You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
848 lines
30 KiB
848 lines
30 KiB
package consensus
|
|
|
|
import (
|
|
"bytes"
|
|
"context"
|
|
"encoding/hex"
|
|
"fmt"
|
|
"sync/atomic"
|
|
"time"
|
|
|
|
bls2 "github.com/harmony-one/bls/ffi/go/bls"
|
|
"github.com/harmony-one/harmony/consensus/signature"
|
|
|
|
nodeconfig "github.com/harmony-one/harmony/internal/configs/node"
|
|
"github.com/harmony-one/harmony/internal/utils"
|
|
|
|
"github.com/rs/zerolog"
|
|
|
|
msg_pb "github.com/harmony-one/harmony/api/proto/message"
|
|
"github.com/harmony-one/harmony/block"
|
|
"github.com/harmony-one/harmony/consensus/quorum"
|
|
"github.com/harmony-one/harmony/core/types"
|
|
"github.com/harmony-one/harmony/crypto/bls"
|
|
vrf_bls "github.com/harmony-one/harmony/crypto/vrf/bls"
|
|
"github.com/harmony-one/harmony/p2p"
|
|
"github.com/harmony-one/harmony/shard"
|
|
"github.com/harmony-one/vdf/src/vdf_go"
|
|
"github.com/pkg/errors"
|
|
"github.com/prometheus/client_golang/prometheus"
|
|
)
|
|
|
|
var (
|
|
errSenderPubKeyNotLeader = errors.New("sender pubkey doesn't match leader")
|
|
errVerifyMessageSignature = errors.New("verify message signature failed")
|
|
errParsingFBFTMessage = errors.New("failed parsing FBFT message")
|
|
)
|
|
|
|
// timeout constant
|
|
const (
|
|
// CommitSigSenderTimeout is the timeout for sending the commit sig to finish block proposal
|
|
CommitSigSenderTimeout = 10 * time.Second
|
|
// CommitSigReceiverTimeout is the timeout for the receiving side of the commit sig
|
|
// if timeout, the receiver should instead ready directly from db for the commit sig
|
|
CommitSigReceiverTimeout = 8 * time.Second
|
|
)
|
|
|
|
// IsViewChangingMode return true if curernt mode is viewchanging
|
|
func (consensus *Consensus) IsViewChangingMode() bool {
|
|
return consensus.current.Mode() == ViewChanging
|
|
}
|
|
|
|
// HandleMessageUpdate will update the consensus state according to received message
|
|
func (consensus *Consensus) HandleMessageUpdate(ctx context.Context, msg *msg_pb.Message, senderKey *bls.SerializedPublicKey) error {
|
|
// when node is in ViewChanging mode, it still accepts normal messages into FBFTLog
|
|
// in order to avoid possible trap forever but drop PREPARE and COMMIT
|
|
// which are message types specifically for a node acting as leader
|
|
// so we just ignore those messages
|
|
if consensus.IsViewChangingMode() &&
|
|
(msg.Type == msg_pb.MessageType_PREPARE ||
|
|
msg.Type == msg_pb.MessageType_COMMIT) {
|
|
return nil
|
|
}
|
|
|
|
// Do easier check before signature check
|
|
if msg.Type == msg_pb.MessageType_ANNOUNCE || msg.Type == msg_pb.MessageType_PREPARED || msg.Type == msg_pb.MessageType_COMMITTED {
|
|
// Only validator needs to check whether the message is from the correct leader
|
|
if !bytes.Equal(senderKey[:], consensus.LeaderPubKey.Bytes[:]) &&
|
|
consensus.current.Mode() == Normal && !consensus.IgnoreViewIDCheck.IsSet() {
|
|
return errSenderPubKeyNotLeader
|
|
}
|
|
}
|
|
|
|
if msg.Type != msg_pb.MessageType_PREPARE && msg.Type != msg_pb.MessageType_COMMIT {
|
|
// Leader doesn't need to check validator's message signature since the consensus signature will be checked
|
|
if !consensus.senderKeySanityChecks(msg, senderKey) {
|
|
return errVerifyMessageSignature
|
|
}
|
|
}
|
|
|
|
// Parse FBFT message
|
|
var fbftMsg *FBFTMessage
|
|
var err error
|
|
switch t := msg.Type; true {
|
|
case t == msg_pb.MessageType_VIEWCHANGE:
|
|
fbftMsg, err = ParseViewChangeMessage(msg)
|
|
case t == msg_pb.MessageType_NEWVIEW:
|
|
members := consensus.Decider.Participants()
|
|
fbftMsg, err = ParseNewViewMessage(msg, members)
|
|
default:
|
|
fbftMsg, err = consensus.ParseFBFTMessage(msg)
|
|
}
|
|
if err != nil || fbftMsg == nil {
|
|
return errors.Wrapf(err, "unable to parse consensus msg with type: %s", msg.Type)
|
|
}
|
|
|
|
canHandleViewChange := true
|
|
intendedForValidator, intendedForLeader :=
|
|
!consensus.IsLeader(),
|
|
consensus.IsLeader()
|
|
|
|
// if in backup normal mode, force ignore view change event and leader event.
|
|
if consensus.current.Mode() == NormalBackup {
|
|
canHandleViewChange = false
|
|
intendedForLeader = false
|
|
}
|
|
|
|
// Route message to handler
|
|
switch t := msg.Type; true {
|
|
// Handle validator intended messages first
|
|
case t == msg_pb.MessageType_ANNOUNCE && intendedForValidator:
|
|
consensus.onAnnounce(msg)
|
|
case t == msg_pb.MessageType_PREPARED && intendedForValidator:
|
|
consensus.onPrepared(fbftMsg)
|
|
case t == msg_pb.MessageType_COMMITTED && intendedForValidator:
|
|
consensus.onCommitted(fbftMsg)
|
|
|
|
// Handle leader intended messages now
|
|
case t == msg_pb.MessageType_PREPARE && intendedForLeader:
|
|
consensus.onPrepare(fbftMsg)
|
|
case t == msg_pb.MessageType_COMMIT && intendedForLeader:
|
|
consensus.onCommit(fbftMsg)
|
|
|
|
// Handle view change messages
|
|
case t == msg_pb.MessageType_VIEWCHANGE && canHandleViewChange:
|
|
consensus.onViewChange(fbftMsg)
|
|
case t == msg_pb.MessageType_NEWVIEW && canHandleViewChange:
|
|
consensus.onNewView(fbftMsg)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
func (consensus *Consensus) finalCommit() {
|
|
numCommits := consensus.Decider.SignersCount(quorum.Commit)
|
|
|
|
consensus.getLogger().Info().
|
|
Int64("NumCommits", numCommits).
|
|
Msg("[finalCommit] Finalizing Consensus")
|
|
beforeCatchupNum := consensus.BlockNum()
|
|
|
|
leaderPriKey, err := consensus.GetConsensusLeaderPrivateKey()
|
|
if err != nil {
|
|
consensus.getLogger().Error().Err(err).Msg("[finalCommit] leader not found")
|
|
return
|
|
}
|
|
// Construct committed message
|
|
network, err := consensus.construct(msg_pb.MessageType_COMMITTED, nil, []*bls.PrivateKeyWrapper{leaderPriKey})
|
|
if err != nil {
|
|
consensus.getLogger().Warn().Err(err).
|
|
Msg("[finalCommit] Unable to construct Committed message")
|
|
return
|
|
}
|
|
msgToSend, FBFTMsg :=
|
|
network.Bytes,
|
|
network.FBFTMsg
|
|
commitSigAndBitmap := FBFTMsg.Payload
|
|
consensus.FBFTLog.AddVerifiedMessage(FBFTMsg)
|
|
// find correct block content
|
|
curBlockHash := consensus.blockHash
|
|
block := consensus.FBFTLog.GetBlockByHash(curBlockHash)
|
|
if block == nil {
|
|
consensus.getLogger().Warn().
|
|
Str("blockHash", hex.EncodeToString(curBlockHash[:])).
|
|
Msg("[finalCommit] Cannot find block by hash")
|
|
return
|
|
}
|
|
|
|
if err := consensus.verifyLastCommitSig(commitSigAndBitmap, block); err != nil {
|
|
consensus.getLogger().Warn().Err(err).Msg("[finalCommit] failed verifying last commit sig")
|
|
return
|
|
}
|
|
consensus.getLogger().Info().Hex("new", commitSigAndBitmap).Msg("[finalCommit] Overriding commit signatures!!")
|
|
consensus.Blockchain().WriteCommitSig(block.NumberU64(), commitSigAndBitmap)
|
|
|
|
// Send committed message before block insertion.
|
|
// if leader successfully finalizes the block, send committed message to validators
|
|
// Note: leader already sent 67% commit in preCommit. The 100% commit won't be sent immediately
|
|
// to save network traffic. It will only be sent in retry if consensus doesn't move forward.
|
|
// Or if the leader is changed for next block, the 100% committed sig will be sent to the next leader immediately.
|
|
if !consensus.IsLeader() || block.IsLastBlockInEpoch() {
|
|
// send immediately
|
|
if err := consensus.msgSender.SendWithRetry(
|
|
block.NumberU64(),
|
|
msg_pb.MessageType_COMMITTED, []nodeconfig.GroupID{
|
|
nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID)),
|
|
},
|
|
p2p.ConstructMessage(msgToSend)); err != nil {
|
|
consensus.getLogger().Warn().Err(err).Msg("[finalCommit] Cannot send committed message")
|
|
} else {
|
|
consensus.getLogger().Info().
|
|
Hex("blockHash", curBlockHash[:]).
|
|
Uint64("blockNum", consensus.BlockNum()).
|
|
Msg("[finalCommit] Sent Committed Message")
|
|
}
|
|
consensus.getLogger().Info().Msg("[finalCommit] Start consensus timer")
|
|
consensus.consensusTimeout[timeoutConsensus].Start()
|
|
} else {
|
|
// delayed send
|
|
consensus.msgSender.DelayedSendWithRetry(
|
|
block.NumberU64(),
|
|
msg_pb.MessageType_COMMITTED, []nodeconfig.GroupID{
|
|
nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID)),
|
|
},
|
|
p2p.ConstructMessage(msgToSend))
|
|
consensus.getLogger().Info().
|
|
Hex("blockHash", curBlockHash[:]).
|
|
Uint64("blockNum", consensus.BlockNum()).
|
|
Hex("lastCommitSig", commitSigAndBitmap).
|
|
Msg("[finalCommit] Queued Committed Message")
|
|
}
|
|
|
|
block.SetCurrentCommitSig(commitSigAndBitmap)
|
|
err = consensus.commitBlock(block, FBFTMsg)
|
|
|
|
if err != nil || consensus.BlockNum()-beforeCatchupNum != 1 {
|
|
consensus.getLogger().Err(err).
|
|
Uint64("beforeCatchupBlockNum", beforeCatchupNum).
|
|
Msg("[finalCommit] Leader failed to commit the confirmed block")
|
|
}
|
|
|
|
// Dump new block into level db
|
|
// In current code, we add signatures in block in tryCatchup, the block dump to explorer does not contains signatures
|
|
// but since explorer doesn't need signatures, it should be fine
|
|
// in future, we will move signatures to next block
|
|
//explorer.GetStorageInstance(consensus.leader.IP, consensus.leader.Port, true).Dump(block, beforeCatchupNum)
|
|
|
|
if consensus.consensusTimeout[timeoutBootstrap].IsActive() {
|
|
consensus.consensusTimeout[timeoutBootstrap].Stop()
|
|
consensus.getLogger().Info().Msg("[finalCommit] stop bootstrap timer only once")
|
|
}
|
|
|
|
consensus.getLogger().Info().
|
|
Uint64("blockNum", block.NumberU64()).
|
|
Uint64("epochNum", block.Epoch().Uint64()).
|
|
Uint64("ViewId", block.Header().ViewID().Uint64()).
|
|
Str("blockHash", block.Hash().String()).
|
|
Int("numTxns", len(block.Transactions())).
|
|
Int("numStakingTxns", len(block.StakingTransactions())).
|
|
Msg("HOORAY!!!!!!! CONSENSUS REACHED!!!!!!!")
|
|
|
|
consensus.UpdateLeaderMetrics(float64(numCommits), float64(block.NumberU64()))
|
|
|
|
// If still the leader, send commit sig/bitmap to finish the new block proposal,
|
|
// else, the block proposal will timeout by itself.
|
|
if consensus.IsLeader() {
|
|
if block.IsLastBlockInEpoch() {
|
|
// No pipelining
|
|
go func() {
|
|
consensus.getLogger().Info().Msg("[finalCommit] sending block proposal signal")
|
|
consensus.ReadySignal <- SyncProposal
|
|
}()
|
|
} else {
|
|
// pipelining
|
|
go func() {
|
|
select {
|
|
case consensus.CommitSigChannel <- commitSigAndBitmap:
|
|
case <-time.After(CommitSigSenderTimeout):
|
|
utils.Logger().Error().Err(err).Msg("[finalCommit] channel not received after 6s for commitSigAndBitmap")
|
|
}
|
|
}()
|
|
}
|
|
}
|
|
}
|
|
|
|
// BlockCommitSigs returns the byte array of aggregated
|
|
// commit signature and bitmap signed on the block
|
|
func (consensus *Consensus) BlockCommitSigs(blockNum uint64) ([]byte, error) {
|
|
if consensus.BlockNum() <= 1 {
|
|
return nil, nil
|
|
}
|
|
lastCommits, err := consensus.Blockchain().ReadCommitSig(blockNum)
|
|
if err != nil ||
|
|
len(lastCommits) < bls.BLSSignatureSizeInBytes {
|
|
msgs := consensus.FBFTLog.GetMessagesByTypeSeq(
|
|
msg_pb.MessageType_COMMITTED, blockNum,
|
|
)
|
|
if len(msgs) != 1 {
|
|
consensus.getLogger().Error().
|
|
Int("numCommittedMsg", len(msgs)).
|
|
Msg("GetLastCommitSig failed with wrong number of committed message")
|
|
return nil, errors.Errorf(
|
|
"GetLastCommitSig failed with wrong number of committed message %d", len(msgs),
|
|
)
|
|
}
|
|
lastCommits = msgs[0].Payload
|
|
}
|
|
|
|
return lastCommits, nil
|
|
}
|
|
|
|
// Start waits for the next new block and run consensus
|
|
func (consensus *Consensus) Start(
|
|
blockChannel chan *types.Block, stopChan, stoppedChan, startChannel chan struct{},
|
|
) {
|
|
go func() {
|
|
toStart := make(chan struct{}, 1)
|
|
isInitialLeader := consensus.IsLeader()
|
|
if isInitialLeader {
|
|
consensus.getLogger().Info().Time("time", time.Now()).Msg("[ConsensusMainLoop] Waiting for consensus start")
|
|
// send a signal to indicate it's ready to run consensus
|
|
// this signal is consumed by node object to create a new block and in turn trigger a new consensus on it
|
|
go func() {
|
|
<-startChannel
|
|
toStart <- struct{}{}
|
|
consensus.getLogger().Info().Time("time", time.Now()).Msg("[ConsensusMainLoop] Send ReadySignal")
|
|
consensus.ReadySignal <- SyncProposal
|
|
}()
|
|
}
|
|
consensus.getLogger().Info().Time("time", time.Now()).Msg("[ConsensusMainLoop] Consensus started")
|
|
defer close(stoppedChan)
|
|
ticker := time.NewTicker(250 * time.Millisecond)
|
|
defer ticker.Stop()
|
|
consensus.consensusTimeout[timeoutBootstrap].Start()
|
|
consensus.getLogger().Info().Msg("[ConsensusMainLoop] Start bootstrap timeout (only once)")
|
|
|
|
// Set up next block due time.
|
|
consensus.NextBlockDue = time.Now().Add(consensus.BlockPeriod)
|
|
start := false
|
|
for {
|
|
select {
|
|
case <-toStart:
|
|
start = true
|
|
case <-ticker.C:
|
|
if !start && isInitialLeader {
|
|
continue
|
|
}
|
|
for k, v := range consensus.consensusTimeout {
|
|
// stop timer in listening mode
|
|
if consensus.current.Mode() == Listening {
|
|
v.Stop()
|
|
continue
|
|
}
|
|
|
|
if consensus.current.Mode() == Syncing {
|
|
// never stop bootstrap timer here in syncing mode as it only starts once
|
|
// if it is stopped, bootstrap will be stopped and nodes
|
|
// can't start view change or join consensus
|
|
// the bootstrap timer will be stopped once consensus is reached or view change
|
|
// is succeeded
|
|
if k != timeoutBootstrap {
|
|
consensus.getLogger().Debug().
|
|
Str("k", k.String()).
|
|
Str("Mode", consensus.current.Mode().String()).
|
|
Msg("[ConsensusMainLoop] consensusTimeout stopped!!!")
|
|
v.Stop()
|
|
continue
|
|
}
|
|
}
|
|
if !v.CheckExpire() {
|
|
continue
|
|
}
|
|
if k != timeoutViewChange {
|
|
consensus.getLogger().Warn().Msg("[ConsensusMainLoop] Ops Consensus Timeout!!!")
|
|
consensus.startViewChange()
|
|
break
|
|
} else {
|
|
consensus.getLogger().Warn().Msg("[ConsensusMainLoop] Ops View Change Timeout!!!")
|
|
consensus.startViewChange()
|
|
break
|
|
}
|
|
}
|
|
|
|
// TODO: Refactor this piece of code to consensus/downloader.go after DNS legacy sync is removed
|
|
case <-consensus.syncReadyChan:
|
|
consensus.getLogger().Info().Msg("[ConsensusMainLoop] syncReadyChan")
|
|
consensus.mutex.Lock()
|
|
if consensus.BlockNum() < consensus.Blockchain().CurrentHeader().Number().Uint64()+1 {
|
|
consensus.SetBlockNum(consensus.Blockchain().CurrentHeader().Number().Uint64() + 1)
|
|
consensus.SetViewIDs(consensus.Blockchain().CurrentHeader().ViewID().Uint64() + 1)
|
|
mode := consensus.UpdateConsensusInformation()
|
|
consensus.current.SetMode(mode)
|
|
consensus.getLogger().Info().Msg("[syncReadyChan] Start consensus timer")
|
|
consensus.consensusTimeout[timeoutConsensus].Start()
|
|
consensus.getLogger().Info().Str("Mode", mode.String()).Msg("Node is IN SYNC")
|
|
consensusSyncCounterVec.With(prometheus.Labels{"consensus": "in_sync"}).Inc()
|
|
} else if consensus.Mode() == Syncing {
|
|
// Corner case where sync is triggered before `onCommitted` and there is a race
|
|
// for block insertion between consensus and downloader.
|
|
mode := consensus.UpdateConsensusInformation()
|
|
consensus.SetMode(mode)
|
|
consensus.getLogger().Info().Msg("[syncReadyChan] Start consensus timer")
|
|
consensus.consensusTimeout[timeoutConsensus].Start()
|
|
consensusSyncCounterVec.With(prometheus.Labels{"consensus": "in_sync"}).Inc()
|
|
}
|
|
consensus.mutex.Unlock()
|
|
|
|
// TODO: Refactor this piece of code to consensus/downloader.go after DNS legacy sync is removed
|
|
case <-consensus.syncNotReadyChan:
|
|
consensus.getLogger().Info().Msg("[ConsensusMainLoop] syncNotReadyChan")
|
|
consensus.SetBlockNum(consensus.Blockchain().CurrentHeader().Number().Uint64() + 1)
|
|
consensus.current.SetMode(Syncing)
|
|
consensus.getLogger().Info().Msg("[ConsensusMainLoop] Node is OUT OF SYNC")
|
|
consensusSyncCounterVec.With(prometheus.Labels{"consensus": "out_of_sync"}).Inc()
|
|
|
|
case newBlock := <-blockChannel:
|
|
//consensus.ReshardingNextLeader(newBlock)
|
|
consensus.getLogger().Info().
|
|
Uint64("MsgBlockNum", newBlock.NumberU64()).
|
|
Msg("[ConsensusMainLoop] Received Proposed New Block!")
|
|
|
|
if newBlock.NumberU64() < consensus.BlockNum() {
|
|
consensus.getLogger().Warn().Uint64("newBlockNum", newBlock.NumberU64()).
|
|
Msg("[ConsensusMainLoop] received old block, abort")
|
|
continue
|
|
}
|
|
// Sleep to wait for the full block time
|
|
consensus.getLogger().Info().Msg("[ConsensusMainLoop] Waiting for Block Time")
|
|
|
|
<-time.After(time.Until(consensus.NextBlockDue))
|
|
consensus.StartFinalityCount()
|
|
|
|
// Update time due for next block
|
|
consensus.NextBlockDue = time.Now().Add(consensus.BlockPeriod)
|
|
|
|
startTime = time.Now()
|
|
consensus.msgSender.Reset(newBlock.NumberU64())
|
|
|
|
consensus.getLogger().Info().
|
|
Int("numTxs", len(newBlock.Transactions())).
|
|
Int("numStakingTxs", len(newBlock.StakingTransactions())).
|
|
Time("startTime", startTime).
|
|
Int64("publicKeys", consensus.Decider.ParticipantsCount()).
|
|
Msg("[ConsensusMainLoop] STARTING CONSENSUS")
|
|
consensus.announce(newBlock)
|
|
case <-stopChan:
|
|
consensus.getLogger().Info().Msg("[ConsensusMainLoop] stopChan")
|
|
return
|
|
}
|
|
}
|
|
}()
|
|
|
|
if consensus.dHelper != nil {
|
|
consensus.dHelper.start()
|
|
}
|
|
}
|
|
|
|
// Close closes the consensus. If current is in normal commit phase, wait until the commit
|
|
// phase end.
|
|
func (consensus *Consensus) Close() error {
|
|
if consensus.dHelper != nil {
|
|
consensus.dHelper.close()
|
|
}
|
|
consensus.waitForCommit()
|
|
return nil
|
|
}
|
|
|
|
// waitForCommit wait extra 2 seconds for commit phase to finish
|
|
func (consensus *Consensus) waitForCommit() {
|
|
if consensus.Mode() != Normal || consensus.phase.Get() != FBFTCommit {
|
|
return
|
|
}
|
|
// We only need to wait consensus is in normal commit phase
|
|
utils.Logger().Warn().Str("phase", consensus.phase.String()).Msg("[shutdown] commit phase has to wait")
|
|
|
|
maxWait := time.Now().Add(2 * consensus.BlockPeriod)
|
|
for time.Now().Before(maxWait) && consensus.GetConsensusPhase() == "Commit" {
|
|
utils.Logger().Warn().Msg("[shutdown] wait for consensus finished")
|
|
time.Sleep(time.Millisecond * 100)
|
|
}
|
|
}
|
|
|
|
// LastMileBlockIter is the iterator to iterate over the last mile blocks in consensus cache.
|
|
// All blocks returned are guaranteed to pass the verification.
|
|
type LastMileBlockIter struct {
|
|
blockCandidates []*types.Block
|
|
fbftLog *FBFTLog
|
|
verify func(*types.Block) error
|
|
curIndex int
|
|
logger *zerolog.Logger
|
|
}
|
|
|
|
// GetLastMileBlockIter get the iterator of the last mile blocks starting from number bnStart
|
|
func (consensus *Consensus) GetLastMileBlockIter(bnStart uint64) (*LastMileBlockIter, error) {
|
|
consensus.mutex.Lock()
|
|
defer consensus.mutex.Unlock()
|
|
|
|
if consensus.BlockVerifier == nil {
|
|
return nil, errors.New("consensus haven't initialized yet")
|
|
}
|
|
blocks, _, err := consensus.getLastMileBlocksAndMsg(bnStart)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
return &LastMileBlockIter{
|
|
blockCandidates: blocks,
|
|
fbftLog: consensus.FBFTLog,
|
|
verify: consensus.BlockVerifier,
|
|
curIndex: 0,
|
|
logger: consensus.getLogger(),
|
|
}, nil
|
|
}
|
|
|
|
// Next iterate to the next last mile block
|
|
func (iter *LastMileBlockIter) Next() *types.Block {
|
|
if iter.curIndex >= len(iter.blockCandidates) {
|
|
return nil
|
|
}
|
|
block := iter.blockCandidates[iter.curIndex]
|
|
iter.curIndex++
|
|
|
|
if !iter.fbftLog.IsBlockVerified(block.Hash()) {
|
|
if err := iter.verify(block); err != nil {
|
|
iter.logger.Debug().Err(err).Msg("block verification failed in consensus last mile block")
|
|
return nil
|
|
}
|
|
iter.fbftLog.MarkBlockVerified(block)
|
|
}
|
|
return block
|
|
}
|
|
|
|
func (consensus *Consensus) getLastMileBlocksAndMsg(bnStart uint64) ([]*types.Block, []*FBFTMessage, error) {
|
|
var (
|
|
blocks []*types.Block
|
|
msgs []*FBFTMessage
|
|
)
|
|
for blockNum := bnStart; ; blockNum++ {
|
|
blk, msg, err := consensus.FBFTLog.GetCommittedBlockAndMsgsFromNumber(blockNum, consensus.getLogger())
|
|
if err != nil {
|
|
if err == errFBFTLogNotFound {
|
|
break
|
|
}
|
|
return nil, nil, err
|
|
}
|
|
blocks = append(blocks, blk)
|
|
msgs = append(msgs, msg)
|
|
}
|
|
return blocks, msgs, nil
|
|
}
|
|
|
|
// preCommitAndPropose commit the current block with 67% commit signatures and start
|
|
// proposing new block which will wait on the full commit signatures to finish
|
|
func (consensus *Consensus) preCommitAndPropose(blk *types.Block) error {
|
|
//fmt.Println("preCommitAndPropose", utils.GetPort(), blk.NumberU64())
|
|
if blk == nil {
|
|
return errors.New("block to pre-commit is nil")
|
|
}
|
|
|
|
leaderPriKey, err := consensus.GetConsensusLeaderPrivateKey()
|
|
if err != nil {
|
|
consensus.getLogger().Error().Err(err).Msg("[preCommitAndPropose] leader not found")
|
|
return err
|
|
}
|
|
|
|
// Construct committed message
|
|
network, err := consensus.construct(msg_pb.MessageType_COMMITTED, nil, []*bls.PrivateKeyWrapper{leaderPriKey})
|
|
if err != nil {
|
|
consensus.getLogger().Warn().Err(err).
|
|
Msg("[preCommitAndPropose] Unable to construct Committed message")
|
|
return err
|
|
}
|
|
|
|
msgToSend, FBFTMsg :=
|
|
network.Bytes,
|
|
network.FBFTMsg
|
|
bareMinimumCommit := FBFTMsg.Payload
|
|
consensus.FBFTLog.AddVerifiedMessage(FBFTMsg)
|
|
|
|
if err := consensus.verifyLastCommitSig(bareMinimumCommit, blk); err != nil {
|
|
return errors.Wrap(err, "[preCommitAndPropose] failed verifying last commit sig")
|
|
}
|
|
|
|
go func() {
|
|
blk.SetCurrentCommitSig(bareMinimumCommit)
|
|
|
|
// Send committed message to validators since 2/3 commit is already collected
|
|
if err := consensus.msgSender.SendWithRetry(
|
|
blk.NumberU64(),
|
|
msg_pb.MessageType_COMMITTED, []nodeconfig.GroupID{
|
|
nodeconfig.NewGroupIDByShardID(nodeconfig.ShardID(consensus.ShardID)),
|
|
},
|
|
p2p.ConstructMessage(msgToSend)); err != nil {
|
|
consensus.getLogger().Warn().Err(err).Msg("[preCommitAndPropose] Cannot send committed message")
|
|
} else {
|
|
consensus.getLogger().Info().
|
|
Str("blockHash", blk.Hash().Hex()).
|
|
Uint64("blockNum", consensus.BlockNum()).
|
|
Hex("lastCommitSig", bareMinimumCommit).
|
|
Msg("[preCommitAndPropose] Sent Committed Message")
|
|
}
|
|
|
|
if _, err := consensus.Blockchain().InsertChain([]*types.Block{blk}, !consensus.FBFTLog.IsBlockVerified(blk.Hash())); err != nil {
|
|
consensus.getLogger().Error().Err(err).Msg("[preCommitAndPropose] Failed to add block to chain")
|
|
return
|
|
}
|
|
|
|
consensus.getLogger().Info().Msg("[preCommitAndPropose] Start consensus timer")
|
|
consensus.consensusTimeout[timeoutConsensus].Start()
|
|
|
|
// Send signal to Node to propose the new block for consensus
|
|
consensus.getLogger().Info().Msg("[preCommitAndPropose] sending block proposal signal")
|
|
|
|
consensus.ReadySignal <- AsyncProposal
|
|
}()
|
|
|
|
return nil
|
|
}
|
|
|
|
func (consensus *Consensus) verifyLastCommitSig(lastCommitSig []byte, blk *types.Block) error {
|
|
if len(lastCommitSig) < bls.BLSSignatureSizeInBytes {
|
|
return errors.New("lastCommitSig not have enough length")
|
|
}
|
|
|
|
aggSigBytes := lastCommitSig[0:bls.BLSSignatureSizeInBytes]
|
|
|
|
aggSig := bls2.Sign{}
|
|
err := aggSig.Deserialize(aggSigBytes)
|
|
|
|
if err != nil {
|
|
return errors.New("unable to deserialize multi-signature from payload")
|
|
}
|
|
aggPubKey := consensus.commitBitmap.AggregatePublic
|
|
|
|
commitPayload := signature.ConstructCommitPayload(consensus.Blockchain(),
|
|
blk.Epoch(), blk.Hash(), blk.NumberU64(), blk.Header().ViewID().Uint64())
|
|
|
|
if !aggSig.VerifyHash(aggPubKey, commitPayload) {
|
|
return errors.New("Failed to verify the multi signature for last commit sig")
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// tryCatchup add the last mile block in PBFT log memory cache to blockchain.
|
|
func (consensus *Consensus) tryCatchup() error {
|
|
// TODO: change this to a more systematic symbol
|
|
if consensus.BlockVerifier == nil {
|
|
return errors.New("consensus haven't finished initialization")
|
|
}
|
|
initBN := consensus.BlockNum()
|
|
defer consensus.postCatchup(initBN)
|
|
|
|
blks, msgs, err := consensus.getLastMileBlocksAndMsg(initBN)
|
|
if err != nil {
|
|
return errors.Wrapf(err, "[TryCatchup] Failed to get last mile blocks: %v", err)
|
|
}
|
|
for i := range blks {
|
|
blk, msg := blks[i], msgs[i]
|
|
if blk == nil {
|
|
return nil
|
|
}
|
|
blk.SetCurrentCommitSig(msg.Payload)
|
|
|
|
if err := consensus.VerifyBlock(blk); err != nil {
|
|
consensus.getLogger().Err(err).Msg("[TryCatchup] failed block verifier")
|
|
return err
|
|
}
|
|
consensus.getLogger().Info().Msg("[TryCatchup] Adding block to chain")
|
|
if err := consensus.commitBlock(blk, msgs[i]); err != nil {
|
|
consensus.getLogger().Error().Err(err).Msg("[TryCatchup] Failed to add block to chain")
|
|
return err
|
|
}
|
|
//fmt.Println("tryCatchup ", utils.GetPort(), blk.NumberU64())
|
|
select {
|
|
// TODO: Remove this when removing dns sync and stream sync is fully up
|
|
case consensus.VerifiedNewBlock <- blk:
|
|
default:
|
|
consensus.getLogger().Info().
|
|
Str("blockHash", blk.Hash().String()).
|
|
Msg("[TryCatchup] consensus verified block send to chan failed")
|
|
continue
|
|
}
|
|
}
|
|
return nil
|
|
}
|
|
|
|
func (consensus *Consensus) commitBlock(blk *types.Block, committedMsg *FBFTMessage) error {
|
|
// this function evaluates for all, leader and validators.
|
|
|
|
if consensus.Blockchain().CurrentBlock().NumberU64() < blk.NumberU64() {
|
|
if _, err := consensus.Blockchain().InsertChain([]*types.Block{blk}, !consensus.FBFTLog.IsBlockVerified(blk.Hash())); err != nil {
|
|
consensus.getLogger().Error().Err(err).Msg("[commitBlock] Failed to add block to chain")
|
|
return err
|
|
}
|
|
}
|
|
|
|
if !committedMsg.HasSingleSender() {
|
|
consensus.getLogger().Error().Msg("[TryCatchup] Leader message can not have multiple sender keys")
|
|
return errIncorrectSender
|
|
}
|
|
|
|
consensus.FinishFinalityCount()
|
|
consensus.PostConsensusJob(blk)
|
|
consensus.SetupForNewConsensus(blk, committedMsg)
|
|
utils.Logger().Info().Uint64("blockNum", blk.NumberU64()).
|
|
Str("hash", blk.Header().Hash().Hex()).
|
|
Msg("Added New Block to Blockchain!!!")
|
|
return nil
|
|
}
|
|
|
|
// SetupForNewConsensus sets the state for new consensus
|
|
func (consensus *Consensus) SetupForNewConsensus(blk *types.Block, committedMsg *FBFTMessage) {
|
|
atomic.StoreUint64(&consensus.blockNum, blk.NumberU64()+1)
|
|
curBlockViewID := consensus.SetCurBlockViewID(committedMsg.ViewID + 1)
|
|
prev := consensus.GetLeaderPubKey()
|
|
idx := consensus.SetLeaderIndex(func(i int) int {
|
|
if curBlockViewID%3 == 0 {
|
|
return i + 1
|
|
}
|
|
return i
|
|
})
|
|
pps := consensus.Decider.Participants()
|
|
consensus.pubKeyLock.Lock()
|
|
consensus.LeaderPubKey = &pps[idx%len(pps)]
|
|
fmt.Printf("SetupForNewConsensus :%d idx: %d future v%d new: %s prev: %s %q\n", utils.GetPort(), idx, curBlockViewID, consensus.LeaderPubKey.Bytes.Hex(), prev.Bytes.Hex(), consensus.isLeader())
|
|
consensus.pubKeyLock.Unlock()
|
|
if consensus.IsLeader() && !consensus.GetLeaderPubKey().Object.IsEqual(prev.Object) {
|
|
// leader changed
|
|
go func() {
|
|
fmt.Printf("ReadySignal :%d for leader %s\n", utils.GetPort(), consensus.GetLeaderPubKey().Bytes.Hex())
|
|
defer fmt.Printf("Defer ReadySignal :%d for leader %s\n", utils.GetPort(), consensus.GetLeaderPubKey().Bytes.Hex())
|
|
consensus.ReadySignal <- SyncProposal
|
|
|
|
}()
|
|
|
|
}
|
|
// Update consensus keys at last so the change of leader status doesn't mess up normal flow
|
|
if blk.IsLastBlockInEpoch() {
|
|
consensus.SetMode(consensus.UpdateConsensusInformation())
|
|
}
|
|
consensus.FBFTLog.PruneCacheBeforeBlock(blk.NumberU64())
|
|
consensus.ResetState()
|
|
}
|
|
|
|
func (consensus *Consensus) postCatchup(initBN uint64) {
|
|
if initBN < consensus.BlockNum() {
|
|
consensus.getLogger().Info().
|
|
Uint64("From", initBN).
|
|
Uint64("To", consensus.BlockNum()).
|
|
Msg("[TryCatchup] Caught up!")
|
|
consensus.switchPhase("TryCatchup", FBFTAnnounce)
|
|
}
|
|
// catch up and skip from view change trap
|
|
if initBN < consensus.BlockNum() && consensus.IsViewChangingMode() {
|
|
consensus.current.SetMode(Normal)
|
|
consensus.consensusTimeout[timeoutViewChange].Stop()
|
|
}
|
|
}
|
|
|
|
// GenerateVrfAndProof generates new VRF/Proof from hash of previous block
|
|
func (consensus *Consensus) GenerateVrfAndProof(newHeader *block.Header) error {
|
|
key, err := consensus.GetConsensusLeaderPrivateKey()
|
|
if err != nil {
|
|
return errors.New("[GenerateVrfAndProof] no leader private key provided")
|
|
}
|
|
sk := vrf_bls.NewVRFSigner(key.Pri)
|
|
previousHeader := consensus.Blockchain().GetHeaderByNumber(
|
|
newHeader.Number().Uint64() - 1,
|
|
)
|
|
if previousHeader == nil {
|
|
return errors.New("[GenerateVrfAndProof] no parent header found")
|
|
}
|
|
|
|
previousHash := previousHeader.Hash()
|
|
vrf, proof := sk.Evaluate(previousHash[:])
|
|
if proof == nil {
|
|
return errors.New("[GenerateVrfAndProof] failed to generate vrf")
|
|
}
|
|
|
|
newHeader.SetVrf(append(vrf[:], proof...))
|
|
|
|
consensus.getLogger().Info().
|
|
Uint64("BlockNum", newHeader.Number().Uint64()).
|
|
Uint64("Epoch", newHeader.Epoch().Uint64()).
|
|
Hex("VRF+Proof", newHeader.Vrf()).
|
|
Msg("[GenerateVrfAndProof] Leader generated a VRF")
|
|
|
|
return nil
|
|
}
|
|
|
|
// GenerateVdfAndProof generates new VDF/Proof from VRFs in the current epoch
|
|
func (consensus *Consensus) GenerateVdfAndProof(newBlock *types.Block, vrfBlockNumbers []uint64) {
|
|
//derive VDF seed from VRFs generated in the current epoch
|
|
seed := [32]byte{}
|
|
for i := 0; i < consensus.VdfSeedSize(); i++ {
|
|
previousVrf := consensus.Blockchain().GetVrfByNumber(vrfBlockNumbers[i])
|
|
for j := 0; j < len(seed); j++ {
|
|
seed[j] = seed[j] ^ previousVrf[j]
|
|
}
|
|
}
|
|
|
|
consensus.getLogger().Info().
|
|
Uint64("MsgBlockNum", newBlock.NumberU64()).
|
|
Uint64("Epoch", newBlock.Header().Epoch().Uint64()).
|
|
Int("Num of VRF", len(vrfBlockNumbers)).
|
|
Msg("[ConsensusMainLoop] VDF computation started")
|
|
|
|
// TODO ek – limit concurrency
|
|
go func() {
|
|
vdf := vdf_go.New(shard.Schedule.VdfDifficulty(), seed)
|
|
outputChannel := vdf.GetOutputChannel()
|
|
start := time.Now()
|
|
vdf.Execute()
|
|
duration := time.Since(start)
|
|
consensus.getLogger().Info().
|
|
Dur("duration", duration).
|
|
Msg("[ConsensusMainLoop] VDF computation finished")
|
|
output := <-outputChannel
|
|
|
|
// The first 516 bytes are the VDF+proof and the last 32 bytes are XORed VRF as seed
|
|
rndBytes := [548]byte{}
|
|
copy(rndBytes[:516], output[:])
|
|
copy(rndBytes[516:], seed[:])
|
|
consensus.RndChannel <- rndBytes
|
|
}()
|
|
}
|
|
|
|
// ValidateVdfAndProof validates the VDF/proof in the current epoch
|
|
func (consensus *Consensus) ValidateVdfAndProof(headerObj *block.Header) bool {
|
|
vrfBlockNumbers, err := consensus.Blockchain().ReadEpochVrfBlockNums(headerObj.Epoch())
|
|
if err != nil {
|
|
consensus.getLogger().Error().Err(err).
|
|
Str("MsgBlockNum", headerObj.Number().String()).
|
|
Msg("[OnAnnounce] failed to read VRF block numbers for VDF computation")
|
|
}
|
|
|
|
//extra check to make sure there's no index out of range error
|
|
//it can happen if epoch is messed up, i.e. VDF ouput is generated in the next epoch
|
|
if consensus.VdfSeedSize() > len(vrfBlockNumbers) {
|
|
return false
|
|
}
|
|
|
|
seed := [32]byte{}
|
|
for i := 0; i < consensus.VdfSeedSize(); i++ {
|
|
previousVrf := consensus.Blockchain().GetVrfByNumber(vrfBlockNumbers[i])
|
|
for j := 0; j < len(seed); j++ {
|
|
seed[j] = seed[j] ^ previousVrf[j]
|
|
}
|
|
}
|
|
|
|
vdfObject := vdf_go.New(shard.Schedule.VdfDifficulty(), seed)
|
|
vdfOutput := [516]byte{}
|
|
copy(vdfOutput[:], headerObj.Vdf())
|
|
if vdfObject.Verify(vdfOutput) {
|
|
consensus.getLogger().Info().
|
|
Str("MsgBlockNum", headerObj.Number().String()).
|
|
Int("Num of VRF", consensus.VdfSeedSize()).
|
|
Msg("[OnAnnounce] validated a new VDF")
|
|
|
|
} else {
|
|
consensus.getLogger().Warn().
|
|
Str("MsgBlockNum", headerObj.Number().String()).
|
|
Uint64("Epoch", headerObj.Epoch().Uint64()).
|
|
Int("Num of VRF", consensus.VdfSeedSize()).
|
|
Msg("[OnAnnounce] VDF proof is not valid")
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}
|
|
|