miner: regenerate mining work every 3 seconds (#17413)

* miner: regenerate mining work every 3 seconds

* miner: polish
This commit is contained in:
gary rong 2018-08-16 19:14:33 +08:00 committed by Péter Szilágyi
parent 5952d962dc
commit 54216811a0
3 changed files with 267 additions and 133 deletions

@ -29,9 +29,6 @@ type PendingLogsEvent struct {
Logs []*types.Log
}
// PendingStateEvent is posted pre mining and notifies of pending state changes.
type PendingStateEvent struct{}
// NewMinedBlockEvent is posted when a block has been imported.
type NewMinedBlockEvent struct{ Block *types.Block }

@ -40,19 +40,27 @@ import (
const (
// resultQueueSize is the size of channel listening to sealing result.
resultQueueSize = 10
// txChanSize is the size of channel listening to NewTxsEvent.
// The number is referenced from the size of tx pool.
txChanSize = 4096
// chainHeadChanSize is the size of channel listening to ChainHeadEvent.
chainHeadChanSize = 10
// chainSideChanSize is the size of channel listening to ChainSideEvent.
chainSideChanSize = 10
// miningLogAtDepth is the number of confirmations before logging successful mining.
miningLogAtDepth = 5
// blockRecommitInterval is the time interval to recreate the mining block with
// any newly arrived transactions.
blockRecommitInterval = 3 * time.Second
)
// Env is the worker's current environment and holds all of the current state information.
type Env struct {
config *params.ChainConfig
// environment is the worker's current environment and holds all of the current state information.
type environment struct {
signer types.Signer
state *state.StateDB // apply state changes here
@ -67,105 +75,6 @@ type Env struct {
receipts []*types.Receipt
}
func (env *Env) commitTransaction(tx *types.Transaction, bc *core.BlockChain, coinbase common.Address, gp *core.GasPool) (error, []*types.Log) {
snap := env.state.Snapshot()
receipt, _, err := core.ApplyTransaction(env.config, bc, &coinbase, gp, env.state, env.header, tx, &env.header.GasUsed, vm.Config{})
if err != nil {
env.state.RevertToSnapshot(snap)
return err, nil
}
env.txs = append(env.txs, tx)
env.receipts = append(env.receipts, receipt)
return nil, receipt.Logs
}
func (env *Env) commitTransactions(mux *event.TypeMux, txs *types.TransactionsByPriceAndNonce, bc *core.BlockChain, coinbase common.Address) {
if env.gasPool == nil {
env.gasPool = new(core.GasPool).AddGas(env.header.GasLimit)
}
var coalescedLogs []*types.Log
for {
// If we don't have enough gas for any further transactions then we're done
if env.gasPool.Gas() < params.TxGas {
log.Trace("Not enough gas for further transactions", "have", env.gasPool, "want", params.TxGas)
break
}
// Retrieve the next transaction and abort if all done
tx := txs.Peek()
if tx == nil {
break
}
// Error may be ignored here. The error has already been checked
// during transaction acceptance is the transaction pool.
//
// We use the eip155 signer regardless of the current hf.
from, _ := types.Sender(env.signer, tx)
// Check whether the tx is replay protected. If we're not in the EIP155 hf
// phase, start ignoring the sender until we do.
if tx.Protected() && !env.config.IsEIP155(env.header.Number) {
log.Trace("Ignoring reply protected transaction", "hash", tx.Hash(), "eip155", env.config.EIP155Block)
txs.Pop()
continue
}
// Start executing the transaction
env.state.Prepare(tx.Hash(), common.Hash{}, env.tcount)
err, logs := env.commitTransaction(tx, bc, coinbase, env.gasPool)
switch err {
case core.ErrGasLimitReached:
// Pop the current out-of-gas transaction without shifting in the next from the account
log.Trace("Gas limit exceeded for current block", "sender", from)
txs.Pop()
case core.ErrNonceTooLow:
// New head notification data race between the transaction pool and miner, shift
log.Trace("Skipping transaction with low nonce", "sender", from, "nonce", tx.Nonce())
txs.Shift()
case core.ErrNonceTooHigh:
// Reorg notification data race between the transaction pool and miner, skip account =
log.Trace("Skipping account with hight nonce", "sender", from, "nonce", tx.Nonce())
txs.Pop()
case nil:
// Everything ok, collect the logs and shift in the next transaction from the same account
coalescedLogs = append(coalescedLogs, logs...)
env.tcount++
txs.Shift()
default:
// Strange error, discard the transaction and get the next in line (note, the
// nonce-too-high clause will prevent us from executing in vain).
log.Debug("Transaction failed, account skipped", "hash", tx.Hash(), "err", err)
txs.Shift()
}
}
if len(coalescedLogs) > 0 || env.tcount > 0 {
// make a copy, the state caches the logs and these logs get "upgraded" from pending to mined
// logs by filling in the block hash when the block was mined by the local miner. This can
// cause a race condition if a log was "upgraded" before the PendingLogsEvent is processed.
cpy := make([]*types.Log, len(coalescedLogs))
for i, l := range coalescedLogs {
cpy[i] = new(types.Log)
*cpy[i] = *l
}
go func(logs []*types.Log, tcount int) {
if len(logs) > 0 {
mux.Post(core.PendingLogsEvent{Logs: logs})
}
if tcount > 0 {
mux.Post(core.PendingStateEvent{})
}
}(cpy, env.tcount)
}
}
// task contains all information for consensus engine sealing and result submitting.
type task struct {
receipts []*types.Receipt
@ -174,6 +83,17 @@ type task struct {
createdAt time.Time
}
const (
commitInterruptNone int32 = iota
commitInterruptNewHead
commitInterruptResubmit
)
type newWorkReq struct {
interrupt *int32
noempty bool
}
// worker is the main object which takes care of submitting new work to consensus engine
// and gathering the sealing result.
type worker struct {
@ -192,12 +112,13 @@ type worker struct {
chainSideSub event.Subscription
// Channels
newWork chan struct{}
newWorkCh chan *newWorkReq
taskCh chan *task
resultCh chan *task
startCh chan struct{}
exitCh chan struct{}
current *Env // An environment for current running cycle.
current *environment // An environment for current running cycle.
possibleUncles map[common.Hash]*types.Block // A set of side blocks as the possible uncle blocks.
unconfirmed *unconfirmedBlocks // A set of locally mined blocks pending canonicalness confirmations.
@ -230,10 +151,11 @@ func newWorker(config *params.ChainConfig, engine consensus.Engine, eth Backend,
txsCh: make(chan core.NewTxsEvent, txChanSize),
chainHeadCh: make(chan core.ChainHeadEvent, chainHeadChanSize),
chainSideCh: make(chan core.ChainSideEvent, chainSideChanSize),
newWork: make(chan struct{}, 1),
newWorkCh: make(chan *newWorkReq),
taskCh: make(chan *task),
resultCh: make(chan *task, resultQueueSize),
exitCh: make(chan struct{}),
startCh: make(chan struct{}, 1),
}
// Subscribe NewTxsEvent for tx pool
worker.txsSub = eth.TxPool().SubscribeNewTxsEvent(worker.txsCh)
@ -242,11 +164,13 @@ func newWorker(config *params.ChainConfig, engine consensus.Engine, eth Backend,
worker.chainSideSub = eth.BlockChain().SubscribeChainSideEvent(worker.chainSideCh)
go worker.mainLoop()
go worker.newWorkLoop()
go worker.resultLoop()
go worker.taskLoop()
// Submit first work to initialize pending state.
worker.newWork <- struct{}{}
worker.startCh <- struct{}{}
return worker
}
@ -286,7 +210,7 @@ func (w *worker) pendingBlock() *types.Block {
// start sets the running status as 1 and triggers new work submitting.
func (w *worker) start() {
atomic.StoreInt32(&w.running, 1)
w.newWork <- struct{}{}
w.startCh <- struct{}{}
}
// stop sets the running status as 0.
@ -313,6 +237,44 @@ func (w *worker) close() {
}
}
// newWorkLoop is a standalone goroutine to submit new mining work upon received events.
func (w *worker) newWorkLoop() {
var interrupt *int32
timer := time.NewTimer(0)
<-timer.C // discard the initial tick
// recommit aborts in-flight transaction execution with given signal and resubmits a new one.
recommit := func(noempty bool, s int32) {
if interrupt != nil {
atomic.StoreInt32(interrupt, s)
}
interrupt = new(int32)
w.newWorkCh <- &newWorkReq{interrupt: interrupt, noempty: noempty}
timer.Reset(blockRecommitInterval)
}
for {
select {
case <-w.startCh:
recommit(false, commitInterruptNewHead)
case <-w.chainHeadCh:
recommit(false, commitInterruptNewHead)
case <-timer.C:
// If mining is running resubmit a new work cycle periodically to pull in
// higher priced transactions. Disable this overhead for pending blocks.
if w.isRunning() && (w.config.Clique == nil || w.config.Clique.Period > 0) {
recommit(true, commitInterruptResubmit)
}
case <-w.exitCh:
return
}
}
}
// mainLoop is a standalone goroutine to regenerate the sealing task based on the received event.
func (w *worker) mainLoop() {
defer w.txsSub.Unsubscribe()
@ -321,13 +283,8 @@ func (w *worker) mainLoop() {
for {
select {
case <-w.newWork:
// Submit a work when the worker is created or started.
w.commitNewWork()
case <-w.chainHeadCh:
// Resubmit a work for new cycle once worker receives chain head event.
w.commitNewWork()
case req := <-w.newWorkCh:
w.commitNewWork(req.interrupt, req.noempty)
case ev := <-w.chainSideCh:
if _, exist := w.possibleUncles[ev.Block.Hash()]; exist {
@ -364,9 +321,9 @@ func (w *worker) mainLoop() {
// already included in the current mining block. These transactions will
// be automatically eliminated.
if !w.isRunning() && w.current != nil {
w.mu.Lock()
w.mu.RLock()
coinbase := w.coinbase
w.mu.Unlock()
w.mu.RUnlock()
txs := make(map[common.Address]types.Transactions)
for _, tx := range ev.Txs {
@ -374,12 +331,12 @@ func (w *worker) mainLoop() {
txs[acc] = append(txs[acc], tx)
}
txset := types.NewTransactionsByPriceAndNonce(w.current.signer, txs)
w.current.commitTransactions(w.mux, txset, w.chain, coinbase)
w.commitTransactions(txset, coinbase, nil)
w.updateSnapshot()
} else {
// If we're mining, but nothing is being processed, wake on new transactions
if w.config.Clique != nil && w.config.Clique.Period == 0 {
w.commitNewWork()
w.commitNewWork(nil, false)
}
}
@ -508,8 +465,7 @@ func (w *worker) makeCurrent(parent *types.Block, header *types.Header) error {
if err != nil {
return err
}
env := &Env{
config: w.config,
env := &environment{
signer: types.NewEIP155Signer(w.config.ChainID),
state: state,
ancestors: mapset.NewSet(),
@ -534,7 +490,7 @@ func (w *worker) makeCurrent(parent *types.Block, header *types.Header) error {
}
// commitUncle adds the given block to uncle block set, returns error if failed to add.
func (w *worker) commitUncle(env *Env, uncle *types.Header) error {
func (w *worker) commitUncle(env *environment, uncle *types.Header) error {
hash := uncle.Hash()
if env.uncles.Contains(hash) {
return fmt.Errorf("uncle not unique")
@ -579,8 +535,120 @@ func (w *worker) updateSnapshot() {
w.snapshotState = w.current.state.Copy()
}
func (w *worker) commitTransaction(tx *types.Transaction, coinbase common.Address) ([]*types.Log, error) {
snap := w.current.state.Snapshot()
receipt, _, err := core.ApplyTransaction(w.config, w.chain, &coinbase, w.current.gasPool, w.current.state, w.current.header, tx, &w.current.header.GasUsed, vm.Config{})
if err != nil {
w.current.state.RevertToSnapshot(snap)
return nil, err
}
w.current.txs = append(w.current.txs, tx)
w.current.receipts = append(w.current.receipts, receipt)
return receipt.Logs, nil
}
func (w *worker) commitTransactions(txs *types.TransactionsByPriceAndNonce, coinbase common.Address, interrupt *int32) bool {
// Short circuit if current is nil
if w.current == nil {
return true
}
if w.current.gasPool == nil {
w.current.gasPool = new(core.GasPool).AddGas(w.current.header.GasLimit)
}
var coalescedLogs []*types.Log
for {
// In the following three cases, we will interrupt the execution of the transaction.
// (1) new head block event arrival, the interrupt signal is 1
// (2) worker start or restart, the interrupt signal is 1
// (3) worker recreate the mining block with any newly arrived transactions, the interrupt signal is 2.
// For the first two cases, the semi-finished work will be discarded.
// For the third case, the semi-finished work will be submitted to the consensus engine.
// TODO(rjl493456442) give feedback to newWorkLoop to adjust resubmit interval if it is too short.
if interrupt != nil && atomic.LoadInt32(interrupt) != commitInterruptNone {
return atomic.LoadInt32(interrupt) == commitInterruptNewHead
}
// If we don't have enough gas for any further transactions then we're done
if w.current.gasPool.Gas() < params.TxGas {
log.Trace("Not enough gas for further transactions", "have", w.current.gasPool, "want", params.TxGas)
break
}
// Retrieve the next transaction and abort if all done
tx := txs.Peek()
if tx == nil {
break
}
// Error may be ignored here. The error has already been checked
// during transaction acceptance is the transaction pool.
//
// We use the eip155 signer regardless of the current hf.
from, _ := types.Sender(w.current.signer, tx)
// Check whether the tx is replay protected. If we're not in the EIP155 hf
// phase, start ignoring the sender until we do.
if tx.Protected() && !w.config.IsEIP155(w.current.header.Number) {
log.Trace("Ignoring reply protected transaction", "hash", tx.Hash(), "eip155", w.config.EIP155Block)
txs.Pop()
continue
}
// Start executing the transaction
w.current.state.Prepare(tx.Hash(), common.Hash{}, w.current.tcount)
logs, err := w.commitTransaction(tx, coinbase)
switch err {
case core.ErrGasLimitReached:
// Pop the current out-of-gas transaction without shifting in the next from the account
log.Trace("Gas limit exceeded for current block", "sender", from)
txs.Pop()
case core.ErrNonceTooLow:
// New head notification data race between the transaction pool and miner, shift
log.Trace("Skipping transaction with low nonce", "sender", from, "nonce", tx.Nonce())
txs.Shift()
case core.ErrNonceTooHigh:
// Reorg notification data race between the transaction pool and miner, skip account =
log.Trace("Skipping account with hight nonce", "sender", from, "nonce", tx.Nonce())
txs.Pop()
case nil:
// Everything ok, collect the logs and shift in the next transaction from the same account
coalescedLogs = append(coalescedLogs, logs...)
w.current.tcount++
txs.Shift()
default:
// Strange error, discard the transaction and get the next in line (note, the
// nonce-too-high clause will prevent us from executing in vain).
log.Debug("Transaction failed, account skipped", "hash", tx.Hash(), "err", err)
txs.Shift()
}
}
if !w.isRunning() && len(coalescedLogs) > 0 {
// We don't push the pendingLogsEvent while we are mining. The reason is that
// when we are mining, the worker will regenerate a mining block every 3 seconds.
// In order to avoid pushing the repeated pendingLog, we disable the pending log pushing.
// make a copy, the state caches the logs and these logs get "upgraded" from pending to mined
// logs by filling in the block hash when the block was mined by the local miner. This can
// cause a race condition if a log was "upgraded" before the PendingLogsEvent is processed.
cpy := make([]*types.Log, len(coalescedLogs))
for i, l := range coalescedLogs {
cpy[i] = new(types.Log)
*cpy[i] = *l
}
go w.mux.Post(core.PendingLogsEvent{Logs: cpy})
}
return false
}
// commitNewWork generates several new sealing tasks based on the parent block.
func (w *worker) commitNewWork() {
func (w *worker) commitNewWork(interrupt *int32, noempty bool) {
w.mu.RLock()
defer w.mu.RUnlock()
@ -666,9 +734,11 @@ func (w *worker) commitNewWork() {
delete(w.possibleUncles, hash)
}
if !noempty {
// Create an empty block based on temporary copied state for sealing in advance without waiting block
// execution finished.
w.commit(uncles, nil, false, tstart)
}
// Fill the block with all available pending transactions.
pending, err := w.eth.TxPool().Pending()
@ -682,7 +752,9 @@ func (w *worker) commitNewWork() {
return
}
txs := types.NewTransactionsByPriceAndNonce(w.current.signer, pending)
env.commitTransactions(w.mux, txs, w.chain, w.coinbase)
if w.commitTransactions(txs, w.coinbase, interrupt) {
return
}
w.commit(uncles, w.fullTaskHook, true, tstart)
}

@ -270,3 +270,68 @@ func TestStreamUncleBlock(t *testing.T) {
t.Error("new task timeout")
}
}
func TestRegenerateMiningBlockEthash(t *testing.T) {
testRegenerateMiningBlock(t, ethashChainConfig, ethash.NewFaker())
}
func TestRegenerateMiningBlockClique(t *testing.T) {
testRegenerateMiningBlock(t, cliqueChainConfig, clique.New(cliqueChainConfig.Clique, ethdb.NewMemDatabase()))
}
func testRegenerateMiningBlock(t *testing.T, chainConfig *params.ChainConfig, engine consensus.Engine) {
defer engine.Close()
w, b := newTestWorker(t, chainConfig, engine)
defer w.close()
var taskCh = make(chan struct{})
taskIndex := 0
w.newTaskHook = func(task *task) {
if task.block.NumberU64() == 1 {
if taskIndex == 2 {
receiptLen, balance := 2, big.NewInt(2000)
if len(task.receipts) != receiptLen {
t.Errorf("receipt number mismatch has %d, want %d", len(task.receipts), receiptLen)
}
if task.state.GetBalance(acc1Addr).Cmp(balance) != 0 {
t.Errorf("account balance mismatch has %d, want %d", task.state.GetBalance(acc1Addr), balance)
}
}
taskCh <- struct{}{}
taskIndex += 1
}
}
w.skipSealHook = func(task *task) bool {
return true
}
w.fullTaskHook = func() {
time.Sleep(100 * time.Millisecond)
}
// Ensure worker has finished initialization
for {
b := w.pendingBlock()
if b != nil && b.NumberU64() == 1 {
break
}
}
w.start()
// Ignore the first two works
for i := 0; i < 2; i += 1 {
select {
case <-taskCh:
case <-time.NewTimer(time.Second).C:
t.Error("new task timeout")
}
}
b.txPool.AddLocals(newTxs)
time.Sleep(3 * time.Second)
select {
case <-taskCh:
case <-time.NewTimer(time.Second).C:
t.Error("new task timeout")
}
}