seaweedfs/weed/shell/command_ec_common.go

341 lines
10 KiB
Go
Raw Normal View History

2019-06-03 09:26:31 +00:00
package shell
import (
"context"
"fmt"
2019-06-11 04:32:56 +00:00
"math"
2019-06-03 09:26:31 +00:00
"sort"
"github.com/chrislusf/seaweedfs/weed/glog"
"github.com/chrislusf/seaweedfs/weed/operation"
"github.com/chrislusf/seaweedfs/weed/pb/master_pb"
"github.com/chrislusf/seaweedfs/weed/pb/volume_server_pb"
"github.com/chrislusf/seaweedfs/weed/storage/erasure_coding"
"github.com/chrislusf/seaweedfs/weed/storage/needle"
"google.golang.org/grpc"
)
2019-06-11 04:32:56 +00:00
func moveMountedShardToEcNode(ctx context.Context, 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
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
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(ctx, commandEnv.option.GrpcDialOption, destinationEcNode, []uint32{uint32(shardId)}, vid, collection, existingLocation.info.Id)
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(ctx, commandEnv.option.GrpcDialOption, vid, existingLocation.info.Id, copiedShardIds)
if err != nil {
return err
}
// ask source node to delete the shard, and maybe the ecx file
err = sourceServerDeleteEcShards(ctx, commandEnv.option.GrpcDialOption, collection, vid, existingLocation.info.Id, copiedShardIds)
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(ctx context.Context, grpcDialOption grpc.DialOption,
targetServer *EcNode, shardIdsToCopy []uint32,
2019-06-03 09:26:31 +00:00
volumeId needle.VolumeId, collection string, existingLocation string) (copiedShardIds []uint32, err error) {
fmt.Printf("allocate %d.%v %s => %s\n", volumeId, shardIdsToCopy, existingLocation, targetServer.info.Id)
err = operation.WithVolumeServerClient(targetServer.info.Id, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
if targetServer.info.Id != existingLocation {
fmt.Printf("copy %d.%v %s => %s\n", volumeId, shardIdsToCopy, existingLocation, targetServer.info.Id)
_, copyErr := volumeServerClient.VolumeEcShardsCopy(ctx, &volume_server_pb.VolumeEcShardsCopyRequest{
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,
2019-06-03 09:26:31 +00:00
SourceDataNode: existingLocation,
})
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(ctx, &volume_server_pb.VolumeEcShardsMountRequest{
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 targetServer.info.Id != existingLocation {
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 sortEcNodesByFreeslotsDecending(ecNodes []*EcNode) {
2019-06-03 09:26:31 +00:00
sort.Slice(ecNodes, func(i, j int) bool {
return ecNodes[i].freeEcSlot > ecNodes[j].freeEcSlot
})
}
func sortEcNodesByFreeslotsAscending(ecNodes []*EcNode) {
sort.Slice(ecNodes, func(i, j int) bool {
return ecNodes[i].freeEcSlot < ecNodes[j].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
}
func countFreeShardSlots(dn *master_pb.DataNodeInfo) (count int) {
return int(dn.MaxVolumeCount-dn.ActiveVolumeCount)*erasure_coding.DataShardsCount - countShards(dn.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
}
2019-06-11 04:32:56 +00:00
type EcRack struct {
ecNodes map[EcNodeId]*EcNode
freeEcSlot int
}
func collectEcNodes(ctx context.Context, commandEnv *CommandEnv, selectedDataCenter string) (ecNodes []*EcNode, totalFreeEcSlots int, err error) {
2019-06-03 09:26:31 +00:00
// list all possible locations
var resp *master_pb.VolumeListResponse
err = commandEnv.MasterClient.WithClient(ctx, func(client master_pb.SeaweedClient) error {
2019-06-03 09:26:31 +00:00
resp, err = client.VolumeList(ctx, &master_pb.VolumeListRequest{})
return err
})
if err != nil {
return nil, 0, err
}
// find out all volume servers with one slot left.
2019-06-11 04:32:56 +00:00
eachDataNode(resp.TopologyInfo, func(dc string, rack RackId, dn *master_pb.DataNodeInfo) {
if selectedDataCenter != "" && selectedDataCenter != dc {
return
}
freeEcSlots := countFreeShardSlots(dn)
ecNodes = append(ecNodes, &EcNode{
info: dn,
dc: dc,
rack: rack,
freeEcSlot: int(freeEcSlots),
})
totalFreeEcSlots += freeEcSlots
2019-06-03 09:26:31 +00:00
})
sortEcNodesByFreeslotsDecending(ecNodes)
2019-06-03 09:26:31 +00:00
return
}
func sourceServerDeleteEcShards(ctx context.Context, grpcDialOption grpc.DialOption,
collection string, volumeId needle.VolumeId, sourceLocation string, toBeDeletedShardIds []uint32) error {
fmt.Printf("delete %d.%v from %s\n", volumeId, toBeDeletedShardIds, sourceLocation)
return operation.WithVolumeServerClient(sourceLocation, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
_, deleteErr := volumeServerClient.VolumeEcShardsDelete(ctx, &volume_server_pb.VolumeEcShardsDeleteRequest{
VolumeId: uint32(volumeId),
Collection: collection,
ShardIds: toBeDeletedShardIds,
})
return deleteErr
})
}
func unmountEcShards(ctx context.Context, grpcDialOption grpc.DialOption,
volumeId needle.VolumeId, sourceLocation string, toBeUnmountedhardIds []uint32) error {
fmt.Printf("unmount %d.%v from %s\n", volumeId, toBeUnmountedhardIds, sourceLocation)
return operation.WithVolumeServerClient(sourceLocation, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
_, deleteErr := volumeServerClient.VolumeEcShardsUnmount(ctx, &volume_server_pb.VolumeEcShardsUnmountRequest{
VolumeId: uint32(volumeId),
ShardIds: toBeUnmountedhardIds,
})
return deleteErr
})
}
func mountEcShards(ctx context.Context, grpcDialOption grpc.DialOption,
collection string, volumeId needle.VolumeId, sourceLocation string, toBeMountedhardIds []uint32) error {
fmt.Printf("mount %d.%v on %s\n", volumeId, toBeMountedhardIds, sourceLocation)
return operation.WithVolumeServerClient(sourceLocation, grpcDialOption, func(volumeServerClient volume_server_pb.VolumeServerClient) error {
_, mountErr := volumeServerClient.VolumeEcShardsMount(ctx, &volume_server_pb.VolumeEcShardsMountRequest{
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 {
for _, shardInfo := range ecNode.info.EcShardInfos {
if needle.VolumeId(shardInfo.Id) == vid {
return erasure_coding.ShardBits(shardInfo.EcIndexBits)
}
}
return 0
}
func (ecNode *EcNode) addEcVolumeShards(vid needle.VolumeId, collection string, shardIds []uint32) *EcNode {
foundVolume := false
for _, shardInfo := range ecNode.info.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
}
}
if !foundVolume {
var newShardBits erasure_coding.ShardBits
for _, shardId := range shardIds {
newShardBits = newShardBits.AddShardId(erasure_coding.ShardId(shardId))
}
ecNode.info.EcShardInfos = append(ecNode.info.EcShardInfos, &master_pb.VolumeEcShardInformationMessage{
Id: uint32(vid),
Collection: collection,
EcIndexBits: uint32(newShardBits),
})
ecNode.freeEcSlot -= len(shardIds)
}
return ecNode
}
func (ecNode *EcNode) deleteEcVolumeShards(vid needle.VolumeId, shardIds []uint32) *EcNode {
for _, shardInfo := range ecNode.info.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()
}
}
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
}