Browse Source

上传和调整过程中将所有的小文件一起处理

gitlink
Sydonian 1 year ago
parent
commit
d4f8540dfa
7 changed files with 392 additions and 250 deletions
  1. +3
    -3
      agent/internal/mq/object.go
  2. +14
    -21
      agent/internal/task/ipfs_pin.go
  3. +6
    -7
      common/pkgs/cmd/create_package.go
  4. +4
    -4
      common/pkgs/mq/agent/object.go
  5. +2
    -2
      common/pkgs/mq/coordinator/package.go
  6. +55
    -18
      scanner/internal/event/check_package_redundancy.go
  7. +308
    -195
      scanner/internal/event/clean_pinned.go

+ 3
- 3
agent/internal/mq/object.go View File

@@ -9,12 +9,12 @@ import (
)

func (svc *Service) PinObject(msg *agtmq.PinObject) (*agtmq.PinObjectResp, *mq.CodeMessage) {
logger.WithField("FileHash", msg.FileHash).Debugf("pin object")
logger.WithField("FileHash", msg.FileHashes).Debugf("pin object")

tsk := svc.taskManager.StartComparable(task.NewIPFSPin(msg.FileHash))
tsk := svc.taskManager.StartNew(task.NewIPFSPin(msg.FileHashes))

if tsk.Error() != nil {
logger.WithField("FileHash", msg.FileHash).
logger.WithField("FileHash", msg.FileHashes).
Warnf("pin object failed, err: %s", tsk.Error().Error())
return nil, mq.Failed(errorcode.OperationFailed, "pin object failed")
}


+ 14
- 21
agent/internal/task/ipfs_pin.go View File

@@ -10,24 +10,15 @@ import (
)

type IPFSPin struct {
FileHash string
FileHashes []string
}

func NewIPFSPin(fileHash string) *IPFSPin {
func NewIPFSPin(fileHashes []string) *IPFSPin {
return &IPFSPin{
FileHash: fileHash,
FileHashes: fileHashes,
}
}

func (t *IPFSPin) Compare(other *Task) bool {
tsk, ok := other.Body().(*IPFSPin)
if !ok {
return false
}

return t.FileHash == tsk.FileHash
}

func (t *IPFSPin) Execute(task *task.Task[TaskContext], ctx TaskContext, complete CompleteFn) {
log := logger.WithType[IPFSPin]("Task")
log.Debugf("begin with %v", logger.FormatStruct(t))
@@ -45,15 +36,17 @@ func (t *IPFSPin) Execute(task *task.Task[TaskContext], ctx TaskContext, complet
}
defer ipfsCli.Close()

err = ipfsCli.Pin(t.FileHash)
if err != nil {
err := fmt.Errorf("pin file failed, err: %w", err)
log.WithField("FileHash", t.FileHash).Warn(err.Error())

complete(err, CompleteOption{
RemovingDelay: time.Minute,
})
return
for _, fileHash := range t.FileHashes {
err = ipfsCli.Pin(fileHash)
if err != nil {
err := fmt.Errorf("pin file failed, err: %w", err)
log.WithField("FileHash", t.FileHashes).Warn(err.Error())

complete(err, CompleteOption{
RemovingDelay: time.Minute,
})
return
}
}

complete(nil, CompleteOption{


+ 6
- 7
common/pkgs/cmd/create_package.go View File

@@ -139,6 +139,10 @@ func uploadAndUpdatePackage(packageID cdssdk.PackageID, objectIter iterator.Uplo
if err != nil {
return nil, fmt.Errorf("new coordinator client: %w", err)
}
defer stgglb.CoordinatorMQPool.Release(coorCli)

// 为所有文件选择相同的上传节点
uploadNode := chooseUploadNode(userNodes, nodeAffinity)

var uploadRets []ObjectUploadResult
//上传文件夹
@@ -154,8 +158,6 @@ func uploadAndUpdatePackage(packageID cdssdk.PackageID, objectIter iterator.Uplo
err = func() error {
defer objInfo.File.Close()

uploadNode := chooseUploadNode(userNodes, nodeAffinity)

fileHash, err := uploadFile(objInfo.File, uploadNode)
if err != nil {
return fmt.Errorf("uploading file: %w", err)
@@ -165,9 +167,6 @@ func uploadAndUpdatePackage(packageID cdssdk.PackageID, objectIter iterator.Uplo
Info: objInfo,
Error: err,
})
if err != nil {
return fmt.Errorf("uploading object: %w", err)
}

adds = append(adds, coormq.NewAddObjectEntry(objInfo.Path, objInfo.Size, fileHash, uploadNode.Node.NodeID))
return nil
@@ -177,7 +176,7 @@ func uploadAndUpdatePackage(packageID cdssdk.PackageID, objectIter iterator.Uplo
}
}

_, err = coorCli.UpdateECPackage(coormq.NewUpdatePackage(packageID, adds, nil))
_, err = coorCli.UpdatePackage(coormq.NewUpdatePackage(packageID, adds, nil))
if err != nil {
return nil, fmt.Errorf("updating package: %w", err)
}
@@ -262,7 +261,7 @@ func pinIPFSFile(nodeID cdssdk.NodeID, fileHash string) error {
defer stgglb.AgentMQPool.Release(agtCli)

// 然后让最近节点pin本地上传的文件
_, err = agtCli.PinObject(agtmq.ReqPinObject(fileHash, false))
_, err = agtCli.PinObject(agtmq.ReqPinObject([]string{fileHash}, false))
if err != nil {
return fmt.Errorf("start pinning object: %w", err)
}


+ 4
- 4
common/pkgs/mq/agent/object.go View File

@@ -11,16 +11,16 @@ var _ = Register(Service.PinObject)

type PinObject struct {
mq.MessageBodyBase
FileHash string `json:"fileHash"`
IsBackground bool `json:"isBackground"`
FileHashes []string `json:"fileHashes"`
IsBackground bool `json:"isBackground"`
}
type PinObjectResp struct {
mq.MessageBodyBase
}

func ReqPinObject(fileHash string, isBackground bool) *PinObject {
func ReqPinObject(fileHashes []string, isBackground bool) *PinObject {
return &PinObject{
FileHash: fileHash,
FileHashes: fileHashes,
IsBackground: isBackground,
}
}


+ 2
- 2
common/pkgs/mq/coordinator/package.go View File

@@ -79,7 +79,7 @@ func (client *Client) CreatePackage(msg *CreatePackage) (*CreatePackageResp, err
return mq.Request(Service.CreatePackage, client.rabbitCli, msg)
}

// 更新EC备份模式的Package
// 更新Package
var _ = Register(Service.UpdatePackage)

type UpdatePackage struct {
@@ -116,7 +116,7 @@ func NewAddObjectEntry(path string, size int64, fileHash string, nodeIDs cdssdk.
NodeID: nodeIDs,
}
}
func (client *Client) UpdateECPackage(msg *UpdatePackage) (*UpdatePackageResp, error) {
func (client *Client) UpdatePackage(msg *UpdatePackage) (*UpdatePackageResp, error) {
return mq.Request(Service.UpdatePackage, client.rabbitCli, msg)
}



+ 55
- 18
scanner/internal/event/check_package_redundancy.go View File

@@ -8,7 +8,7 @@ import (
"github.com/samber/lo"
"gitlink.org.cn/cloudream/common/pkgs/logger"
cdssdk "gitlink.org.cn/cloudream/common/sdks/storage"
"gitlink.org.cn/cloudream/common/utils/sort"
"gitlink.org.cn/cloudream/common/utils/sort2"
stgglb "gitlink.org.cn/cloudream/storage/common/globals"
stgmod "gitlink.org.cn/cloudream/storage/common/models"
"gitlink.org.cn/cloudream/storage/common/pkgs/distlock/reqbuilder"
@@ -110,7 +110,10 @@ func (t *CheckPackageRedundancy) Execute(execCtx ExecuteContext) {
defRep := cdssdk.DefaultRepRedundancy
defEC := cdssdk.DefaultECRedundancy

// TODO 目前rep的备份数量固定为2,所以这里直接选出两个节点
mostBlockNodeIDs := t.summaryRepObjectBlockNodes(getObjs.Objects, 2)
newRepNodes := t.chooseNewNodesForRep(&defRep, allNodes)
rechoosedRepNodes := t.rechooseNodesForRep(mostBlockNodeIDs, &defRep, allNodes)
newECNodes := t.chooseNewNodesForEC(&defEC, allNodes)

// 加锁
@@ -149,8 +152,7 @@ func (t *CheckPackageRedundancy) Execute(execCtx ExecuteContext) {
log.WithField("ObjectID", obj.Object.ObjectID).Debugf("redundancy: rep -> ec")
entry, err = t.repToEC(obj, &defEC, newECNodes)
} else {
uploadNodes := t.rechooseNodesForRep(obj, red, allNodes)
entry, err = t.repToRep(obj, &defRep, uploadNodes)
entry, err = t.repToRep(obj, &defRep, rechoosedRepNodes)
}

case *cdssdk.ECRedundancy:
@@ -183,8 +185,43 @@ func (t *CheckPackageRedundancy) Execute(execCtx ExecuteContext) {
}
}

// 统计每个对象块所在的节点,选出块最多的不超过nodeCnt个节点
func (t *CheckPackageRedundancy) summaryRepObjectBlockNodes(objs []stgmod.ObjectDetail, nodeCnt int) []cdssdk.NodeID {
type nodeBlocks struct {
NodeID cdssdk.NodeID
Count int
}

nodeBlocksMap := make(map[cdssdk.NodeID]*nodeBlocks)
for _, obj := range objs {
shouldUseEC := obj.Object.Size > config.Cfg().ECFileSizeThreshold
if _, ok := obj.Object.Redundancy.(*cdssdk.RepRedundancy); ok && !shouldUseEC {
for _, block := range obj.Blocks {
if _, ok := nodeBlocksMap[block.NodeID]; !ok {
nodeBlocksMap[block.NodeID] = &nodeBlocks{
NodeID: block.NodeID,
Count: 0,
}
}
nodeBlocksMap[block.NodeID].Count++
}
}
}

nodes := lo.Values(nodeBlocksMap)
sort2.Sort(nodes, func(left *nodeBlocks, right *nodeBlocks) int {
return right.Count - left.Count
})

ids := lo.Map(nodes, func(item *nodeBlocks, idx int) cdssdk.NodeID { return item.NodeID })
if len(ids) > nodeCnt {
ids = ids[:nodeCnt]
}
return ids
}

func (t *CheckPackageRedundancy) chooseNewNodesForRep(red *cdssdk.RepRedundancy, allNodes map[cdssdk.NodeID]*NodeLoadInfo) []*NodeLoadInfo {
sortedNodes := sort.Sort(lo.Values(allNodes), func(left *NodeLoadInfo, right *NodeLoadInfo) int {
sortedNodes := sort2.Sort(lo.Values(allNodes), func(left *NodeLoadInfo, right *NodeLoadInfo) int {
dm := right.LoadsRecentMonth - left.LoadsRecentMonth
if dm != 0 {
return dm
@@ -197,7 +234,7 @@ func (t *CheckPackageRedundancy) chooseNewNodesForRep(red *cdssdk.RepRedundancy,
}

func (t *CheckPackageRedundancy) chooseNewNodesForEC(red *cdssdk.ECRedundancy, allNodes map[cdssdk.NodeID]*NodeLoadInfo) []*NodeLoadInfo {
sortedNodes := sort.Sort(lo.Values(allNodes), func(left *NodeLoadInfo, right *NodeLoadInfo) int {
sortedNodes := sort2.Sort(lo.Values(allNodes), func(left *NodeLoadInfo, right *NodeLoadInfo) int {
dm := right.LoadsRecentMonth - left.LoadsRecentMonth
if dm != 0 {
return dm
@@ -209,36 +246,36 @@ func (t *CheckPackageRedundancy) chooseNewNodesForEC(red *cdssdk.ECRedundancy, a
return t.chooseSoManyNodes(red.N, sortedNodes)
}

func (t *CheckPackageRedundancy) rechooseNodesForRep(obj stgmod.ObjectDetail, red *cdssdk.RepRedundancy, allNodes map[cdssdk.NodeID]*NodeLoadInfo) []*NodeLoadInfo {
func (t *CheckPackageRedundancy) rechooseNodesForRep(mostBlockNodeIDs []cdssdk.NodeID, red *cdssdk.RepRedundancy, allNodes map[cdssdk.NodeID]*NodeLoadInfo) []*NodeLoadInfo {
type rechooseNode struct {
*NodeLoadInfo
CachedBlockIndex int
HasBlock bool
}

var rechooseNodes []*rechooseNode
for _, node := range allNodes {
cachedBlockIndex := -1
for _, block := range obj.Blocks {
if block.NodeID == node.Node.NodeID {
cachedBlockIndex = block.Index
hasBlock := false
for _, id := range mostBlockNodeIDs {
if id == node.Node.NodeID {
hasBlock = true
break
}
}

rechooseNodes = append(rechooseNodes, &rechooseNode{
NodeLoadInfo: node,
CachedBlockIndex: cachedBlockIndex,
NodeLoadInfo: node,
HasBlock: hasBlock,
})
}

sortedNodes := sort.Sort(rechooseNodes, func(left *rechooseNode, right *rechooseNode) int {
sortedNodes := sort2.Sort(rechooseNodes, func(left *rechooseNode, right *rechooseNode) int {
dm := right.LoadsRecentMonth - left.LoadsRecentMonth
if dm != 0 {
return dm
}

// 已经缓存了文件块的节点优先选择
v := sort.CmpBool(right.CachedBlockIndex > -1, left.CachedBlockIndex > -1)
v := sort2.CmpBool(right.HasBlock, left.HasBlock)
if v != 0 {
return v
}
@@ -271,14 +308,14 @@ func (t *CheckPackageRedundancy) rechooseNodesForEC(obj stgmod.ObjectDetail, red
})
}

sortedNodes := sort.Sort(rechooseNodes, func(left *rechooseNode, right *rechooseNode) int {
sortedNodes := sort2.Sort(rechooseNodes, func(left *rechooseNode, right *rechooseNode) int {
dm := right.LoadsRecentMonth - left.LoadsRecentMonth
if dm != 0 {
return dm
}

// 已经缓存了文件块的节点优先选择
v := sort.CmpBool(right.CachedBlockIndex > -1, left.CachedBlockIndex > -1)
v := sort2.CmpBool(right.CachedBlockIndex > -1, left.CachedBlockIndex > -1)
if v != 0 {
return v
}
@@ -627,7 +664,7 @@ func (t *CheckPackageRedundancy) pinObject(nodeID cdssdk.NodeID, fileHash string
}
defer stgglb.AgentMQPool.Release(agtCli)

_, err = agtCli.PinObject(agtmq.ReqPinObject(fileHash, false))
_, err = agtCli.PinObject(agtmq.ReqPinObject([]string{fileHash}, false))
if err != nil {
return fmt.Errorf("start pinning object: %w", err)
}


+ 308
- 195
scanner/internal/event/clean_pinned.go View File

@@ -4,23 +4,23 @@ import (
"fmt"
"math"
"math/rand"
"strconv"
"sync"

"github.com/samber/lo"
"gitlink.org.cn/cloudream/common/pkgs/bitmap"
"gitlink.org.cn/cloudream/common/pkgs/logger"
cdssdk "gitlink.org.cn/cloudream/common/sdks/storage"
mylo "gitlink.org.cn/cloudream/common/utils/lo"
"gitlink.org.cn/cloudream/common/utils/lo2"
mymath "gitlink.org.cn/cloudream/common/utils/math"
myref "gitlink.org.cn/cloudream/common/utils/reflect"
mysort "gitlink.org.cn/cloudream/common/utils/sort"
"gitlink.org.cn/cloudream/common/utils/sort2"
"gitlink.org.cn/cloudream/storage/common/consts"
stgglb "gitlink.org.cn/cloudream/storage/common/globals"
stgmod "gitlink.org.cn/cloudream/storage/common/models"
"gitlink.org.cn/cloudream/storage/common/pkgs/distlock/reqbuilder"
"gitlink.org.cn/cloudream/storage/common/pkgs/ioswitch/plans"
agtmq "gitlink.org.cn/cloudream/storage/common/pkgs/mq/agent"
coormq "gitlink.org.cn/cloudream/storage/common/pkgs/mq/coordinator"
scevt "gitlink.org.cn/cloudream/storage/common/pkgs/mq/scanner/event"
"gitlink.org.cn/cloudream/storage/scanner/internal/config"
)

type CleanPinned struct {
@@ -67,20 +67,134 @@ func (t *CleanPinned) Execute(execCtx ExecuteContext) {
}
readerNodeIDs := lo.Map(getLoadLog.Logs, func(item coormq.PackageLoadLogDetail, idx int) cdssdk.NodeID { return item.Storage.NodeID })

var changeRedEntries []coormq.ChangeObjectRedundancyEntry
var allNodeID []cdssdk.NodeID
for _, obj := range getObjs.Objects {
entry, err := t.doOne(execCtx, readerNodeIDs, coorCli, obj)
for _, block := range obj.Blocks {
allNodeID = append(allNodeID, block.NodeID)
}
allNodeID = append(allNodeID, obj.PinnedAt...)
}

getNodeResp, err := coorCli.GetNodes(coormq.NewGetNodes(lo.Union(allNodeID)))
if err != nil {
log.Warnf("getting nodes: %s", err.Error())
return
}

allNodeInfos := make(map[cdssdk.NodeID]*cdssdk.Node)
for _, node := range getNodeResp.Nodes {
n := node
allNodeInfos[node.NodeID] = &n
}

// 只对ec和rep对象进行处理
var ecObjects []stgmod.ObjectDetail
var repObjects []stgmod.ObjectDetail
for _, obj := range getObjs.Objects {
if _, ok := obj.Object.Redundancy.(*cdssdk.ECRedundancy); ok {
ecObjects = append(ecObjects, obj)
} else if _, ok := obj.Object.Redundancy.(*cdssdk.RepRedundancy); ok {
repObjects = append(repObjects, obj)
}
}

planBld := plans.NewPlanBuilder()
pinPlans := make(map[cdssdk.NodeID]*[]string)

// 对于rep对象,统计出所有对象块分布最多的两个节点,用这两个节点代表所有rep对象块的分布,去进行退火算法
var repObjectsUpdateEntries []coormq.ChangeObjectRedundancyEntry
repMostNodeIDs := t.summaryRepObjectBlockNodes(repObjects)
solu := t.startAnnealing(allNodeInfos, readerNodeIDs, annealingObject{
totalBlockCount: 1,
minBlockCnt: 1,
pinnedAt: repMostNodeIDs,
blocks: nil,
})
for _, obj := range repObjects {
repObjectsUpdateEntries = append(repObjectsUpdateEntries, t.makePlansForRepObject(solu, obj, pinPlans))
}

// 对于ec对象,则每个对象单独进行退火算法
var ecObjectsUpdateEntries []coormq.ChangeObjectRedundancyEntry
for _, obj := range ecObjects {
ecRed := obj.Object.Redundancy.(*cdssdk.ECRedundancy)
solu := t.startAnnealing(allNodeInfos, readerNodeIDs, annealingObject{
totalBlockCount: ecRed.N,
minBlockCnt: ecRed.K,
pinnedAt: obj.PinnedAt,
blocks: obj.Blocks,
})
ecObjectsUpdateEntries = append(ecObjectsUpdateEntries, t.makePlansForECObject(allNodeInfos, solu, obj, &planBld))
}

wg := sync.WaitGroup{}

// 执行pin操作
var anyPinErr error
for nodeID, pin := range pinPlans {
wg.Add(1)
go func(nodeID cdssdk.NodeID, pin *[]string) {
defer wg.Done()

agtCli, err := stgglb.AgentMQPool.Acquire(nodeID)
if err != nil {
log.Warnf("new agent client: %s", err.Error())
return
}
defer stgglb.AgentMQPool.Release(agtCli)

_, err = agtCli.PinObject(agtmq.ReqPinObject(*pin, false))
if err != nil {
log.Warnf("pinning object: %s", err.Error())
anyPinErr = err
}
}(nodeID, pin)
}

// 执行IO计划
var ioSwRets map[string]any
var ioSwErr error
wg.Add(1)
go func() {
defer wg.Done()
plan, err := planBld.Build()
if err != nil {
log.WithField("PackageID", obj).Warn(err.Error())
continue
ioSwErr = fmt.Errorf("building io switch plan: %w", err)
return
}
if entry != nil {
changeRedEntries = append(changeRedEntries, *entry)

exec, err := plans.Execute(*plan)
if err != nil {
ioSwErr = fmt.Errorf("executing io switch plan: %w", err)
return
}
ret, err := exec.Wait()
if err != nil {
ioSwErr = fmt.Errorf("waiting io switch plan: %w", err)
return
}
ioSwRets = ret.ResultValues
}()

wg.Wait()

if anyPinErr != nil {
log.Warn(anyPinErr.Error())
return
}
if ioSwErr != nil {
log.Warn(ioSwErr.Error())
return
}

if len(changeRedEntries) > 0 {
_, err = coorCli.ChangeObjectRedundancy(coormq.ReqChangeObjectRedundancy(changeRedEntries))
// 根据按照方案进行调整的结果,填充更新元数据的命令
for i := range ecObjectsUpdateEntries {
t.populateECObjectEntry(&ecObjectsUpdateEntries[i], ecObjects[i], ioSwRets)
}

finalEntries := append(repObjectsUpdateEntries, ecObjectsUpdateEntries...)
if len(finalEntries) > 0 {
_, err = coorCli.ChangeObjectRedundancy(coormq.ReqChangeObjectRedundancy(finalEntries))
if err != nil {
log.Warnf("changing object redundancy: %s", err.Error())
return
@@ -88,15 +202,52 @@ func (t *CleanPinned) Execute(execCtx ExecuteContext) {
}
}

type doingContext struct {
execCtx ExecuteContext
readerNodeIDs []cdssdk.NodeID // 近期可能访问此对象的节点
nodesSortedByReader map[cdssdk.NodeID][]nodeDist // 拥有数据的节点到每个可能访问对象的节点按距离排序
nodeInfos map[cdssdk.NodeID]*cdssdk.Node
// 统计每个对象块所在的节点,选出块最多的不超过nodeCnt个节点
func (t *CleanPinned) summaryRepObjectBlockNodes(objs []stgmod.ObjectDetail) []cdssdk.NodeID {
type nodeBlocks struct {
NodeID cdssdk.NodeID
Count int
}

nodeBlocksMap := make(map[cdssdk.NodeID]*nodeBlocks)
for _, obj := range objs {
shouldUseEC := obj.Object.Size > config.Cfg().ECFileSizeThreshold
if _, ok := obj.Object.Redundancy.(*cdssdk.RepRedundancy); ok && !shouldUseEC {
for _, block := range obj.Blocks {
if _, ok := nodeBlocksMap[block.NodeID]; !ok {
nodeBlocksMap[block.NodeID] = &nodeBlocks{
NodeID: block.NodeID,
Count: 0,
}
}
nodeBlocksMap[block.NodeID].Count++
}
}
}

nodes := lo.Values(nodeBlocksMap)
sort2.Sort(nodes, func(left *nodeBlocks, right *nodeBlocks) int {
return right.Count - left.Count
})

// 只选出块数超过一半的节点,但要保证至少有两个节点
for i := 2; i < len(nodes); i++ {
if nodes[i].Count < len(objs)/2 {
nodes = nodes[:i]
break
}
}

return lo.Map(nodes, func(item *nodeBlocks, idx int) cdssdk.NodeID { return item.NodeID })
}

type annealingState struct {
allNodeInfos map[cdssdk.NodeID]*cdssdk.Node // 所有节点的信息
readerNodeIDs []cdssdk.NodeID // 近期可能访问此对象的节点
nodesSortedByReader map[cdssdk.NodeID][]nodeDist // 拥有数据的节点到每个可能访问对象的节点按距离排序
object annealingObject // 进行退火的对象
blockList []objectBlock // 排序后的块分布情况
nodeBlockBitmaps map[cdssdk.NodeID]*bitmap.Bitmap64 // 用位图的形式表示每一个节点上有哪些块
allBlockTypeCount int // object总共被分成了几块
minBlockTypeCount int // 最少要几块才能恢复出完整的object
nodeCombTree combinatorialTree // 节点组合树,用于加速计算容灾度

maxScore float64 // 搜索过程中得到过的最大分数
@@ -127,6 +278,13 @@ type combinatorialTree struct {
localNodeIDToNodeID []cdssdk.NodeID
}

type annealingObject struct {
totalBlockCount int
minBlockCnt int
pinnedAt []cdssdk.NodeID
blocks []stgmod.ObjectBlock
}

const (
iterActionNone = 0
iterActionSkip = 1
@@ -331,125 +489,84 @@ type combinatorialTreeNode struct {
blocksBitmap bitmap.Bitmap64 // 选择了这个中心之后,所有中心一共包含多少种块
}

func (t *CleanPinned) doOne(execCtx ExecuteContext, readerNodeIDs []cdssdk.NodeID, coorCli *coormq.Client, obj stgmod.ObjectDetail) (*coormq.ChangeObjectRedundancyEntry, error) {
if len(obj.PinnedAt) == 0 && len(obj.Blocks) == 0 {
return nil, nil
}
type annealingSolution struct {
blockList []objectBlock // 所有节点的块分布情况
rmBlocks []bool // 要删除哪些块
}

ctx := doingContext{
execCtx: execCtx,
func (t *CleanPinned) startAnnealing(allNodeInfos map[cdssdk.NodeID]*cdssdk.Node, readerNodeIDs []cdssdk.NodeID, object annealingObject) annealingSolution {
state := &annealingState{
allNodeInfos: allNodeInfos,
readerNodeIDs: readerNodeIDs,
nodesSortedByReader: make(map[cdssdk.NodeID][]nodeDist),
nodeInfos: make(map[cdssdk.NodeID]*cdssdk.Node),
object: object,
nodeBlockBitmaps: make(map[cdssdk.NodeID]*bitmap.Bitmap64),
}

err := t.getNodeInfos(&ctx, coorCli, obj)
if err != nil {
return nil, err
}

err = t.makeBlockList(&ctx, obj)
if err != nil {
return nil, err
t.initBlockList(state)
if state.blockList == nil {
return annealingSolution{}
}

if ctx.blockList == nil {
return nil, nil
}

t.makeNodeBlockBitmap(&ctx)
t.initNodeBlockBitmap(state)

t.sortNodeByReaderDistance(&ctx)
t.sortNodeByReaderDistance(state)

ctx.rmBlocks = make([]bool, len(ctx.blockList))
ctx.inversedIndex = -1
ctx.nodeCombTree = newCombinatorialTree(ctx.nodeBlockBitmaps)
state.rmBlocks = make([]bool, len(state.blockList))
state.inversedIndex = -1
state.nodeCombTree = newCombinatorialTree(state.nodeBlockBitmaps)

ctx.lastScore = t.calcScore(&ctx)
ctx.maxScore = ctx.lastScore
ctx.maxScoreRmBlocks = mylo.ArrayClone(ctx.rmBlocks)
state.lastScore = t.calcScore(state)
state.maxScore = state.lastScore
state.maxScoreRmBlocks = lo2.ArrayClone(state.rmBlocks)

// 模拟退火算法的温度
curTemp := ctx.lastScore
curTemp := state.lastScore
// 结束温度
finalTemp := curTemp * 0.2
// 冷却率
coolingRate := 0.95

for curTemp > finalTemp {
ctx.inversedIndex = rand.Intn(len(ctx.rmBlocks))
block := ctx.blockList[ctx.inversedIndex]
ctx.rmBlocks[ctx.inversedIndex] = !ctx.rmBlocks[ctx.inversedIndex]
ctx.nodeBlockBitmaps[block.NodeID].Set(block.Index, !ctx.rmBlocks[ctx.inversedIndex])
ctx.nodeCombTree.UpdateBitmap(block.NodeID, *ctx.nodeBlockBitmaps[block.NodeID], ctx.minBlockTypeCount)
state.inversedIndex = rand.Intn(len(state.rmBlocks))
block := state.blockList[state.inversedIndex]
state.rmBlocks[state.inversedIndex] = !state.rmBlocks[state.inversedIndex]
state.nodeBlockBitmaps[block.NodeID].Set(block.Index, !state.rmBlocks[state.inversedIndex])
state.nodeCombTree.UpdateBitmap(block.NodeID, *state.nodeBlockBitmaps[block.NodeID], state.object.minBlockCnt)

curScore := t.calcScore(&ctx)
curScore := t.calcScore(state)

dScore := curScore - ctx.lastScore
dScore := curScore - state.lastScore
// 如果新方案比旧方案得分低,且没有要求强制接受新方案,那么就将变化改回去
if curScore == 0 || (dScore < 0 && !t.alwaysAccept(curTemp, dScore, coolingRate)) {
ctx.rmBlocks[ctx.inversedIndex] = !ctx.rmBlocks[ctx.inversedIndex]
ctx.nodeBlockBitmaps[block.NodeID].Set(block.Index, !ctx.rmBlocks[ctx.inversedIndex])
ctx.nodeCombTree.UpdateBitmap(block.NodeID, *ctx.nodeBlockBitmaps[block.NodeID], ctx.minBlockTypeCount)
fmt.Printf("\n")
state.rmBlocks[state.inversedIndex] = !state.rmBlocks[state.inversedIndex]
state.nodeBlockBitmaps[block.NodeID].Set(block.Index, !state.rmBlocks[state.inversedIndex])
state.nodeCombTree.UpdateBitmap(block.NodeID, *state.nodeBlockBitmaps[block.NodeID], state.object.minBlockCnt)
// fmt.Printf("\n")
} else {
fmt.Printf(" accept!\n")
ctx.lastScore = curScore
if ctx.maxScore < curScore {
ctx.maxScore = ctx.lastScore
ctx.maxScoreRmBlocks = mylo.ArrayClone(ctx.rmBlocks)
// fmt.Printf(" accept!\n")
state.lastScore = curScore
if state.maxScore < curScore {
state.maxScore = state.lastScore
state.maxScoreRmBlocks = lo2.ArrayClone(state.rmBlocks)
}
}
curTemp *= coolingRate
}

return t.applySolution(ctx, obj)
}

func (t *CleanPinned) getNodeInfos(ctx *doingContext, coorCli *coormq.Client, obj stgmod.ObjectDetail) error {
var nodeIDs []cdssdk.NodeID
for _, b := range obj.Blocks {
nodeIDs = append(nodeIDs, b.NodeID)
return annealingSolution{
blockList: state.blockList,
rmBlocks: state.maxScoreRmBlocks,
}
nodeIDs = append(nodeIDs, obj.PinnedAt...)

nodeIDs = append(nodeIDs, ctx.readerNodeIDs...)

getNode, err := coorCli.GetNodes(coormq.NewGetNodes(lo.Uniq(nodeIDs)))
if err != nil {
return fmt.Errorf("requesting to coordinator: %w", err)
}

for _, n := range getNode.Nodes {
ctx.nodeInfos[n.NodeID] = &n
}

return nil
}

func (t *CleanPinned) makeBlockList(ctx *doingContext, obj stgmod.ObjectDetail) error {
blockCnt := 1
minBlockCnt := 1
switch red := obj.Object.Redundancy.(type) {
case *cdssdk.NoneRedundancy:
return nil
case *cdssdk.RepRedundancy:
blockCnt = 1
minBlockCnt = 1
case *cdssdk.ECRedundancy:
blockCnt = red.N
minBlockCnt = red.K
default:
return fmt.Errorf("unknow redundancy type: %v", myref.TypeOfValue(obj.Object.Redundancy))
}

func (t *CleanPinned) initBlockList(ctx *annealingState) {
blocksMap := make(map[cdssdk.NodeID][]objectBlock)

// 先生成所有的影子块
for _, pinned := range obj.PinnedAt {
blocks := make([]objectBlock, 0, blockCnt)
for i := 0; i < blockCnt; i++ {
for _, pinned := range ctx.object.pinnedAt {
blocks := make([]objectBlock, 0, ctx.object.totalBlockCount)
for i := 0; i < ctx.object.totalBlockCount; i++ {
blocks = append(blocks, objectBlock{
Index: i,
NodeID: pinned,
@@ -460,7 +577,7 @@ func (t *CleanPinned) makeBlockList(ctx *doingContext, obj stgmod.ObjectDetail)
}

// 再填充实际块
for _, b := range obj.Blocks {
for _, b := range ctx.object.blocks {
blocks := blocksMap[b.NodeID]

has := false
@@ -490,7 +607,7 @@ func (t *CleanPinned) makeBlockList(ctx *doingContext, obj stgmod.ObjectDetail)
for _, bs := range blocksMap {
sortedBlocks = append(sortedBlocks, bs...)
}
sortedBlocks = mysort.Sort(sortedBlocks, func(left objectBlock, right objectBlock) int {
sortedBlocks = sort2.Sort(sortedBlocks, func(left objectBlock, right objectBlock) int {
d := left.NodeID - right.NodeID
if d != 0 {
return int(d)
@@ -499,36 +616,33 @@ func (t *CleanPinned) makeBlockList(ctx *doingContext, obj stgmod.ObjectDetail)
return left.Index - right.Index
})

ctx.allBlockTypeCount = blockCnt
ctx.minBlockTypeCount = minBlockCnt
ctx.blockList = sortedBlocks
return nil
}

func (t *CleanPinned) makeNodeBlockBitmap(ctx *doingContext) {
for _, b := range ctx.blockList {
mp, ok := ctx.nodeBlockBitmaps[b.NodeID]
func (t *CleanPinned) initNodeBlockBitmap(state *annealingState) {
for _, b := range state.blockList {
mp, ok := state.nodeBlockBitmaps[b.NodeID]
if !ok {
nb := bitmap.Bitmap64(0)
mp = &nb
ctx.nodeBlockBitmaps[b.NodeID] = mp
state.nodeBlockBitmaps[b.NodeID] = mp
}
mp.Set(b.Index, true)
}
}

func (t *CleanPinned) sortNodeByReaderDistance(ctx *doingContext) {
for _, r := range ctx.readerNodeIDs {
func (t *CleanPinned) sortNodeByReaderDistance(state *annealingState) {
for _, r := range state.readerNodeIDs {
var nodeDists []nodeDist

for n := range ctx.nodeBlockBitmaps {
for n := range state.nodeBlockBitmaps {
if r == n {
// 同节点时距离视为0.1
nodeDists = append(nodeDists, nodeDist{
NodeID: n,
Distance: consts.NodeDistanceSameNode,
})
} else if ctx.nodeInfos[r].LocationID == ctx.nodeInfos[n].LocationID {
} else if state.allNodeInfos[r].LocationID == state.allNodeInfos[n].LocationID {
// 同地区时距离视为1
nodeDists = append(nodeDists, nodeDist{
NodeID: n,
@@ -543,14 +657,14 @@ func (t *CleanPinned) sortNodeByReaderDistance(ctx *doingContext) {
}
}

ctx.nodesSortedByReader[r] = mysort.Sort(nodeDists, func(left, right nodeDist) int { return mysort.Cmp(left.Distance, right.Distance) })
state.nodesSortedByReader[r] = sort2.Sort(nodeDists, func(left, right nodeDist) int { return sort2.Cmp(left.Distance, right.Distance) })
}
}

func (t *CleanPinned) calcScore(ctx *doingContext) float64 {
dt := t.calcDisasterTolerance(ctx)
ac := t.calcMinAccessCost(ctx)
sc := t.calcSpaceCost(ctx)
func (t *CleanPinned) calcScore(state *annealingState) float64 {
dt := t.calcDisasterTolerance(state)
ac := t.calcMinAccessCost(state)
sc := t.calcSpaceCost(state)

dtSc := 1.0
if dt < 1 {
@@ -566,42 +680,43 @@ func (t *CleanPinned) calcScore(ctx *doingContext) float64 {
newSc = dtSc / (sc * ac)
}

fmt.Printf("solu: %v, cur: %v, dt: %v, ac: %v, sc: %v ", ctx.rmBlocks, newSc, dt, ac, sc)
// fmt.Printf("solu: %v, cur: %v, dt: %v, ac: %v, sc: %v ", ctx.rmBlocks, newSc, dt, ac, sc)
return newSc
}

// 计算容灾度
func (t *CleanPinned) calcDisasterTolerance(ctx *doingContext) float64 {
if ctx.inversedIndex != -1 {
node := ctx.blockList[ctx.inversedIndex]
ctx.nodeCombTree.UpdateBitmap(node.NodeID, *ctx.nodeBlockBitmaps[node.NodeID], ctx.minBlockTypeCount)
func (t *CleanPinned) calcDisasterTolerance(state *annealingState) float64 {
if state.inversedIndex != -1 {
node := state.blockList[state.inversedIndex]
state.nodeCombTree.UpdateBitmap(node.NodeID, *state.nodeBlockBitmaps[node.NodeID], state.object.minBlockCnt)
}
return float64(len(ctx.nodeBlockBitmaps) - ctx.nodeCombTree.FindKBlocksMaxDepth(ctx.minBlockTypeCount))
return float64(len(state.nodeBlockBitmaps) - state.nodeCombTree.FindKBlocksMaxDepth(state.object.minBlockCnt))
}

// 计算最小访问数据的代价
func (t *CleanPinned) calcMinAccessCost(ctx *doingContext) float64 {
func (t *CleanPinned) calcMinAccessCost(state *annealingState) float64 {
cost := math.MaxFloat64
for _, reader := range ctx.readerNodeIDs {
tarNodes := ctx.nodesSortedByReader[reader]
for _, reader := range state.readerNodeIDs {
tarNodes := state.nodesSortedByReader[reader]
gotBlocks := bitmap.Bitmap64(0)
thisCost := 0.0

for _, tar := range tarNodes {
tarNodeMp := ctx.nodeBlockBitmaps[tar.NodeID]
tarNodeMp := state.nodeBlockBitmaps[tar.NodeID]

// 只需要从目的节点上获得缺少的块
curWeigth := gotBlocks.Weight()
// 下面的if会在拿到k个块之后跳出循环,所以or多了块也没关系
gotBlocks.Or(tarNodeMp)
willGetBlocks := mymath.Min(gotBlocks.Weight()-curWeigth, ctx.minBlockTypeCount-curWeigth)
// 但是算读取块的消耗时,不能多算,最多算读了k个块的消耗
willGetBlocks := mymath.Min(gotBlocks.Weight()-curWeigth, state.object.minBlockCnt-curWeigth)
thisCost += float64(willGetBlocks) * float64(tar.Distance)

if gotBlocks.Weight() >= ctx.minBlockTypeCount {
if gotBlocks.Weight() >= state.object.minBlockCnt {
break
}
}
if gotBlocks.Weight() >= ctx.minBlockTypeCount {
if gotBlocks.Weight() >= state.object.minBlockCnt {
cost = math.Min(cost, thisCost)
}
}
@@ -610,7 +725,7 @@ func (t *CleanPinned) calcMinAccessCost(ctx *doingContext) float64 {
}

// 计算冗余度
func (t *CleanPinned) calcSpaceCost(ctx *doingContext) float64 {
func (t *CleanPinned) calcSpaceCost(ctx *annealingState) float64 {
blockCount := 0
for i, b := range ctx.blockList {
if ctx.rmBlocks[i] {
@@ -625,26 +740,58 @@ func (t *CleanPinned) calcSpaceCost(ctx *doingContext) float64 {
}
}
// 所有算力中心上拥有的块的总数 / 一个对象被分成了几个块
return float64(blockCount) / float64(ctx.minBlockTypeCount)
return float64(blockCount) / float64(ctx.object.minBlockCnt)
}

// 如果新方案得分比旧方案小,那么在一定概率内也接受新方案
func (t *CleanPinned) alwaysAccept(curTemp float64, dScore float64, coolingRate float64) bool {
v := math.Exp(dScore / curTemp / coolingRate)
fmt.Printf(" -- chance: %v, temp: %v", v, curTemp)
// fmt.Printf(" -- chance: %v, temp: %v", v, curTemp)
return v > rand.Float64()
}

func (t *CleanPinned) applySolution(ctx doingContext, obj stgmod.ObjectDetail) (*coormq.ChangeObjectRedundancyEntry, error) {
func (t *CleanPinned) makePlansForRepObject(solu annealingSolution, obj stgmod.ObjectDetail, pinPlans map[cdssdk.NodeID]*[]string) coormq.ChangeObjectRedundancyEntry {
entry := coormq.ChangeObjectRedundancyEntry{
ObjectID: obj.Object.ObjectID,
Redundancy: obj.Object.Redundancy,
}

for i, f := range solu.rmBlocks {
hasCache := lo.ContainsBy(obj.Blocks, func(b stgmod.ObjectBlock) bool { return b.NodeID == solu.blockList[i].NodeID }) ||
lo.ContainsBy(obj.PinnedAt, func(n cdssdk.NodeID) bool { return n == solu.blockList[i].NodeID })
willRm := f

if !willRm {
// 如果对象在退火后要保留副本的节点没有副本,则需要在这个节点创建副本
if !hasCache {
pinPlan, ok := pinPlans[solu.blockList[i].NodeID]
if !ok {
pinPlan = &[]string{}
pinPlans[solu.blockList[i].NodeID] = pinPlan
}
*pinPlan = append(*pinPlan, obj.Object.FileHash)
}
entry.Blocks = append(entry.Blocks, stgmod.ObjectBlock{
ObjectID: obj.Object.ObjectID,
Index: solu.blockList[i].Index,
NodeID: solu.blockList[i].NodeID,
FileHash: obj.Object.FileHash,
})
}
}

return entry
}

func (t *CleanPinned) makePlansForECObject(allNodeInfos map[cdssdk.NodeID]*cdssdk.Node, solu annealingSolution, obj stgmod.ObjectDetail, planBld *plans.PlanBuilder) coormq.ChangeObjectRedundancyEntry {
entry := coormq.ChangeObjectRedundancyEntry{
ObjectID: obj.Object.ObjectID,
Redundancy: obj.Object.Redundancy,
}
fmt.Printf("final solu: %v, score: %v\n", ctx.maxScoreRmBlocks, ctx.maxScore)

reconstrct := make(map[cdssdk.NodeID]*[]int)
for i, f := range ctx.maxScoreRmBlocks {
block := ctx.blockList[i]
for i, f := range solu.rmBlocks {
block := solu.blockList[i]
if !f {
entry.Blocks = append(entry.Blocks, stgmod.ObjectBlock{
ObjectID: obj.Object.ObjectID,
@@ -666,64 +813,30 @@ func (t *CleanPinned) applySolution(ctx doingContext, obj stgmod.ObjectDetail) (
}
}

bld := reqbuilder.NewBuilder()
for id := range reconstrct {
bld.IPFS().Buzy(id)
}

mutex, err := bld.MutexLock(ctx.execCtx.Args.DistLock)
if err != nil {
return nil, fmt.Errorf("acquiring distlock: %w", err)
}
defer mutex.Unlock()

if ecRed, ok := obj.Object.Redundancy.(*cdssdk.ECRedundancy); ok {
for id, idxs := range reconstrct {
bld := plans.NewPlanBuilder()
agt := bld.AtAgent(*ctx.nodeInfos[id])

strs := agt.IPFSRead(obj.Object.FileHash).ChunkedSplit(ecRed.ChunkSize, ecRed.K, true)
ss := agt.ECReconstructAny(*ecRed, lo.Range(ecRed.K), *idxs, strs.Streams...)
for i, s := range ss.Streams {
s.IPFSWrite(fmt.Sprintf("%d", (*idxs)[i]))
}

plan, err := bld.Build()
if err != nil {
return nil, fmt.Errorf("building io switch plan: %w", err)
}

exec, err := plans.Execute(*plan)
if err != nil {
return nil, fmt.Errorf("executing io switch plan: %w", err)
}
ret, err := exec.Wait()
if err != nil {
return nil, fmt.Errorf("executing io switch plan: %w", err)
}
ecRed := obj.Object.Redundancy.(*cdssdk.ECRedundancy)

for k, v := range ret.ResultValues {
idx, err := strconv.ParseInt(k, 10, 32)
if err != nil {
return nil, fmt.Errorf("parsing plan result: %w", err)
}
for id, idxs := range reconstrct {
agt := planBld.AtAgent(*allNodeInfos[id])

for i := range entry.Blocks {
if entry.Blocks[i].NodeID == id && entry.Blocks[i].Index == int(idx) {
entry.Blocks[i].FileHash = v.(string)
}
}
}

}
} else if _, ok := obj.Object.Redundancy.(*cdssdk.RepRedundancy); ok {
// rep模式不分块,所以每一个Block的FileHash就是完整文件的FileHash
for i := range entry.Blocks {
entry.Blocks[i].FileHash = obj.Object.FileHash
strs := agt.IPFSRead(obj.Object.FileHash).ChunkedSplit(ecRed.ChunkSize, ecRed.K, true)
ss := agt.ECReconstructAny(*ecRed, lo.Range(ecRed.K), *idxs, strs.Streams...)
for i, s := range ss.Streams {
s.IPFSWrite(fmt.Sprintf("%d.%d", obj.Object.ObjectID, (*idxs)[i]))
}
}
return entry
}

return &entry, nil
func (t *CleanPinned) populateECObjectEntry(entry *coormq.ChangeObjectRedundancyEntry, obj stgmod.ObjectDetail, ioRets map[string]any) {
for i := range entry.Blocks {
if entry.Blocks[i].FileHash != "" {
continue
}

key := fmt.Sprintf("%d.%d", obj.Object.ObjectID, entry.Blocks[i].Index)
// 不应该出现key不存在的情况
entry.Blocks[i].FileHash = ioRets[key].(string)
}
}

func init() {


Loading…
Cancel
Save