staged stream sync v1.0

.gitignore

@@ -94,3 +94,6 @@ explorer_storage_*
 
 # pprof profiles
 profiles/*.pb.gz
+
+# cache db
+cache_*_db

api/service/manager.go

@@ -23,6 +23,7 @@ const (
 	Prometheus
 	Synchronize
 	CrosslinkSending
+	StagedStreamSync
 )
 
 func (t Type) String() string {
@@ -45,6 +46,8 @@ func (t Type) String() string {
 		return "Synchronize"
 	case CrosslinkSending:
 		return "CrosslinkSending"
+	case StagedStreamSync:
+		return "StagedStreamSync"
 	default:
 		return "Unknown"
 	}

api/service/stagedstreamsync/adapter.go

@@ -0,0 +1,34 @@
+package stagedstreamsync
+
+import (
+	"context"
+
+	"github.com/ethereum/go-ethereum/common"
+	"github.com/ethereum/go-ethereum/event"
+	"github.com/harmony-one/harmony/consensus/engine"
+	"github.com/harmony-one/harmony/core/types"
+	"github.com/harmony-one/harmony/p2p/stream/common/streammanager"
+	syncproto "github.com/harmony-one/harmony/p2p/stream/protocols/sync"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+)
+
+type syncProtocol interface {
+	GetCurrentBlockNumber(ctx context.Context, opts ...syncproto.Option) (uint64, sttypes.StreamID, error)
+	GetBlocksByNumber(ctx context.Context, bns []uint64, opts ...syncproto.Option) ([]*types.Block, sttypes.StreamID, error)
+	GetRawBlocksByNumber(ctx context.Context, bns []uint64, opts ...syncproto.Option) ([][]byte, [][]byte, sttypes.StreamID, error)
+	GetBlockHashes(ctx context.Context, bns []uint64, opts ...syncproto.Option) ([]common.Hash, sttypes.StreamID, error)
+	GetBlocksByHashes(ctx context.Context, hs []common.Hash, opts ...syncproto.Option) ([]*types.Block, sttypes.StreamID, error)
+
+	RemoveStream(stID sttypes.StreamID) // If a stream delivers invalid data, remove the stream
+	StreamFailed(stID sttypes.StreamID, reason string)
+	SubscribeAddStreamEvent(ch chan<- streammanager.EvtStreamAdded) event.Subscription
+	NumStreams() int
+}
+
+type blockChain interface {
+	engine.ChainReader
+	Engine() engine.Engine
+
+	InsertChain(chain types.Blocks, verifyHeaders bool) (int, error)
+	WriteCommitSig(blockNum uint64, lastCommits []byte) error
+}

api/service/stagedstreamsync/beacon_helper.go

@@ -0,0 +1,156 @@
+package stagedstreamsync
+
+import (
+	"time"
+
+	"github.com/harmony-one/harmony/core/types"
+	"github.com/harmony-one/harmony/internal/utils"
+	"github.com/rs/zerolog"
+)
+
+// lastMileCache keeps the last 50 number blocks in memory cache
+const lastMileCap = 50
+
+type (
+	// beaconHelper is the helper for the beacon downloader. The beaconHelper is only started
+	// when node is running on side chain, listening to beacon client pub-sub message and
+	// insert the latest blocks to the beacon chain.
+	beaconHelper struct {
+		bc     blockChain
+		blockC <-chan *types.Block
+		// TODO: refactor this hook to consensus module. We'd better put it in
+		//   consensus module under a subscription.
+		insertHook func()
+
+		lastMileCache *blocksByNumber
+		insertC       chan insertTask
+		closeC        chan struct{}
+		logger        zerolog.Logger
+	}
+
+	insertTask struct {
+		doneC chan struct{}
+	}
+)
+
+func newBeaconHelper(bc blockChain, blockC <-chan *types.Block, insertHook func()) *beaconHelper {
+	return &beaconHelper{
+		bc:            bc,
+		blockC:        blockC,
+		insertHook:    insertHook,
+		lastMileCache: newBlocksByNumber(lastMileCap),
+		insertC:       make(chan insertTask, 1),
+		closeC:        make(chan struct{}),
+		logger: utils.Logger().With().
+			Str("module", "downloader").
+			Str("sub-module", "beacon helper").
+			Logger(),
+	}
+}
+
+func (bh *beaconHelper) start() {
+	go bh.loop()
+}
+
+func (bh *beaconHelper) close() {
+	close(bh.closeC)
+}
+
+func (bh *beaconHelper) loop() {
+	t := time.NewTicker(10 * time.Second)
+	defer t.Stop()
+	for {
+		select {
+		case <-t.C:
+			bh.insertAsync()
+
+		case b, ok := <-bh.blockC:
+			if !ok {
+				return // blockC closed. Node exited
+			}
+			if b == nil {
+				continue
+			}
+			bh.lastMileCache.push(b)
+			bh.insertAsync()
+
+		case it := <-bh.insertC:
+			inserted, bn, err := bh.insertLastMileBlocks()
+			numBlocksInsertedBeaconHelperCounter.Add(float64(inserted))
+			if err != nil {
+				bh.logger.Error().Err(err).Msg(WrapStagedSyncMsg("insert last mile blocks error"))
+				continue
+			}
+			bh.logger.Info().Int("inserted", inserted).
+				Uint64("end height", bn).
+				Uint32("shard", bh.bc.ShardID()).
+				Msg(WrapStagedSyncMsg("insert last mile blocks"))
+
+			close(it.doneC)
+
+		case <-bh.closeC:
+			return
+		}
+	}
+}
+
+// insertAsync triggers the insert last mile without blocking
+func (bh *beaconHelper) insertAsync() {
+	select {
+	case bh.insertC <- insertTask{
+		doneC: make(chan struct{}),
+	}:
+	default:
+	}
+}
+
+// insertSync triggers the insert last mile while blocking
+func (bh *beaconHelper) insertSync() {
+	task := insertTask{
+		doneC: make(chan struct{}),
+	}
+	bh.insertC <- task
+	<-task.doneC
+}
+
+func (bh *beaconHelper) insertLastMileBlocks() (inserted int, bn uint64, err error) {
+	bn = bh.bc.CurrentBlock().NumberU64() + 1
+	for {
+		b := bh.getNextBlock(bn)
+		if b == nil {
+			bn--
+			return
+		}
+		// TODO: Instruct the beacon helper to verify signatures. This may require some forks
+		//       in pub-sub message (add commit sigs in node.block.sync messages)
+		if _, err = bh.bc.InsertChain(types.Blocks{b}, true); err != nil {
+			bn--
+			return
+		}
+		bh.logger.Info().Uint64("number", b.NumberU64()).Msg(WrapStagedSyncMsg("Inserted block from beacon pub-sub"))
+
+		if bh.insertHook != nil {
+			bh.insertHook()
+		}
+		inserted++
+		bn++
+	}
+}
+
+func (bh *beaconHelper) getNextBlock(expBN uint64) *types.Block {
+	for bh.lastMileCache.len() > 0 {
+		b := bh.lastMileCache.pop()
+		if b == nil {
+			return nil
+		}
+		if b.NumberU64() < expBN {
+			continue
+		}
+		if b.NumberU64() > expBN {
+			bh.lastMileCache.push(b)
+			return nil
+		}
+		return b
+	}
+	return nil
+}

api/service/stagedstreamsync/block_by_hash_manager.go

@@ -0,0 +1,133 @@
+package stagedstreamsync
+
+import (
+	"sync"
+
+	"github.com/ethereum/go-ethereum/common"
+	"github.com/harmony-one/harmony/core/types"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/pkg/errors"
+)
+
+type getBlocksByHashManager struct {
+	hashes    []common.Hash
+	pendings  map[common.Hash]struct{}
+	results   map[common.Hash]blockResult
+	whitelist []sttypes.StreamID
+
+	lock sync.Mutex
+}
+
+func newGetBlocksByHashManager(hashes []common.Hash, whitelist []sttypes.StreamID) *getBlocksByHashManager {
+	return &getBlocksByHashManager{
+		hashes:    hashes,
+		pendings:  make(map[common.Hash]struct{}),
+		results:   make(map[common.Hash]blockResult),
+		whitelist: whitelist,
+	}
+}
+
+func (m *getBlocksByHashManager) getNextHashes() ([]common.Hash, []sttypes.StreamID, error) {
+	m.lock.Lock()
+	defer m.lock.Unlock()
+
+	num := m.numBlocksPerRequest()
+	hashes := make([]common.Hash, 0, num)
+	if len(m.whitelist) == 0 {
+		return nil, nil, errors.New("empty white list")
+	}
+
+	for _, hash := range m.hashes {
+		if len(hashes) == num {
+			break
+		}
+		_, ok1 := m.pendings[hash]
+		_, ok2 := m.results[hash]
+		if !ok1 && !ok2 {
+			hashes = append(hashes, hash)
+		}
+	}
+	sts := make([]sttypes.StreamID, len(m.whitelist))
+	copy(sts, m.whitelist)
+	return hashes, sts, nil
+}
+
+func (m *getBlocksByHashManager) numBlocksPerRequest() int {
+	val := divideCeil(len(m.hashes), len(m.whitelist))
+	if val < BlockByHashesLowerCap {
+		val = BlockByHashesLowerCap
+	}
+	if val > BlockByHashesUpperCap {
+		val = BlockByHashesUpperCap
+	}
+	return val
+}
+
+func (m *getBlocksByHashManager) numRequests() int {
+	return divideCeil(len(m.hashes), m.numBlocksPerRequest())
+}
+
+func (m *getBlocksByHashManager) addResult(hashes []common.Hash, blocks []*types.Block, stid sttypes.StreamID) {
+	m.lock.Lock()
+	defer m.lock.Unlock()
+
+	for i, hash := range hashes {
+		block := blocks[i]
+		delete(m.pendings, hash)
+		m.results[hash] = blockResult{
+			block: block,
+			stid:  stid,
+		}
+	}
+}
+
+func (m *getBlocksByHashManager) handleResultError(hashes []common.Hash, stid sttypes.StreamID) {
+	m.lock.Lock()
+	defer m.lock.Unlock()
+
+	m.removeStreamID(stid)
+
+	for _, hash := range hashes {
+		delete(m.pendings, hash)
+	}
+}
+
+func (m *getBlocksByHashManager) getResults() ([]*types.Block, []sttypes.StreamID, error) {
+	m.lock.Lock()
+	defer m.lock.Unlock()
+
+	blocks := make([]*types.Block, 0, len(m.hashes))
+	stids := make([]sttypes.StreamID, 0, len(m.hashes))
+	for _, hash := range m.hashes {
+		if m.results[hash].block == nil {
+			return nil, nil, errors.New("SANITY: nil block found")
+		}
+		blocks = append(blocks, m.results[hash].block)
+		stids = append(stids, m.results[hash].stid)
+	}
+	return blocks, stids, nil
+}
+
+func (m *getBlocksByHashManager) isDone() bool {
+	m.lock.Lock()
+	defer m.lock.Unlock()
+
+	return len(m.results) == len(m.hashes)
+}
+
+func (m *getBlocksByHashManager) removeStreamID(target sttypes.StreamID) {
+	// O(n^2) complexity. But considering the whitelist size is small, should not
+	// have performance issue.
+loop:
+	for i, stid := range m.whitelist {
+		if stid == target {
+			if i == len(m.whitelist) {
+				m.whitelist = m.whitelist[:i]
+			} else {
+				m.whitelist = append(m.whitelist[:i], m.whitelist[i+1:]...)
+			}
+			goto loop
+		}
+	}
+	return
+}

api/service/stagedstreamsync/block_hash_result.go

@@ -0,0 +1,75 @@
+package stagedstreamsync
+
+import (
+	"sync"
+
+	"github.com/ethereum/go-ethereum/common"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+)
+
+type (
+	blockHashResults struct {
+		bns     []uint64
+		results []map[sttypes.StreamID]common.Hash
+
+		lock sync.Mutex
+	}
+)
+
+func newBlockHashResults(bns []uint64) *blockHashResults {
+	results := make([]map[sttypes.StreamID]common.Hash, 0, len(bns))
+	for range bns {
+		results = append(results, make(map[sttypes.StreamID]common.Hash))
+	}
+	return &blockHashResults{
+		bns:     bns,
+		results: results,
+	}
+}
+
+func (res *blockHashResults) addResult(hashes []common.Hash, stid sttypes.StreamID) {
+	res.lock.Lock()
+	defer res.lock.Unlock()
+
+	for i, h := range hashes {
+		if h == emptyHash {
+			return // nil block hash reached
+		}
+		res.results[i][stid] = h
+	}
+	return
+}
+
+func (res *blockHashResults) computeLongestHashChain() ([]common.Hash, []sttypes.StreamID) {
+	var (
+		whitelist map[sttypes.StreamID]struct{}
+		hashChain []common.Hash
+	)
+	for _, result := range res.results {
+		hash, nextWl := countHashMaxVote(result, whitelist)
+		if hash == emptyHash {
+			break
+		}
+		hashChain = append(hashChain, hash)
+		whitelist = nextWl
+	}
+
+	sts := make([]sttypes.StreamID, 0, len(whitelist))
+	for st := range whitelist {
+		sts = append(sts, st)
+	}
+	return hashChain, sts
+}
+
+func (res *blockHashResults) numBlocksWithResults() int {
+	res.lock.Lock()
+	defer res.lock.Unlock()
+
+	cnt := 0
+	for _, result := range res.results {
+		if len(result) != 0 {
+			cnt++
+		}
+	}
+	return cnt
+}

api/service/stagedstreamsync/block_manager.go

@@ -0,0 +1,172 @@
+package stagedstreamsync
+
+import (
+	"sync"
+
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/ledgerwatch/erigon-lib/kv"
+	"github.com/rs/zerolog"
+)
+
+type BlockDownloadDetails struct {
+	loopID   int
+	streamID sttypes.StreamID
+}
+
+// blockDownloadManager is the helper structure for get blocks request management
+type blockDownloadManager struct {
+	chain blockChain
+	tx    kv.RwTx
+
+	targetBN   uint64
+	requesting map[uint64]struct{}             // block numbers that have been assigned to workers but not received
+	processing map[uint64]struct{}             // block numbers received requests but not inserted
+	retries    *prioritizedNumbers             // requests where error happens
+	rq         *resultQueue                    // result queue wait to be inserted into blockchain
+	bdd        map[uint64]BlockDownloadDetails // details about how this block was downloaded
+
+	logger zerolog.Logger
+	lock   sync.Mutex
+}
+
+func newBlockDownloadManager(tx kv.RwTx, chain blockChain, targetBN uint64, logger zerolog.Logger) *blockDownloadManager {
+	return &blockDownloadManager{
+		chain:      chain,
+		tx:         tx,
+		targetBN:   targetBN,
+		requesting: make(map[uint64]struct{}),
+		processing: make(map[uint64]struct{}),
+		retries:    newPrioritizedNumbers(),
+		rq:         newResultQueue(),
+		bdd:        make(map[uint64]BlockDownloadDetails),
+		logger:     logger,
+	}
+}
+
+// GetNextBatch get the next block numbers batch
+func (gbm *blockDownloadManager) GetNextBatch() []uint64 {
+	gbm.lock.Lock()
+	defer gbm.lock.Unlock()
+
+	cap := BlocksPerRequest
+
+	bns := gbm.getBatchFromRetries(cap)
+	if len(bns) > 0 {
+		cap -= len(bns)
+		gbm.addBatchToRequesting(bns)
+	}
+
+	if gbm.availableForMoreTasks() {
+		addBNs := gbm.getBatchFromUnprocessed(cap)
+		gbm.addBatchToRequesting(addBNs)
+		bns = append(bns, addBNs...)
+	}
+
+	return bns
+}
+
+// HandleRequestError handles the error result
+func (gbm *blockDownloadManager) HandleRequestError(bns []uint64, err error, streamID sttypes.StreamID) {
+	gbm.lock.Lock()
+	defer gbm.lock.Unlock()
+
+	// add requested block numbers to retries
+	for _, bn := range bns {
+		delete(gbm.requesting, bn)
+		gbm.retries.push(bn)
+	}
+}
+
+// HandleRequestResult handles get blocks result
+func (gbm *blockDownloadManager) HandleRequestResult(bns []uint64, blockBytes [][]byte, sigBytes [][]byte, loopID int, streamID sttypes.StreamID) error {
+	gbm.lock.Lock()
+	defer gbm.lock.Unlock()
+
+	for i, bn := range bns {
+		delete(gbm.requesting, bn)
+		if len(blockBytes[i]) <= 1 {
+			gbm.retries.push(bn)
+		} else {
+			gbm.processing[bn] = struct{}{}
+			gbm.bdd[bn] = BlockDownloadDetails{
+				loopID:   loopID,
+				streamID: streamID,
+			}
+		}
+	}
+	return nil
+}
+
+// SetDownloadDetails sets the download details for a batch of blocks
+func (gbm *blockDownloadManager) SetDownloadDetails(bns []uint64, loopID int, streamID sttypes.StreamID) error {
+	gbm.lock.Lock()
+	defer gbm.lock.Unlock()
+
+	for _, bn := range bns {
+		gbm.bdd[bn] = BlockDownloadDetails{
+			loopID:   loopID,
+			streamID: streamID,
+		}
+	}
+	return nil
+}
+
+// GetDownloadDetails returns the download details for a block
+func (gbm *blockDownloadManager) GetDownloadDetails(blockNumber uint64) (loopID int, streamID sttypes.StreamID) {
+	gbm.lock.Lock()
+	defer gbm.lock.Unlock()
+
+	return gbm.bdd[blockNumber].loopID, gbm.bdd[blockNumber].streamID
+}
+
+// getBatchFromRetries get the block number batch to be requested from retries.
+func (gbm *blockDownloadManager) getBatchFromRetries(cap int) []uint64 {
+	var (
+		requestBNs []uint64
+		curHeight  = gbm.chain.CurrentBlock().NumberU64()
+	)
+	for cnt := 0; cnt < cap; cnt++ {
+		bn := gbm.retries.pop()
+		if bn == 0 {
+			break // no more retries
+		}
+		if bn <= curHeight {
+			continue
+		}
+		requestBNs = append(requestBNs, bn)
+	}
+	return requestBNs
+}
+
+// getBatchFromUnprocessed returns a batch of block numbers to be requested from unprocessed.
+func (gbm *blockDownloadManager) getBatchFromUnprocessed(cap int) []uint64 {
+	var (
+		requestBNs []uint64
+		curHeight  = gbm.chain.CurrentBlock().NumberU64()
+	)
+	bn := curHeight + 1
+	// TODO: this algorithm can be potentially optimized.
+	for cnt := 0; cnt < cap && bn <= gbm.targetBN; cnt++ {
+		for bn <= gbm.targetBN {
+			_, ok1 := gbm.requesting[bn]
+			_, ok2 := gbm.processing[bn]
+			if !ok1 && !ok2 {
+				requestBNs = append(requestBNs, bn)
+				bn++
+				break
+			}
+			bn++
+		}
+	}
+	return requestBNs
+}
+
+func (gbm *blockDownloadManager) availableForMoreTasks() bool {
+	return gbm.rq.results.Len() < SoftQueueCap
+}
+
+func (gbm *blockDownloadManager) addBatchToRequesting(bns []uint64) {
+	for _, bn := range bns {
+		gbm.requesting[bn] = struct{}{}
+	}
+}

api/service/stagedstreamsync/const.go

@@ -0,0 +1,81 @@
+package stagedstreamsync
+
+import (
+	"time"
+
+	"github.com/harmony-one/harmony/core/types"
+	nodeconfig "github.com/harmony-one/harmony/internal/configs/node"
+)
+
+const (
+	BlocksPerRequest      int = 10 // number of blocks for each request
+	BlocksPerInsertion    int = 50 // number of blocks for each insert batch
+	BlockHashesPerRequest int = 20 // number of get block hashes for short range sync
+	BlockByHashesUpperCap int = 10 // number of get blocks by hashes upper cap
+	BlockByHashesLowerCap int = 3  // number of get blocks by hashes lower cap
+
+	LastMileBlocksThreshold int = 10
+
+	// SoftQueueCap is the soft cap of size in resultQueue. When the queue size is larger than this limit,
+	// no more request will be assigned to workers to wait for InsertChain to finish.
+	SoftQueueCap int = 100
+
+	// DefaultConcurrency is the default settings for concurrency
+	DefaultConcurrency = 4
+
+	// ShortRangeTimeout is the timeout for each short range sync, which allow short range sync
+	// to restart automatically when stuck in `getBlockHashes`
+	ShortRangeTimeout = 1 * time.Minute
+)
+
+type (
+	// Config is the downloader config
+	Config struct {
+		// Only run stream sync protocol as a server.
+		// TODO: remove this when stream sync is fully up.
+		ServerOnly bool
+
+		// parameters
+		Network     nodeconfig.NetworkType
+		Concurrency int // Number of concurrent sync requests
+		MinStreams  int // Minimum number of streams to do sync
+		InitStreams int // Number of streams requirement for initial bootstrap
+
+		// stream manager config
+		SmSoftLowCap int
+		SmHardLowCap int
+		SmHiCap      int
+		SmDiscBatch  int
+
+		// config for beacon config
+		BHConfig *BeaconHelperConfig
+
+		// log the stage progress
+		LogProgress bool
+	}
+
+	// BeaconHelperConfig is the extra config used for beaconHelper which uses
+	// pub-sub block message to do sync.
+	BeaconHelperConfig struct {
+		BlockC     <-chan *types.Block
+		InsertHook func()
+	}
+)
+
+func (c *Config) fixValues() {
+	if c.Concurrency == 0 {
+		c.Concurrency = DefaultConcurrency
+	}
+	if c.Concurrency > c.MinStreams {
+		c.MinStreams = c.Concurrency
+	}
+	if c.MinStreams > c.InitStreams {
+		c.InitStreams = c.MinStreams
+	}
+	if c.MinStreams > c.SmSoftLowCap {
+		c.SmSoftLowCap = c.MinStreams
+	}
+	if c.MinStreams > c.SmHardLowCap {
+		c.SmHardLowCap = c.MinStreams
+	}
+}

api/service/stagedstreamsync/default_stages.go

@@ -0,0 +1,87 @@
+package stagedstreamsync
+
+import (
+	"context"
+)
+
+type ForwardOrder []SyncStageID
+type RevertOrder []SyncStageID
+type CleanUpOrder []SyncStageID
+
+var DefaultForwardOrder = ForwardOrder{
+	Heads,
+	SyncEpoch,
+	ShortRange,
+	BlockBodies,
+	// Stages below don't use Internet
+	States,
+	Finish,
+}
+
+var DefaultRevertOrder = RevertOrder{
+	Finish,
+	States,
+	BlockBodies,
+	ShortRange,
+	SyncEpoch,
+	Heads,
+}
+
+var DefaultCleanUpOrder = CleanUpOrder{
+	Finish,
+	States,
+	BlockBodies,
+	ShortRange,
+	SyncEpoch,
+	Heads,
+}
+
+func DefaultStages(ctx context.Context,
+	headsCfg StageHeadsCfg,
+	seCfg StageEpochCfg,
+	srCfg StageShortRangeCfg,
+	bodiesCfg StageBodiesCfg,
+	statesCfg StageStatesCfg,
+	finishCfg StageFinishCfg,
+) []*Stage {
+
+	handlerStageHeads := NewStageHeads(headsCfg)
+	handlerStageShortRange := NewStageShortRange(srCfg)
+	handlerStageEpochSync := NewStageEpoch(seCfg)
+	handlerStageBodies := NewStageBodies(bodiesCfg)
+	handlerStageStates := NewStageStates(statesCfg)
+	handlerStageFinish := NewStageFinish(finishCfg)
+
+	return []*Stage{
+		{
+			ID:          Heads,
+			Description: "Retrieve Chain Heads",
+			Handler:     handlerStageHeads,
+		},
+		{
+			ID:          SyncEpoch,
+			Description: "Sync only Last Block of Epoch",
+			Handler:     handlerStageEpochSync,
+		},
+		{
+			ID:          ShortRange,
+			Description: "Short Range Sync",
+			Handler:     handlerStageShortRange,
+		},
+		{
+			ID:          BlockBodies,
+			Description: "Retrieve Block Bodies",
+			Handler:     handlerStageBodies,
+		},
+		{
+			ID:          States,
+			Description: "Update Blockchain State",
+			Handler:     handlerStageStates,
+		},
+		{
+			ID:          Finish,
+			Description: "Finalize Changes",
+			Handler:     handlerStageFinish,
+		},
+	}
+}

api/service/stagedstreamsync/downloader.go

@@ -0,0 +1,303 @@
+package stagedstreamsync
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/ethereum/go-ethereum/event"
+	"github.com/pkg/errors"
+	"github.com/rs/zerolog"
+
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/core/types"
+	"github.com/harmony-one/harmony/crypto/bls"
+	"github.com/harmony-one/harmony/internal/chain"
+	nodeconfig "github.com/harmony-one/harmony/internal/configs/node"
+	"github.com/harmony-one/harmony/internal/utils"
+	"github.com/harmony-one/harmony/p2p"
+	"github.com/harmony-one/harmony/p2p/stream/common/streammanager"
+	"github.com/harmony-one/harmony/p2p/stream/protocols/sync"
+	"github.com/harmony-one/harmony/shard"
+)
+
+type (
+	// Downloader is responsible for sync task of one shard
+	Downloader struct {
+		bc                 blockChain
+		syncProtocol       syncProtocol
+		bh                 *beaconHelper
+		stagedSyncInstance *StagedStreamSync
+
+		downloadC chan struct{}
+		closeC    chan struct{}
+		ctx       context.Context
+		cancel    func()
+
+		config Config
+		logger zerolog.Logger
+	}
+)
+
+// NewDownloader creates a new downloader
+func NewDownloader(host p2p.Host, bc core.BlockChain, config Config) *Downloader {
+	config.fixValues()
+
+	sp := sync.NewProtocol(sync.Config{
+		Chain:     bc,
+		Host:      host.GetP2PHost(),
+		Discovery: host.GetDiscovery(),
+		ShardID:   nodeconfig.ShardID(bc.ShardID()),
+		Network:   config.Network,
+
+		SmSoftLowCap: config.SmSoftLowCap,
+		SmHardLowCap: config.SmHardLowCap,
+		SmHiCap:      config.SmHiCap,
+		DiscBatch:    config.SmDiscBatch,
+	})
+
+	host.AddStreamProtocol(sp)
+
+	var bh *beaconHelper
+	if config.BHConfig != nil && bc.ShardID() == shard.BeaconChainShardID {
+		bh = newBeaconHelper(bc, config.BHConfig.BlockC, config.BHConfig.InsertHook)
+	}
+
+	logger := utils.Logger().With().Str("module", "StagedStreamSync").Uint32("ShardID", bc.ShardID()).Logger()
+
+	ctx, cancel := context.WithCancel(context.Background())
+
+	//TODO: use mem db should be in config file
+	stagedSyncInstance, err := CreateStagedSync(ctx, bc, false, sp, config, logger, config.LogProgress)
+	if err != nil {
+		return nil
+	}
+
+	return &Downloader{
+		bc:                 bc,
+		syncProtocol:       sp,
+		bh:                 bh,
+		stagedSyncInstance: stagedSyncInstance,
+
+		downloadC: make(chan struct{}),
+		closeC:    make(chan struct{}),
+		ctx:       ctx,
+		cancel:    cancel,
+
+		config: config,
+		logger: logger,
+	}
+}
+
+// Start starts the downloader
+func (d *Downloader) Start() {
+	go func() {
+		d.waitForBootFinish()
+		fmt.Printf("boot completed for shard %d, %d streams are connected\n", d.bc.ShardID(), d.syncProtocol.NumStreams())
+		d.loop()
+	}()
+
+	if d.bh != nil {
+		d.bh.start()
+	}
+}
+
+// Close closes the downloader
+func (d *Downloader) Close() {
+	close(d.closeC)
+	d.cancel()
+
+	if d.bh != nil {
+		d.bh.close()
+	}
+}
+
+// DownloadAsync triggers the download async.
+func (d *Downloader) DownloadAsync() {
+	select {
+	case d.downloadC <- struct{}{}:
+		consensusTriggeredDownloadCounterVec.With(d.promLabels()).Inc()
+
+	case <-time.After(100 * time.Millisecond):
+	}
+}
+
+// NumPeers returns the number of peers connected of a specific shard.
+func (d *Downloader) NumPeers() int {
+	return d.syncProtocol.NumStreams()
+}
+
+// IsSyncing returns the current sync status
+func (d *Downloader) SyncStatus() (bool, uint64, uint64) {
+	syncing, target := d.stagedSyncInstance.status.get()
+	if !syncing {
+		target = d.bc.CurrentBlock().NumberU64()
+	}
+	return syncing, target, 0
+}
+
+// SubscribeDownloadStarted subscribes download started
+func (d *Downloader) SubscribeDownloadStarted(ch chan struct{}) event.Subscription {
+	d.stagedSyncInstance.evtDownloadStartedSubscribed = true
+	return d.stagedSyncInstance.evtDownloadStarted.Subscribe(ch)
+}
+
+// SubscribeDownloadFinished subscribes the download finished
+func (d *Downloader) SubscribeDownloadFinished(ch chan struct{}) event.Subscription {
+	d.stagedSyncInstance.evtDownloadFinishedSubscribed = true
+	return d.stagedSyncInstance.evtDownloadFinished.Subscribe(ch)
+}
+
+// waitForBootFinish waits for stream manager to finish the initial discovery and have
+// enough peers to start downloader
+func (d *Downloader) waitForBootFinish() {
+	evtCh := make(chan streammanager.EvtStreamAdded, 1)
+	sub := d.syncProtocol.SubscribeAddStreamEvent(evtCh)
+	defer sub.Unsubscribe()
+
+	checkCh := make(chan struct{}, 1)
+	trigger := func() {
+		select {
+		case checkCh <- struct{}{}:
+		default:
+		}
+	}
+	trigger()
+
+	t := time.NewTicker(10 * time.Second)
+	defer t.Stop()
+	for {
+		select {
+		case <-t.C:
+			trigger()
+
+		case <-evtCh:
+			trigger()
+
+		case <-checkCh:
+			if d.syncProtocol.NumStreams() >= d.config.InitStreams {
+				return
+			}
+		case <-d.closeC:
+			return
+		}
+	}
+}
+
+func (d *Downloader) loop() {
+	ticker := time.NewTicker(10 * time.Second)
+	defer ticker.Stop()
+	initSync := d.bc.ShardID() != shard.BeaconChainShardID
+	trigger := func() {
+		select {
+		case d.downloadC <- struct{}{}:
+		case <-time.After(100 * time.Millisecond):
+		}
+	}
+	go trigger()
+
+	for {
+		select {
+		case <-ticker.C:
+			go trigger()
+
+		case <-d.downloadC:
+			addedBN, err := d.stagedSyncInstance.doSync(d.ctx, initSync)
+			if err != nil {
+				//TODO: if there is a bad block which can't be resolved
+				if d.stagedSyncInstance.invalidBlock.Active {
+					numTriedStreams := len(d.stagedSyncInstance.invalidBlock.StreamID)
+					// if many streams couldn't solve it, then that's an unresolvable bad block
+					if numTriedStreams >= d.config.InitStreams {
+						if !d.stagedSyncInstance.invalidBlock.IsLogged {
+							fmt.Println("unresolvable bad block:", d.stagedSyncInstance.invalidBlock.Number)
+							d.stagedSyncInstance.invalidBlock.IsLogged = true
+						}
+						//TODO: if we don't have any new or untried stream in the list, sleep or panic
+					}
+				}
+
+				// If error happens, sleep 5 seconds and retry
+				d.logger.Error().
+					Err(err).
+					Bool("initSync", initSync).
+					Msg(WrapStagedSyncMsg("sync loop failed"))
+				go func() {
+					time.Sleep(5 * time.Second)
+					trigger()
+				}()
+				time.Sleep(1 * time.Second)
+				continue
+			}
+			d.logger.Info().Int("block added", addedBN).
+				Uint64("current height", d.bc.CurrentBlock().NumberU64()).
+				Bool("initSync", initSync).
+				Uint32("shard", d.bc.ShardID()).
+				Msg(WrapStagedSyncMsg("sync finished"))
+
+			if addedBN != 0 {
+				// If block number has been changed, trigger another sync
+				// and try to add last mile from pub-sub (blocking)
+				go trigger()
+				if d.bh != nil {
+					d.bh.insertSync()
+				}
+			}
+			d.stagedSyncInstance.initSync = false
+			initSync = false
+
+		case <-d.closeC:
+			return
+		}
+	}
+}
+
+var emptySigVerifyErr *sigVerifyErr
+
+type sigVerifyErr struct {
+	err error
+}
+
+func (e *sigVerifyErr) Error() string {
+	return fmt.Sprintf("[VerifyHeaderSignature] %v", e.err.Error())
+}
+
+func verifyAndInsertBlocks(bc blockChain, blocks types.Blocks) (int, error) {
+	for i, block := range blocks {
+		if err := verifyAndInsertBlock(bc, block, blocks[i+1:]...); err != nil {
+			return i, err
+		}
+	}
+	return len(blocks), nil
+}
+
+func verifyAndInsertBlock(bc blockChain, block *types.Block, nextBlocks ...*types.Block) error {
+	var (
+		sigBytes bls.SerializedSignature
+		bitmap   []byte
+		err      error
+	)
+	if len(nextBlocks) > 0 {
+		// get commit sig from the next block
+		next := nextBlocks[0]
+		sigBytes = next.Header().LastCommitSignature()
+		bitmap = next.Header().LastCommitBitmap()
+	} else {
+		// get commit sig from current block
+		sigBytes, bitmap, err = chain.ParseCommitSigAndBitmap(block.GetCurrentCommitSig())
+		if err != nil {
+			return errors.Wrap(err, "parse commitSigAndBitmap")
+		}
+	}
+
+	if err := bc.Engine().VerifyHeaderSignature(bc, block.Header(), sigBytes, bitmap); err != nil {
+		return &sigVerifyErr{err}
+	}
+	if err := bc.Engine().VerifyHeader(bc, block.Header(), true); err != nil {
+		return errors.Wrap(err, "[VerifyHeader]")
+	}
+	if _, err := bc.InsertChain(types.Blocks{block}, false); err != nil {
+		return errors.Wrap(err, "[InsertChain]")
+	}
+	return nil
+}

api/service/stagedstreamsync/downloaders.go

@@ -0,0 +1,96 @@
+package stagedstreamsync
+
+import (
+	"github.com/harmony-one/abool"
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/p2p"
+)
+
+// Downloaders is the set of downloaders
+type Downloaders struct {
+	ds     map[uint32]*Downloader
+	active *abool.AtomicBool
+
+	config Config
+}
+
+// NewDownloaders creates Downloaders for sync of multiple blockchains
+func NewDownloaders(host p2p.Host, bcs []core.BlockChain, config Config) *Downloaders {
+	ds := make(map[uint32]*Downloader)
+
+	for _, bc := range bcs {
+		if bc == nil {
+			continue
+		}
+		if _, ok := ds[bc.ShardID()]; ok {
+			continue
+		}
+		ds[bc.ShardID()] = NewDownloader(host, bc, config)
+	}
+	return &Downloaders{
+		ds:     ds,
+		active: abool.New(),
+		config: config,
+	}
+}
+
+// Start starts the downloaders
+func (ds *Downloaders) Start() {
+	if ds.config.ServerOnly {
+		// Run in server only mode. Do not start downloaders.
+		return
+	}
+	ds.active.Set()
+	for _, d := range ds.ds {
+		d.Start()
+	}
+}
+
+// Close closes the downloaders
+func (ds *Downloaders) Close() {
+	if ds.config.ServerOnly {
+		// Run in server only mode. Downloaders not started.
+		return
+	}
+	ds.active.UnSet()
+	for _, d := range ds.ds {
+		d.Close()
+	}
+}
+
+// DownloadAsync triggers a download
+func (ds *Downloaders) DownloadAsync(shardID uint32) {
+	d, ok := ds.ds[shardID]
+	if !ok && d != nil {
+		d.DownloadAsync()
+	}
+}
+
+// GetShardDownloader returns the downloader with the given shard ID
+func (ds *Downloaders) GetShardDownloader(shardID uint32) *Downloader {
+	return ds.ds[shardID]
+}
+
+// NumPeers returns the connected peers for each shard
+func (ds *Downloaders) NumPeers() map[uint32]int {
+	res := make(map[uint32]int)
+
+	for sid, d := range ds.ds {
+		res[sid] = d.NumPeers()
+	}
+	return res
+}
+
+// SyncStatus returns whether the given shard is doing syncing task and the target block number
+func (ds *Downloaders) SyncStatus(shardID uint32) (bool, uint64, uint64) {
+	d, ok := ds.ds[shardID]
+	if !ok {
+		return false, 0, 0
+	}
+	return d.SyncStatus()
+}
+
+// IsActive returns whether the downloader is active
+func (ds *Downloaders) IsActive() bool {
+	return ds.active.IsSet()
+}

api/service/stagedstreamsync/errors.go

@@ -0,0 +1,56 @@
+package stagedstreamsync
+
+import (
+	"fmt"
+)
+
+// Errors ...
+var (
+	ErrRegistrationFail                   = WrapStagedSyncError("registration failed")
+	ErrGetBlock                           = WrapStagedSyncError("get block failed")
+	ErrGetBlockHash                       = WrapStagedSyncError("get block hash failed")
+	ErrGetConsensusHashes                 = WrapStagedSyncError("get consensus hashes failed")
+	ErrGenStateSyncTaskQueue              = WrapStagedSyncError("generate state sync task queue failed")
+	ErrDownloadBlocks                     = WrapStagedSyncError("get download blocks failed")
+	ErrUpdateBlockAndStatus               = WrapStagedSyncError("update block and status failed")
+	ErrGenerateNewState                   = WrapStagedSyncError("get generate new state failed")
+	ErrFetchBlockHashProgressFail         = WrapStagedSyncError("fetch cache progress for block hashes stage failed")
+	ErrFetchCachedBlockHashFail           = WrapStagedSyncError("fetch cached block hashes failed")
+	ErrNotEnoughBlockHashes               = WrapStagedSyncError("peers haven't sent all requested block hashes")
+	ErrRetrieveCachedProgressFail         = WrapStagedSyncError("retrieving cache progress for block hashes stage failed")
+	ErrRetrieveCachedHashProgressFail     = WrapStagedSyncError("retrieving cache progress for block hashes stage failed")
+	ErrSaveBlockHashesProgressFail        = WrapStagedSyncError("saving progress for block hashes stage failed")
+	ErrSaveCachedBlockHashesProgressFail  = WrapStagedSyncError("saving cache progress for block hashes stage failed")
+	ErrSavingCacheLastBlockHashFail       = WrapStagedSyncError("saving cache last block hash for block hashes stage failed")
+	ErrCachingBlockHashFail               = WrapStagedSyncError("caching downloaded block hashes failed")
+	ErrCommitTransactionFail              = WrapStagedSyncError("failed to write db commit")
+	ErrUnexpectedNumberOfBlocks           = WrapStagedSyncError("unexpected number of block delivered")
+	ErrSavingBodiesProgressFail           = WrapStagedSyncError("saving progress for block bodies stage failed")
+	ErrAddTasksToQueueFail                = WrapStagedSyncError("cannot add task to queue")
+	ErrSavingCachedBodiesProgressFail     = WrapStagedSyncError("saving cache progress for blocks stage failed")
+	ErrRetrievingCachedBodiesProgressFail = WrapStagedSyncError("retrieving cache progress for blocks stage failed")
+	ErrNoConnectedPeers                   = WrapStagedSyncError("haven't connected to any peer yet!")
+	ErrNotEnoughConnectedPeers            = WrapStagedSyncError("not enough connected peers")
+	ErrSaveStateProgressFail              = WrapStagedSyncError("saving progress for block States stage failed")
+	ErrPruningCursorCreationFail          = WrapStagedSyncError("failed to create cursor for pruning")
+	ErrInvalidBlockNumber                 = WrapStagedSyncError("invalid block number")
+	ErrInvalidBlockBytes                  = WrapStagedSyncError("invalid block bytes to insert into chain")
+	ErrAddTaskFailed                      = WrapStagedSyncError("cannot add task to queue")
+	ErrNodeNotEnoughBlockHashes           = WrapStagedSyncError("some of the nodes didn't provide all block hashes")
+	ErrCachingBlocksFail                  = WrapStagedSyncError("caching downloaded block bodies failed")
+	ErrSaveBlocksFail                     = WrapStagedSyncError("save downloaded block bodies failed")
+	ErrStageNotFound                      = WrapStagedSyncError("stage not found")
+	ErrSomeNodesNotReady                  = WrapStagedSyncError("some nodes are not ready")
+	ErrSomeNodesBlockHashFail             = WrapStagedSyncError("some nodes failed to download block hashes")
+	ErrMaxPeerHeightFail                  = WrapStagedSyncError("get max peer height failed")
+)
+
+// WrapStagedSyncError wraps errors for staged sync and returns error object
+func WrapStagedSyncError(context string) error {
+	return fmt.Errorf("[STAGED_STREAM_SYNC]: %s", context)
+}
+
+// WrapStagedSyncMsg wraps message for staged sync and returns string
+func WrapStagedSyncMsg(context string) string {
+	return fmt.Sprintf("[STAGED_STREAM_SYNC]: %s", context)
+}

api/service/stagedstreamsync/helpers.go

@@ -0,0 +1,114 @@
+package stagedstreamsync
+
+import (
+	"encoding/binary"
+	"fmt"
+	"math"
+
+	"github.com/ethereum/go-ethereum/common"
+	"github.com/harmony-one/harmony/core/types"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/pkg/errors"
+)
+
+func marshalData(blockNumber uint64) []byte {
+	return encodeBigEndian(blockNumber)
+}
+
+func unmarshalData(data []byte) (uint64, error) {
+	if len(data) == 0 {
+		return 0, nil
+	}
+	if len(data) < 8 {
+		return 0, fmt.Errorf("value must be at least 8 bytes, got %d", len(data))
+	}
+	return binary.BigEndian.Uint64(data[:8]), nil
+}
+
+func encodeBigEndian(n uint64) []byte {
+	var v [8]byte
+	binary.BigEndian.PutUint64(v[:], n)
+	return v[:]
+}
+
+func divideCeil(x, y int) int {
+	fVal := float64(x) / float64(y)
+	return int(math.Ceil(fVal))
+}
+
+// computeBlockNumberByMaxVote computes the target block number by max vote.
+func computeBlockNumberByMaxVote(votes map[sttypes.StreamID]uint64) uint64 {
+	var (
+		nm     = make(map[uint64]int)
+		res    uint64
+		maxCnt int
+	)
+	for _, bn := range votes {
+		_, ok := nm[bn]
+		if !ok {
+			nm[bn] = 0
+		}
+		nm[bn]++
+		cnt := nm[bn]
+
+		if cnt > maxCnt || (cnt == maxCnt && bn > res) {
+			res = bn
+			maxCnt = cnt
+		}
+	}
+	return res
+}
+
+func checkGetBlockByHashesResult(blocks []*types.Block, hashes []common.Hash) error {
+	if len(blocks) != len(hashes) {
+		return errors.New("unexpected number of getBlocksByHashes result")
+	}
+	for i, block := range blocks {
+		if block == nil {
+			return errors.New("nil block found")
+		}
+		if block.Hash() != hashes[i] {
+			return fmt.Errorf("unexpected block hash: %x / %x", block.Hash(), hashes[i])
+		}
+	}
+	return nil
+}
+
+func countHashMaxVote(m map[sttypes.StreamID]common.Hash, whitelist map[sttypes.StreamID]struct{}) (common.Hash, map[sttypes.StreamID]struct{}) {
+	var (
+		voteM  = make(map[common.Hash]int)
+		res    common.Hash
+		maxCnt = 0
+	)
+
+	for st, h := range m {
+		if len(whitelist) != 0 {
+			if _, ok := whitelist[st]; !ok {
+				continue
+			}
+		}
+		if _, ok := voteM[h]; !ok {
+			voteM[h] = 0
+		}
+		voteM[h]++
+		if voteM[h] > maxCnt {
+			maxCnt = voteM[h]
+			res = h
+		}
+	}
+
+	nextWl := make(map[sttypes.StreamID]struct{})
+	for st, h := range m {
+		if h != res {
+			continue
+		}
+		if len(whitelist) != 0 {
+			if _, ok := whitelist[st]; ok {
+				nextWl[st] = struct{}{}
+			}
+		} else {
+			nextWl[st] = struct{}{}
+		}
+	}
+	return res, nextWl
+}

api/service/stagedstreamsync/metric.go

@@ -0,0 +1,98 @@
+package stagedstreamsync
+
+import (
+	"fmt"
+
+	prom "github.com/harmony-one/harmony/api/service/prometheus"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+func init() {
+	prom.PromRegistry().MustRegister(
+		consensusTriggeredDownloadCounterVec,
+		longRangeSyncedBlockCounterVec,
+		longRangeFailInsertedBlockCounterVec,
+		numShortRangeCounterVec,
+		numFailedDownloadCounterVec,
+		numBlocksInsertedShortRangeHistogramVec,
+		numBlocksInsertedBeaconHelperCounter,
+	)
+}
+
+var (
+	consensusTriggeredDownloadCounterVec = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "hmy",
+			Subsystem: "downloader",
+			Name:      "consensus_trigger",
+			Help:      "number of times consensus triggered download task",
+		},
+		[]string{"ShardID"},
+	)
+
+	longRangeSyncedBlockCounterVec = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "hmy",
+			Subsystem: "downloader",
+			Name:      "num_blocks_synced_long_range",
+			Help:      "number of blocks synced in long range sync",
+		},
+		[]string{"ShardID"},
+	)
+
+	longRangeFailInsertedBlockCounterVec = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "hmy",
+			Subsystem: "downloader",
+			Name:      "num_blocks_failed_long_range",
+			Help:      "number of blocks failed to insert into change in long range sync",
+		},
+		[]string{"ShardID", "error"},
+	)
+
+	numShortRangeCounterVec = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "hmy",
+			Subsystem: "downloader",
+			Name:      "num_short_range",
+			Help:      "number of short range sync is triggered",
+		},
+		[]string{"ShardID"},
+	)
+
+	numFailedDownloadCounterVec = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "hmy",
+			Subsystem: "downloader",
+			Name:      "failed_download",
+			Help:      "number of downloading is failed",
+		},
+		[]string{"ShardID", "error"},
+	)
+
+	numBlocksInsertedShortRangeHistogramVec = prometheus.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Namespace: "hmy",
+			Subsystem: "downloader",
+			Name:      "num_blocks_inserted_short_range",
+			Help:      "number of blocks inserted for each short range sync",
+			// Buckets: 0, 1, 2, 4, +INF (capped at 10)
+			Buckets: prometheus.ExponentialBuckets(0.5, 2, 5),
+		},
+		[]string{"ShardID"},
+	)
+
+	numBlocksInsertedBeaconHelperCounter = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Namespace: "hmy",
+			Subsystem: "downloader",
+			Name:      "num_blocks_inserted_beacon_helper",
+			Help:      "number of blocks inserted from beacon helper",
+		},
+	)
+)
+
+func (d *Downloader) promLabels() prometheus.Labels {
+	sid := d.bc.ShardID()
+	return prometheus.Labels{"ShardID": fmt.Sprintf("%d", sid)}
+}

api/service/stagedstreamsync/service.go

@@ -0,0 +1,30 @@
+package stagedstreamsync
+
+import (
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/p2p"
+)
+
+// StagedStreamSyncService is simply a adapter of downloaders, which support block synchronization
+type StagedStreamSyncService struct {
+	Downloaders *Downloaders
+}
+
+// NewService creates a new downloader service
+func NewService(host p2p.Host, bcs []core.BlockChain, config Config) *StagedStreamSyncService {
+	return &StagedStreamSyncService{
+		Downloaders: NewDownloaders(host, bcs, config),
+	}
+}
+
+// Start starts the service
+func (s *StagedStreamSyncService) Start() error {
+	s.Downloaders.Start()
+	return nil
+}
+
+// Stop stops the service
+func (s *StagedStreamSyncService) Stop() error {
+	s.Downloaders.Close()
+	return nil
+}

api/service/stagedstreamsync/short_range_helper.go

@@ -0,0 +1,221 @@
+package stagedstreamsync
+
+import (
+	"context"
+	"sync"
+	"time"
+
+	"github.com/ethereum/go-ethereum/common"
+	"github.com/harmony-one/harmony/core/types"
+	syncProto "github.com/harmony-one/harmony/p2p/stream/protocols/sync"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/pkg/errors"
+	"github.com/rs/zerolog"
+)
+
+type srHelper struct {
+	syncProtocol syncProtocol
+
+	ctx    context.Context
+	config Config
+	logger zerolog.Logger
+}
+
+func (sh *srHelper) getHashChain(bns []uint64) ([]common.Hash, []sttypes.StreamID, error) {
+	results := newBlockHashResults(bns)
+
+	var wg sync.WaitGroup
+	wg.Add(sh.config.Concurrency)
+
+	for i := 0; i != sh.config.Concurrency; i++ {
+		go func(index int) {
+			defer wg.Done()
+
+			hashes, stid, err := sh.doGetBlockHashesRequest(bns)
+			if err != nil {
+				sh.logger.Warn().Err(err).Str("StreamID", string(stid)).
+					Msg(WrapStagedSyncMsg("doGetBlockHashes return error"))
+				return
+			}
+			results.addResult(hashes, stid)
+		}(i)
+	}
+	wg.Wait()
+
+	select {
+	case <-sh.ctx.Done():
+		sh.logger.Info().Err(sh.ctx.Err()).Int("num blocks", results.numBlocksWithResults()).
+			Msg(WrapStagedSyncMsg("short range sync get hashes timed out"))
+		return nil, nil, sh.ctx.Err()
+	default:
+	}
+
+	hashChain, wl := results.computeLongestHashChain()
+	sh.logger.Info().Int("hashChain size", len(hashChain)).Int("whitelist", len(wl)).
+		Msg(WrapStagedSyncMsg("computeLongestHashChain result"))
+	return hashChain, wl, nil
+}
+
+func (sh *srHelper) getBlocksChain(bns []uint64) ([]*types.Block, sttypes.StreamID, error) {
+	return sh.doGetBlocksByNumbersRequest(bns)
+}
+
+func (sh *srHelper) getBlocksByHashes(hashes []common.Hash, whitelist []sttypes.StreamID) ([]*types.Block, []sttypes.StreamID, error) {
+	ctx, cancel := context.WithCancel(sh.ctx)
+	defer cancel()
+	m := newGetBlocksByHashManager(hashes, whitelist)
+
+	var (
+		wg      sync.WaitGroup
+		gErr    error
+		errLock sync.Mutex
+	)
+
+	concurrency := sh.config.Concurrency
+	if concurrency > m.numRequests() {
+		concurrency = m.numRequests()
+	}
+
+	wg.Add(concurrency)
+	for i := 0; i != concurrency; i++ {
+		go func(index int) {
+			defer wg.Done()
+			defer cancel() // it's ok to cancel context more than once
+
+			for {
+				if m.isDone() {
+					return
+				}
+				hashes, wl, err := m.getNextHashes()
+				if err != nil {
+					errLock.Lock()
+					gErr = err
+					errLock.Unlock()
+					return
+				}
+				if len(hashes) == 0 {
+					select {
+					case <-time.After(200 * time.Millisecond):
+						continue
+					case <-ctx.Done():
+						return
+					}
+				}
+				blocks, stid, err := sh.doGetBlocksByHashesRequest(ctx, hashes, wl)
+				if err != nil {
+					sh.logger.Warn().Err(err).
+						Str("StreamID", string(stid)).
+						Int("hashes", len(hashes)).
+						Int("index", index).
+						Msg(WrapStagedSyncMsg("getBlocksByHashes worker failed"))
+					m.handleResultError(hashes, stid)
+				} else {
+					m.addResult(hashes, blocks, stid)
+				}
+			}
+		}(i)
+	}
+	wg.Wait()
+
+	if gErr != nil {
+		return nil, nil, gErr
+	}
+	select {
+	case <-sh.ctx.Done():
+		res, _, _ := m.getResults()
+		sh.logger.Info().Err(sh.ctx.Err()).Int("num blocks", len(res)).
+			Msg(WrapStagedSyncMsg("short range sync get blocks timed out"))
+		return nil, nil, sh.ctx.Err()
+	default:
+	}
+
+	return m.getResults()
+}
+
+func (sh *srHelper) checkPrerequisites() error {
+	if sh.syncProtocol.NumStreams() < sh.config.Concurrency {
+		return errors.New("not enough streams")
+	}
+	return nil
+}
+
+func (sh *srHelper) prepareBlockHashNumbers(curNumber uint64, count int) []uint64 {
+
+	n := count
+	if count > BlockHashesPerRequest {
+		n = BlockHashesPerRequest
+	}
+	res := make([]uint64, 0, n)
+
+	for bn := curNumber + 1; bn <= curNumber+uint64(n); bn++ {
+		res = append(res, bn)
+	}
+	return res
+}
+
+func (sh *srHelper) doGetBlockHashesRequest(bns []uint64) ([]common.Hash, sttypes.StreamID, error) {
+	ctx, cancel := context.WithTimeout(sh.ctx, 1*time.Second)
+	defer cancel()
+
+	hashes, stid, err := sh.syncProtocol.GetBlockHashes(ctx, bns)
+	if err != nil {
+		sh.logger.Warn().Err(err).
+			Interface("block numbers", bns).
+			Str("stream", string(stid)).
+			Msg(WrapStagedSyncMsg("failed to doGetBlockHashesRequest"))
+		return nil, stid, err
+	}
+	if len(hashes) != len(bns) {
+		err := errors.New("unexpected get block hashes result delivered")
+		sh.logger.Warn().Err(err).
+			Str("stream", string(stid)).
+			Msg(WrapStagedSyncMsg("failed to doGetBlockHashesRequest"))
+		sh.syncProtocol.StreamFailed(stid, "unexpected get block hashes result delivered")
+		return nil, stid, err
+	}
+	return hashes, stid, nil
+}
+
+func (sh *srHelper) doGetBlocksByNumbersRequest(bns []uint64) ([]*types.Block, sttypes.StreamID, error) {
+	ctx, cancel := context.WithTimeout(sh.ctx, 10*time.Second)
+	defer cancel()
+
+	blocks, stid, err := sh.syncProtocol.GetBlocksByNumber(ctx, bns)
+	if err != nil {
+		sh.logger.Warn().Err(err).Str("stream", string(stid)).Msg(WrapStagedSyncMsg("failed to doGetBlockHashesRequest"))
+		return nil, stid, err
+	}
+	return blocks, stid, nil
+}
+
+func (sh *srHelper) doGetBlocksByHashesRequest(ctx context.Context, hashes []common.Hash, wl []sttypes.StreamID) ([]*types.Block, sttypes.StreamID, error) {
+	ctx, cancel := context.WithTimeout(sh.ctx, 10*time.Second)
+	defer cancel()
+
+	blocks, stid, err := sh.syncProtocol.GetBlocksByHashes(ctx, hashes,
+		syncProto.WithWhitelist(wl))
+	if err != nil {
+		sh.logger.Warn().Err(err).Str("stream", string(stid)).Msg("failed to getBlockByHashes")
+		return nil, stid, err
+	}
+	if err := checkGetBlockByHashesResult(blocks, hashes); err != nil {
+		sh.logger.Warn().Err(err).Str("stream", string(stid)).Msg(WrapStagedSyncMsg("failed to getBlockByHashes"))
+		sh.syncProtocol.StreamFailed(stid, "failed to getBlockByHashes")
+		return nil, stid, err
+	}
+	return blocks, stid, nil
+}
+
+func (sh *srHelper) removeStreams(sts []sttypes.StreamID) {
+	for _, st := range sts {
+		sh.syncProtocol.RemoveStream(st)
+	}
+}
+
+// blameAllStreams only not to blame all whitelisted streams when the it's not the last block signature verification failed.
+func (sh *srHelper) blameAllStreams(blocks types.Blocks, errIndex int, err error) bool {
+	if errors.As(err, &emptySigVerifyErr) && errIndex == len(blocks)-1 {
+		return false
+	}
+	return true
+}

api/service/stagedstreamsync/stage.go

@@ -0,0 +1,112 @@
+package stagedstreamsync
+
+import (
+	"context"
+	"errors"
+
+	"github.com/ethereum/go-ethereum/common"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/ledgerwatch/erigon-lib/kv"
+)
+
+type ExecFunc func(firstCycle bool, invalidBlockRevert bool, s *StageState, reverter Reverter, tx kv.RwTx) error
+
+type StageHandler interface {
+	// Exec is the execution function for the stage to move forward.
+	// * firstCycle - is it the first cycle of syncing.
+	// * invalidBlockRevert - whether the execution is to solve the invalid block
+	// * s - is the current state of the stage and contains stage data.
+	// * reverter - if the stage needs to cause reverting, `reverter` methods can be used.
+	Exec(firstCycle bool, invalidBlockRevert bool, s *StageState, reverter Reverter, tx kv.RwTx) error
+
+	// Revert is the reverting logic of the stage.
+	// * firstCycle - is it the first cycle of syncing.
+	// * u - contains information about the revert itself.
+	// * s - represents the state of this stage at the beginning of revert.
+	Revert(firstCycle bool, u *RevertState, s *StageState, tx kv.RwTx) error
+
+	// CleanUp is the execution function for the stage to prune old data.
+	// * firstCycle - is it the first cycle of syncing.
+	// * p - is the current state of the stage and contains stage data.
+	CleanUp(firstCycle bool, p *CleanUpState, tx kv.RwTx) error
+
+	// SetStageContext updates the context for stage
+	SetStageContext(ctx context.Context)
+}
+
+// Stage is a single sync stage in staged sync.
+type Stage struct {
+	// ID of the sync stage. Should not be empty and should be unique. It is recommended to prefix it with reverse domain to avoid clashes (`com.example.my-stage`).
+	ID SyncStageID
+	// Handler handles the logic for the stage
+	Handler StageHandler
+	// Description is a string that is shown in the logs.
+	Description string
+	// DisabledDescription shows in the log with a message if the stage is disabled. Here, you can show which command line flags should be provided to enable the page.
+	DisabledDescription string
+	// Disabled defines if the stage is disabled. It sets up when the stage is build by its `StageBuilder`.
+	Disabled bool
+}
+
+var ErrStopped = errors.New("stopped")
+var ErrRevert = errors.New("unwound")
+
+// StageState is the state of the stage.
+type StageState struct {
+	state       *StagedStreamSync
+	ID          SyncStageID
+	BlockNumber uint64 // BlockNumber is the current block number of the stage at the beginning of the state execution.
+}
+
+func (s *StageState) LogPrefix() string { return s.state.LogPrefix() }
+
+func (s *StageState) CurrentStageProgress(db kv.Getter) (uint64, error) {
+	return GetStageProgress(db, s.ID, s.state.isBeacon)
+}
+
+func (s *StageState) StageProgress(db kv.Getter, id SyncStageID) (uint64, error) {
+	return GetStageProgress(db, id, s.state.isBeacon)
+}
+
+// Update updates the stage state (current block number) in the database. Can be called multiple times during stage execution.
+func (s *StageState) Update(db kv.Putter, newBlockNum uint64) error {
+	return SaveStageProgress(db, s.ID, s.state.isBeacon, newBlockNum)
+}
+func (s *StageState) UpdateCleanUp(db kv.Putter, blockNum uint64) error {
+	return SaveStageCleanUpProgress(db, s.ID, s.state.isBeacon, blockNum)
+}
+
+// Reverter allows the stage to cause an revert.
+type Reverter interface {
+	// RevertTo begins staged sync revert to the specified block.
+	RevertTo(revertPoint uint64, invalidBlockNumber uint64, invalidBlockHash common.Hash, invalidBlockStreamID sttypes.StreamID)
+}
+
+// RevertState contains the information about revert.
+type RevertState struct {
+	ID          SyncStageID
+	RevertPoint uint64 // RevertPoint is the block to revert to.
+	state       *StagedStreamSync
+}
+
+func (u *RevertState) LogPrefix() string { return u.state.LogPrefix() }
+
+// Done updates the DB state of the stage.
+func (u *RevertState) Done(db kv.Putter) error {
+	return SaveStageProgress(db, u.ID, u.state.isBeacon, u.RevertPoint)
+}
+
+type CleanUpState struct {
+	ID              SyncStageID
+	ForwardProgress uint64 // progress of stage forward move
+	CleanUpProgress uint64 // progress of stage prune move. after sync cycle it become equal to ForwardProgress by Done() method
+	state           *StagedStreamSync
+}
+
+func (s *CleanUpState) LogPrefix() string { return s.state.LogPrefix() + " CleanUp" }
+func (s *CleanUpState) Done(db kv.Putter) error {
+	return SaveStageCleanUpProgress(db, s.ID, s.state.isBeacon, s.ForwardProgress)
+}
+func (s *CleanUpState) DoneAt(db kv.Putter, blockNum uint64) error {
+	return SaveStageCleanUpProgress(db, s.ID, s.state.isBeacon, blockNum)
+}

api/service/stagedstreamsync/stage_bodies.go

@@ -0,0 +1,420 @@
+package stagedstreamsync
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"time"
+
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/core/types"
+	"github.com/harmony-one/harmony/internal/utils"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/ledgerwatch/erigon-lib/kv"
+	"github.com/pkg/errors"
+)
+
+type StageBodies struct {
+	configs StageBodiesCfg
+}
+type StageBodiesCfg struct {
+	ctx         context.Context
+	bc          core.BlockChain
+	db          kv.RwDB
+	blockDBs    []kv.RwDB
+	concurrency int
+	protocol    syncProtocol
+	isBeacon    bool
+	logProgress bool
+}
+
+func NewStageBodies(cfg StageBodiesCfg) *StageBodies {
+	return &StageBodies{
+		configs: cfg,
+	}
+}
+
+func NewStageBodiesCfg(ctx context.Context, bc core.BlockChain, db kv.RwDB, blockDBs []kv.RwDB, concurrency int, protocol syncProtocol, isBeacon bool, logProgress bool) StageBodiesCfg {
+	return StageBodiesCfg{
+		ctx:         ctx,
+		bc:          bc,
+		db:          db,
+		blockDBs:    blockDBs,
+		concurrency: concurrency,
+		protocol:    protocol,
+		isBeacon:    isBeacon,
+		logProgress: logProgress,
+	}
+}
+
+func (b *StageBodies) SetStageContext(ctx context.Context) {
+	b.configs.ctx = ctx
+}
+
+// Exec progresses Bodies stage in the forward direction
+func (b *StageBodies) Exec(firstCycle bool, invalidBlockRevert bool, s *StageState, reverter Reverter, tx kv.RwTx) (err error) {
+
+	useInternalTx := tx == nil
+
+	if invalidBlockRevert {
+		return b.redownloadBadBlock(s)
+	}
+
+	// for short range sync, skip this stage
+	if !s.state.initSync {
+		return nil
+	}
+
+	maxHeight := s.state.status.targetBN
+	currentHead := b.configs.bc.CurrentBlock().NumberU64()
+	if currentHead >= maxHeight {
+		return nil
+	}
+	currProgress := uint64(0)
+	targetHeight := s.state.currentCycle.TargetHeight
+	// isBeacon := s.state.isBeacon
+	// isLastCycle := targetHeight >= maxHeight
+
+	if errV := CreateView(b.configs.ctx, b.configs.db, tx, func(etx kv.Tx) error {
+		if currProgress, err = s.CurrentStageProgress(etx); err != nil {
+			return err
+		}
+		return nil
+	}); errV != nil {
+		return errV
+	}
+
+	if currProgress == 0 {
+		if err := b.cleanAllBlockDBs(); err != nil {
+			return err
+		}
+		currProgress = currentHead
+	}
+
+	if currProgress >= targetHeight {
+		return nil
+	}
+
+	// size := uint64(0)
+	startTime := time.Now()
+	// startBlock := currProgress
+	if b.configs.logProgress {
+		fmt.Print("\033[s") // save the cursor position
+	}
+
+	if useInternalTx {
+		var err error
+		tx, err = b.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	// Fetch blocks from neighbors
+	s.state.gbm = newBlockDownloadManager(tx, b.configs.bc, targetHeight, s.state.logger)
+
+	// Setup workers to fetch blocks from remote node
+	var wg sync.WaitGroup
+
+	for i := 0; i != s.state.config.Concurrency; i++ {
+		wg.Add(1)
+		go b.runBlockWorkerLoop(s.state.gbm, &wg, i, startTime)
+	}
+
+	wg.Wait()
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// runBlockWorkerLoop creates a work loop for download blocks
+func (b *StageBodies) runBlockWorkerLoop(gbm *blockDownloadManager, wg *sync.WaitGroup, loopID int, startTime time.Time) {
+
+	currentBlock := int(b.configs.bc.CurrentBlock().NumberU64())
+
+	defer wg.Done()
+
+	for {
+		select {
+		case <-b.configs.ctx.Done():
+			return
+		default:
+		}
+		batch := gbm.GetNextBatch()
+		if len(batch) == 0 {
+			select {
+			case <-b.configs.ctx.Done():
+				return
+			case <-time.After(100 * time.Millisecond):
+				return
+			}
+		}
+
+		blockBytes, sigBytes, stid, err := b.downloadRawBlocks(batch)
+		if err != nil {
+			if !errors.Is(err, context.Canceled) {
+				b.configs.protocol.StreamFailed(stid, "downloadRawBlocks failed")
+			}
+			utils.Logger().Error().
+				Err(err).
+				Str("stream", string(stid)).
+				Interface("block numbers", batch).
+				Msg(WrapStagedSyncMsg("downloadRawBlocks failed"))
+			err = errors.Wrap(err, "request error")
+			gbm.HandleRequestError(batch, err, stid)
+		} else {
+			if err = b.saveBlocks(gbm.tx, batch, blockBytes, sigBytes, loopID, stid); err != nil {
+				panic("[STAGED_STREAM_SYNC] saving downloaded blocks to db failed.")
+			}
+			gbm.HandleRequestResult(batch, blockBytes, sigBytes, loopID, stid)
+			if b.configs.logProgress {
+				//calculating block download speed
+				dt := time.Now().Sub(startTime).Seconds()
+				speed := float64(0)
+				if dt > 0 {
+					speed = float64(len(gbm.bdd)) / dt
+				}
+				blockSpeed := fmt.Sprintf("%.2f", speed)
+
+				fmt.Print("\033[u\033[K") // restore the cursor position and clear the line
+				fmt.Println("downloaded blocks:", currentBlock+len(gbm.bdd), "/", int(gbm.targetBN), "(", blockSpeed, "blocks/s", ")")
+			}
+		}
+	}
+}
+
+// redownloadBadBlock tries to redownload the bad block from other streams
+func (b *StageBodies) redownloadBadBlock(s *StageState) error {
+
+	batch := make([]uint64, 1)
+	batch = append(batch, s.state.invalidBlock.Number)
+
+	for {
+		if b.configs.protocol.NumStreams() == 0 {
+			return errors.Errorf("re-download bad block from all streams failed")
+		}
+		blockBytes, sigBytes, stid, err := b.downloadRawBlocks(batch)
+		if err != nil {
+			if !errors.Is(err, context.Canceled) {
+				b.configs.protocol.StreamFailed(stid, "tried to re-download bad block from this stream, but downloadRawBlocks failed")
+			}
+			continue
+		}
+		isOneOfTheBadStreams := false
+		for _, id := range s.state.invalidBlock.StreamID {
+			if id == stid {
+				b.configs.protocol.RemoveStream(stid)
+				isOneOfTheBadStreams = true
+				break
+			}
+		}
+		if isOneOfTheBadStreams {
+			continue
+		}
+		s.state.gbm.SetDownloadDetails(batch, 0, stid)
+		if errU := b.configs.blockDBs[0].Update(context.Background(), func(tx kv.RwTx) error {
+			if err = b.saveBlocks(tx, batch, blockBytes, sigBytes, 0, stid); err != nil {
+				return errors.Errorf("[STAGED_STREAM_SYNC] saving re-downloaded bad block to db failed.")
+			}
+			return nil
+		}); errU != nil {
+			continue
+		}
+		break
+	}
+	return nil
+}
+
+func (b *StageBodies) downloadBlocks(bns []uint64) ([]*types.Block, sttypes.StreamID, error) {
+	ctx, cancel := context.WithTimeout(b.configs.ctx, 10*time.Second)
+	defer cancel()
+
+	blocks, stid, err := b.configs.protocol.GetBlocksByNumber(ctx, bns)
+	if err != nil {
+		return nil, stid, err
+	}
+	if err := validateGetBlocksResult(bns, blocks); err != nil {
+		return nil, stid, err
+	}
+	return blocks, stid, nil
+}
+
+func (b *StageBodies) downloadRawBlocks(bns []uint64) ([][]byte, [][]byte, sttypes.StreamID, error) {
+	ctx, cancel := context.WithTimeout(b.configs.ctx, 10*time.Second)
+	defer cancel()
+
+	return b.configs.protocol.GetRawBlocksByNumber(ctx, bns)
+}
+
+func validateGetBlocksResult(requested []uint64, result []*types.Block) error {
+	if len(result) != len(requested) {
+		return fmt.Errorf("unexpected number of blocks delivered: %v / %v", len(result), len(requested))
+	}
+	for i, block := range result {
+		if block != nil && block.NumberU64() != requested[i] {
+			return fmt.Errorf("block with unexpected number delivered: %v / %v", block.NumberU64(), requested[i])
+		}
+	}
+	return nil
+}
+
+// saveBlocks saves the blocks into db
+func (b *StageBodies) saveBlocks(tx kv.RwTx, bns []uint64, blockBytes [][]byte, sigBytes [][]byte, loopID int, stid sttypes.StreamID) error {
+
+	tx, err := b.configs.blockDBs[loopID].BeginRw(context.Background())
+	if err != nil {
+		return err
+	}
+	defer tx.Rollback()
+
+	for i := uint64(0); i < uint64(len(blockBytes)); i++ {
+		block := blockBytes[i]
+		sig := sigBytes[i]
+		if block == nil {
+			continue
+		}
+
+		blkKey := marshalData(bns[i])
+
+		if err := tx.Put(BlocksBucket, blkKey, block); err != nil {
+			utils.Logger().Error().
+				Err(err).
+				Uint64("block height", bns[i]).
+				Msg("[STAGED_STREAM_SYNC] adding block to db failed")
+			return err
+		}
+		// sigKey := []byte("s" + string(bns[i]))
+		if err := tx.Put(BlockSignaturesBucket, blkKey, sig); err != nil {
+			utils.Logger().Error().
+				Err(err).
+				Uint64("block height", bns[i]).
+				Msg("[STAGED_STREAM_SYNC] adding block sig to db failed")
+			return err
+		}
+	}
+
+	if err := tx.Commit(); err != nil {
+		return err
+	}
+
+	return nil
+}
+
+func (b *StageBodies) saveProgress(s *StageState, progress uint64, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		var err error
+		tx, err = b.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	// save progress
+	if err = s.Update(tx, progress); err != nil {
+		utils.Logger().Error().
+			Err(err).
+			Msgf("[STAGED_SYNC] saving progress for block bodies stage failed")
+		return ErrSavingBodiesProgressFail
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (b *StageBodies) cleanBlocksDB(loopID int) (err error) {
+
+	tx, errb := b.configs.blockDBs[loopID].BeginRw(b.configs.ctx)
+	if errb != nil {
+		return errb
+	}
+	defer tx.Rollback()
+
+	// clean block bodies db
+	if err = tx.ClearBucket(BlocksBucket); err != nil {
+		utils.Logger().Error().
+			Err(err).
+			Msgf("[STAGED_STREAM_SYNC] clear blocks bucket after revert failed")
+		return err
+	}
+	// clean block signatures db
+	if err = tx.ClearBucket(BlockSignaturesBucket); err != nil {
+		utils.Logger().Error().
+			Err(err).
+			Msgf("[STAGED_STREAM_SYNC] clear block signatures bucket after revert failed")
+		return err
+	}
+
+	if err = tx.Commit(); err != nil {
+		return err
+	}
+
+	return nil
+}
+
+func (b *StageBodies) cleanAllBlockDBs() (err error) {
+	//clean all blocks DBs
+	for i := 0; i < b.configs.concurrency; i++ {
+		if err := b.cleanBlocksDB(i); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (b *StageBodies) Revert(firstCycle bool, u *RevertState, s *StageState, tx kv.RwTx) (err error) {
+
+	//clean all blocks DBs
+	if err := b.cleanAllBlockDBs(); err != nil {
+		return err
+	}
+
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = b.configs.db.BeginRw(b.configs.ctx)
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+	// save progress
+	currentHead := b.configs.bc.CurrentBlock().NumberU64()
+	if err = s.Update(tx, currentHead); err != nil {
+		utils.Logger().Error().
+			Err(err).
+			Msgf("[STAGED_SYNC] saving progress for block bodies stage after revert failed")
+		return err
+	}
+
+	if err = u.Done(tx); err != nil {
+		return err
+	}
+
+	if useInternalTx {
+		if err = tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (b *StageBodies) CleanUp(firstCycle bool, p *CleanUpState, tx kv.RwTx) (err error) {
+
+	//clean all blocks DBs
+	if err := b.cleanAllBlockDBs(); err != nil {
+		return err
+	}
+
+	return nil
+}

api/service/stagedstreamsync/stage_epoch.go

@@ -0,0 +1,198 @@
+package stagedstreamsync
+
+import (
+	"context"
+
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/internal/utils"
+	"github.com/harmony-one/harmony/shard"
+	"github.com/ledgerwatch/erigon-lib/kv"
+	"github.com/pkg/errors"
+)
+
+type StageEpoch struct {
+	configs StageEpochCfg
+}
+
+type StageEpochCfg struct {
+	ctx context.Context
+	bc  core.BlockChain
+	db  kv.RwDB
+}
+
+func NewStageEpoch(cfg StageEpochCfg) *StageEpoch {
+	return &StageEpoch{
+		configs: cfg,
+	}
+}
+
+func NewStageEpochCfg(ctx context.Context, bc core.BlockChain, db kv.RwDB) StageEpochCfg {
+	return StageEpochCfg{
+		ctx: ctx,
+		bc:  bc,
+		db:  db,
+	}
+}
+
+func (sr *StageEpoch) SetStageContext(ctx context.Context) {
+	sr.configs.ctx = ctx
+}
+
+func (sr *StageEpoch) Exec(firstCycle bool, invalidBlockRevert bool, s *StageState, reverter Reverter, tx kv.RwTx) error {
+
+	// no need to update epoch chain if we are redoing the stages because of bad block
+	if invalidBlockRevert {
+		return nil
+	}
+	// for long range sync, skip this stage
+	if s.state.initSync {
+		return nil
+	}
+
+	if _, ok := sr.configs.bc.(*core.EpochChain); !ok {
+		return nil
+	}
+
+	// doShortRangeSyncForEpochSync
+	n, err := sr.doShortRangeSyncForEpochSync(s)
+	s.state.inserted = n
+	if err != nil {
+		return err
+	}
+
+	useInternalTx := tx == nil
+	if useInternalTx {
+		var err error
+		tx, err = sr.configs.db.BeginRw(sr.configs.ctx)
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func (sr *StageEpoch) doShortRangeSyncForEpochSync(s *StageState) (int, error) {
+
+	numShortRangeCounterVec.With(s.state.promLabels()).Inc()
+
+	srCtx, cancel := context.WithTimeout(s.state.ctx, ShortRangeTimeout)
+	defer cancel()
+
+	//TODO: merge srHelper with StageEpochConfig
+	sh := &srHelper{
+		syncProtocol: s.state.protocol,
+		ctx:          srCtx,
+		config:       s.state.config,
+		logger:       utils.Logger().With().Str("mode", "epoch chain short range").Logger(),
+	}
+
+	if err := sh.checkPrerequisites(); err != nil {
+		return 0, errors.Wrap(err, "prerequisite")
+	}
+	curBN := s.state.bc.CurrentBlock().NumberU64()
+	bns := make([]uint64, 0, BlocksPerRequest)
+	// in epoch chain, we have only the last block of each epoch, so, the current
+	// block's epoch number shows the last epoch we have. We should start
+	// from next epoch then
+	loopEpoch := s.state.bc.CurrentHeader().Epoch().Uint64() + 1
+	for len(bns) < BlocksPerRequest {
+		blockNum := shard.Schedule.EpochLastBlock(loopEpoch)
+		if blockNum > curBN {
+			bns = append(bns, blockNum)
+		}
+		loopEpoch = loopEpoch + 1
+	}
+
+	if len(bns) == 0 {
+		return 0, nil
+	}
+
+	////////////////////////////////////////////////////////
+	hashChain, whitelist, err := sh.getHashChain(bns)
+	if err != nil {
+		return 0, errors.Wrap(err, "getHashChain")
+	}
+	if len(hashChain) == 0 {
+		// short circuit for no sync is needed
+		return 0, nil
+	}
+	blocks, streamID, err := sh.getBlocksByHashes(hashChain, whitelist)
+	if err != nil {
+		utils.Logger().Warn().Err(err).Msg("epoch sync getBlocksByHashes failed")
+		if !errors.Is(err, context.Canceled) {
+			sh.removeStreams(whitelist) // Remote nodes cannot provide blocks with target hashes
+		}
+		return 0, errors.Wrap(err, "epoch sync getBlocksByHashes")
+	}
+	///////////////////////////////////////////////////////
+	// TODO: check this
+	// blocks, streamID, err := sh.getBlocksChain(bns)
+	// if err != nil {
+	// 	return 0, errors.Wrap(err, "getHashChain")
+	// }
+	///////////////////////////////////////////////////////
+	if len(blocks) == 0 {
+		// short circuit for no sync is needed
+		return 0, nil
+	}
+
+	n, err := s.state.bc.InsertChain(blocks, true)
+	numBlocksInsertedShortRangeHistogramVec.With(s.state.promLabels()).Observe(float64(n))
+	if err != nil {
+		utils.Logger().Info().Err(err).Int("blocks inserted", n).Msg("Insert block failed")
+		sh.removeStreams(streamID) // Data provided by remote nodes is corrupted
+		return n, err
+	}
+	if n > 0 {
+		utils.Logger().Info().Int("blocks inserted", n).Msg("Insert block success")
+	}
+	return n, nil
+}
+
+func (sr *StageEpoch) Revert(firstCycle bool, u *RevertState, s *StageState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = sr.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if err = u.Done(tx); err != nil {
+		return err
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (sr *StageEpoch) CleanUp(firstCycle bool, p *CleanUpState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = sr.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if useInternalTx {
+		if err = tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}

api/service/stagedstreamsync/stage_finish.go

@@ -0,0 +1,114 @@
+package stagedstreamsync
+
+import (
+	"context"
+
+	"github.com/ledgerwatch/erigon-lib/kv"
+)
+
+type StageFinish struct {
+	configs StageFinishCfg
+}
+
+type StageFinishCfg struct {
+	ctx context.Context
+	db  kv.RwDB
+}
+
+func NewStageFinish(cfg StageFinishCfg) *StageFinish {
+	return &StageFinish{
+		configs: cfg,
+	}
+}
+
+func NewStageFinishCfg(ctx context.Context, db kv.RwDB) StageFinishCfg {
+	return StageFinishCfg{
+		ctx: ctx,
+		db:  db,
+	}
+}
+
+func (finish *StageFinish) SetStageContext(ctx context.Context) {
+	finish.configs.ctx = ctx
+}
+
+func (finish *StageFinish) Exec(firstCycle bool, invalidBlockRevert bool, s *StageState, reverter Reverter, tx kv.RwTx) error {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		var err error
+		tx, err = finish.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	// TODO: prepare indices (useful for RPC) and finalize
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func (bh *StageFinish) clearBucket(tx kv.RwTx, isBeacon bool) error {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		var err error
+		tx, err = bh.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (finish *StageFinish) Revert(firstCycle bool, u *RevertState, s *StageState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = finish.configs.db.BeginRw(finish.configs.ctx)
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if err = u.Done(tx); err != nil {
+		return err
+	}
+
+	if useInternalTx {
+		if err = tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (finish *StageFinish) CleanUp(firstCycle bool, p *CleanUpState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = finish.configs.db.BeginRw(finish.configs.ctx)
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if useInternalTx {
+		if err = tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}

api/service/stagedstreamsync/stage_heads.go

@@ -0,0 +1,157 @@
+package stagedstreamsync
+
+import (
+	"context"
+
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/internal/utils"
+	"github.com/ledgerwatch/erigon-lib/kv"
+)
+
+type StageHeads struct {
+	configs StageHeadsCfg
+}
+
+type StageHeadsCfg struct {
+	ctx context.Context
+	bc  core.BlockChain
+	db  kv.RwDB
+}
+
+func NewStageHeads(cfg StageHeadsCfg) *StageHeads {
+	return &StageHeads{
+		configs: cfg,
+	}
+}
+
+func NewStageHeadersCfg(ctx context.Context, bc core.BlockChain, db kv.RwDB) StageHeadsCfg {
+	return StageHeadsCfg{
+		ctx: ctx,
+		bc:  bc,
+		db:  db,
+	}
+}
+
+func (heads *StageHeads) SetStageContext(ctx context.Context) {
+	heads.configs.ctx = ctx
+}
+
+func (heads *StageHeads) Exec(firstCycle bool, invalidBlockRevert bool, s *StageState, reverter Reverter, tx kv.RwTx) error {
+
+	// no need to update target if we are redoing the stages because of bad block
+	if invalidBlockRevert {
+		return nil
+	}
+
+	// no need for short range sync
+	if !s.state.initSync {
+		return nil
+	}
+
+	useInternalTx := tx == nil
+	if useInternalTx {
+		var err error
+		tx, err = heads.configs.db.BeginRw(heads.configs.ctx)
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	maxHeight := s.state.status.targetBN
+	maxBlocksPerSyncCycle := uint64(1024) // TODO: should be in config -> s.state.MaxBlocksPerSyncCycle
+	currentHeight := heads.configs.bc.CurrentBlock().NumberU64()
+	s.state.currentCycle.TargetHeight = maxHeight
+	targetHeight := uint64(0)
+	if errV := CreateView(heads.configs.ctx, heads.configs.db, tx, func(etx kv.Tx) (err error) {
+		if targetHeight, err = s.CurrentStageProgress(etx); err != nil {
+			return err
+		}
+		return nil
+	}); errV != nil {
+		return errV
+	}
+
+	if currentHeight >= maxHeight {
+		utils.Logger().Info().Uint64("current number", currentHeight).Uint64("target number", maxHeight).
+			Msg(WrapStagedSyncMsg("early return of long range sync"))
+		return nil
+	}
+
+	// if current height is ahead of target height, we need recalculate target height
+	if currentHeight >= targetHeight {
+		if maxHeight <= currentHeight {
+			return nil
+		}
+		utils.Logger().Info().
+			Uint64("max blocks per sync cycle", maxBlocksPerSyncCycle).
+			Uint64("maxPeersHeight", maxHeight).
+			Msgf(WrapStagedSyncMsg("current height is ahead of target height, target height is readjusted to max peers height"))
+		targetHeight = maxHeight
+	}
+
+	if targetHeight > maxHeight {
+		targetHeight = maxHeight
+	}
+
+	if maxBlocksPerSyncCycle > 0 && targetHeight-currentHeight > maxBlocksPerSyncCycle {
+		targetHeight = currentHeight + maxBlocksPerSyncCycle
+	}
+
+	s.state.currentCycle.TargetHeight = targetHeight
+
+	if err := s.Update(tx, targetHeight); err != nil {
+		utils.Logger().Error().
+			Err(err).
+			Msgf(WrapStagedSyncMsg("saving progress for headers stage failed"))
+		return err
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func (heads *StageHeads) Revert(firstCycle bool, u *RevertState, s *StageState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = heads.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if err = u.Done(tx); err != nil {
+		return err
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (heads *StageHeads) CleanUp(firstCycle bool, p *CleanUpState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = heads.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if useInternalTx {
+		if err = tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}

api/service/stagedstreamsync/stage_short_range.go

@@ -0,0 +1,205 @@
+package stagedstreamsync
+
+import (
+	"context"
+
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/internal/utils"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/ledgerwatch/erigon-lib/kv"
+	"github.com/pkg/errors"
+)
+
+type StageShortRange struct {
+	configs StageShortRangeCfg
+}
+
+type StageShortRangeCfg struct {
+	ctx context.Context
+	bc  core.BlockChain
+	db  kv.RwDB
+}
+
+func NewStageShortRange(cfg StageShortRangeCfg) *StageShortRange {
+	return &StageShortRange{
+		configs: cfg,
+	}
+}
+
+func NewStageShortRangeCfg(ctx context.Context, bc core.BlockChain, db kv.RwDB) StageShortRangeCfg {
+	return StageShortRangeCfg{
+		ctx: ctx,
+		bc:  bc,
+		db:  db,
+	}
+}
+
+func (sr *StageShortRange) SetStageContext(ctx context.Context) {
+	sr.configs.ctx = ctx
+}
+
+func (sr *StageShortRange) Exec(firstCycle bool, invalidBlockRevert bool, s *StageState, reverter Reverter, tx kv.RwTx) error {
+
+	// no need to do short range if we are redoing the stages because of bad block
+	if invalidBlockRevert {
+		return nil
+	}
+
+	// for long range sync, skip this stage
+	if s.state.initSync {
+		return nil
+	}
+
+	if _, ok := sr.configs.bc.(*core.EpochChain); ok {
+		return nil
+	}
+
+	curBN := sr.configs.bc.CurrentBlock().NumberU64()
+	if curBN >= s.state.status.targetBN {
+		return nil
+	}
+
+	// do short range sync
+	n, err := sr.doShortRangeSync(s)
+	s.state.inserted = n
+	if err != nil {
+		return err
+	}
+
+	useInternalTx := tx == nil
+	if useInternalTx {
+		var err error
+		tx, err = sr.configs.db.BeginRw(sr.configs.ctx)
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// doShortRangeSync does the short range sync.
+// Compared with long range sync, short range sync is more focused on syncing to the latest block.
+// It consist of 3 steps:
+// 1. Obtain the block hashes and compute the longest hash chain..
+// 2. Get blocks by hashes from computed hash chain.
+// 3. Insert the blocks to blockchain.
+func (sr *StageShortRange) doShortRangeSync(s *StageState) (int, error) {
+
+	numShortRangeCounterVec.With(s.state.promLabels()).Inc()
+
+	srCtx, cancel := context.WithTimeout(s.state.ctx, ShortRangeTimeout)
+	defer cancel()
+
+	sh := &srHelper{
+		syncProtocol: s.state.protocol,
+		ctx:          srCtx,
+		config:       s.state.config,
+		logger:       utils.Logger().With().Str("mode", "short range").Logger(),
+	}
+
+	if err := sh.checkPrerequisites(); err != nil {
+		return 0, errors.Wrap(err, "prerequisite")
+	}
+	curBN := sr.configs.bc.CurrentBlock().NumberU64()
+	blkCount := int(s.state.status.targetBN) - int(curBN)
+	blkNums := sh.prepareBlockHashNumbers(curBN, blkCount)
+	hashChain, whitelist, err := sh.getHashChain(blkNums)
+	if err != nil {
+		return 0, errors.Wrap(err, "getHashChain")
+	}
+
+	if len(hashChain) == 0 {
+		// short circuit for no sync is needed
+		return 0, nil
+	}
+
+	expEndBN := curBN + uint64(len(hashChain))
+	utils.Logger().Info().Uint64("current number", curBN).
+		Uint64("target number", expEndBN).
+		Interface("hashChain", hashChain).
+		Msg("short range start syncing")
+
+	s.state.status.setTargetBN(expEndBN)
+
+	s.state.status.startSyncing()
+	defer func() {
+		utils.Logger().Info().Msg("short range finished syncing")
+		s.state.status.finishSyncing()
+	}()
+
+	blocks, stids, err := sh.getBlocksByHashes(hashChain, whitelist)
+	if err != nil {
+		utils.Logger().Warn().Err(err).Msg("getBlocksByHashes failed")
+		if !errors.Is(err, context.Canceled) {
+			sh.removeStreams(whitelist) // Remote nodes cannot provide blocks with target hashes
+		}
+		return 0, errors.Wrap(err, "getBlocksByHashes")
+	}
+
+	utils.Logger().Info().Int("num blocks", len(blocks)).Msg("getBlockByHashes result")
+
+	n, err := verifyAndInsertBlocks(sr.configs.bc, blocks)
+	numBlocksInsertedShortRangeHistogramVec.With(s.state.promLabels()).Observe(float64(n))
+	if err != nil {
+		utils.Logger().Warn().Err(err).Int("blocks inserted", n).Msg("Insert block failed")
+		if sh.blameAllStreams(blocks, n, err) {
+			sh.removeStreams(whitelist) // Data provided by remote nodes is corrupted
+		} else {
+			// It is the last block gives a wrong commit sig. Blame the provider of the last block.
+			st2Blame := stids[len(stids)-1]
+			sh.removeStreams([]sttypes.StreamID{st2Blame})
+		}
+		return n, err
+	}
+	utils.Logger().Info().Err(err).Int("blocks inserted", n).Msg("Insert block success")
+
+	return n, nil
+}
+
+func (sr *StageShortRange) Revert(firstCycle bool, u *RevertState, s *StageState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = sr.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if err = u.Done(tx); err != nil {
+		return err
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (sr *StageShortRange) CleanUp(firstCycle bool, p *CleanUpState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = sr.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if useInternalTx {
+		if err = tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}

api/service/stagedstreamsync/stage_state.go

@@ -0,0 +1,295 @@
+package stagedstreamsync
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/ethereum/go-ethereum/common"
+	"github.com/ethereum/go-ethereum/rlp"
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/core/types"
+	"github.com/harmony-one/harmony/internal/utils"
+	"github.com/ledgerwatch/erigon-lib/kv"
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/rs/zerolog"
+)
+
+type StageStates struct {
+	configs StageStatesCfg
+}
+type StageStatesCfg struct {
+	ctx         context.Context
+	bc          core.BlockChain
+	db          kv.RwDB
+	blockDBs    []kv.RwDB
+	concurrency int
+	logger      zerolog.Logger
+	logProgress bool
+}
+
+func NewStageStates(cfg StageStatesCfg) *StageStates {
+	return &StageStates{
+		configs: cfg,
+	}
+}
+
+func NewStageStatesCfg(ctx context.Context,
+	bc core.BlockChain,
+	db kv.RwDB,
+	blockDBs []kv.RwDB,
+	concurrency int,
+	logger zerolog.Logger,
+	logProgress bool) StageStatesCfg {
+
+	return StageStatesCfg{
+		ctx:         ctx,
+		bc:          bc,
+		db:          db,
+		blockDBs:    blockDBs,
+		concurrency: concurrency,
+		logger:      logger,
+		logProgress: logProgress,
+	}
+}
+
+func (stg *StageStates) SetStageContext(ctx context.Context) {
+	stg.configs.ctx = ctx
+}
+
+// Exec progresses States stage in the forward direction
+func (stg *StageStates) Exec(firstCycle bool, invalidBlockRevert bool, s *StageState, reverter Reverter, tx kv.RwTx) (err error) {
+
+	// for short range sync, skip this step
+	if !s.state.initSync {
+		return nil
+	}
+
+	maxHeight := s.state.status.targetBN
+	currentHead := stg.configs.bc.CurrentBlock().NumberU64()
+	if currentHead >= maxHeight {
+		return nil
+	}
+	currProgress := stg.configs.bc.CurrentBlock().NumberU64()
+	targetHeight := s.state.currentCycle.TargetHeight
+	if currProgress >= targetHeight {
+		return nil
+	}
+	useInternalTx := tx == nil
+	if useInternalTx {
+		var err error
+		tx, err = stg.configs.db.BeginRw(stg.configs.ctx)
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	// isLastCycle := targetHeight >= maxHeight
+	startTime := time.Now()
+	startBlock := currProgress
+	pl := s.state.promLabels()
+	gbm := s.state.gbm
+
+	// prepare db transactions
+	txs := make([]kv.RwTx, stg.configs.concurrency)
+	for i := 0; i < stg.configs.concurrency; i++ {
+		txs[i], err = stg.configs.blockDBs[i].BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+	}
+
+	defer func() {
+		for i := 0; i < stg.configs.concurrency; i++ {
+			txs[i].Rollback()
+		}
+	}()
+
+	if stg.configs.logProgress {
+		fmt.Print("\033[s") // save the cursor position
+	}
+
+	for i := currProgress + 1; i <= targetHeight; i++ {
+		blkKey := marshalData(i)
+		loopID, streamID := gbm.GetDownloadDetails(i)
+
+		blockBytes, err := txs[loopID].GetOne(BlocksBucket, blkKey)
+		if err != nil {
+			return err
+		}
+		sigBytes, err := txs[loopID].GetOne(BlockSignaturesBucket, blkKey)
+		if err != nil {
+			return err
+		}
+
+		// if block size is invalid, we have to break the updating state loop
+		// we don't need to do rollback, because the latest batch haven't added to chain yet
+		sz := len(blockBytes)
+		if sz <= 1 {
+			utils.Logger().Error().
+				Uint64("block number", i).
+				Msg("block size invalid")
+			invalidBlockHash := common.Hash{}
+			s.state.protocol.StreamFailed(streamID, "zero bytes block is received from stream")
+			reverter.RevertTo(stg.configs.bc.CurrentBlock().NumberU64(), i, invalidBlockHash, streamID)
+			return ErrInvalidBlockBytes
+		}
+
+		var block *types.Block
+		if err := rlp.DecodeBytes(blockBytes, &block); err != nil {
+			utils.Logger().Error().
+				Uint64("block number", i).
+				Msg("block size invalid")
+			s.state.protocol.StreamFailed(streamID, "invalid block is received from stream")
+			invalidBlockHash := common.Hash{}
+			reverter.RevertTo(stg.configs.bc.CurrentBlock().NumberU64(), i, invalidBlockHash, streamID)
+			return ErrInvalidBlockBytes
+		}
+		if sigBytes != nil {
+			block.SetCurrentCommitSig(sigBytes)
+		}
+
+		if block.NumberU64() != i {
+			s.state.protocol.StreamFailed(streamID, "invalid block with unmatched number is received from stream")
+			invalidBlockHash := block.Hash()
+			reverter.RevertTo(stg.configs.bc.CurrentBlock().NumberU64(), i, invalidBlockHash, streamID)
+			return ErrInvalidBlockNumber
+		}
+
+		if err := verifyAndInsertBlock(stg.configs.bc, block); err != nil {
+			stg.configs.logger.Warn().Err(err).Uint64("cycle target block", targetHeight).
+				Uint64("block number", block.NumberU64()).
+				Msg(WrapStagedSyncMsg("insert blocks failed in long range"))
+			s.state.protocol.StreamFailed(streamID, "unverifiable invalid block is received from stream")
+			invalidBlockHash := block.Hash()
+			reverter.RevertTo(stg.configs.bc.CurrentBlock().NumberU64(), block.NumberU64(), invalidBlockHash, streamID)
+			pl["error"] = err.Error()
+			longRangeFailInsertedBlockCounterVec.With(pl).Inc()
+			return err
+		}
+
+		if invalidBlockRevert {
+			if s.state.invalidBlock.Number == i {
+				s.state.invalidBlock.resolve()
+			}
+		}
+
+		s.state.inserted++
+		longRangeSyncedBlockCounterVec.With(pl).Inc()
+
+		utils.Logger().Info().
+			Uint64("blockHeight", block.NumberU64()).
+			Uint64("blockEpoch", block.Epoch().Uint64()).
+			Str("blockHex", block.Hash().Hex()).
+			Uint32("ShardID", block.ShardID()).
+			Msg("[STAGED_STREAM_SYNC] New Block Added to Blockchain")
+
+		// update cur progress
+		currProgress = stg.configs.bc.CurrentBlock().NumberU64()
+
+		for i, tx := range block.StakingTransactions() {
+			utils.Logger().Info().
+				Msgf(
+					"StakingTxn %d: %s, %v", i, tx.StakingType().String(), tx.StakingMessage(),
+				)
+		}
+
+		// log the stage progress in console
+		if stg.configs.logProgress {
+			//calculating block speed
+			dt := time.Now().Sub(startTime).Seconds()
+			speed := float64(0)
+			if dt > 0 {
+				speed = float64(currProgress-startBlock) / dt
+			}
+			blockSpeed := fmt.Sprintf("%.2f", speed)
+			fmt.Print("\033[u\033[K") // restore the cursor position and clear the line
+			fmt.Println("insert blocks progress:", currProgress, "/", targetHeight, "(", blockSpeed, "blocks/s", ")")
+		}
+
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func (stg *StageStates) insertChain(gbm *blockDownloadManager,
+	protocol syncProtocol,
+	lbls prometheus.Labels,
+	targetBN uint64) {
+
+}
+
+func (stg *StageStates) saveProgress(s *StageState, tx kv.RwTx) (err error) {
+
+	useInternalTx := tx == nil
+	if useInternalTx {
+		var err error
+		tx, err = stg.configs.db.BeginRw(context.Background())
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	// save progress
+	if err = s.Update(tx, stg.configs.bc.CurrentBlock().NumberU64()); err != nil {
+		utils.Logger().Error().
+			Err(err).
+			Msgf("[STAGED_SYNC] saving progress for block States stage failed")
+		return ErrSaveStateProgressFail
+	}
+
+	if useInternalTx {
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (stg *StageStates) Revert(firstCycle bool, u *RevertState, s *StageState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = stg.configs.db.BeginRw(stg.configs.ctx)
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if err = u.Done(tx); err != nil {
+		return err
+	}
+
+	if useInternalTx {
+		if err = tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (stg *StageStates) CleanUp(firstCycle bool, p *CleanUpState, tx kv.RwTx) (err error) {
+	useInternalTx := tx == nil
+	if useInternalTx {
+		tx, err = stg.configs.db.BeginRw(stg.configs.ctx)
+		if err != nil {
+			return err
+		}
+		defer tx.Rollback()
+	}
+
+	if useInternalTx {
+		if err = tx.Commit(); err != nil {
+			return err
+		}
+	}
+	return nil
+}

api/service/stagedstreamsync/staged_stream_sync.go

@@ -0,0 +1,597 @@
+package stagedstreamsync
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"time"
+
+	"github.com/ethereum/go-ethereum/common"
+	"github.com/ethereum/go-ethereum/event"
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/core/types"
+	"github.com/harmony-one/harmony/internal/utils"
+	syncproto "github.com/harmony-one/harmony/p2p/stream/protocols/sync"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/ledgerwatch/erigon-lib/kv"
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/rs/zerolog"
+)
+
+type InvalidBlock struct {
+	Active   bool
+	Number   uint64
+	Hash     common.Hash
+	IsLogged bool
+	StreamID []sttypes.StreamID
+}
+
+func (ib *InvalidBlock) set(num uint64, hash common.Hash, resetBadStreams bool) {
+	ib.Active = true
+	ib.IsLogged = false
+	ib.Number = num
+	ib.Hash = hash
+	if resetBadStreams {
+		ib.StreamID = make([]sttypes.StreamID, 0)
+	}
+}
+
+func (ib *InvalidBlock) resolve() {
+	ib.Active = false
+	ib.IsLogged = false
+	ib.Number = 0
+	ib.Hash = common.Hash{}
+	ib.StreamID = ib.StreamID[:0]
+}
+
+func (ib *InvalidBlock) addBadStream(bsID sttypes.StreamID) {
+	// only add uniques IDs
+	for _, stID := range ib.StreamID {
+		if stID == bsID {
+			return
+		}
+	}
+	ib.StreamID = append(ib.StreamID, bsID)
+}
+
+type StagedStreamSync struct {
+	ctx        context.Context
+	bc         core.BlockChain
+	isBeacon   bool
+	isExplorer bool
+	db         kv.RwDB
+	protocol   syncProtocol
+	gbm        *blockDownloadManager // initialized when finished get block number
+	inserted   int
+	config     Config
+	logger     zerolog.Logger
+	status     status //TODO: merge this with currentSyncCycle
+	initSync   bool   // if sets to true, node start long range syncing
+	UseMemDB   bool
+
+	revertPoint     *uint64 // used to run stages
+	prevRevertPoint *uint64 // used to get value from outside of staged sync after cycle (for example to notify RPCDaemon)
+	invalidBlock    InvalidBlock
+	currentStage    uint
+	LogProgress     bool
+	currentCycle    SyncCycle // current cycle
+	stages          []*Stage
+	revertOrder     []*Stage
+	pruningOrder    []*Stage
+	timings         []Timing
+	logPrefixes     []string
+
+	evtDownloadFinished           event.Feed // channel for each download task finished
+	evtDownloadFinishedSubscribed bool
+	evtDownloadStarted            event.Feed // channel for each download has started
+	evtDownloadStartedSubscribed  bool
+}
+
+// BlockWithSig the serialization structure for request DownloaderRequest_BLOCKWITHSIG
+// The block is encoded as block + commit signature
+type BlockWithSig struct {
+	Block              *types.Block
+	CommitSigAndBitmap []byte
+}
+
+type Timing struct {
+	isRevert  bool
+	isCleanUp bool
+	stage     SyncStageID
+	took      time.Duration
+}
+
+type SyncCycle struct {
+	Number       uint64
+	TargetHeight uint64
+	lock         sync.RWMutex
+}
+
+func (s *StagedStreamSync) Len() int                    { return len(s.stages) }
+func (s *StagedStreamSync) Context() context.Context    { return s.ctx }
+func (s *StagedStreamSync) Blockchain() core.BlockChain { return s.bc }
+func (s *StagedStreamSync) DB() kv.RwDB                 { return s.db }
+func (s *StagedStreamSync) IsBeacon() bool              { return s.isBeacon }
+func (s *StagedStreamSync) IsExplorer() bool            { return s.isExplorer }
+func (s *StagedStreamSync) LogPrefix() string {
+	if s == nil {
+		return ""
+	}
+	return s.logPrefixes[s.currentStage]
+}
+func (s *StagedStreamSync) PrevRevertPoint() *uint64 { return s.prevRevertPoint }
+
+func (s *StagedStreamSync) NewRevertState(id SyncStageID, revertPoint uint64) *RevertState {
+	return &RevertState{id, revertPoint, s}
+}
+
+func (s *StagedStreamSync) CleanUpStageState(id SyncStageID, forwardProgress uint64, tx kv.Tx, db kv.RwDB) (*CleanUpState, error) {
+	var pruneProgress uint64
+	var err error
+
+	if errV := CreateView(context.Background(), db, tx, func(tx kv.Tx) error {
+		pruneProgress, err = GetStageCleanUpProgress(tx, id, s.isBeacon)
+		if err != nil {
+			return err
+		}
+		return nil
+	}); errV != nil {
+		return nil, errV
+	}
+
+	return &CleanUpState{id, forwardProgress, pruneProgress, s}, nil
+}
+
+func (s *StagedStreamSync) NextStage() {
+	if s == nil {
+		return
+	}
+	s.currentStage++
+}
+
+// IsBefore returns true if stage1 goes before stage2 in staged sync
+func (s *StagedStreamSync) IsBefore(stage1, stage2 SyncStageID) bool {
+	idx1 := -1
+	idx2 := -1
+	for i, stage := range s.stages {
+		if stage.ID == stage1 {
+			idx1 = i
+		}
+
+		if stage.ID == stage2 {
+			idx2 = i
+		}
+	}
+
+	return idx1 < idx2
+}
+
+// IsAfter returns true if stage1 goes after stage2 in staged sync
+func (s *StagedStreamSync) IsAfter(stage1, stage2 SyncStageID) bool {
+	idx1 := -1
+	idx2 := -1
+	for i, stage := range s.stages {
+		if stage.ID == stage1 {
+			idx1 = i
+		}
+
+		if stage.ID == stage2 {
+			idx2 = i
+		}
+	}
+
+	return idx1 > idx2
+}
+
+func (s *StagedStreamSync) RevertTo(revertPoint uint64, invalidBlockNumber uint64, invalidBlockHash common.Hash, invalidBlockStreamID sttypes.StreamID) {
+	utils.Logger().Info().
+		Interface("invalidBlockNumber", invalidBlockNumber).
+		Interface("invalidBlockHash", invalidBlockHash).
+		Interface("invalidBlockStreamID", invalidBlockStreamID).
+		Uint64("revertPoint", revertPoint).
+		Msgf(WrapStagedSyncMsg("Reverting blocks"))
+	s.revertPoint = &revertPoint
+	if invalidBlockNumber > 0 || invalidBlockHash != (common.Hash{}) {
+		resetBadStreams := !s.invalidBlock.Active
+		s.invalidBlock.set(invalidBlockNumber, invalidBlockHash, resetBadStreams)
+		s.invalidBlock.addBadStream(invalidBlockStreamID)
+	}
+}
+
+func (s *StagedStreamSync) Done() {
+	s.currentStage = uint(len(s.stages))
+	s.revertPoint = nil
+}
+
+func (s *StagedStreamSync) IsDone() bool {
+	return s.currentStage >= uint(len(s.stages)) && s.revertPoint == nil
+}
+
+func (s *StagedStreamSync) SetCurrentStage(id SyncStageID) error {
+	for i, stage := range s.stages {
+		if stage.ID == id {
+			s.currentStage = uint(i)
+			return nil
+		}
+	}
+
+	return ErrStageNotFound
+}
+
+func (s *StagedStreamSync) StageState(stage SyncStageID, tx kv.Tx, db kv.RwDB) (*StageState, error) {
+	var blockNum uint64
+	var err error
+	if errV := CreateView(context.Background(), db, tx, func(rtx kv.Tx) error {
+		blockNum, err = GetStageProgress(rtx, stage, s.isBeacon)
+		if err != nil {
+			return err
+		}
+		return nil
+	}); errV != nil {
+		return nil, errV
+	}
+
+	return &StageState{s, stage, blockNum}, nil
+}
+
+func (s *StagedStreamSync) cleanUp(fromStage int, db kv.RwDB, tx kv.RwTx, firstCycle bool) error {
+	found := false
+	for i := 0; i < len(s.pruningOrder); i++ {
+		if s.pruningOrder[i].ID == s.stages[fromStage].ID {
+			found = true
+		}
+		if !found || s.pruningOrder[i] == nil || s.pruningOrder[i].Disabled {
+			continue
+		}
+		if err := s.pruneStage(firstCycle, s.pruningOrder[i], db, tx); err != nil {
+			panic(err)
+		}
+	}
+	return nil
+}
+
+func New(ctx context.Context,
+	bc core.BlockChain,
+	db kv.RwDB,
+	stagesList []*Stage,
+	isBeacon bool,
+	protocol syncProtocol,
+	useMemDB bool,
+	config Config,
+	logger zerolog.Logger,
+) *StagedStreamSync {
+
+	revertStages := make([]*Stage, len(stagesList))
+	for i, stageIndex := range DefaultRevertOrder {
+		for _, s := range stagesList {
+			if s.ID == stageIndex {
+				revertStages[i] = s
+				break
+			}
+		}
+	}
+	pruneStages := make([]*Stage, len(stagesList))
+	for i, stageIndex := range DefaultCleanUpOrder {
+		for _, s := range stagesList {
+			if s.ID == stageIndex {
+				pruneStages[i] = s
+				break
+			}
+		}
+	}
+
+	logPrefixes := make([]string, len(stagesList))
+	for i := range stagesList {
+		logPrefixes[i] = fmt.Sprintf("%d/%d %s", i+1, len(stagesList), stagesList[i].ID)
+	}
+
+	status := newStatus()
+
+	return &StagedStreamSync{
+		ctx:          ctx,
+		bc:           bc,
+		isBeacon:     isBeacon,
+		db:           db,
+		protocol:     protocol,
+		gbm:          nil,
+		status:       status,
+		inserted:     0,
+		config:       config,
+		logger:       logger,
+		stages:       stagesList,
+		currentStage: 0,
+		revertOrder:  revertStages,
+		pruningOrder: pruneStages,
+		logPrefixes:  logPrefixes,
+		UseMemDB:     useMemDB,
+	}
+}
+
+func (s *StagedStreamSync) doGetCurrentNumberRequest() (uint64, sttypes.StreamID, error) {
+	ctx, cancel := context.WithTimeout(s.ctx, 10*time.Second)
+	defer cancel()
+
+	bn, stid, err := s.protocol.GetCurrentBlockNumber(ctx, syncproto.WithHighPriority())
+	if err != nil {
+		return 0, stid, err
+	}
+	return bn, stid, nil
+}
+
+func (s *StagedStreamSync) promLabels() prometheus.Labels {
+	sid := s.bc.ShardID()
+	return prometheus.Labels{"ShardID": fmt.Sprintf("%d", sid)}
+}
+
+func (s *StagedStreamSync) checkHaveEnoughStreams() error {
+	numStreams := s.protocol.NumStreams()
+	if numStreams < s.config.MinStreams {
+		return fmt.Errorf("number of streams smaller than minimum: %v < %v",
+			numStreams, s.config.MinStreams)
+	}
+	return nil
+}
+
+func (s *StagedStreamSync) SetNewContext(ctx context.Context) error {
+	for _, s := range s.stages {
+		s.Handler.SetStageContext(ctx)
+	}
+	return nil
+}
+
+func (s *StagedStreamSync) Run(db kv.RwDB, tx kv.RwTx, firstCycle bool) error {
+	s.prevRevertPoint = nil
+	s.timings = s.timings[:0]
+
+	for !s.IsDone() {
+		if s.revertPoint != nil {
+			s.prevRevertPoint = s.revertPoint
+			s.revertPoint = nil
+			if !s.invalidBlock.Active {
+				for j := 0; j < len(s.revertOrder); j++ {
+					if s.revertOrder[j] == nil || s.revertOrder[j].Disabled {
+						continue
+					}
+					if err := s.revertStage(firstCycle, s.revertOrder[j], db, tx); err != nil {
+						utils.Logger().Error().
+							Err(err).
+							Interface("stage id", s.revertOrder[j].ID).
+							Msgf(WrapStagedSyncMsg("revert stage failed"))
+						return err
+					}
+				}
+			}
+			if err := s.SetCurrentStage(s.stages[0].ID); err != nil {
+				return err
+			}
+			firstCycle = false
+		}
+
+		stage := s.stages[s.currentStage]
+
+		if stage.Disabled {
+			utils.Logger().Trace().
+				Msgf(WrapStagedSyncMsg(fmt.Sprintf("%s disabled. %s", stage.ID, stage.DisabledDescription)))
+
+			s.NextStage()
+			continue
+		}
+
+		if err := s.runStage(stage, db, tx, firstCycle, s.invalidBlock.Active); err != nil {
+			utils.Logger().Error().
+				Err(err).
+				Interface("stage id", stage.ID).
+				Msgf(WrapStagedSyncMsg("stage failed"))
+			return err
+		}
+		s.NextStage()
+	}
+
+	if err := s.cleanUp(0, db, tx, firstCycle); err != nil {
+		utils.Logger().Error().
+			Err(err).
+			Msgf(WrapStagedSyncMsg("stages cleanup failed"))
+		return err
+	}
+	if err := s.SetCurrentStage(s.stages[0].ID); err != nil {
+		return err
+	}
+	if err := printLogs(tx, s.timings); err != nil {
+		return err
+	}
+	s.currentStage = 0
+	return nil
+}
+
+func CreateView(ctx context.Context, db kv.RwDB, tx kv.Tx, f func(tx kv.Tx) error) error {
+	if tx != nil {
+		return f(tx)
+	}
+	return db.View(context.Background(), func(etx kv.Tx) error {
+		return f(etx)
+	})
+}
+
+func ByteCount(b uint64) string {
+	const unit = 1024
+	if b < unit {
+		return fmt.Sprintf("%dB", b)
+	}
+	div, exp := uint64(unit), 0
+	for n := b / unit; n >= unit; n /= unit {
+		div *= unit
+		exp++
+	}
+	return fmt.Sprintf("%.1f%cB",
+		float64(b)/float64(div), "KMGTPE"[exp])
+}
+
+func printLogs(tx kv.RwTx, timings []Timing) error {
+	var logCtx []interface{}
+	count := 0
+	for i := range timings {
+		if timings[i].took < 50*time.Millisecond {
+			continue
+		}
+		count++
+		if count == 50 {
+			break
+		}
+		if timings[i].isRevert {
+			logCtx = append(logCtx, "Revert "+string(timings[i].stage), timings[i].took.Truncate(time.Millisecond).String())
+		} else if timings[i].isCleanUp {
+			logCtx = append(logCtx, "CleanUp "+string(timings[i].stage), timings[i].took.Truncate(time.Millisecond).String())
+		} else {
+			logCtx = append(logCtx, string(timings[i].stage), timings[i].took.Truncate(time.Millisecond).String())
+		}
+	}
+	if len(logCtx) > 0 {
+		utils.Logger().Info().
+			Msgf(WrapStagedSyncMsg(fmt.Sprintf("Timings (slower than 50ms) %v", logCtx...)))
+	}
+
+	if tx == nil {
+		return nil
+	}
+
+	if len(logCtx) > 0 { // also don't print this logs if everything is fast
+		buckets := Buckets
+		bucketSizes := make([]interface{}, 0, 2*len(buckets))
+		for _, bucket := range buckets {
+			sz, err1 := tx.BucketSize(bucket)
+			if err1 != nil {
+				return err1
+			}
+			bucketSizes = append(bucketSizes, bucket, ByteCount(sz))
+		}
+		utils.Logger().Info().
+			Msgf(WrapStagedSyncMsg(fmt.Sprintf("Tables %v", bucketSizes...)))
+	}
+	tx.CollectMetrics()
+	return nil
+}
+
+func (s *StagedStreamSync) runStage(stage *Stage, db kv.RwDB, tx kv.RwTx, firstCycle bool, invalidBlockRevert bool) (err error) {
+	start := time.Now()
+	stageState, err := s.StageState(stage.ID, tx, db)
+	if err != nil {
+		return err
+	}
+
+	if err = stage.Handler.Exec(firstCycle, invalidBlockRevert, stageState, s, tx); err != nil {
+		utils.Logger().Error().
+			Err(err).
+			Interface("stage id", stage.ID).
+			Msgf(WrapStagedSyncMsg("stage failed"))
+		return fmt.Errorf("[%s] %w", s.LogPrefix(), err)
+	}
+
+	took := time.Since(start)
+	if took > 60*time.Second {
+		logPrefix := s.LogPrefix()
+		utils.Logger().Info().
+			Msgf(WrapStagedSyncMsg(fmt.Sprintf("%s:  DONE in %d", logPrefix, took)))
+
+	}
+	s.timings = append(s.timings, Timing{stage: stage.ID, took: took})
+	return nil
+}
+
+func (s *StagedStreamSync) revertStage(firstCycle bool, stage *Stage, db kv.RwDB, tx kv.RwTx) error {
+	start := time.Now()
+	stageState, err := s.StageState(stage.ID, tx, db)
+	if err != nil {
+		return err
+	}
+
+	revert := s.NewRevertState(stage.ID, *s.revertPoint)
+
+	if stageState.BlockNumber <= revert.RevertPoint {
+		return nil
+	}
+
+	if err = s.SetCurrentStage(stage.ID); err != nil {
+		return err
+	}
+
+	err = stage.Handler.Revert(firstCycle, revert, stageState, tx)
+	if err != nil {
+		return fmt.Errorf("[%s] %w", s.LogPrefix(), err)
+	}
+
+	took := time.Since(start)
+	if took > 60*time.Second {
+		logPrefix := s.LogPrefix()
+		utils.Logger().Info().
+			Msgf(WrapStagedSyncMsg(fmt.Sprintf("%s: Revert done in %d", logPrefix, took)))
+	}
+	s.timings = append(s.timings, Timing{isRevert: true, stage: stage.ID, took: took})
+	return nil
+}
+
+func (s *StagedStreamSync) pruneStage(firstCycle bool, stage *Stage, db kv.RwDB, tx kv.RwTx) error {
+	start := time.Now()
+
+	stageState, err := s.StageState(stage.ID, tx, db)
+	if err != nil {
+		return err
+	}
+
+	prune, err := s.CleanUpStageState(stage.ID, stageState.BlockNumber, tx, db)
+	if err != nil {
+		return err
+	}
+	if err = s.SetCurrentStage(stage.ID); err != nil {
+		return err
+	}
+
+	err = stage.Handler.CleanUp(firstCycle, prune, tx)
+	if err != nil {
+		return fmt.Errorf("[%s] %w", s.LogPrefix(), err)
+	}
+
+	took := time.Since(start)
+	if took > 60*time.Second {
+		logPrefix := s.LogPrefix()
+		utils.Logger().Info().
+			Msgf(WrapStagedSyncMsg(fmt.Sprintf("%s: CleanUp done in %d", logPrefix, took)))
+	}
+	s.timings = append(s.timings, Timing{isCleanUp: true, stage: stage.ID, took: took})
+	return nil
+}
+
+// DisableAllStages disables all stages including their reverts
+func (s *StagedStreamSync) DisableAllStages() []SyncStageID {
+	var backupEnabledIds []SyncStageID
+	for i := range s.stages {
+		if !s.stages[i].Disabled {
+			backupEnabledIds = append(backupEnabledIds, s.stages[i].ID)
+		}
+	}
+	for i := range s.stages {
+		s.stages[i].Disabled = true
+	}
+	return backupEnabledIds
+}
+
+func (s *StagedStreamSync) DisableStages(ids ...SyncStageID) {
+	for i := range s.stages {
+		for _, id := range ids {
+			if s.stages[i].ID != id {
+				continue
+			}
+			s.stages[i].Disabled = true
+		}
+	}
+}
+
+func (s *StagedStreamSync) EnableStages(ids ...SyncStageID) {
+	for i := range s.stages {
+		for _, id := range ids {
+			if s.stages[i].ID != id {
+				continue
+			}
+			s.stages[i].Disabled = false
+		}
+	}
+}

api/service/stagedstreamsync/stages.go

@@ -0,0 +1,71 @@
+package stagedstreamsync
+
+import (
+	"github.com/ledgerwatch/erigon-lib/kv"
+)
+
+// SyncStageID represents the stages in the Mode.StagedSync mode
+type SyncStageID string
+
+const (
+	Heads       SyncStageID = "Heads"       // Heads are downloaded
+	ShortRange  SyncStageID = "ShortRange"  // short range
+	SyncEpoch   SyncStageID = "SyncEpoch"   // epoch sync
+	BlockBodies SyncStageID = "BlockBodies" // Block bodies are downloaded, TxHash and UncleHash are getting verified
+	States      SyncStageID = "States"      // will construct most recent state from downloaded blocks
+	Finish      SyncStageID = "Finish"      // Nominal stage after all other stages
+)
+
+func GetStageName(stage string, isBeacon bool, prune bool) string {
+	name := stage
+	if isBeacon {
+		name = "beacon_" + name
+	}
+	if prune {
+		name = "prune_" + name
+	}
+	return name
+}
+
+func GetStageID(stage SyncStageID, isBeacon bool, prune bool) []byte {
+	return []byte(GetStageName(string(stage), isBeacon, prune))
+}
+
+func GetBucketName(bucketName string, isBeacon bool) string {
+	name := bucketName
+	if isBeacon {
+		name = "Beacon" + name
+	}
+	return name
+}
+
+// GetStageProgress retrieves saved progress of a given sync stage from the database
+func GetStageProgress(db kv.Getter, stage SyncStageID, isBeacon bool) (uint64, error) {
+	stgID := GetStageID(stage, isBeacon, false)
+	v, err := db.GetOne(kv.SyncStageProgress, stgID)
+	if err != nil {
+		return 0, err
+	}
+	return unmarshalData(v)
+}
+
+// SaveStageProgress saves progress of given sync stage
+func SaveStageProgress(db kv.Putter, stage SyncStageID, isBeacon bool, progress uint64) error {
+	stgID := GetStageID(stage, isBeacon, false)
+	return db.Put(kv.SyncStageProgress, stgID, marshalData(progress))
+}
+
+// GetStageCleanUpProgress retrieves saved progress of given sync stage from the database
+func GetStageCleanUpProgress(db kv.Getter, stage SyncStageID, isBeacon bool) (uint64, error) {
+	stgID := GetStageID(stage, isBeacon, true)
+	v, err := db.GetOne(kv.SyncStageProgress, stgID)
+	if err != nil {
+		return 0, err
+	}
+	return unmarshalData(v)
+}
+
+func SaveStageCleanUpProgress(db kv.Putter, stage SyncStageID, isBeacon bool, progress uint64) error {
+	stgID := GetStageID(stage, isBeacon, true)
+	return db.Put(kv.SyncStageProgress, stgID, marshalData(progress))
+}

api/service/stagedstreamsync/syncing.go

@@ -0,0 +1,320 @@
+package stagedstreamsync
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"time"
+
+	"github.com/harmony-one/harmony/core"
+	"github.com/harmony-one/harmony/internal/utils"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/harmony-one/harmony/shard"
+	"github.com/ledgerwatch/erigon-lib/kv"
+	"github.com/ledgerwatch/erigon-lib/kv/mdbx"
+	"github.com/ledgerwatch/erigon-lib/kv/memdb"
+	"github.com/ledgerwatch/log/v3"
+	"github.com/pkg/errors"
+	"github.com/rs/zerolog"
+)
+
+const (
+	BlocksBucket          = "BlockBodies"
+	BlockSignaturesBucket = "BlockSignatures"
+	StageProgressBucket   = "StageProgress"
+
+	// cache db keys
+	LastBlockHeight = "LastBlockHeight"
+	LastBlockHash   = "LastBlockHash"
+)
+
+var Buckets = []string{
+	BlocksBucket,
+	BlockSignaturesBucket,
+	StageProgressBucket,
+}
+
+// CreateStagedSync creates an instance of staged sync
+func CreateStagedSync(ctx context.Context,
+	bc core.BlockChain,
+	UseMemDB bool,
+	protocol syncProtocol,
+	config Config,
+	logger zerolog.Logger,
+	logProgress bool,
+) (*StagedStreamSync, error) {
+
+	isBeacon := bc.ShardID() == shard.BeaconChainShardID
+
+	var mainDB kv.RwDB
+	dbs := make([]kv.RwDB, config.Concurrency)
+	if UseMemDB {
+		mainDB = memdb.New()
+		for i := 0; i < config.Concurrency; i++ {
+			dbs[i] = memdb.New()
+		}
+	} else {
+		mainDB = mdbx.NewMDBX(log.New()).Path(GetBlockDbPath(isBeacon, -1)).MustOpen()
+		for i := 0; i < config.Concurrency; i++ {
+			dbPath := GetBlockDbPath(isBeacon, i)
+			dbs[i] = mdbx.NewMDBX(log.New()).Path(dbPath).MustOpen()
+		}
+	}
+
+	if errInitDB := initDB(ctx, mainDB, dbs, config.Concurrency); errInitDB != nil {
+		return nil, errInitDB
+	}
+
+	stageHeadsCfg := NewStageHeadersCfg(ctx, bc, mainDB)
+	stageShortRangeCfg := NewStageShortRangeCfg(ctx, bc, mainDB)
+	stageSyncEpochCfg := NewStageEpochCfg(ctx, bc, mainDB)
+	stageBodiesCfg := NewStageBodiesCfg(ctx, bc, mainDB, dbs, config.Concurrency, protocol, isBeacon, logProgress)
+	stageStatesCfg := NewStageStatesCfg(ctx, bc, mainDB, dbs, config.Concurrency, logger, logProgress)
+	stageFinishCfg := NewStageFinishCfg(ctx, mainDB)
+
+	stages := DefaultStages(ctx,
+		stageHeadsCfg,
+		stageSyncEpochCfg,
+		stageShortRangeCfg,
+		stageBodiesCfg,
+		stageStatesCfg,
+		stageFinishCfg,
+	)
+
+	return New(ctx,
+		bc,
+		mainDB,
+		stages,
+		isBeacon,
+		protocol,
+		UseMemDB,
+		config,
+		logger,
+	), nil
+}
+
+// initDB inits the sync loop main database and create buckets
+func initDB(ctx context.Context, mainDB kv.RwDB, dbs []kv.RwDB, concurrency int) error {
+
+	// create buckets for mainDB
+	tx, errRW := mainDB.BeginRw(ctx)
+	if errRW != nil {
+		return errRW
+	}
+	defer tx.Rollback()
+
+	for _, name := range Buckets {
+		if err := tx.CreateBucket(GetStageName(name, false, false)); err != nil {
+			return err
+		}
+	}
+	if err := tx.Commit(); err != nil {
+		return err
+	}
+
+	// create buckets for block cache DBs
+	for _, db := range dbs {
+		tx, errRW := db.BeginRw(ctx)
+		if errRW != nil {
+			return errRW
+		}
+
+		if err := tx.CreateBucket(BlocksBucket); err != nil {
+			return err
+		}
+		if err := tx.CreateBucket(BlockSignaturesBucket); err != nil {
+			return err
+		}
+
+		if err := tx.Commit(); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func GetBlockDbPath(beacon bool, loopID int) string {
+	if beacon {
+		if loopID >= 0 {
+			return fmt.Sprintf("%s_%d", "cache/beacon_blocks_db", loopID)
+		} else {
+			return "cache/beacon_blocks_db_main"
+		}
+	} else {
+		if loopID >= 0 {
+			return fmt.Sprintf("%s_%d", "cache/blocks_db", loopID)
+		} else {
+			return "cache/blocks_db_main"
+		}
+	}
+}
+
+// doSync does the long range sync.
+// One LongRangeSync consists of several iterations.
+// For each iteration, estimate the current block number, then fetch block & insert to blockchain
+func (s *StagedStreamSync) doSync(downloaderContext context.Context, initSync bool) (int, error) {
+
+	var totalInserted int
+
+	s.initSync = initSync
+
+	if err := s.checkPrerequisites(); err != nil {
+		return 0, err
+	}
+
+	var estimatedHeight uint64
+	if initSync {
+		if h, err := s.estimateCurrentNumber(); err != nil {
+			return 0, err
+		} else {
+			estimatedHeight = h
+			//TODO: use directly currentCycle var
+			s.status.setTargetBN(estimatedHeight)
+		}
+		if curBN := s.bc.CurrentBlock().NumberU64(); estimatedHeight <= curBN {
+			s.logger.Info().Uint64("current number", curBN).Uint64("target number", estimatedHeight).
+				Msg(WrapStagedSyncMsg("early return of long range sync"))
+			return 0, nil
+		}
+
+		s.startSyncing()
+		defer s.finishSyncing()
+	}
+
+	for {
+		ctx, cancel := context.WithCancel(downloaderContext)
+		s.ctx = ctx
+		s.SetNewContext(ctx)
+
+		n, err := s.doSyncCycle(ctx, initSync)
+		if err != nil {
+			pl := s.promLabels()
+			pl["error"] = err.Error()
+			numFailedDownloadCounterVec.With(pl).Inc()
+
+			cancel()
+			return totalInserted + n, err
+		}
+		cancel()
+
+		totalInserted += n
+
+		// if it's not long range sync, skip loop
+		if n < LastMileBlocksThreshold || !initSync {
+			return totalInserted, nil
+		}
+	}
+
+}
+
+func (s *StagedStreamSync) doSyncCycle(ctx context.Context, initSync bool) (int, error) {
+
+	// TODO: initSync=true means currentCycleNumber==0, so we can remove initSync
+
+	var totalInserted int
+
+	s.inserted = 0
+	startHead := s.bc.CurrentBlock().NumberU64()
+	canRunCycleInOneTransaction := false
+
+	var tx kv.RwTx
+	if canRunCycleInOneTransaction {
+		var err error
+		if tx, err = s.DB().BeginRw(context.Background()); err != nil {
+			return totalInserted, err
+		}
+		defer tx.Rollback()
+	}
+
+	startTime := time.Now()
+
+	// Do one cycle of staged sync
+	initialCycle := s.currentCycle.Number == 0
+	if err := s.Run(s.DB(), tx, initialCycle); err != nil {
+		utils.Logger().Error().
+			Err(err).
+			Bool("isBeacon", s.isBeacon).
+			Uint32("shard", s.bc.ShardID()).
+			Uint64("currentHeight", startHead).
+			Msgf(WrapStagedSyncMsg("sync cycle failed"))
+		return totalInserted, err
+	}
+
+	totalInserted += s.inserted
+
+	s.currentCycle.lock.Lock()
+	s.currentCycle.Number++
+	s.currentCycle.lock.Unlock()
+
+	// calculating sync speed (blocks/second)
+	if s.LogProgress && s.inserted > 0 {
+		dt := time.Now().Sub(startTime).Seconds()
+		speed := float64(0)
+		if dt > 0 {
+			speed = float64(s.inserted) / dt
+		}
+		syncSpeed := fmt.Sprintf("%.2f", speed)
+		fmt.Println("sync speed:", syncSpeed, "blocks/s")
+	}
+
+	return totalInserted, nil
+}
+
+func (s *StagedStreamSync) startSyncing() {
+	s.status.startSyncing()
+	if s.evtDownloadStartedSubscribed {
+		s.evtDownloadStarted.Send(struct{}{})
+	}
+}
+
+func (s *StagedStreamSync) finishSyncing() {
+	s.status.finishSyncing()
+	if s.evtDownloadFinishedSubscribed {
+		s.evtDownloadFinished.Send(struct{}{})
+	}
+}
+
+func (s *StagedStreamSync) checkPrerequisites() error {
+	return s.checkHaveEnoughStreams()
+}
+
+// estimateCurrentNumber roughly estimates the current block number.
+// The block number does not need to be exact, but just a temporary target of the iteration
+func (s *StagedStreamSync) estimateCurrentNumber() (uint64, error) {
+	var (
+		cnResults = make(map[sttypes.StreamID]uint64)
+		lock      sync.Mutex
+		wg        sync.WaitGroup
+	)
+	wg.Add(s.config.Concurrency)
+	for i := 0; i != s.config.Concurrency; i++ {
+		go func() {
+			defer wg.Done()
+			bn, stid, err := s.doGetCurrentNumberRequest()
+			if err != nil {
+				s.logger.Err(err).Str("streamID", string(stid)).
+					Msg(WrapStagedSyncMsg("getCurrentNumber request failed"))
+				if !errors.Is(err, context.Canceled) {
+					s.protocol.StreamFailed(stid, "getCurrentNumber request failed")
+				}
+				return
+			}
+			lock.Lock()
+			cnResults[stid] = bn
+			lock.Unlock()
+		}()
+	}
+	wg.Wait()
+
+	if len(cnResults) == 0 {
+		select {
+		case <-s.ctx.Done():
+			return 0, s.ctx.Err()
+		default:
+		}
+		return 0, errors.New("zero block number response from remote nodes")
+	}
+	bn := computeBlockNumberByMaxVote(cnResults)
+	return bn, nil
+}

api/service/stagedstreamsync/types.go

@@ -0,0 +1,287 @@
+package stagedstreamsync
+
+import (
+	"container/heap"
+	"sync"
+
+	"github.com/ethereum/go-ethereum/common"
+	"github.com/harmony-one/harmony/core/types"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+)
+
+var (
+	emptyHash common.Hash
+)
+
+type status struct {
+	isSyncing bool
+	targetBN  uint64
+	lock      sync.Mutex
+}
+
+func newStatus() status {
+	return status{}
+}
+
+func (s *status) startSyncing() {
+	s.lock.Lock()
+	defer s.lock.Unlock()
+
+	s.isSyncing = true
+}
+
+func (s *status) setTargetBN(val uint64) {
+	s.lock.Lock()
+	defer s.lock.Unlock()
+
+	s.targetBN = val
+}
+
+func (s *status) finishSyncing() {
+	s.lock.Lock()
+	defer s.lock.Unlock()
+
+	s.isSyncing = false
+	s.targetBN = 0
+}
+
+func (s *status) get() (bool, uint64) {
+	s.lock.Lock()
+	defer s.lock.Unlock()
+
+	return s.isSyncing, s.targetBN
+}
+
+type getBlocksResult struct {
+	bns    []uint64
+	blocks []*types.Block
+	stid   sttypes.StreamID
+}
+
+type resultQueue struct {
+	results *priorityQueue
+	lock    sync.Mutex
+}
+
+func newResultQueue() *resultQueue {
+	pq := make(priorityQueue, 0, 200) // 200 - rough estimate
+	heap.Init(&pq)
+	return &resultQueue{
+		results: &pq,
+	}
+}
+
+// addBlockResults adds the blocks to the result queue to be processed by insertChainLoop.
+// If a nil block is detected in the block list, will not process further blocks.
+func (rq *resultQueue) addBlockResults(blocks []*types.Block, stid sttypes.StreamID) {
+	rq.lock.Lock()
+	defer rq.lock.Unlock()
+
+	for _, block := range blocks {
+		if block == nil {
+			continue
+		}
+		heap.Push(rq.results, &blockResult{
+			block: block,
+			stid:  stid,
+		})
+	}
+	return
+}
+
+// popBlockResults pop a continuous list of blocks starting at expStartBN with capped size.
+// Return the stale block numbers as the second return value
+func (rq *resultQueue) popBlockResults(expStartBN uint64, cap int) ([]*blockResult, []uint64) {
+	rq.lock.Lock()
+	defer rq.lock.Unlock()
+
+	var (
+		res    = make([]*blockResult, 0, cap)
+		stales []uint64
+	)
+
+	for cnt := 0; rq.results.Len() > 0 && cnt < cap; cnt++ {
+		br := heap.Pop(rq.results).(*blockResult)
+		// stale block number
+		if br.block.NumberU64() < expStartBN {
+			stales = append(stales, br.block.NumberU64())
+			continue
+		}
+		if br.block.NumberU64() != expStartBN {
+			heap.Push(rq.results, br)
+			return res, stales
+		}
+		res = append(res, br)
+		expStartBN++
+	}
+	return res, stales
+}
+
+// removeResultsByStreamID removes the block results of the given stream, returns the block
+// number removed from the queue
+func (rq *resultQueue) removeResultsByStreamID(stid sttypes.StreamID) []uint64 {
+	rq.lock.Lock()
+	defer rq.lock.Unlock()
+
+	var removed []uint64
+
+Loop:
+	for {
+		for i, res := range *rq.results {
+			blockRes := res.(*blockResult)
+			if blockRes.stid == stid {
+				rq.removeByIndex(i)
+				removed = append(removed, blockRes.block.NumberU64())
+				goto Loop
+			}
+		}
+		break
+	}
+	return removed
+}
+
+func (rq *resultQueue) length() int {
+	return len(*rq.results)
+}
+
+func (rq *resultQueue) removeByIndex(index int) {
+	heap.Remove(rq.results, index)
+}
+
+// bnPrioritizedItem is the item which uses block number to determine its priority
+type bnPrioritizedItem interface {
+	getBlockNumber() uint64
+}
+
+type blockResult struct {
+	block *types.Block
+	stid  sttypes.StreamID
+}
+
+func (br *blockResult) getBlockNumber() uint64 {
+	return br.block.NumberU64()
+}
+
+func blockResultsToBlocks(results []*blockResult) []*types.Block {
+	blocks := make([]*types.Block, 0, len(results))
+
+	for _, result := range results {
+		blocks = append(blocks, result.block)
+	}
+	return blocks
+}
+
+type (
+	prioritizedNumber uint64
+
+	prioritizedNumbers struct {
+		q *priorityQueue
+	}
+)
+
+func (b prioritizedNumber) getBlockNumber() uint64 {
+	return uint64(b)
+}
+
+func newPrioritizedNumbers() *prioritizedNumbers {
+	pqs := make(priorityQueue, 0)
+	heap.Init(&pqs)
+	return &prioritizedNumbers{
+		q: &pqs,
+	}
+}
+
+func (pbs *prioritizedNumbers) push(bn uint64) {
+	heap.Push(pbs.q, prioritizedNumber(bn))
+}
+
+func (pbs *prioritizedNumbers) pop() uint64 {
+	if pbs.q.Len() == 0 {
+		return 0
+	}
+	item := heap.Pop(pbs.q)
+	return uint64(item.(prioritizedNumber))
+}
+
+func (pbs *prioritizedNumbers) length() int {
+	return len(*pbs.q)
+}
+
+type (
+	blockByNumber types.Block
+
+	// blocksByNumber is the priority queue ordered by number
+	blocksByNumber struct {
+		q   *priorityQueue
+		cap int
+	}
+)
+
+func (b *blockByNumber) getBlockNumber() uint64 {
+	raw := (*types.Block)(b)
+	return raw.NumberU64()
+}
+
+func newBlocksByNumber(cap int) *blocksByNumber {
+	pqs := make(priorityQueue, 0)
+	heap.Init(&pqs)
+	return &blocksByNumber{
+		q:   &pqs,
+		cap: cap,
+	}
+}
+
+func (bs *blocksByNumber) push(b *types.Block) {
+	heap.Push(bs.q, (*blockByNumber)(b))
+	for bs.q.Len() > bs.cap {
+		heap.Pop(bs.q)
+	}
+}
+
+func (bs *blocksByNumber) pop() *types.Block {
+	if bs.q.Len() == 0 {
+		return nil
+	}
+	item := heap.Pop(bs.q)
+	return (*types.Block)(item.(*blockByNumber))
+}
+
+func (bs *blocksByNumber) len() int {
+	return bs.q.Len()
+}
+
+// priorityQueue is a priority queue with lowest block number with highest priority
+type priorityQueue []bnPrioritizedItem
+
+func (q priorityQueue) Len() int {
+	return len(q)
+}
+
+func (q priorityQueue) Less(i, j int) bool {
+	bn1 := q[i].getBlockNumber()
+	bn2 := q[j].getBlockNumber()
+	return bn1 < bn2 // small block number has higher priority
+}
+
+func (q priorityQueue) Swap(i, j int) {
+	q[i], q[j] = q[j], q[i]
+}
+
+func (q *priorityQueue) Push(x interface{}) {
+	item, ok := x.(bnPrioritizedItem)
+	if !ok {
+		panic("wrong type of getBlockNumber interface")
+	}
+	*q = append(*q, item)
+}
+
+func (q *priorityQueue) Pop() interface{} {
+	prev := *q
+	n := len(prev)
+	if n == 0 {
+		return nil
+	}
+	res := prev[n-1]
+	*q = prev[0 : n-1]
+	return res
+}

api/service/stagedstreamsync/types_test.go

@@ -0,0 +1,266 @@
+package stagedstreamsync
+
+import (
+	"container/heap"
+	"fmt"
+	"math/big"
+	"strings"
+	"testing"
+
+	"github.com/harmony-one/harmony/block"
+	headerV3 "github.com/harmony-one/harmony/block/v3"
+	"github.com/harmony-one/harmony/core/types"
+	bls_cosi "github.com/harmony-one/harmony/crypto/bls"
+	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+)
+
+func TestResultQueue_AddBlockResults(t *testing.T) {
+	tests := []struct {
+		initBNs []uint64
+		addBNs  []uint64
+		expSize int
+	}{
+		{
+			initBNs: []uint64{},
+			addBNs:  []uint64{1, 2, 3, 4},
+			expSize: 4,
+		},
+		{
+			initBNs: []uint64{1, 2, 3, 4},
+			addBNs:  []uint64{5, 6, 7, 8},
+			expSize: 8,
+		},
+	}
+	for i, test := range tests {
+		rq := makeTestResultQueue(test.initBNs)
+		rq.addBlockResults(makeTestBlocks(test.addBNs), "")
+
+		if rq.results.Len() != test.expSize {
+			t.Errorf("Test %v: unexpected size: %v / %v", i, rq.results.Len(), test.expSize)
+		}
+	}
+}
+
+func TestResultQueue_PopBlockResults(t *testing.T) {
+	tests := []struct {
+		initBNs   []uint64
+		cap       int
+		expStart  uint64
+		expSize   int
+		staleSize int
+	}{
+		{
+			initBNs:   []uint64{1, 2, 3, 4, 5},
+			cap:       3,
+			expStart:  1,
+			expSize:   3,
+			staleSize: 0,
+		},
+		{
+			initBNs:   []uint64{1, 2, 3, 4, 5},
+			cap:       10,
+			expStart:  1,
+			expSize:   5,
+			staleSize: 0,
+		},
+		{
+			initBNs:   []uint64{1, 3, 4, 5},
+			cap:       10,
+			expStart:  1,
+			expSize:   1,
+			staleSize: 0,
+		},
+		{
+			initBNs:   []uint64{1, 2, 3, 4, 5},
+			cap:       10,
+			expStart:  0,
+			expSize:   0,
+			staleSize: 0,
+		},
+		{
+			initBNs:   []uint64{1, 1, 1, 1, 2},
+			cap:       10,
+			expStart:  1,
+			expSize:   2,
+			staleSize: 3,
+		},
+		{
+			initBNs:   []uint64{1, 2, 3, 4, 5},
+			cap:       10,
+			expStart:  2,
+			expSize:   4,
+			staleSize: 1,
+		},
+	}
+	for i, test := range tests {
+		rq := makeTestResultQueue(test.initBNs)
+		res, stales := rq.popBlockResults(test.expStart, test.cap)
+		if len(res) != test.expSize {
+			t.Errorf("Test %v: unexpect size %v / %v", i, len(res), test.expSize)
+		}
+		if len(stales) != test.staleSize {
+			t.Errorf("Test %v: unexpect stale size %v / %v", i, len(stales), test.staleSize)
+		}
+	}
+}
+
+func TestResultQueue_RemoveResultsByStreamID(t *testing.T) {
+	tests := []struct {
+		rq         *resultQueue
+		rmStreamID sttypes.StreamID
+		removed    int
+		expSize    int
+	}{
+		{
+			rq:         makeTestResultQueue([]uint64{1, 2, 3, 4}),
+			rmStreamID: "test stream id",
+			removed:    4,
+			expSize:    0,
+		},
+		{
+			rq: func() *resultQueue {
+				rq := makeTestResultQueue([]uint64{2, 3, 4, 5})
+				rq.addBlockResults([]*types.Block{
+					makeTestBlock(1),
+					makeTestBlock(5),
+					makeTestBlock(6),
+				}, "another test stream id")
+				return rq
+			}(),
+			rmStreamID: "test stream id",
+			removed:    4,
+			expSize:    3,
+		},
+		{
+			rq: func() *resultQueue {
+				rq := makeTestResultQueue([]uint64{2, 3, 4, 5})
+				rq.addBlockResults([]*types.Block{
+					makeTestBlock(1),
+					makeTestBlock(5),
+					makeTestBlock(6),
+				}, "another test stream id")
+				return rq
+			}(),
+			rmStreamID: "another test stream id",
+			removed:    3,
+			expSize:    4,
+		},
+	}
+	for i, test := range tests {
+		res := test.rq.removeResultsByStreamID(test.rmStreamID)
+		if len(res) != test.removed {
+			t.Errorf("Test %v: unexpected number removed %v / %v", i, len(res), test.removed)
+		}
+		if gotSize := test.rq.results.Len(); gotSize != test.expSize {
+			t.Errorf("Test %v: unexpected number after removal %v / %v", i, gotSize, test.expSize)
+		}
+	}
+}
+
+func makeTestResultQueue(bns []uint64) *resultQueue {
+	rq := newResultQueue()
+	for _, bn := range bns {
+		heap.Push(rq.results, &blockResult{
+			block: makeTestBlock(bn),
+			stid:  "test stream id",
+		})
+	}
+	return rq
+}
+
+func TestPrioritizedBlocks(t *testing.T) {
+	addBNs := []uint64{4, 7, 6, 9}
+
+	bns := newPrioritizedNumbers()
+	for _, bn := range addBNs {
+		bns.push(bn)
+	}
+	prevBN := uint64(0)
+	for len(*bns.q) > 0 {
+		b := bns.pop()
+		if b < prevBN {
+			t.Errorf("number not incrementing")
+		}
+		prevBN = b
+	}
+	if last := bns.pop(); last != 0 {
+		t.Errorf("last elem is not 0")
+	}
+}
+
+func TestBlocksByNumber(t *testing.T) {
+	addBNs := []uint64{4, 7, 6, 9}
+
+	bns := newBlocksByNumber(10)
+	for _, bn := range addBNs {
+		bns.push(makeTestBlock(bn))
+	}
+	if bns.len() != len(addBNs) {
+		t.Errorf("size unexpected: %v / %v", bns.len(), len(addBNs))
+	}
+	prevBN := uint64(0)
+	for len(*bns.q) > 0 {
+		b := bns.pop()
+		if b.NumberU64() < prevBN {
+			t.Errorf("number not incrementing")
+		}
+		prevBN = b.NumberU64()
+	}
+	if lastBlock := bns.pop(); lastBlock != nil {
+		t.Errorf("last block is not nil")
+	}
+}
+
+func TestPriorityQueue(t *testing.T) {
+	testBNs := []uint64{1, 9, 2, 4, 5, 12}
+	pq := make(priorityQueue, 0, 10)
+	heap.Init(&pq)
+	for _, bn := range testBNs {
+		heap.Push(&pq, &blockResult{
+			block: makeTestBlock(bn),
+			stid:  "",
+		})
+	}
+	cmpBN := uint64(0)
+	for pq.Len() > 0 {
+		bn := heap.Pop(&pq).(*blockResult).block.NumberU64()
+		if bn < cmpBN {
+			t.Errorf("not incrementing")
+		}
+		cmpBN = bn
+	}
+	if pq.Len() != 0 {
+		t.Errorf("after poping, size not 0")
+	}
+}
+
+func makeTestBlocks(bns []uint64) []*types.Block {
+	blocks := make([]*types.Block, 0, len(bns))
+	for _, bn := range bns {
+		blocks = append(blocks, makeTestBlock(bn))
+	}
+	return blocks
+}
+
+func makeTestBlock(bn uint64) *types.Block {
+	testHeader := &block.Header{Header: headerV3.NewHeader()}
+	testHeader.SetNumber(big.NewInt(int64(bn)))
+	testHeader.SetLastCommitSignature(bls_cosi.SerializedSignature{})
+	testHeader.SetLastCommitBitmap(make([]byte, 10))
+	block := types.NewBlockWithHeader(testHeader)
+	block.SetCurrentCommitSig(make([]byte, 106))
+	return block
+}
+
+func assertError(got, expect error) error {
+	if (got == nil) != (expect == nil) {
+		return fmt.Errorf("unexpected error [%v] / [%v]", got, expect)
+	}
+	if (got == nil) || (expect == nil) {
+		return nil
+	}
+	if !strings.Contains(got.Error(), expect.Error()) {
+		return fmt.Errorf("unexpected error [%v] / [%v]", got, expect)
+	}
+	return nil
+}

cmd/harmony/main.go

@@ -39,6 +39,7 @@ import (
 	"github.com/harmony-one/harmony/api/service"
 	"github.com/harmony-one/harmony/api/service/pprof"
 	"github.com/harmony-one/harmony/api/service/prometheus"
+	"github.com/harmony-one/harmony/api/service/stagedstreamsync"
 	"github.com/harmony-one/harmony/api/service/synchronize"
 	"github.com/harmony-one/harmony/common/fdlimit"
 	"github.com/harmony-one/harmony/common/ntp"
@@ -415,8 +416,12 @@ func setupNodeAndRun(hc harmonyconfig.HarmonyConfig) {
 
 	// Setup services
 	if hc.Sync.Enabled {
+		if hc.Sync.StagedSync {
+			setupStagedSyncService(currentNode, myHost, hc)
+		} else {
 			setupSyncService(currentNode, myHost, hc)
 		}
+	}
 	if currentNode.NodeConfig.Role() == nodeconfig.Validator {
 		currentNode.RegisterValidatorServices()
 	} else if currentNode.NodeConfig.Role() == nodeconfig.ExplorerNode {
@@ -906,6 +911,45 @@ func setupSyncService(node *node.Node, host p2p.Host, hc harmonyconfig.HarmonyCo
 	}
 }
 
+func setupStagedSyncService(node *node.Node, host p2p.Host, hc harmonyconfig.HarmonyConfig) {
+	blockchains := []core.BlockChain{node.Blockchain()}
+	if !node.IsRunningBeaconChain() {
+		blockchains = append(blockchains, node.Beaconchain())
+	}
+
+	sConfig := stagedstreamsync.Config{
+		ServerOnly:   !hc.Sync.Downloader,
+		Network:      nodeconfig.NetworkType(hc.Network.NetworkType),
+		Concurrency:  hc.Sync.Concurrency,
+		MinStreams:   hc.Sync.MinPeers,
+		InitStreams:  hc.Sync.InitStreams,
+		SmSoftLowCap: hc.Sync.DiscSoftLowCap,
+		SmHardLowCap: hc.Sync.DiscHardLowCap,
+		SmHiCap:      hc.Sync.DiscHighCap,
+		SmDiscBatch:  hc.Sync.DiscBatch,
+		LogProgress:  node.NodeConfig.LogProgress,
+	}
+
+	// If we are running side chain, we will need to do some extra works for beacon
+	// sync.
+	if !node.IsRunningBeaconChain() {
+		sConfig.BHConfig = &stagedstreamsync.BeaconHelperConfig{
+			BlockC:     node.BeaconBlockChannel,
+			InsertHook: node.BeaconSyncHook,
+		}
+	}
+
+	//Setup stream sync service
+	s := stagedstreamsync.NewService(host, blockchains, sConfig)
+
+	node.RegisterService(service.StagedStreamSync, s)
+
+	d := s.Downloaders.GetShardDownloader(node.Blockchain().ShardID())
+	if hc.Sync.Downloader && hc.General.NodeType != nodeTypeExplorer {
+		node.Consensus.SetDownloader(d) // Set downloader when stream client is active
+	}
+}
+
 func setupBlacklist(hc harmonyconfig.HarmonyConfig) (map[ethCommon.Address]struct{}, error) {
 	rosetta_common.InitRosettaFile(hc.TxPool.RosettaFixFile)
 

go.mod

@@ -165,6 +165,7 @@ require (
 	github.com/libp2p/go-cidranger v1.1.0 // indirect
 	github.com/libp2p/go-flow-metrics v0.1.0 // indirect
 	github.com/libp2p/go-libp2p-asn-util v0.2.0 // indirect
+	github.com/libp2p/go-libp2p-core v0.20.1 // indirect
 	github.com/libp2p/go-libp2p-kbucket v0.5.0 // indirect
 	github.com/libp2p/go-libp2p-record v0.2.0 // indirect
 	github.com/libp2p/go-msgio v0.2.0 // indirect

go.sum

@@ -594,6 +594,8 @@ github.com/libp2p/go-libp2p v0.24.0 h1:DQk/5bBon+yUVIGTeRVBmOYpZzoBHx/VTC0xoLgJG
 github.com/libp2p/go-libp2p v0.24.0/go.mod h1:28t24CYDlnBs23rIs1OclU89YbhgibrBq2LFbMe+cFw=
 github.com/libp2p/go-libp2p-asn-util v0.2.0 h1:rg3+Os8jbnO5DxkC7K/Utdi+DkY3q/d1/1q+8WeNAsw=
 github.com/libp2p/go-libp2p-asn-util v0.2.0/go.mod h1:WoaWxbHKBymSN41hWSq/lGKJEca7TNm58+gGJi2WsLI=
+github.com/libp2p/go-libp2p-core v0.20.1 h1:fQz4BJyIFmSZAiTbKV8qoYhEH5Dtv/cVhZbG3Ib/+Cw=
+github.com/libp2p/go-libp2p-core v0.20.1/go.mod h1:6zR8H7CvQWgYLsbG4on6oLNSGcyKaYFSEYyDt51+bIY=
 github.com/libp2p/go-libp2p-kad-dht v0.19.0 h1:2HuiInHZTm9ZvQajaqdaPLHr0PCKKigWiflakimttE0=
 github.com/libp2p/go-libp2p-kad-dht v0.19.0/go.mod h1:qPIXdiZsLczhV4/+4EO1jE8ae0YCW4ZOogc4WVIyTEU=
 github.com/libp2p/go-libp2p-kbucket v0.5.0 h1:g/7tVm8ACHDxH29BGrpsQlnNeu+6OF1A9bno/4/U1oA=

hmy/downloader/const.go

@@ -35,7 +35,8 @@ type (
 		// Only run stream sync protocol as a server.
 		// TODO: remove this when stream sync is fully up.
 		ServerOnly bool
-
+		// use staged sync
+		Staged bool
 		// parameters
 		Network     nodeconfig.NetworkType
 		Concurrency int // Number of concurrent sync requests

hmy/downloader/metric.go

@@ -23,7 +23,7 @@ var (
 	consensusTriggeredDownloadCounterVec = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
 			Namespace: "hmy",
-			Subsystem: "downloader",
+			Subsystem: "StreamSync",
 			Name:      "consensus_trigger",
 			Help:      "number of times consensus triggered download task",
 		},
@@ -33,7 +33,7 @@ var (
 	longRangeSyncedBlockCounterVec = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
 			Namespace: "hmy",
-			Subsystem: "downloader",
+			Subsystem: "StreamSync",
 			Name:      "num_blocks_synced_long_range",
 			Help:      "number of blocks synced in long range sync",
 		},
@@ -43,7 +43,7 @@ var (
 	longRangeFailInsertedBlockCounterVec = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
 			Namespace: "hmy",
-			Subsystem: "downloader",
+			Subsystem: "StreamSync",
 			Name:      "num_blocks_failed_long_range",
 			Help:      "number of blocks failed to insert into change in long range sync",
 		},
@@ -53,7 +53,7 @@ var (
 	numShortRangeCounterVec = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
 			Namespace: "hmy",
-			Subsystem: "downloader",
+			Subsystem: "StreamSync",
 			Name:      "num_short_range",
 			Help:      "number of short range sync is triggered",
 		},
@@ -63,7 +63,7 @@ var (
 	numFailedDownloadCounterVec = prometheus.NewCounterVec(
 		prometheus.CounterOpts{
 			Namespace: "hmy",
-			Subsystem: "downloader",
+			Subsystem: "StreamSync",
 			Name:      "failed_download",
 			Help:      "number of downloading is failed",
 		},
@@ -73,7 +73,7 @@ var (
 	numBlocksInsertedShortRangeHistogramVec = prometheus.NewHistogramVec(
 		prometheus.HistogramOpts{
 			Namespace: "hmy",
-			Subsystem: "downloader",
+			Subsystem: "StreamSync",
 			Name:      "num_blocks_inserted_short_range",
 			Help:      "number of blocks inserted for each short range sync",
 			// Buckets: 0, 1, 2, 4, +INF (capped at 10)
@@ -85,7 +85,7 @@ var (
 	numBlocksInsertedBeaconHelperCounter = prometheus.NewCounter(
 		prometheus.CounterOpts{
 			Namespace: "hmy",
-			Subsystem: "downloader",
+			Subsystem: "StreamSync",
 			Name:      "num_blocks_inserted_beacon_helper",
 			Help:      "number of blocks inserted from beacon helper",
 		},

hmy/downloader/shortrange.go

@@ -20,7 +20,7 @@ import (
 // doShortRangeSync does the short range sync.
 // Compared with long range sync, short range sync is more focused on syncing to the latest block.
 // It consist of 3 steps:
-// 1. Obtain the block hashes and ompute the longest hash chain..
+// 1. Obtain the block hashes and compute the longest hash chain..
 // 2. Get blocks by hashes from computed hash chain.
 // 3. Insert the blocks to blockchain.
 func (d *Downloader) doShortRangeSync() (int, error) {

node/node_syncing.go

@@ -22,11 +22,11 @@ import (
 	"github.com/harmony-one/harmony/api/service/legacysync"
 	legdownloader "github.com/harmony-one/harmony/api/service/legacysync/downloader"
 	downloader_pb "github.com/harmony-one/harmony/api/service/legacysync/downloader/proto"
+	"github.com/harmony-one/harmony/api/service/stagedstreamsync"
 	"github.com/harmony-one/harmony/api/service/stagedsync"
 	"github.com/harmony-one/harmony/api/service/synchronize"
 	"github.com/harmony-one/harmony/core"
 	"github.com/harmony-one/harmony/core/types"
-	"github.com/harmony-one/harmony/hmy/downloader"
 	nodeconfig "github.com/harmony-one/harmony/internal/configs/node"
 	"github.com/harmony-one/harmony/internal/utils"
 	"github.com/harmony-one/harmony/node/worker"
@@ -811,7 +811,7 @@ func (node *Node) legacySyncStatus(shardID uint32) (bool, uint64, uint64) {
 	}
 }
 
-// IsOutOfSync return whether the node is out of sync of the given hsardID
+// IsOutOfSync return whether the node is out of sync of the given shardID
 func (node *Node) IsOutOfSync(shardID uint32) bool {
 	ds := node.getDownloaders()
 	if ds == nil || !ds.IsActive() {
@@ -859,7 +859,28 @@ func (node *Node) SyncPeers() map[string]int {
 	return res
 }
 
-func (node *Node) getDownloaders() *downloader.Downloaders {
+type Downloaders interface {
+	Start()
+	Close()
+	DownloadAsync(shardID uint32)
+	// GetShardDownloader(shardID uint32) *Downloader
+	NumPeers() map[uint32]int
+	SyncStatus(shardID uint32) (bool, uint64, uint64)
+	IsActive() bool
+}
+
+func (node *Node) getDownloaders() Downloaders {
+	if node.NodeConfig.StagedSync {
+		syncService := node.serviceManager.GetService(service.StagedStreamSync)
+		if syncService == nil {
+			return nil
+		}
+		dsService, ok := syncService.(*stagedstreamsync.StagedStreamSyncService)
+		if !ok {
+			return nil
+		}
+		return dsService.Downloaders
+	} else {
 		syncService := node.serviceManager.GetService(service.Synchronize)
 		if syncService == nil {
 			return nil
@@ -869,4 +890,5 @@ func (node *Node) getDownloaders() *downloader.Downloaders {
 			return nil
 		}
 		return dsService.Downloaders
+	}
 }

p2p/stream/common/requestmanager/interface_test.go

@@ -114,6 +114,18 @@ func (st *testStream) CloseOnExit() error {
 	return nil
 }
 
+func (st *testStream) FailedTimes() int {
+	return 0
+}
+
+func (st *testStream) AddFailedTimes() {
+	return
+}
+
+func (st *testStream) ResetFailedRimes() {
+	return
+}
+
 func makeDummyTestStreams(indexes []int) []sttypes.Stream {
 	sts := make([]sttypes.Stream, 0, len(indexes))
 

p2p/stream/common/streammanager/interface_test.go

@@ -70,6 +70,18 @@ func (st *testStream) ReadBytes() ([]byte, error) {
 	return nil, nil
 }
 
+func (st *testStream) FailedTimes() int {
+	return 0
+}
+
+func (st *testStream) AddFailedTimes() {
+	return
+}
+
+func (st *testStream) ResetFailedRimes() {
+	return
+}
+
 func (st *testStream) Close() error {
 	if st.closed {
 		return errors.New("already closed")

p2p/stream/common/streammanager/streammanager.go

@@ -10,6 +10,7 @@ import (
 	"github.com/harmony-one/abool"
 	"github.com/harmony-one/harmony/internal/utils"
 	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/harmony-one/harmony/shard"
 	"github.com/libp2p/go-libp2p/core/network"
 	libp2p_peer "github.com/libp2p/go-libp2p/core/peer"
 	"github.com/libp2p/go-libp2p/core/protocol"
@@ -73,6 +74,9 @@ func newStreamManager(pid sttypes.ProtoID, host host, pf peerFinder, handleStrea
 
 	protoSpec, _ := sttypes.ProtoIDToProtoSpec(pid)
 
+	fmt.Println("my peer id: ", host.ID().String())
+	fmt.Println("my proto id: ", pid)
+
 	return &streamManager{
 		myProtoID:     pid,
 		myProtoSpec:   protoSpec,
@@ -234,6 +238,9 @@ func (sm *streamManager) sanityCheckStream(st sttypes.Stream) error {
 	if mySpec.ShardID != rmSpec.ShardID {
 		return fmt.Errorf("unexpected shard ID: %v/%v", rmSpec.ShardID, mySpec.ShardID)
 	}
+	if mySpec.ShardID == shard.BeaconChainShardID && !rmSpec.BeaconNode {
+		return fmt.Errorf("unexpected beacon node with shard ID: %v/%v", rmSpec.ShardID, mySpec.ShardID)
+	}
 	return nil
 }
 
@@ -323,7 +330,7 @@ func (sm *streamManager) discoverAndSetupStream(discCtx context.Context) (int, e
 }
 
 func (sm *streamManager) discover(ctx context.Context) (<-chan libp2p_peer.AddrInfo, error) {
-	protoID := string(sm.myProtoID)
+	protoID := sm.targetProtoID()
 	discBatch := sm.config.DiscBatch
 	if sm.config.HiCap-sm.streams.size() < sm.config.DiscBatch {
 		discBatch = sm.config.HiCap - sm.streams.size()
@@ -340,6 +347,14 @@ func (sm *streamManager) discover(ctx context.Context) (<-chan libp2p_peer.AddrI
 	return sm.pf.FindPeers(ctx2, protoID, discBatch)
 }
 
+func (sm *streamManager) targetProtoID() string {
+	targetSpec := sm.myProtoSpec
+	if targetSpec.ShardID == shard.BeaconChainShardID { // for beacon chain, only connect to beacon nodes
+		targetSpec.BeaconNode = true
+	}
+	return string(targetSpec.ToProtoID())
+}
+
 func (sm *streamManager) setupStreamWithPeer(ctx context.Context, pid libp2p_peer.ID) error {
 	timer := prometheus.NewTimer(setupStreamDuration.With(prometheus.Labels{"topic": string(sm.myProtoID)}))
 	defer timer.ObserveDuration()
@@ -347,7 +362,7 @@ func (sm *streamManager) setupStreamWithPeer(ctx context.Context, pid libp2p_pee
 	nCtx, cancel := context.WithTimeout(ctx, connectTimeout)
 	defer cancel()
 
-	st, err := sm.host.NewStream(nCtx, pid, protocol.ID(sm.myProtoID))
+	st, err := sm.host.NewStream(nCtx, pid, protocol.ID(sm.targetProtoID()))
 	if err != nil {
 		return err
 	}
@@ -392,6 +407,10 @@ func (ss *streamSet) get(id sttypes.StreamID) (sttypes.Stream, bool) {
 	ss.lock.RLock()
 	defer ss.lock.RUnlock()
 
+	if id == "" {
+		return nil, false
+	}
+	
 	st, ok := ss.streams[id]
 	return st, ok
 }

p2p/stream/protocols/sync/client.go

@@ -43,6 +43,36 @@ func (p *Protocol) GetBlocksByNumber(ctx context.Context, bns []uint64, opts ...
 	return
 }
 
+func (p *Protocol) GetRawBlocksByNumber(ctx context.Context, bns []uint64, opts ...Option) (blockBytes [][]byte, sigBytes [][]byte, stid sttypes.StreamID, err error) {
+	timer := p.doMetricClientRequest("getBlocksByNumber")
+	defer p.doMetricPostClientRequest("getBlocksByNumber", err, timer)
+
+	if len(bns) == 0 {
+		err = fmt.Errorf("zero block numbers requested")
+		return
+	}
+	if len(bns) > GetBlocksByNumAmountCap {
+		err = fmt.Errorf("number of blocks exceed cap of %v", GetBlocksByNumAmountCap)
+		return
+	}
+	req := newGetBlocksByNumberRequest(bns)
+	resp, stid, err := p.rm.DoRequest(ctx, req, opts...)
+	if err != nil {
+		// At this point, error can be context canceled, context timed out, or waiting queue
+		// is already full.
+		return
+	}
+
+	// Parse and return blocks
+	sResp, ok := resp.(*syncResponse)
+	if !ok || sResp == nil {
+		err = errors.New("not sync response")
+		return
+	}
+	blockBytes, sigBytes, err = req.parseBlockBytesAndSigs(sResp)
+	return
+}
+
 // GetCurrentBlockNumber get the current block number from remote node
 func (p *Protocol) GetCurrentBlockNumber(ctx context.Context, opts ...Option) (bn uint64, stid sttypes.StreamID, err error) {
 	timer := p.doMetricClientRequest("getBlockNumber")

p2p/stream/protocols/sync/const.go

@@ -17,6 +17,9 @@ const (
 	// See comments for GetBlocksByNumAmountCap.
 	GetBlocksByHashesAmountCap = 10
 
+	// MaxStreamFailures is the maximum allowed failures before stream gets removed
+	MaxStreamFailures = 3
+
 	// minAdvertiseInterval is the minimum advertise interval
 	minAdvertiseInterval = 1 * time.Minute
 

p2p/stream/protocols/sync/protocol.go

@@ -2,11 +2,13 @@ package sync
 
 import (
 	"context"
+	"fmt"
 	"strconv"
 	"time"
 
 	"github.com/ethereum/go-ethereum/event"
 	"github.com/harmony-one/harmony/consensus/engine"
+	"github.com/harmony-one/harmony/core"
 	nodeconfig "github.com/harmony-one/harmony/internal/configs/node"
 	shardingconfig "github.com/harmony-one/harmony/internal/configs/sharding"
 	"github.com/harmony-one/harmony/internal/utils"
@@ -15,6 +17,7 @@ import (
 	"github.com/harmony-one/harmony/p2p/stream/common/requestmanager"
 	"github.com/harmony-one/harmony/p2p/stream/common/streammanager"
 	sttypes "github.com/harmony-one/harmony/p2p/stream/types"
+	"github.com/harmony-one/harmony/shard"
 	"github.com/hashicorp/go-version"
 	libp2p_host "github.com/libp2p/go-libp2p/core/host"
 	libp2p_network "github.com/libp2p/go-libp2p/core/network"
@@ -40,6 +43,7 @@ type (
 	// Protocol is the protocol for sync streaming
 	Protocol struct {
 		chain      engine.ChainReader            // provide SYNC data
+		beaconNode bool                          // is beacon node or shard chain node
 		schedule   shardingconfig.Schedule       // provide schedule information
 		rl         ratelimiter.RateLimiter       // limit the incoming request rate
 		sm         streammanager.StreamManager   // stream management
@@ -74,8 +78,13 @@ type (
 func NewProtocol(config Config) *Protocol {
 	ctx, cancel := context.WithCancel(context.Background())
 
+	isBeaconNode := config.Chain.ShardID() == shard.BeaconChainShardID
+	if _, ok := config.Chain.(*core.EpochChain); ok {
+		isBeaconNode = false
+	}
 	sp := &Protocol{
 		chain:      config.Chain,
+		beaconNode: isBeaconNode,
 		disc:       config.Discovery,
 		config:     config,
 		ctx:        ctx,
@@ -162,6 +171,7 @@ func (p *Protocol) HandleStream(raw libp2p_network.Stream) {
 			Msg("failed to add new stream")
 		return
 	}
+	fmt.Println("Node connected to", raw.Conn().RemotePeer().String(), "(", st.ProtoID(), ")")
 	st.run()
 }
 
@@ -219,18 +229,40 @@ func (p *Protocol) protoIDByVersion(v *version.Version) sttypes.ProtoID {
 		NetworkType: p.config.Network,
 		ShardID:     p.config.ShardID,
 		Version:     v,
+		BeaconNode:  p.beaconNode,
 	}
 	return spec.ToProtoID()
 }
 
 // RemoveStream removes the stream of the given stream ID
+// TODO: add reason to parameters
 func (p *Protocol) RemoveStream(stID sttypes.StreamID) {
-	if stID == "" {
-		return
+	st, exist := p.sm.GetStreamByID(stID)
+	if exist && st != nil {
+		//TODO: log this incident with reason
+		st.Close()
+		// stream manager removes this stream from the list and triggers discovery if number of streams are not enough
+		p.sm.RemoveStream(stID) //TODO: double check to see if this part is needed
 	}
+}
+
+func (p *Protocol) StreamFailed(stID sttypes.StreamID, reason string) {
 	st, exist := p.sm.GetStreamByID(stID)
 	if exist && st != nil {
+		st.AddFailedTimes()
+		p.logger.Info().
+			Str("stream ID", string(st.ID())).
+			Int("num failures", st.FailedTimes()).
+			Str("reason", reason).
+			Msg("stream failed")
+		if st.FailedTimes() >= MaxStreamFailures {
 			st.Close()
+			// stream manager removes this stream from the list and triggers discovery if number of streams are not enough
+			p.sm.RemoveStream(stID) //TODO: double check to see if this part is needed
+			p.logger.Warn().
+				Str("stream ID", string(st.ID())).
+				Msg("stream removed")
+		}
 	}
 }
 

p2p/stream/types/stream.go

@@ -21,6 +21,9 @@ type Stream interface {
 	ReadBytes() ([]byte, error)
 	Close() error
 	CloseOnExit() error
+	FailedTimes() int
+	AddFailedTimes()
+	ResetFailedRimes()
 }
 
 // BaseStream is the wrapper around
@@ -34,6 +37,8 @@ type BaseStream struct {
 	spec     ProtoSpec
 	specErr  error
 	specOnce sync.Once
+
+	failedTimes int
 }
 
 // NewBaseStream creates BaseStream as the wrapper of libp2p Stream
@@ -42,6 +47,7 @@ func NewBaseStream(st libp2p_network.Stream) *BaseStream {
 	return &BaseStream{
 		raw:         st,
 		reader:      reader,
+		failedTimes: 0,
 	}
 }
 
@@ -72,6 +78,18 @@ func (st *BaseStream) Close() error {
 	return st.raw.Reset()
 }
 
+func (st *BaseStream) FailedTimes() int {
+	return st.failedTimes
+}
+
+func (st *BaseStream) AddFailedTimes() {
+	st.failedTimes++
+}
+
+func (st *BaseStream) ResetFailedRimes() {
+	st.failedTimes = 0
+}
+
 const (
 	maxMsgBytes = 20 * 1024 * 1024 // 20MB
 	sizeBytes   = 4                // uint32

p2p/stream/types/utils.go

@@ -11,7 +11,7 @@ import (
 
 	nodeconfig "github.com/harmony-one/harmony/internal/configs/node"
 	"github.com/hashicorp/go-version"
-	libp2p_proto "github.com/libp2p/go-libp2p/core/protocol"
+	libp2p_proto "github.com/libp2p/go-libp2p-core/protocol"
 	"github.com/pkg/errors"
 )
 
@@ -20,10 +20,10 @@ const (
 	ProtoIDCommonPrefix = "harmony"
 
 	// ProtoIDFormat is the format of stream protocol ID
-	ProtoIDFormat = "%s/%s/%s/%d/%s"
+	ProtoIDFormat = "%s/%s/%s/%d/%s/%d"
 
 	// protoIDNumElem is the number of elements of the ProtoID. See comments in ProtoID
-	protoIDNumElem = 5
+	protoIDNumElem = 6
 )
 
 // ProtoID is the protocol id for streaming, an alias of libp2p stream protocol ID。
@@ -32,6 +32,7 @@ const (
 // 2. NetworkType - mainnet, testnet, stn, e.t.c.
 // 3. ShardID - shard ID of the current protocol.
 // 4. Version - Stream protocol version for backward compatibility.
+// 5. BeaconNode - whether stream is from a beacon chain node or shard chain node
 type ProtoID libp2p_proto.ID
 
 // ProtoSpec is the un-serialized stream proto id specification
@@ -43,12 +44,13 @@ type ProtoSpec struct {
 	NetworkType nodeconfig.NetworkType
 	ShardID     nodeconfig.ShardID
 	Version     *version.Version
+	BeaconNode  bool
 }
 
 // ToProtoID convert a ProtoSpec to ProtoID.
 func (spec ProtoSpec) ToProtoID() ProtoID {
 	s := fmt.Sprintf(ProtoIDFormat, ProtoIDCommonPrefix, spec.Service,
-		spec.NetworkType, spec.ShardID, spec.Version.String())
+		spec.NetworkType, spec.ShardID, spec.Version.String(), bool2int(spec.BeaconNode))
 	return ProtoID(s)
 }
 
@@ -64,6 +66,7 @@ func ProtoIDToProtoSpec(id ProtoID) (ProtoSpec, error) {
 		networkType   = comps[2]
 		shardIDStr    = comps[3]
 		versionStr    = comps[4]
+		beaconnodeStr = comps[5]
 	)
 	shardID, err := strconv.Atoi(shardIDStr)
 	if err != nil {
@@ -76,11 +79,16 @@ func ProtoIDToProtoSpec(id ProtoID) (ProtoSpec, error) {
 	if err != nil {
 		return ProtoSpec{}, errors.Wrap(err, "unexpected version string")
 	}
+	isBeaconNode, err := strconv.Atoi(beaconnodeStr)
+	if err != nil {
+		return ProtoSpec{}, errors.Wrap(err, "invalid beacon node flag")
+	}
 	return ProtoSpec{
 		Service:     service,
 		NetworkType: nodeconfig.NetworkType(networkType),
 		ShardID:     nodeconfig.ShardID(uint32(shardID)),
 		Version:     version,
+		BeaconNode:  int2bool(isBeaconNode),
 	}, nil
 }
 
@@ -90,3 +98,14 @@ func GenReqID() uint64 {
 	rand.Read(rnd[:])
 	return binary.BigEndian.Uint64(rnd[:])
 }
+
+func bool2int(b bool) int {
+	if b {
+		return 1
+	}
+	return 0
+}
+
+func int2bool(i int) bool {
+	return i > 0
+}