mirror of
https://github.com/tendermint/tendermint.git
synced 2026-01-06 05:25:35 +00:00
config: backport the rename of fastsync to blocksync (#9259)
This is largely a cherry pick of #6755 with some additional fixups added where detected. This change moves the blockchain package to a package called blocksync. Additionally, it renames the relevant uses of the term `fastsync` to `blocksync`. closes: #9227 #### PR checklist - [ ] Tests written/updated, or no tests needed - [x] `CHANGELOG_PENDING.md` updated, or no changelog entry needed - [x] Updated relevant documentation (`docs/`) and code comments, or no documentation updates needed
This commit is contained in:
William Banfield
GitHub
436
blocksync/reactor.go
Normal file
436
blocksync/reactor.go
Normal file
@@ -0,0 +1,436 @@
|
||||
package blocksync
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
"reflect"
|
||||
"time"
|
||||
|
||||
"github.com/tendermint/tendermint/libs/log"
|
||||
"github.com/tendermint/tendermint/p2p"
|
||||
bcproto "github.com/tendermint/tendermint/proto/tendermint/blocksync"
|
||||
sm "github.com/tendermint/tendermint/state"
|
||||
"github.com/tendermint/tendermint/store"
|
||||
"github.com/tendermint/tendermint/types"
|
||||
)
|
||||
|
||||
const (
|
||||
// BlocksyncChannel is a channel for blocks and status updates (`BlockStore` height)
|
||||
BlocksyncChannel = byte(0x40)
|
||||
|
||||
trySyncIntervalMS = 10
|
||||
|
||||
// stop syncing when last block's time is
|
||||
// within this much of the system time.
|
||||
// stopSyncingDurationMinutes = 10
|
||||
|
||||
// ask for best height every 10s
|
||||
statusUpdateIntervalSeconds = 10
|
||||
// check if we should switch to consensus reactor
|
||||
switchToConsensusIntervalSeconds = 1
|
||||
)
|
||||
|
||||
type consensusReactor interface {
|
||||
// for when we switch from blockchain reactor and block sync to
|
||||
// the consensus machine
|
||||
SwitchToConsensus(state sm.State, skipWAL bool)
|
||||
}
|
||||
|
||||
type peerError struct {
|
||||
err error
|
||||
peerID p2p.ID
|
||||
}
|
||||
|
||||
func (e peerError) Error() string {
|
||||
return fmt.Sprintf("error with peer %v: %s", e.peerID, e.err.Error())
|
||||
}
|
||||
|
||||
// Reactor handles long-term catchup syncing.
|
||||
type Reactor struct {
|
||||
p2p.BaseReactor
|
||||
|
||||
// immutable
|
||||
initialState sm.State
|
||||
|
||||
blockExec *sm.BlockExecutor
|
||||
store *store.BlockStore
|
||||
pool *BlockPool
|
||||
blockSync bool
|
||||
|
||||
requestsCh <-chan BlockRequest
|
||||
errorsCh <-chan peerError
|
||||
}
|
||||
|
||||
// NewReactor returns new reactor instance.
|
||||
func NewReactor(state sm.State, blockExec *sm.BlockExecutor, store *store.BlockStore,
|
||||
blockSync bool) *Reactor {
|
||||
|
||||
if state.LastBlockHeight != store.Height() {
|
||||
panic(fmt.Sprintf("state (%v) and store (%v) height mismatch", state.LastBlockHeight,
|
||||
store.Height()))
|
||||
}
|
||||
|
||||
requestsCh := make(chan BlockRequest, maxTotalRequesters)
|
||||
|
||||
const capacity = 1000 // must be bigger than peers count
|
||||
errorsCh := make(chan peerError, capacity) // so we don't block in #Receive#pool.AddBlock
|
||||
|
||||
startHeight := store.Height() + 1
|
||||
if startHeight == 1 {
|
||||
startHeight = state.InitialHeight
|
||||
}
|
||||
pool := NewBlockPool(startHeight, requestsCh, errorsCh)
|
||||
|
||||
bcR := &Reactor{
|
||||
initialState: state,
|
||||
blockExec: blockExec,
|
||||
store: store,
|
||||
pool: pool,
|
||||
blockSync: blockSync,
|
||||
requestsCh: requestsCh,
|
||||
errorsCh: errorsCh,
|
||||
}
|
||||
bcR.BaseReactor = *p2p.NewBaseReactor("Reactor", bcR)
|
||||
return bcR
|
||||
}
|
||||
|
||||
// SetLogger implements service.Service by setting the logger on reactor and pool.
|
||||
func (bcR *Reactor) SetLogger(l log.Logger) {
|
||||
bcR.BaseService.Logger = l
|
||||
bcR.pool.Logger = l
|
||||
}
|
||||
|
||||
// OnStart implements service.Service.
|
||||
func (bcR *Reactor) OnStart() error {
|
||||
if bcR.blockSync {
|
||||
err := bcR.pool.Start()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
go bcR.poolRoutine(false)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// SwitchToBlockSync is called by the state sync reactor when switching to block sync.
|
||||
func (bcR *Reactor) SwitchToBlockSync(state sm.State) error {
|
||||
bcR.blockSync = true
|
||||
bcR.initialState = state
|
||||
|
||||
bcR.pool.height = state.LastBlockHeight + 1
|
||||
err := bcR.pool.Start()
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
go bcR.poolRoutine(true)
|
||||
return nil
|
||||
}
|
||||
|
||||
// OnStop implements service.Service.
|
||||
func (bcR *Reactor) OnStop() {
|
||||
if bcR.blockSync {
|
||||
if err := bcR.pool.Stop(); err != nil {
|
||||
bcR.Logger.Error("Error stopping pool", "err", err)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// GetChannels implements Reactor
|
||||
func (bcR *Reactor) GetChannels() []*p2p.ChannelDescriptor {
|
||||
return []*p2p.ChannelDescriptor{
|
||||
{
|
||||
ID: BlocksyncChannel,
|
||||
Priority: 5,
|
||||
SendQueueCapacity: 1000,
|
||||
RecvBufferCapacity: 50 * 4096,
|
||||
RecvMessageCapacity: MaxMsgSize,
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
// AddPeer implements Reactor by sending our state to peer.
|
||||
func (bcR *Reactor) AddPeer(peer p2p.Peer) {
|
||||
msgBytes, err := EncodeMsg(&bcproto.StatusResponse{
|
||||
Base: bcR.store.Base(),
|
||||
Height: bcR.store.Height()})
|
||||
if err != nil {
|
||||
bcR.Logger.Error("could not convert msg to protobuf", "err", err)
|
||||
return
|
||||
}
|
||||
|
||||
peer.Send(BlocksyncChannel, msgBytes)
|
||||
// it's OK if send fails. will try later in poolRoutine
|
||||
|
||||
// peer is added to the pool once we receive the first
|
||||
// bcStatusResponseMessage from the peer and call pool.SetPeerRange
|
||||
}
|
||||
|
||||
// RemovePeer implements Reactor by removing peer from the pool.
|
||||
func (bcR *Reactor) RemovePeer(peer p2p.Peer, reason interface{}) {
|
||||
bcR.pool.RemovePeer(peer.ID())
|
||||
}
|
||||
|
||||
// respondToPeer loads a block and sends it to the requesting peer,
|
||||
// if we have it. Otherwise, we'll respond saying we don't have it.
|
||||
func (bcR *Reactor) respondToPeer(msg *bcproto.BlockRequest,
|
||||
src p2p.Peer) (queued bool) {
|
||||
|
||||
block := bcR.store.LoadBlock(msg.Height)
|
||||
if block != nil {
|
||||
bl, err := block.ToProto()
|
||||
if err != nil {
|
||||
bcR.Logger.Error("could not convert msg to protobuf", "err", err)
|
||||
return false
|
||||
}
|
||||
|
||||
msgBytes, err := EncodeMsg(&bcproto.BlockResponse{Block: bl})
|
||||
if err != nil {
|
||||
bcR.Logger.Error("could not marshal msg", "err", err)
|
||||
return false
|
||||
}
|
||||
|
||||
return src.TrySend(BlocksyncChannel, msgBytes)
|
||||
}
|
||||
|
||||
bcR.Logger.Info("Peer asking for a block we don't have", "src", src, "height", msg.Height)
|
||||
|
||||
msgBytes, err := EncodeMsg(&bcproto.NoBlockResponse{Height: msg.Height})
|
||||
if err != nil {
|
||||
bcR.Logger.Error("could not convert msg to protobuf", "err", err)
|
||||
return false
|
||||
}
|
||||
|
||||
return src.TrySend(BlocksyncChannel, msgBytes)
|
||||
}
|
||||
|
||||
// Receive implements Reactor by handling 4 types of messages (look below).
|
||||
func (bcR *Reactor) Receive(chID byte, src p2p.Peer, msgBytes []byte) {
|
||||
msg, err := DecodeMsg(msgBytes)
|
||||
if err != nil {
|
||||
bcR.Logger.Error("Error decoding message", "src", src, "chId", chID, "err", err)
|
||||
bcR.Switch.StopPeerForError(src, err)
|
||||
return
|
||||
}
|
||||
|
||||
if err = ValidateMsg(msg); err != nil {
|
||||
bcR.Logger.Error("Peer sent us invalid msg", "peer", src, "msg", msg, "err", err)
|
||||
bcR.Switch.StopPeerForError(src, err)
|
||||
return
|
||||
}
|
||||
|
||||
bcR.Logger.Debug("Receive", "src", src, "chID", chID, "msg", msg)
|
||||
|
||||
switch msg := msg.(type) {
|
||||
case *bcproto.BlockRequest:
|
||||
bcR.respondToPeer(msg, src)
|
||||
case *bcproto.BlockResponse:
|
||||
bi, err := types.BlockFromProto(msg.Block)
|
||||
if err != nil {
|
||||
bcR.Logger.Error("Block content is invalid", "err", err)
|
||||
return
|
||||
}
|
||||
bcR.pool.AddBlock(src.ID(), bi, len(msgBytes))
|
||||
case *bcproto.StatusRequest:
|
||||
// Send peer our state.
|
||||
msgBytes, err := EncodeMsg(&bcproto.StatusResponse{
|
||||
Height: bcR.store.Height(),
|
||||
Base: bcR.store.Base(),
|
||||
})
|
||||
if err != nil {
|
||||
bcR.Logger.Error("could not convert msg to protobut", "err", err)
|
||||
return
|
||||
}
|
||||
src.TrySend(BlocksyncChannel, msgBytes)
|
||||
case *bcproto.StatusResponse:
|
||||
// Got a peer status. Unverified.
|
||||
bcR.pool.SetPeerRange(src.ID(), msg.Base, msg.Height)
|
||||
case *bcproto.NoBlockResponse:
|
||||
bcR.Logger.Debug("Peer does not have requested block", "peer", src, "height", msg.Height)
|
||||
default:
|
||||
bcR.Logger.Error(fmt.Sprintf("Unknown message type %v", reflect.TypeOf(msg)))
|
||||
}
|
||||
}
|
||||
|
||||
// Handle messages from the poolReactor telling the reactor what to do.
|
||||
// NOTE: Don't sleep in the FOR_LOOP or otherwise slow it down!
|
||||
func (bcR *Reactor) poolRoutine(stateSynced bool) {
|
||||
|
||||
trySyncTicker := time.NewTicker(trySyncIntervalMS * time.Millisecond)
|
||||
defer trySyncTicker.Stop()
|
||||
|
||||
statusUpdateTicker := time.NewTicker(statusUpdateIntervalSeconds * time.Second)
|
||||
defer statusUpdateTicker.Stop()
|
||||
|
||||
switchToConsensusTicker := time.NewTicker(switchToConsensusIntervalSeconds * time.Second)
|
||||
defer switchToConsensusTicker.Stop()
|
||||
|
||||
blocksSynced := uint64(0)
|
||||
|
||||
chainID := bcR.initialState.ChainID
|
||||
state := bcR.initialState
|
||||
|
||||
lastHundred := time.Now()
|
||||
lastRate := 0.0
|
||||
|
||||
didProcessCh := make(chan struct{}, 1)
|
||||
|
||||
go func() {
|
||||
for {
|
||||
select {
|
||||
case <-bcR.Quit():
|
||||
return
|
||||
case <-bcR.pool.Quit():
|
||||
return
|
||||
case request := <-bcR.requestsCh:
|
||||
peer := bcR.Switch.Peers().Get(request.PeerID)
|
||||
if peer == nil {
|
||||
continue
|
||||
}
|
||||
msgBytes, err := EncodeMsg(&bcproto.BlockRequest{Height: request.Height})
|
||||
if err != nil {
|
||||
bcR.Logger.Error("could not convert msg to proto", "err", err)
|
||||
continue
|
||||
}
|
||||
|
||||
queued := peer.TrySend(BlocksyncChannel, msgBytes)
|
||||
if !queued {
|
||||
bcR.Logger.Debug("Send queue is full, drop block request", "peer", peer.ID(), "height", request.Height)
|
||||
}
|
||||
case err := <-bcR.errorsCh:
|
||||
peer := bcR.Switch.Peers().Get(err.peerID)
|
||||
if peer != nil {
|
||||
bcR.Switch.StopPeerForError(peer, err)
|
||||
}
|
||||
|
||||
case <-statusUpdateTicker.C:
|
||||
// ask for status updates
|
||||
go bcR.BroadcastStatusRequest() // nolint: errcheck
|
||||
|
||||
}
|
||||
}
|
||||
}()
|
||||
|
||||
FOR_LOOP:
|
||||
for {
|
||||
select {
|
||||
case <-switchToConsensusTicker.C:
|
||||
height, numPending, lenRequesters := bcR.pool.GetStatus()
|
||||
outbound, inbound, _ := bcR.Switch.NumPeers()
|
||||
bcR.Logger.Debug("Consensus ticker", "numPending", numPending, "total", lenRequesters,
|
||||
"outbound", outbound, "inbound", inbound)
|
||||
if bcR.pool.IsCaughtUp() {
|
||||
bcR.Logger.Info("Time to switch to consensus reactor!", "height", height)
|
||||
if err := bcR.pool.Stop(); err != nil {
|
||||
bcR.Logger.Error("Error stopping pool", "err", err)
|
||||
}
|
||||
conR, ok := bcR.Switch.Reactor("CONSENSUS").(consensusReactor)
|
||||
if ok {
|
||||
conR.SwitchToConsensus(state, blocksSynced > 0 || stateSynced)
|
||||
}
|
||||
// else {
|
||||
// should only happen during testing
|
||||
// }
|
||||
|
||||
break FOR_LOOP
|
||||
}
|
||||
|
||||
case <-trySyncTicker.C: // chan time
|
||||
select {
|
||||
case didProcessCh <- struct{}{}:
|
||||
default:
|
||||
}
|
||||
|
||||
case <-didProcessCh:
|
||||
// NOTE: It is a subtle mistake to process more than a single block
|
||||
// at a time (e.g. 10) here, because we only TrySend 1 request per
|
||||
// loop. The ratio mismatch can result in starving of blocks, a
|
||||
// sudden burst of requests and responses, and repeat.
|
||||
// Consequently, it is better to split these routines rather than
|
||||
// coupling them as it's written here. TODO uncouple from request
|
||||
// routine.
|
||||
|
||||
// See if there are any blocks to sync.
|
||||
first, second := bcR.pool.PeekTwoBlocks()
|
||||
// bcR.Logger.Info("TrySync peeked", "first", first, "second", second)
|
||||
if first == nil || second == nil {
|
||||
// We need both to sync the first block.
|
||||
continue FOR_LOOP
|
||||
} else {
|
||||
// Try again quickly next loop.
|
||||
didProcessCh <- struct{}{}
|
||||
}
|
||||
|
||||
firstParts := first.MakePartSet(types.BlockPartSizeBytes)
|
||||
firstPartSetHeader := firstParts.Header()
|
||||
firstID := types.BlockID{Hash: first.Hash(), PartSetHeader: firstPartSetHeader}
|
||||
// Finally, verify the first block using the second's commit
|
||||
// NOTE: we can probably make this more efficient, but note that calling
|
||||
// first.Hash() doesn't verify the tx contents, so MakePartSet() is
|
||||
// currently necessary.
|
||||
err := state.Validators.VerifyCommitLight(
|
||||
chainID, firstID, first.Height, second.LastCommit)
|
||||
|
||||
if err == nil {
|
||||
// validate the block before we persist it
|
||||
err = bcR.blockExec.ValidateBlock(state, first)
|
||||
}
|
||||
|
||||
if err != nil {
|
||||
bcR.Logger.Error("Error in validation", "err", err)
|
||||
peerID := bcR.pool.RedoRequest(first.Height)
|
||||
peer := bcR.Switch.Peers().Get(peerID)
|
||||
if peer != nil {
|
||||
// NOTE: we've already removed the peer's request, but we
|
||||
// still need to clean up the rest.
|
||||
bcR.Switch.StopPeerForError(peer, fmt.Errorf("Reactor validation error: %v", err))
|
||||
}
|
||||
peerID2 := bcR.pool.RedoRequest(second.Height)
|
||||
peer2 := bcR.Switch.Peers().Get(peerID2)
|
||||
if peer2 != nil && peer2 != peer {
|
||||
// NOTE: we've already removed the peer's request, but we
|
||||
// still need to clean up the rest.
|
||||
bcR.Switch.StopPeerForError(peer2, fmt.Errorf("Reactor validation error: %v", err))
|
||||
}
|
||||
continue FOR_LOOP
|
||||
}
|
||||
|
||||
bcR.pool.PopRequest()
|
||||
|
||||
// TODO: batch saves so we dont persist to disk every block
|
||||
bcR.store.SaveBlock(first, firstParts, second.LastCommit)
|
||||
|
||||
// TODO: same thing for app - but we would need a way to
|
||||
// get the hash without persisting the state
|
||||
state, _, err = bcR.blockExec.ApplyBlock(state, firstID, first)
|
||||
if err != nil {
|
||||
// TODO This is bad, are we zombie?
|
||||
panic(fmt.Sprintf("Failed to process committed block (%d:%X): %v", first.Height, first.Hash(), err))
|
||||
}
|
||||
blocksSynced++
|
||||
|
||||
if blocksSynced%100 == 0 {
|
||||
lastRate = 0.9*lastRate + 0.1*(100/time.Since(lastHundred).Seconds())
|
||||
bcR.Logger.Info("Block Sync Rate", "height", bcR.pool.height,
|
||||
"max_peer_height", bcR.pool.MaxPeerHeight(), "blocks/s", lastRate)
|
||||
lastHundred = time.Now()
|
||||
}
|
||||
|
||||
continue FOR_LOOP
|
||||
|
||||
case <-bcR.Quit():
|
||||
break FOR_LOOP
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// BroadcastStatusRequest broadcasts `BlockStore` base and height.
|
||||
func (bcR *Reactor) BroadcastStatusRequest() error {
|
||||
bm, err := EncodeMsg(&bcproto.StatusRequest{})
|
||||
if err != nil {
|
||||
bcR.Logger.Error("could not convert msg to proto", "err", err)
|
||||
return fmt.Errorf("could not convert msg to proto: %w", err)
|
||||
}
|
||||
|
||||
bcR.Switch.Broadcast(BlocksyncChannel, bm)
|
||||
|
||||
return nil
|
||||
}
|
||||
Reference in New Issue
Block a user