-06-11 04:32:56 +00:00
}
}
2021-02-16 13:33:38 +00:00
} else {
diskInfo = &master_pb.DiskInfo{
Type: string(types.HardDriveType),
}
ecNode.info.DiskInfos[string(types.HardDriveType)] = diskInfo
2019-06-11 04:32:56 +00:00
}
if !foundVolume {
var newShardBits erasure_coding.ShardBits
for _, shardId := range shardIds {
newShardBits = newShardBits.AddShardId(erasure_coding.ShardId(shardId))
}
2021-02-16 10:47:02 +00:00
diskInfo.EcShardInfos = append(diskInfo.EcShardInfos, &master_pb.VolumeEcShardInformationMessage{
2019-06-11 04:32:56 +00:00
Id: uint32(vid),
Collection: collection,
EcIndexBits: uint32(newShardBits),
2021-02-16 10:47:02 +00:00
DiskType: string(types.HardDriveType),
2019-06-11 04:32:56 +00:00
})
ecNode.freeEcSlot -= len(shardIds)
}
return ecNode
}
func (ecNode *EcNode) deleteEcVolumeShards(vid needle.VolumeId, shardIds []uint32) *EcNode {
2021-02-16 13:13:48 +00:00
if diskInfo, found := ecNode.info.DiskInfos[string(types.HardDriveType)]; found {
for _, shardInfo := range diskInfo.EcShardInfos {
if needle.VolumeId(shardInfo.Id) == vid {
oldShardBits := erasure_coding.ShardBits(shardInfo.EcIndexBits)
newShardBits := oldShardBits
for _, shardId := range shardIds {
newShardBits = newShardBits.RemoveShardId(erasure_coding.ShardId(shardId))
}
shardInfo.EcIndexBits = uint32(newShardBits)
ecNode.freeEcSlot -= newShardBits.ShardIdCount() - oldShardBits.ShardIdCount()
2019-06-11 04:32:56 +00:00
}
}
}
return ecNode
}
func groupByCount(data []*EcNode, identifierFn func(*EcNode) (id string, count int)) map[string]int {
countMap := make(map[string]int)
for _, d := range data {
id, count := identifierFn(d)
countMap[id] += count
}
return countMap
}
func groupBy(data []*EcNode, identifierFn func(*EcNode) (id string)) map[string][]*EcNode {
groupMap := make(map[string][]*EcNode)
for _, d := range data {
id := identifierFn(d)
groupMap[id] = append(groupMap[id], d)
}
return groupMap
}