2017-11-10 03:30:20 +03:00
|
|
|
package bitcoindnotify
|
|
|
|
|
|
|
|
import (
|
|
|
|
"errors"
|
|
|
|
"fmt"
|
|
|
|
"sync"
|
|
|
|
"sync/atomic"
|
|
|
|
"time"
|
|
|
|
|
2018-06-05 04:34:16 +03:00
|
|
|
"github.com/btcsuite/btcd/btcjson"
|
|
|
|
"github.com/btcsuite/btcd/chaincfg"
|
|
|
|
"github.com/btcsuite/btcd/chaincfg/chainhash"
|
|
|
|
"github.com/btcsuite/btcd/rpcclient"
|
|
|
|
"github.com/btcsuite/btcd/wire"
|
|
|
|
"github.com/btcsuite/btcutil"
|
|
|
|
"github.com/btcsuite/btcwallet/chain"
|
|
|
|
"github.com/btcsuite/btcwallet/wtxmgr"
|
2018-07-17 10:13:06 +03:00
|
|
|
"github.com/lightningnetwork/lnd/chainntnfs"
|
2017-11-10 03:30:20 +03:00
|
|
|
)
|
|
|
|
|
|
|
|
const (
|
|
|
|
|
|
|
|
// notifierType uniquely identifies this concrete implementation of the
|
|
|
|
// ChainNotifier interface.
|
|
|
|
notifierType = "bitcoind"
|
|
|
|
|
|
|
|
// reorgSafetyLimit is assumed maximum depth of a chain reorganization.
|
|
|
|
// After this many confirmation, transaction confirmation info will be
|
|
|
|
// pruned.
|
|
|
|
reorgSafetyLimit = 100
|
|
|
|
)
|
|
|
|
|
|
|
|
var (
|
|
|
|
// ErrChainNotifierShuttingDown is used when we are trying to
|
|
|
|
// measure a spend notification when notifier is already stopped.
|
|
|
|
ErrChainNotifierShuttingDown = errors.New("chainntnfs: system interrupt " +
|
|
|
|
"while attempting to register for spend notification.")
|
|
|
|
)
|
|
|
|
|
|
|
|
// chainUpdate encapsulates an update to the current main chain. This struct is
|
|
|
|
// used as an element within an unbounded queue in order to avoid blocking the
|
|
|
|
// main rpc dispatch rule.
|
|
|
|
type chainUpdate struct {
|
|
|
|
blockHash *chainhash.Hash
|
|
|
|
blockHeight int32
|
|
|
|
}
|
|
|
|
|
|
|
|
// TODO(roasbeef): generalize struct below:
|
2018-02-26 21:37:58 +03:00
|
|
|
// * move chans to config
|
|
|
|
// * extract common code
|
|
|
|
// * allow outside callers to handle send conditions
|
2017-11-10 03:30:20 +03:00
|
|
|
|
|
|
|
// BitcoindNotifier implements the ChainNotifier interface using a bitcoind
|
|
|
|
// chain client. Multiple concurrent clients are supported. All notifications
|
|
|
|
// are achieved via non-blocking sends on client channels.
|
|
|
|
type BitcoindNotifier struct {
|
|
|
|
spendClientCounter uint64 // To be used atomically.
|
|
|
|
epochClientCounter uint64 // To be used atomically.
|
|
|
|
|
|
|
|
started int32 // To be used atomically.
|
|
|
|
stopped int32 // To be used atomically.
|
|
|
|
|
|
|
|
chainConn *chain.BitcoindClient
|
|
|
|
|
|
|
|
notificationCancels chan interface{}
|
|
|
|
notificationRegistry chan interface{}
|
|
|
|
|
|
|
|
spendNotifications map[wire.OutPoint]map[uint64]*spendNotification
|
|
|
|
|
|
|
|
txConfNotifier *chainntnfs.TxConfNotifier
|
|
|
|
|
|
|
|
blockEpochClients map[uint64]*blockEpochRegistration
|
|
|
|
|
|
|
|
wg sync.WaitGroup
|
|
|
|
quit chan struct{}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Ensure BitcoindNotifier implements the ChainNotifier interface at compile
|
|
|
|
// time.
|
|
|
|
var _ chainntnfs.ChainNotifier = (*BitcoindNotifier)(nil)
|
|
|
|
|
|
|
|
// New returns a new BitcoindNotifier instance. This function assumes the
|
|
|
|
// bitcoind node detailed in the passed configuration is already running, and
|
|
|
|
// willing to accept RPC requests and new zmq clients.
|
|
|
|
func New(config *rpcclient.ConnConfig, zmqConnect string,
|
|
|
|
params chaincfg.Params) (*BitcoindNotifier, error) {
|
|
|
|
notifier := &BitcoindNotifier{
|
|
|
|
notificationCancels: make(chan interface{}),
|
|
|
|
notificationRegistry: make(chan interface{}),
|
|
|
|
|
|
|
|
blockEpochClients: make(map[uint64]*blockEpochRegistration),
|
|
|
|
|
|
|
|
spendNotifications: make(map[wire.OutPoint]map[uint64]*spendNotification),
|
|
|
|
|
|
|
|
quit: make(chan struct{}),
|
|
|
|
}
|
|
|
|
|
|
|
|
// Disable connecting to bitcoind within the rpcclient.New method. We
|
|
|
|
// defer establishing the connection to our .Start() method.
|
|
|
|
config.DisableConnectOnNew = true
|
|
|
|
config.DisableAutoReconnect = false
|
|
|
|
chainConn, err := chain.NewBitcoindClient(¶ms, config.Host,
|
|
|
|
config.User, config.Pass, zmqConnect, 100*time.Millisecond)
|
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
notifier.chainConn = chainConn
|
|
|
|
|
|
|
|
return notifier, nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Start connects to the running bitcoind node over websockets, registers for
|
|
|
|
// block notifications, and finally launches all related helper goroutines.
|
|
|
|
func (b *BitcoindNotifier) Start() error {
|
|
|
|
// Already started?
|
|
|
|
if atomic.AddInt32(&b.started, 1) != 1 {
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Connect to bitcoind, and register for notifications on connected,
|
|
|
|
// and disconnected blocks.
|
|
|
|
if err := b.chainConn.Start(); err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
if err := b.chainConn.NotifyBlocks(); err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
|
|
|
|
_, currentHeight, err := b.chainConn.GetBestBlock()
|
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
|
|
|
|
b.txConfNotifier = chainntnfs.NewTxConfNotifier(
|
|
|
|
uint32(currentHeight), reorgSafetyLimit)
|
|
|
|
|
|
|
|
b.wg.Add(1)
|
2018-04-07 04:25:41 +03:00
|
|
|
go b.notificationDispatcher(currentHeight)
|
2017-11-10 03:30:20 +03:00
|
|
|
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Stop shutsdown the BitcoindNotifier.
|
|
|
|
func (b *BitcoindNotifier) Stop() error {
|
|
|
|
// Already shutting down?
|
|
|
|
if atomic.AddInt32(&b.stopped, 1) != 1 {
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// Shutdown the rpc client, this gracefully disconnects from bitcoind,
|
|
|
|
// and cleans up all related resources.
|
|
|
|
b.chainConn.Stop()
|
|
|
|
|
|
|
|
close(b.quit)
|
|
|
|
b.wg.Wait()
|
|
|
|
|
|
|
|
// Notify all pending clients of our shutdown by closing the related
|
|
|
|
// notification channels.
|
|
|
|
for _, spendClients := range b.spendNotifications {
|
|
|
|
for _, spendClient := range spendClients {
|
|
|
|
close(spendClient.spendChan)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for _, epochClient := range b.blockEpochClients {
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
close(epochClient.cancelChan)
|
|
|
|
epochClient.wg.Wait()
|
|
|
|
|
2017-11-10 03:30:20 +03:00
|
|
|
close(epochClient.epochChan)
|
|
|
|
}
|
|
|
|
b.txConfNotifier.TearDown()
|
|
|
|
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// blockNtfn packages a notification of a connected/disconnected block along
|
|
|
|
// with its height at the time.
|
|
|
|
type blockNtfn struct {
|
|
|
|
sha *chainhash.Hash
|
|
|
|
height int32
|
|
|
|
}
|
|
|
|
|
|
|
|
// notificationDispatcher is the primary goroutine which handles client
|
|
|
|
// notification registrations, as well as notification dispatches.
|
2018-04-07 04:25:41 +03:00
|
|
|
func (b *BitcoindNotifier) notificationDispatcher(bestHeight int32) {
|
2017-11-10 03:30:20 +03:00
|
|
|
out:
|
|
|
|
for {
|
|
|
|
select {
|
|
|
|
case cancelMsg := <-b.notificationCancels:
|
|
|
|
switch msg := cancelMsg.(type) {
|
|
|
|
case *spendCancel:
|
|
|
|
chainntnfs.Log.Infof("Cancelling spend "+
|
|
|
|
"notification for out_point=%v, "+
|
|
|
|
"spend_id=%v", msg.op, msg.spendID)
|
|
|
|
|
|
|
|
// Before we attempt to close the spendChan,
|
|
|
|
// ensure that the notification hasn't already
|
|
|
|
// yet been dispatched.
|
|
|
|
if outPointClients, ok := b.spendNotifications[msg.op]; ok {
|
|
|
|
close(outPointClients[msg.spendID].spendChan)
|
|
|
|
delete(b.spendNotifications[msg.op], msg.spendID)
|
|
|
|
}
|
|
|
|
|
|
|
|
case *epochCancel:
|
|
|
|
chainntnfs.Log.Infof("Cancelling epoch "+
|
|
|
|
"notification, epoch_id=%v", msg.epochID)
|
|
|
|
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
// First, we'll lookup the original
|
|
|
|
// registration in order to stop the active
|
|
|
|
// queue goroutine.
|
|
|
|
reg := b.blockEpochClients[msg.epochID]
|
|
|
|
reg.epochQueue.Stop()
|
|
|
|
|
|
|
|
// Next, close the cancel channel for this
|
2017-11-10 03:30:20 +03:00
|
|
|
// specific client, and wait for the client to
|
|
|
|
// exit.
|
|
|
|
close(b.blockEpochClients[msg.epochID].cancelChan)
|
|
|
|
b.blockEpochClients[msg.epochID].wg.Wait()
|
|
|
|
|
|
|
|
// Once the client has exited, we can then
|
|
|
|
// safely close the channel used to send epoch
|
|
|
|
// notifications, in order to notify any
|
|
|
|
// listeners that the intent has been
|
|
|
|
// cancelled.
|
|
|
|
close(b.blockEpochClients[msg.epochID].epochChan)
|
|
|
|
delete(b.blockEpochClients, msg.epochID)
|
|
|
|
|
|
|
|
}
|
|
|
|
case registerMsg := <-b.notificationRegistry:
|
|
|
|
switch msg := registerMsg.(type) {
|
|
|
|
case *spendNotification:
|
|
|
|
chainntnfs.Log.Infof("New spend subscription: "+
|
|
|
|
"utxo=%v", msg.targetOutpoint)
|
|
|
|
op := *msg.targetOutpoint
|
|
|
|
|
|
|
|
if _, ok := b.spendNotifications[op]; !ok {
|
|
|
|
b.spendNotifications[op] = make(map[uint64]*spendNotification)
|
|
|
|
}
|
|
|
|
b.spendNotifications[op][msg.spendID] = msg
|
|
|
|
b.chainConn.NotifySpent([]*wire.OutPoint{&op})
|
2018-02-26 21:37:58 +03:00
|
|
|
case *confirmationNotification:
|
|
|
|
chainntnfs.Log.Infof("New confirmation "+
|
2017-11-10 03:30:20 +03:00
|
|
|
"subscription: txid=%v, numconfs=%v",
|
|
|
|
msg.TxID, msg.NumConfirmations)
|
|
|
|
|
2018-02-26 21:37:58 +03:00
|
|
|
_, currentHeight, err := b.chainConn.GetBestBlock()
|
|
|
|
if err != nil {
|
|
|
|
chainntnfs.Log.Error(err)
|
|
|
|
}
|
|
|
|
|
2017-11-10 03:30:20 +03:00
|
|
|
// Lookup whether the transaction is already included in the
|
|
|
|
// active chain.
|
2018-02-26 21:37:58 +03:00
|
|
|
txConf, err := b.historicalConfDetails(
|
|
|
|
msg.TxID, msg.heightHint, uint32(currentHeight),
|
|
|
|
)
|
2017-11-10 03:30:20 +03:00
|
|
|
if err != nil {
|
|
|
|
chainntnfs.Log.Error(err)
|
|
|
|
}
|
2018-02-26 21:37:58 +03:00
|
|
|
|
2017-11-10 03:30:20 +03:00
|
|
|
err = b.txConfNotifier.Register(&msg.ConfNtfn, txConf)
|
|
|
|
if err != nil {
|
|
|
|
chainntnfs.Log.Error(err)
|
|
|
|
}
|
|
|
|
case *blockEpochRegistration:
|
|
|
|
chainntnfs.Log.Infof("New block epoch subscription")
|
|
|
|
b.blockEpochClients[msg.epochID] = msg
|
2018-04-07 04:25:41 +03:00
|
|
|
case chain.RelevantTx:
|
|
|
|
b.handleRelevantTx(msg, bestHeight)
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
case ntfn := <-b.chainConn.Notifications():
|
|
|
|
switch item := ntfn.(type) {
|
|
|
|
case chain.BlockConnected:
|
2018-04-07 04:25:41 +03:00
|
|
|
if item.Height != bestHeight+1 {
|
2017-11-10 03:30:20 +03:00
|
|
|
chainntnfs.Log.Warnf("Received blocks out of order: "+
|
|
|
|
"current height=%d, new height=%d",
|
2018-04-07 04:25:41 +03:00
|
|
|
bestHeight, item.Height)
|
2017-11-10 03:30:20 +03:00
|
|
|
continue
|
|
|
|
}
|
2018-04-07 04:25:41 +03:00
|
|
|
bestHeight = item.Height
|
2017-11-10 03:30:20 +03:00
|
|
|
|
|
|
|
rawBlock, err := b.chainConn.GetBlock(&item.Hash)
|
|
|
|
if err != nil {
|
|
|
|
chainntnfs.Log.Errorf("Unable to get block: %v", err)
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
|
|
|
|
chainntnfs.Log.Infof("New block: height=%v, sha=%v",
|
|
|
|
item.Height, item.Hash)
|
|
|
|
|
|
|
|
b.notifyBlockEpochs(item.Height, &item.Hash)
|
|
|
|
|
|
|
|
txns := btcutil.NewBlock(rawBlock).Transactions()
|
|
|
|
err = b.txConfNotifier.ConnectTip(&item.Hash,
|
|
|
|
uint32(item.Height), txns)
|
|
|
|
if err != nil {
|
|
|
|
chainntnfs.Log.Error(err)
|
|
|
|
}
|
|
|
|
continue
|
|
|
|
|
|
|
|
case chain.BlockDisconnected:
|
2018-04-07 04:25:41 +03:00
|
|
|
if item.Height != bestHeight {
|
2017-11-10 03:30:20 +03:00
|
|
|
chainntnfs.Log.Warnf("Received blocks "+
|
|
|
|
"out of order: current height="+
|
|
|
|
"%d, disconnected height=%d",
|
2018-04-07 04:25:41 +03:00
|
|
|
bestHeight, item.Height)
|
2017-11-10 03:30:20 +03:00
|
|
|
continue
|
|
|
|
}
|
2018-04-07 04:25:41 +03:00
|
|
|
bestHeight = item.Height - 1
|
2017-11-10 03:30:20 +03:00
|
|
|
|
|
|
|
chainntnfs.Log.Infof("Block disconnected from "+
|
|
|
|
"main chain: height=%v, sha=%v",
|
|
|
|
item.Height, item.Hash)
|
|
|
|
|
|
|
|
err := b.txConfNotifier.DisconnectTip(
|
|
|
|
uint32(item.Height))
|
|
|
|
if err != nil {
|
|
|
|
chainntnfs.Log.Error(err)
|
|
|
|
}
|
|
|
|
|
|
|
|
case chain.RelevantTx:
|
2018-04-07 04:25:41 +03:00
|
|
|
b.handleRelevantTx(item, bestHeight)
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
case <-b.quit:
|
|
|
|
break out
|
|
|
|
}
|
|
|
|
}
|
|
|
|
b.wg.Done()
|
|
|
|
}
|
|
|
|
|
2018-04-07 04:25:41 +03:00
|
|
|
// handleRelevantTx notifies any clients of a relevant transaction.
|
|
|
|
func (b *BitcoindNotifier) handleRelevantTx(tx chain.RelevantTx, bestHeight int32) {
|
|
|
|
msgTx := tx.TxRecord.MsgTx
|
2018-07-17 12:18:18 +03:00
|
|
|
|
|
|
|
// We only care about notifying on confirmed spends, so in case this is
|
|
|
|
// a mempool spend, we can continue, and wait for the spend to appear
|
|
|
|
// in chain.
|
|
|
|
if tx.Block == nil {
|
|
|
|
return
|
|
|
|
}
|
|
|
|
|
2018-04-07 04:25:41 +03:00
|
|
|
// First, check if this transaction spends an output
|
|
|
|
// that has an existing spend notification for it.
|
|
|
|
for i, txIn := range msgTx.TxIn {
|
|
|
|
prevOut := txIn.PreviousOutPoint
|
|
|
|
|
|
|
|
// If this transaction indeed does spend an
|
|
|
|
// output which we have a registered
|
|
|
|
// notification for, then create a spend
|
|
|
|
// summary, finally sending off the details to
|
|
|
|
// the notification subscriber.
|
|
|
|
if clients, ok := b.spendNotifications[prevOut]; ok {
|
|
|
|
spenderSha := msgTx.TxHash()
|
|
|
|
spendDetails := &chainntnfs.SpendDetail{
|
|
|
|
SpentOutPoint: &prevOut,
|
|
|
|
SpenderTxHash: &spenderSha,
|
|
|
|
SpendingTx: &msgTx,
|
|
|
|
SpenderInputIndex: uint32(i),
|
|
|
|
}
|
2018-07-17 12:18:18 +03:00
|
|
|
spendDetails.SpendingHeight = tx.Block.Height
|
2018-04-07 04:25:41 +03:00
|
|
|
|
2018-07-17 12:18:18 +03:00
|
|
|
for _, ntfn := range clients {
|
|
|
|
chainntnfs.Log.Infof("Dispatching confirmed "+
|
|
|
|
"spend notification for outpoint=%v "+
|
|
|
|
"at height %v", ntfn.targetOutpoint,
|
2018-07-13 13:35:29 +03:00
|
|
|
spendDetails.SpendingHeight)
|
2018-04-07 04:25:41 +03:00
|
|
|
ntfn.spendChan <- spendDetails
|
|
|
|
|
2018-07-17 12:18:18 +03:00
|
|
|
// Close spendChan to ensure that any calls to
|
|
|
|
// Cancel will not block. This is safe to do
|
|
|
|
// since the channel is buffered, and the
|
2018-04-07 04:25:41 +03:00
|
|
|
// message can still be read by the receiver.
|
|
|
|
close(ntfn.spendChan)
|
|
|
|
}
|
|
|
|
delete(b.spendNotifications, prevOut)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-11-10 03:30:20 +03:00
|
|
|
// historicalConfDetails looks up whether a transaction is already included in a
|
|
|
|
// block in the active chain and, if so, returns details about the confirmation.
|
|
|
|
func (b *BitcoindNotifier) historicalConfDetails(txid *chainhash.Hash,
|
2018-02-26 21:37:58 +03:00
|
|
|
heightHint, currentHeight uint32) (*chainntnfs.TxConfirmation, error) {
|
|
|
|
|
|
|
|
// First, we'll attempt to retrieve the transaction details using the
|
|
|
|
// backend node's transaction index.
|
|
|
|
txConf, err := b.confDetailsFromTxIndex(txid)
|
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
|
|
|
|
if txConf != nil {
|
|
|
|
return txConf, nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// If the backend node's transaction index is not enabled, then we'll
|
|
|
|
// fall back to manually scanning the chain's blocks, looking for the
|
|
|
|
// block where the transaction was included in.
|
|
|
|
return b.confDetailsManually(txid, heightHint, currentHeight)
|
|
|
|
}
|
|
|
|
|
|
|
|
// confDetailsFromTxIndex looks up whether a transaction is already included
|
|
|
|
// in a block in the active chain by using the backend node's transaction index.
|
|
|
|
// If the transaction is found, its confirmation details are returned.
|
|
|
|
// Otherwise, nil is returned.
|
|
|
|
func (b *BitcoindNotifier) confDetailsFromTxIndex(txid *chainhash.Hash,
|
2017-11-10 03:30:20 +03:00
|
|
|
) (*chainntnfs.TxConfirmation, error) {
|
|
|
|
|
2018-02-26 21:37:58 +03:00
|
|
|
// If the transaction has some or all of its confirmations required,
|
2017-11-10 03:30:20 +03:00
|
|
|
// then we may be able to dispatch it immediately.
|
|
|
|
tx, err := b.chainConn.GetRawTransactionVerbose(txid)
|
2018-02-26 21:37:58 +03:00
|
|
|
if err != nil {
|
|
|
|
// Avoid returning an error if the transaction index is not
|
|
|
|
// enabled to proceed with fallback methods.
|
|
|
|
jsonErr, ok := err.(*btcjson.RPCError)
|
|
|
|
if !ok || jsonErr.Code != btcjson.ErrRPCNoTxInfo {
|
|
|
|
return nil, fmt.Errorf("unable to query for txid "+
|
|
|
|
"%v: %v", txid, err)
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
2018-02-26 21:37:58 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// Make sure we actually retrieved a transaction that is included in a
|
|
|
|
// block. Without this, we won't be able to retrieve its confirmation
|
|
|
|
// details.
|
|
|
|
if tx == nil || tx.BlockHash == "" {
|
|
|
|
return nil, nil
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// As we need to fully populate the returned TxConfirmation struct,
|
|
|
|
// grab the block in which the transaction was confirmed so we can
|
|
|
|
// locate its exact index within the block.
|
|
|
|
blockHash, err := chainhash.NewHashFromStr(tx.BlockHash)
|
|
|
|
if err != nil {
|
2018-02-26 21:37:58 +03:00
|
|
|
return nil, fmt.Errorf("unable to get block hash %v for "+
|
|
|
|
"historical dispatch: %v", tx.BlockHash, err)
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
2018-02-26 21:37:58 +03:00
|
|
|
|
2017-11-10 03:30:20 +03:00
|
|
|
block, err := b.chainConn.GetBlockVerbose(blockHash)
|
|
|
|
if err != nil {
|
2018-02-26 21:37:58 +03:00
|
|
|
return nil, fmt.Errorf("unable to get block with hash %v for "+
|
|
|
|
"historical dispatch: %v", blockHash, err)
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
|
2018-02-26 21:37:58 +03:00
|
|
|
// If the block was obtained, locate the transaction's index within the
|
2017-11-10 03:30:20 +03:00
|
|
|
// block so we can give the subscriber full confirmation details.
|
|
|
|
targetTxidStr := txid.String()
|
2018-02-26 21:37:58 +03:00
|
|
|
for txIndex, txHash := range block.Tx {
|
2017-11-10 03:30:20 +03:00
|
|
|
if txHash == targetTxidStr {
|
2018-02-26 21:37:58 +03:00
|
|
|
return &chainntnfs.TxConfirmation{
|
|
|
|
BlockHash: blockHash,
|
|
|
|
BlockHeight: uint32(block.Height),
|
|
|
|
TxIndex: uint32(txIndex),
|
|
|
|
}, nil
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-02-26 21:37:58 +03:00
|
|
|
// We return an error because we should have found the transaction
|
|
|
|
// within the block, but didn't.
|
|
|
|
return nil, fmt.Errorf("unable to locate tx %v in block %v", txid,
|
|
|
|
blockHash)
|
|
|
|
}
|
2017-11-10 03:30:20 +03:00
|
|
|
|
2018-02-26 21:37:58 +03:00
|
|
|
// confDetailsManually looks up whether a transaction is already included in a
|
|
|
|
// block in the active chain by scanning the chain's blocks, starting from the
|
|
|
|
// earliest height the transaction could have been included in, to the current
|
|
|
|
// height in the chain. If the transaction is found, its confirmation details
|
|
|
|
// are returned. Otherwise, nil is returned.
|
|
|
|
func (b *BitcoindNotifier) confDetailsManually(txid *chainhash.Hash,
|
|
|
|
heightHint, currentHeight uint32) (*chainntnfs.TxConfirmation, error) {
|
|
|
|
|
|
|
|
targetTxidStr := txid.String()
|
|
|
|
|
|
|
|
// Begin scanning blocks at every height to determine where the
|
|
|
|
// transaction was included in.
|
|
|
|
for height := heightHint; height <= currentHeight; height++ {
|
|
|
|
blockHash, err := b.chainConn.GetBlockHash(int64(height))
|
|
|
|
if err != nil {
|
|
|
|
return nil, fmt.Errorf("unable to get hash from block "+
|
|
|
|
"with height %d", height)
|
|
|
|
}
|
|
|
|
|
|
|
|
block, err := b.chainConn.GetBlockVerbose(blockHash)
|
|
|
|
if err != nil {
|
|
|
|
return nil, fmt.Errorf("unable to get block with hash "+
|
|
|
|
"%v: %v", blockHash, err)
|
|
|
|
}
|
|
|
|
|
|
|
|
for txIndex, txHash := range block.Tx {
|
|
|
|
// If we're able to find the transaction in this block,
|
|
|
|
// return its confirmation details.
|
|
|
|
if txHash == targetTxidStr {
|
|
|
|
return &chainntnfs.TxConfirmation{
|
|
|
|
BlockHash: blockHash,
|
|
|
|
BlockHeight: height,
|
|
|
|
TxIndex: uint32(txIndex),
|
|
|
|
}, nil
|
|
|
|
}
|
|
|
|
}
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
2018-02-26 21:37:58 +03:00
|
|
|
|
|
|
|
// If we reach here, then we were not able to find the transaction
|
|
|
|
// within a block, so we avoid returning an error.
|
|
|
|
return nil, nil
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// notifyBlockEpochs notifies all registered block epoch clients of the newly
|
|
|
|
// connected block to the main chain.
|
|
|
|
func (b *BitcoindNotifier) notifyBlockEpochs(newHeight int32, newSha *chainhash.Hash) {
|
|
|
|
epoch := &chainntnfs.BlockEpoch{
|
|
|
|
Height: newHeight,
|
|
|
|
Hash: newSha,
|
|
|
|
}
|
|
|
|
|
|
|
|
for _, epochClient := range b.blockEpochClients {
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
select {
|
2017-11-10 03:30:20 +03:00
|
|
|
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
case epochClient.epochQueue.ChanIn() <- epoch:
|
2017-11-10 03:30:20 +03:00
|
|
|
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
case <-epochClient.cancelChan:
|
2017-11-10 03:30:20 +03:00
|
|
|
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
case <-b.quit:
|
|
|
|
}
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// spendNotification couples a target outpoint along with the channel used for
|
|
|
|
// notifications once a spend of the outpoint has been detected.
|
|
|
|
type spendNotification struct {
|
|
|
|
targetOutpoint *wire.OutPoint
|
|
|
|
|
|
|
|
spendChan chan *chainntnfs.SpendDetail
|
|
|
|
|
|
|
|
spendID uint64
|
2018-02-26 21:37:58 +03:00
|
|
|
|
|
|
|
heightHint uint32
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// spendCancel is a message sent to the BitcoindNotifier when a client wishes
|
|
|
|
// to cancel an outstanding spend notification that has yet to be dispatched.
|
|
|
|
type spendCancel struct {
|
|
|
|
// op is the target outpoint of the notification to be cancelled.
|
|
|
|
op wire.OutPoint
|
|
|
|
|
|
|
|
// spendID the ID of the notification to cancel.
|
|
|
|
spendID uint64
|
|
|
|
}
|
|
|
|
|
|
|
|
// RegisterSpendNtfn registers an intent to be notified once the target
|
|
|
|
// outpoint has been spent by a transaction on-chain. Once a spend of the target
|
|
|
|
// outpoint has been detected, the details of the spending event will be sent
|
2018-02-26 21:37:58 +03:00
|
|
|
// across the 'Spend' channel. The heightHint should represent the earliest
|
|
|
|
// height in the chain where the transaction could have been spent in.
|
2017-11-10 03:30:20 +03:00
|
|
|
func (b *BitcoindNotifier) RegisterSpendNtfn(outpoint *wire.OutPoint,
|
2018-07-17 10:13:06 +03:00
|
|
|
heightHint uint32) (*chainntnfs.SpendEvent, error) {
|
2017-11-10 03:30:20 +03:00
|
|
|
|
|
|
|
ntfn := &spendNotification{
|
|
|
|
targetOutpoint: outpoint,
|
|
|
|
spendChan: make(chan *chainntnfs.SpendDetail, 1),
|
|
|
|
spendID: atomic.AddUint64(&b.spendClientCounter, 1),
|
|
|
|
}
|
|
|
|
|
|
|
|
select {
|
|
|
|
case <-b.quit:
|
|
|
|
return nil, ErrChainNotifierShuttingDown
|
|
|
|
case b.notificationRegistry <- ntfn:
|
|
|
|
}
|
|
|
|
|
2018-03-28 11:22:41 +03:00
|
|
|
if err := b.chainConn.NotifySpent([]*wire.OutPoint{outpoint}); err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
|
2018-02-26 21:37:58 +03:00
|
|
|
// The following conditional checks to ensure that when a spend
|
|
|
|
// notification is registered, the output hasn't already been spent. If
|
|
|
|
// the output is no longer in the UTXO set, the chain will be rescanned
|
|
|
|
// from the point where the output was added. The rescan will dispatch
|
|
|
|
// the notification.
|
|
|
|
txOut, err := b.chainConn.GetTxOut(&outpoint.Hash, outpoint.Index, true)
|
2017-11-10 03:30:20 +03:00
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
|
2018-02-26 21:37:58 +03:00
|
|
|
if txOut == nil {
|
|
|
|
// First, we'll attempt to retrieve the transaction's block hash
|
|
|
|
// using the backend's transaction index.
|
|
|
|
tx, err := b.chainConn.GetRawTransactionVerbose(&outpoint.Hash)
|
2017-11-10 03:30:20 +03:00
|
|
|
if err != nil {
|
2018-02-26 21:37:58 +03:00
|
|
|
// Avoid returning an error if the transaction was not
|
|
|
|
// found to proceed with fallback methods.
|
2017-11-10 03:30:20 +03:00
|
|
|
jsonErr, ok := err.(*btcjson.RPCError)
|
|
|
|
if !ok || jsonErr.Code != btcjson.ErrRPCNoTxInfo {
|
2018-02-26 21:37:58 +03:00
|
|
|
return nil, fmt.Errorf("unable to query for "+
|
|
|
|
"txid %v: %v", outpoint.Hash, err)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
var blockHash *chainhash.Hash
|
|
|
|
if tx != nil && tx.BlockHash != "" {
|
|
|
|
// If we're able to retrieve a valid block hash from the
|
|
|
|
// transaction, then we'll use it as our rescan starting
|
|
|
|
// point.
|
|
|
|
blockHash, err = chainhash.NewHashFromStr(tx.BlockHash)
|
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
// Otherwise, we'll attempt to retrieve the hash for the
|
|
|
|
// block at the heightHint.
|
|
|
|
blockHash, err = b.chainConn.GetBlockHash(
|
|
|
|
int64(heightHint),
|
|
|
|
)
|
|
|
|
if err != nil {
|
2017-11-10 03:30:20 +03:00
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-04-07 04:25:41 +03:00
|
|
|
// We'll only scan old blocks if the transaction has actually
|
2018-01-29 01:46:54 +03:00
|
|
|
// been included within a block. Otherwise, we'll encounter an
|
|
|
|
// error when scanning for blocks. This can happens in the case
|
|
|
|
// of a race condition, wherein the output itself is unspent,
|
|
|
|
// and only arrives in the mempool after the getxout call.
|
2018-02-26 21:37:58 +03:00
|
|
|
if blockHash != nil {
|
2018-04-07 04:25:41 +03:00
|
|
|
// Rescan all the blocks until the current one.
|
2018-04-18 06:08:50 +03:00
|
|
|
startHeight, err := b.chainConn.GetBlockHeight(
|
|
|
|
blockHash,
|
|
|
|
)
|
2017-11-10 03:30:20 +03:00
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
2018-04-07 04:25:41 +03:00
|
|
|
|
|
|
|
_, endHeight, err := b.chainConn.GetBestBlock()
|
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
for i := startHeight; i <= endHeight; i++ {
|
|
|
|
blockHash, err := b.chainConn.GetBlockHash(int64(i))
|
2017-11-10 03:30:20 +03:00
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
2018-04-07 04:25:41 +03:00
|
|
|
block, err := b.chainConn.GetBlock(blockHash)
|
|
|
|
if err != nil {
|
|
|
|
return nil, err
|
|
|
|
}
|
|
|
|
for _, tx := range block.Transactions {
|
|
|
|
for _, in := range tx.TxIn {
|
|
|
|
if in.PreviousOutPoint == *outpoint {
|
|
|
|
relTx := chain.RelevantTx{
|
|
|
|
TxRecord: &wtxmgr.TxRecord{
|
|
|
|
MsgTx: *tx,
|
|
|
|
Hash: tx.TxHash(),
|
|
|
|
Received: block.Header.Timestamp,
|
|
|
|
},
|
|
|
|
Block: &wtxmgr.BlockMeta{
|
|
|
|
Block: wtxmgr.Block{
|
|
|
|
Hash: block.BlockHash(),
|
|
|
|
Height: i,
|
|
|
|
},
|
|
|
|
Time: block.Header.Timestamp,
|
|
|
|
},
|
|
|
|
}
|
|
|
|
select {
|
|
|
|
case <-b.quit:
|
|
|
|
return nil, ErrChainNotifierShuttingDown
|
|
|
|
case b.notificationRegistry <- relTx:
|
|
|
|
}
|
|
|
|
break out
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return &chainntnfs.SpendEvent{
|
|
|
|
Spend: ntfn.spendChan,
|
|
|
|
Cancel: func() {
|
|
|
|
cancel := &spendCancel{
|
|
|
|
op: *outpoint,
|
|
|
|
spendID: ntfn.spendID,
|
|
|
|
}
|
|
|
|
|
|
|
|
// Submit spend cancellation to notification dispatcher.
|
|
|
|
select {
|
|
|
|
case b.notificationCancels <- cancel:
|
|
|
|
// Cancellation is being handled, drain the spend chan until it is
|
|
|
|
// closed before yielding to the caller.
|
|
|
|
for {
|
|
|
|
select {
|
|
|
|
case _, ok := <-ntfn.spendChan:
|
|
|
|
if !ok {
|
|
|
|
return
|
|
|
|
}
|
|
|
|
case <-b.quit:
|
|
|
|
return
|
|
|
|
}
|
|
|
|
}
|
|
|
|
case <-b.quit:
|
|
|
|
}
|
|
|
|
},
|
|
|
|
}, nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// confirmationNotification represents a client's intent to receive a
|
|
|
|
// notification once the target txid reaches numConfirmations confirmations.
|
2018-02-26 21:37:58 +03:00
|
|
|
type confirmationNotification struct {
|
2017-11-10 03:30:20 +03:00
|
|
|
chainntnfs.ConfNtfn
|
2018-02-26 21:37:58 +03:00
|
|
|
heightHint uint32
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// RegisterConfirmationsNtfn registers a notification with BitcoindNotifier
|
|
|
|
// which will be triggered once the txid reaches numConfs number of
|
|
|
|
// confirmations.
|
|
|
|
func (b *BitcoindNotifier) RegisterConfirmationsNtfn(txid *chainhash.Hash,
|
2018-02-26 21:37:58 +03:00
|
|
|
numConfs, heightHint uint32) (*chainntnfs.ConfirmationEvent, error) {
|
2017-11-10 03:30:20 +03:00
|
|
|
|
2018-02-26 21:37:58 +03:00
|
|
|
ntfn := &confirmationNotification{
|
|
|
|
ConfNtfn: chainntnfs.ConfNtfn{
|
2017-11-10 03:30:20 +03:00
|
|
|
TxID: txid,
|
|
|
|
NumConfirmations: numConfs,
|
2018-03-19 21:48:44 +03:00
|
|
|
Event: chainntnfs.NewConfirmationEvent(numConfs),
|
2017-11-10 03:30:20 +03:00
|
|
|
},
|
2018-02-26 21:37:58 +03:00
|
|
|
heightHint: heightHint,
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
select {
|
|
|
|
case <-b.quit:
|
|
|
|
return nil, ErrChainNotifierShuttingDown
|
|
|
|
case b.notificationRegistry <- ntfn:
|
|
|
|
return ntfn.Event, nil
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// blockEpochRegistration represents a client's intent to receive a
|
|
|
|
// notification with each newly connected block.
|
|
|
|
type blockEpochRegistration struct {
|
|
|
|
epochID uint64
|
|
|
|
|
|
|
|
epochChan chan *chainntnfs.BlockEpoch
|
|
|
|
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
epochQueue *chainntnfs.ConcurrentQueue
|
|
|
|
|
2017-11-10 03:30:20 +03:00
|
|
|
cancelChan chan struct{}
|
|
|
|
|
|
|
|
wg sync.WaitGroup
|
|
|
|
}
|
|
|
|
|
|
|
|
// epochCancel is a message sent to the BitcoindNotifier when a client wishes
|
|
|
|
// to cancel an outstanding epoch notification that has yet to be dispatched.
|
|
|
|
type epochCancel struct {
|
|
|
|
epochID uint64
|
|
|
|
}
|
|
|
|
|
|
|
|
// RegisterBlockEpochNtfn returns a BlockEpochEvent which subscribes the
|
|
|
|
// caller to receive notifications, of each new block connected to the main
|
|
|
|
// chain.
|
|
|
|
func (b *BitcoindNotifier) RegisterBlockEpochNtfn() (*chainntnfs.BlockEpochEvent, error) {
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
reg := &blockEpochRegistration{
|
|
|
|
epochQueue: chainntnfs.NewConcurrentQueue(20),
|
2017-11-10 03:30:20 +03:00
|
|
|
epochChan: make(chan *chainntnfs.BlockEpoch, 20),
|
|
|
|
cancelChan: make(chan struct{}),
|
|
|
|
epochID: atomic.AddUint64(&b.epochClientCounter, 1),
|
|
|
|
}
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
reg.epochQueue.Start()
|
|
|
|
|
|
|
|
// Before we send the request to the main goroutine, we'll launch a new
|
|
|
|
// goroutine to proxy items added to our queue to the client itself.
|
|
|
|
// This ensures that all notifications are received *in order*.
|
|
|
|
reg.wg.Add(1)
|
|
|
|
go func() {
|
|
|
|
defer reg.wg.Done()
|
|
|
|
|
|
|
|
for {
|
|
|
|
select {
|
|
|
|
case ntfn := <-reg.epochQueue.ChanOut():
|
|
|
|
blockNtfn := ntfn.(*chainntnfs.BlockEpoch)
|
|
|
|
select {
|
|
|
|
case reg.epochChan <- blockNtfn:
|
|
|
|
|
|
|
|
case <-reg.cancelChan:
|
|
|
|
return
|
|
|
|
|
|
|
|
case <-b.quit:
|
|
|
|
return
|
|
|
|
}
|
|
|
|
|
|
|
|
case <-reg.cancelChan:
|
|
|
|
return
|
|
|
|
|
|
|
|
case <-b.quit:
|
|
|
|
return
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}()
|
2017-11-10 03:30:20 +03:00
|
|
|
|
|
|
|
select {
|
|
|
|
case <-b.quit:
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
// As we're exiting before the registration could be sent,
|
|
|
|
// we'll stop the queue now ourselves.
|
|
|
|
reg.epochQueue.Stop()
|
|
|
|
|
2017-11-10 03:30:20 +03:00
|
|
|
return nil, errors.New("chainntnfs: system interrupt while " +
|
|
|
|
"attempting to register for block epoch notification.")
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
case b.notificationRegistry <- reg:
|
2017-11-10 03:30:20 +03:00
|
|
|
return &chainntnfs.BlockEpochEvent{
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
Epochs: reg.epochChan,
|
2017-11-10 03:30:20 +03:00
|
|
|
Cancel: func() {
|
|
|
|
cancel := &epochCancel{
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
epochID: reg.epochID,
|
2017-11-10 03:30:20 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// Submit epoch cancellation to notification dispatcher.
|
|
|
|
select {
|
|
|
|
case b.notificationCancels <- cancel:
|
|
|
|
// Cancellation is being handled, drain the epoch channel until it is
|
|
|
|
// closed before yielding to caller.
|
|
|
|
for {
|
|
|
|
select {
|
chainntnfs: ensure all block epoch notifications are sent *in order*
In this commit, we fix a lingering bug related to the way that we
deliver block epoch notifications to end users. Before this commit, we
would launch a new goroutine for *each block*. This was done in order
to ensure that the notification dispatch wouldn’t block the main
goroutine that was dispatching the notifications. This method archived
the goal, but had a nasty side effect that the goroutines could be
re-ordered during scheduling, meaning that in the case of fast
successive blocks, then notifications would be delivered out of order.
Receiving out of order notifications is either disallowed, or can cause
sub-systems that rely on these notifications to get into weird states.
In order to fix this issue, we’ll no longer launch a new goroutine to
deliver each notification to an awaiting client. Instead, each client
will now gain a concurrent in-order queue for notification delivery.
Due to the internal design of chainntnfs.ConcurrentQueue, the caller
should never block, yet the receivers will receive notifications in
order. This change solves the re-ordering issue and also minimizes the
number of goroutines that we’ll create in order to deliver block epoch
notifications.
2018-02-10 03:13:21 +03:00
|
|
|
case _, ok := <-reg.epochChan:
|
2017-11-10 03:30:20 +03:00
|
|
|
if !ok {
|
|
|
|
return
|
|
|
|
}
|
|
|
|
case <-b.quit:
|
|
|
|
return
|
|
|
|
}
|
|
|
|
}
|
|
|
|
case <-b.quit:
|
|
|
|
}
|
|
|
|
},
|
|
|
|
}, nil
|
|
|
|
}
|
|
|
|
}
|