aboutsummaryrefslogtreecommitdiffstats
path: root/eth/downloader/downloader.go
diff options
context:
space:
mode:
Diffstat (limited to 'eth/downloader/downloader.go')
-rw-r--r--eth/downloader/downloader.go310
1 files changed, 134 insertions, 176 deletions
diff --git a/eth/downloader/downloader.go b/eth/downloader/downloader.go
index 153427ee4..5fa18a2e3 100644
--- a/eth/downloader/downloader.go
+++ b/eth/downloader/downloader.go
@@ -74,7 +74,6 @@ var (
errBadPeer = errors.New("action from bad peer ignored")
errStallingPeer = errors.New("peer is stalling")
errNoPeers = errors.New("no peers to keep download active")
- errPendingQueue = errors.New("pending items in queue")
errTimeout = errors.New("timeout")
errEmptyHashSet = errors.New("empty hash set by peer")
errEmptyHeaderSet = errors.New("empty header set by peer")
@@ -90,6 +89,7 @@ var (
errCancelBodyFetch = errors.New("block body download canceled (requested)")
errCancelReceiptFetch = errors.New("receipt download canceled (requested)")
errCancelStateFetch = errors.New("state data download canceled (requested)")
+ errCancelProcessing = errors.New("processing canceled (requested)")
errNoSyncActive = errors.New("no sync active")
)
@@ -129,7 +129,6 @@ type Downloader struct {
// Status
synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing
synchronising int32
- processing int32
notified int32
// Channels
@@ -215,7 +214,7 @@ func (d *Downloader) Progress() (uint64, uint64, uint64) {
// Synchronising returns whether the downloader is currently retrieving blocks.
func (d *Downloader) Synchronising() bool {
- return atomic.LoadInt32(&d.synchronising) > 0 || atomic.LoadInt32(&d.processing) > 0
+ return atomic.LoadInt32(&d.synchronising) > 0
}
// RegisterPeer injects a new download peer into the set of block source to be
@@ -263,9 +262,6 @@ func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode
glog.V(logger.Debug).Infof("Removing peer %v: %v", id, err)
d.dropPeer(id)
- case errPendingQueue:
- glog.V(logger.Debug).Infoln("Synchronisation aborted:", err)
-
default:
glog.V(logger.Warn).Infof("Synchronisation failed: %v", err)
}
@@ -290,10 +286,6 @@ func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode
if atomic.CompareAndSwapInt32(&d.notified, 0, 1) {
glog.V(logger.Info).Infoln("Block synchronisation started")
}
- // Abort if the queue still contains some leftover data
- if d.queue.GetHeadResult() != nil {
- return errPendingQueue
- }
// Reset the queue, peer set and wake channels to clean any internal leftover state
d.queue.Reset()
d.peers.Reset()
@@ -335,7 +327,6 @@ func (d *Downloader) syncWithPeer(p *peer, hash common.Hash, td *big.Int) (err e
defer func() {
// reset on error
if err != nil {
- d.cancel()
d.mux.Post(FailedEvent{err})
} else {
d.mux.Post(DoneEvent{})
@@ -365,23 +356,15 @@ func (d *Downloader) syncWithPeer(p *peer, hash common.Hash, td *big.Int) (err e
d.syncStatsChainHeight = latest
d.syncStatsLock.Unlock()
- // Initiate the sync using a concurrent hash and block retrieval algorithm
+ // Initiate the sync using a concurrent hash and block retrieval algorithm
+ d.queue.Prepare(origin+1, d.mode, 0)
if d.syncInitHook != nil {
d.syncInitHook(origin, latest)
}
- d.queue.Prepare(origin+1, d.mode, 0)
-
- errc := make(chan error, 2)
- go func() { errc <- d.fetchHashes61(p, td, origin+1) }()
- go func() { errc <- d.fetchBlocks61(origin + 1) }()
-
- // If any fetcher fails, cancel the other
- if err := <-errc; err != nil {
- d.cancel()
- <-errc
- return err
- }
- return <-errc
+ return d.spawnSync(
+ func() error { return d.fetchHashes61(p, td, origin+1) },
+ func() error { return d.fetchBlocks61(origin + 1) },
+ )
case p.version >= 62:
// Look up the sync boundaries: the common ancestor and the target block
@@ -405,7 +388,6 @@ func (d *Downloader) syncWithPeer(p *peer, hash common.Hash, td *big.Int) (err e
switch d.mode {
case LightSync:
pivot = latest
-
case FastSync:
// Calculate the new fast/slow sync pivot point
pivotOffset, err := rand.Int(rand.Reader, big.NewInt(int64(fsPivotInterval)))
@@ -426,34 +408,51 @@ func (d *Downloader) syncWithPeer(p *peer, hash common.Hash, td *big.Int) (err e
glog.V(logger.Debug).Infof("Fast syncing until pivot block #%d", pivot)
}
d.queue.Prepare(origin+1, d.mode, pivot)
-
if d.syncInitHook != nil {
d.syncInitHook(origin, latest)
}
- errc := make(chan error, 4)
- go func() { errc <- d.fetchHeaders(p, td, origin+1) }() // Headers are always retrieved
- go func() { errc <- d.fetchBodies(origin + 1) }() // Bodies are retrieved during normal and fast sync
- go func() { errc <- d.fetchReceipts(origin + 1) }() // Receipts are retrieved during fast sync
- go func() { errc <- d.fetchNodeData() }() // Node state data is retrieved during fast sync
-
- // If any fetcher fails, cancel the others
- var fail error
- for i := 0; i < cap(errc); i++ {
- if err := <-errc; err != nil {
- if fail == nil {
- fail = err
- d.cancel()
- }
- }
- }
- return fail
+ return d.spawnSync(
+ func() error { return d.fetchHeaders(p, td, origin+1) }, // Headers are always retrieved
+ func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync
+ func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync
+ func() error { return d.fetchNodeData() }, // Node state data is retrieved during fast sync
+ )
default:
// Something very wrong, stop right here
glog.V(logger.Error).Infof("Unsupported eth protocol: %d", p.version)
return errBadPeer
}
- return nil
+}
+
+// spawnSync runs d.process and all given fetcher functions to completion in
+// separate goroutines, returning the first error that appears.
+func (d *Downloader) spawnSync(fetchers ...func() error) error {
+ var wg sync.WaitGroup
+ errc := make(chan error, len(fetchers)+1)
+ wg.Add(len(fetchers) + 1)
+ go func() { defer wg.Done(); errc <- d.process() }()
+ for _, fn := range fetchers {
+ fn := fn
+ go func() { defer wg.Done(); errc <- fn() }()
+ }
+ // Wait for the first error, then terminate the others.
+ var err error
+ for i := 0; i < len(fetchers)+1; i++ {
+ if i == len(fetchers) {
+ // Close the queue when all fetchers have exited.
+ // This will cause the block processor to end when
+ // it has processed the queue.
+ d.queue.Close()
+ }
+ if err = <-errc; err != nil {
+ break
+ }
+ }
+ d.queue.Close()
+ d.cancel()
+ wg.Wait()
+ return err
}
// cancel cancels all of the operations and resets the queue. It returns true
@@ -470,12 +469,10 @@ func (d *Downloader) cancel() {
}
}
d.cancelLock.Unlock()
-
- // Reset the queue
- d.queue.Reset()
}
// Terminate interrupts the downloader, canceling all pending operations.
+// The downloader cannot be reused after calling Terminate.
func (d *Downloader) Terminate() {
atomic.StoreInt32(&d.interrupt, 1)
d.cancel()
@@ -495,15 +492,6 @@ func (d *Downloader) fetchHeight61(p *peer) (uint64, error) {
case <-d.cancelCh:
return 0, errCancelBlockFetch
- case <-d.headerCh:
- // Out of bounds eth/62 block headers received, ignore them
-
- case <-d.bodyCh:
- // Out of bounds eth/62 block bodies received, ignore them
-
- case <-d.hashCh:
- // Out of bounds hashes received, ignore them
-
case packet := <-d.blockCh:
// Discard anything not from the origin peer
if packet.PeerId() != p.id {
@@ -521,6 +509,16 @@ func (d *Downloader) fetchHeight61(p *peer) (uint64, error) {
case <-timeout:
glog.V(logger.Debug).Infof("%v: head block timeout", p)
return 0, errTimeout
+
+ case <-d.hashCh:
+ // Out of bounds hashes received, ignore them
+
+ case <-d.headerCh:
+ case <-d.bodyCh:
+ case <-d.stateCh:
+ case <-d.receiptCh:
+ // Ignore eth/{62,63} packets because this is eth/61.
+ // These can arrive as a late delivery from a previous sync.
}
}
}
@@ -571,18 +569,19 @@ func (d *Downloader) findAncestor61(p *peer) (uint64, error) {
}
}
+ case <-timeout:
+ glog.V(logger.Debug).Infof("%v: head hash timeout", p)
+ return 0, errTimeout
+
case <-d.blockCh:
// Out of bounds blocks received, ignore them
case <-d.headerCh:
- // Out of bounds eth/62 block headers received, ignore them
-
case <-d.bodyCh:
- // Out of bounds eth/62 block bodies received, ignore them
-
- case <-timeout:
- glog.V(logger.Debug).Infof("%v: head hash timeout", p)
- return 0, errTimeout
+ case <-d.stateCh:
+ case <-d.receiptCh:
+ // Ignore eth/{62,63} packets because this is eth/61.
+ // These can arrive as a late delivery from a previous sync.
}
}
// If the head fetch already found an ancestor, return
@@ -631,18 +630,19 @@ func (d *Downloader) findAncestor61(p *peer) (uint64, error) {
}
start = check
+ case <-timeout:
+ glog.V(logger.Debug).Infof("%v: search hash timeout", p)
+ return 0, errTimeout
+
case <-d.blockCh:
// Out of bounds blocks received, ignore them
case <-d.headerCh:
- // Out of bounds eth/62 block headers received, ignore them
-
case <-d.bodyCh:
- // Out of bounds eth/62 block bodies received, ignore them
-
- case <-timeout:
- glog.V(logger.Debug).Infof("%v: search hash timeout", p)
- return 0, errTimeout
+ case <-d.stateCh:
+ case <-d.receiptCh:
+ // Ignore eth/{62,63} packets because this is eth/61.
+ // These can arrive as a late delivery from a previous sync.
}
}
}
@@ -676,12 +676,6 @@ func (d *Downloader) fetchHashes61(p *peer, td *big.Int, from uint64) error {
case <-d.cancelCh:
return errCancelHashFetch
- case <-d.headerCh:
- // Out of bounds eth/62 block headers received, ignore them
-
- case <-d.bodyCh:
- // Out of bounds eth/62 block bodies received, ignore them
-
case packet := <-d.hashCh:
// Make sure the active peer is giving us the hashes
if packet.PeerId() != p.id {
@@ -750,6 +744,13 @@ func (d *Downloader) fetchHashes61(p *peer, td *big.Int, from uint64) error {
glog.V(logger.Debug).Infof("%v: hash request timed out", p)
hashTimeoutMeter.Mark(1)
return errTimeout
+
+ case <-d.headerCh:
+ case <-d.bodyCh:
+ case <-d.stateCh:
+ case <-d.receiptCh:
+ // Ignore eth/{62,63} packets because this is eth/61.
+ // These can arrive as a late delivery from a previous sync.
}
}
}
@@ -774,12 +775,6 @@ func (d *Downloader) fetchBlocks61(from uint64) error {
case <-d.cancelCh:
return errCancelBlockFetch
- case <-d.headerCh:
- // Out of bounds eth/62 block headers received, ignore them
-
- case <-d.bodyCh:
- // Out of bounds eth/62 block bodies received, ignore them
-
case packet := <-d.blockCh:
// If the peer was previously banned and failed to deliver it's pack
// in a reasonable time frame, ignore it's message.
@@ -800,7 +795,6 @@ func (d *Downloader) fetchBlocks61(from uint64) error {
peer.Promote()
peer.SetBlocksIdle()
glog.V(logger.Detail).Infof("%s: delivered %d blocks", peer, len(blocks))
- go d.process()
case errInvalidChain:
// The hash chain is invalid (blocks are not ordered properly), abort
@@ -826,7 +820,6 @@ func (d *Downloader) fetchBlocks61(from uint64) error {
peer.Demote()
peer.SetBlocksIdle()
glog.V(logger.Detail).Infof("%s: delivery partially failed: %v", peer, err)
- go d.process()
}
}
// Blocks arrived, try to update the progress
@@ -909,6 +902,13 @@ func (d *Downloader) fetchBlocks61(from uint64) error {
if !throttled && !d.queue.InFlightBlocks() && len(idles) == total {
return errPeersUnavailable
}
+
+ case <-d.headerCh:
+ case <-d.bodyCh:
+ case <-d.stateCh:
+ case <-d.receiptCh:
+ // Ignore eth/{62,63} packets because this is eth/61.
+ // These can arrive as a late delivery from a previous sync.
}
}
}
@@ -941,18 +941,19 @@ func (d *Downloader) fetchHeight(p *peer) (uint64, error) {
}
return headers[0].Number.Uint64(), nil
+ case <-timeout:
+ glog.V(logger.Debug).Infof("%v: head header timeout", p)
+ return 0, errTimeout
+
case <-d.bodyCh:
- // Out of bounds block bodies received, ignore them
+ case <-d.stateCh:
+ case <-d.receiptCh:
+ // Out of bounds delivery, ignore
case <-d.hashCh:
- // Out of bounds eth/61 hashes received, ignore them
-
case <-d.blockCh:
- // Out of bounds eth/61 blocks received, ignore them
-
- case <-timeout:
- glog.V(logger.Debug).Infof("%v: head header timeout", p)
- return 0, errTimeout
+ // Ignore eth/61 packets because this is eth/62+.
+ // These can arrive as a late delivery from a previous sync.
}
}
}
@@ -1008,18 +1009,19 @@ func (d *Downloader) findAncestor(p *peer) (uint64, error) {
}
}
+ case <-timeout:
+ glog.V(logger.Debug).Infof("%v: head header timeout", p)
+ return 0, errTimeout
+
case <-d.bodyCh:
- // Out of bounds block bodies received, ignore them
+ case <-d.stateCh:
+ case <-d.receiptCh:
+ // Out of bounds delivery, ignore
case <-d.hashCh:
- // Out of bounds eth/61 hashes received, ignore them
-
case <-d.blockCh:
- // Out of bounds eth/61 blocks received, ignore them
-
- case <-timeout:
- glog.V(logger.Debug).Infof("%v: head header timeout", p)
- return 0, errTimeout
+ // Ignore eth/61 packets because this is eth/62+.
+ // These can arrive as a late delivery from a previous sync.
}
}
// If the head fetch already found an ancestor, return
@@ -1068,18 +1070,19 @@ func (d *Downloader) findAncestor(p *peer) (uint64, error) {
}
start = check
+ case <-timeout:
+ glog.V(logger.Debug).Infof("%v: search header timeout", p)
+ return 0, errTimeout
+
case <-d.bodyCh:
- // Out of bounds block bodies received, ignore them
+ case <-d.stateCh:
+ case <-d.receiptCh:
+ // Out of bounds delivery, ignore
case <-d.hashCh:
- // Out of bounds eth/61 hashes received, ignore them
-
case <-d.blockCh:
- // Out of bounds eth/61 blocks received, ignore them
-
- case <-timeout:
- glog.V(logger.Debug).Infof("%v: search header timeout", p)
- return 0, errTimeout
+ // Ignore eth/61 packets because this is eth/62+.
+ // These can arrive as a late delivery from a previous sync.
}
}
}
@@ -1141,12 +1144,6 @@ func (d *Downloader) fetchHeaders(p *peer, td *big.Int, from uint64) error {
case <-d.cancelCh:
return errCancelHeaderFetch
- case <-d.hashCh:
- // Out of bounds eth/61 hashes received, ignore them
-
- case <-d.blockCh:
- // Out of bounds eth/61 blocks received, ignore them
-
case packet := <-d.headerCh:
// Make sure the active peer is giving us the headers
if packet.PeerId() != p.id {
@@ -1268,6 +1265,11 @@ func (d *Downloader) fetchHeaders(p *peer, td *big.Int, from uint64) error {
}
}
return nil
+
+ case <-d.hashCh:
+ case <-d.blockCh:
+ // Ignore eth/61 packets because this is eth/62+.
+ // These can arrive as a late delivery from a previous sync.
}
}
}
@@ -1336,10 +1338,8 @@ func (d *Downloader) fetchNodeData() error {
d.cancel()
return
}
- // Processing succeeded, notify state fetcher and processor of continuation
- if d.queue.PendingNodeData() == 0 {
- go d.process()
- } else {
+ // Processing succeeded, notify state fetcher of continuation
+ if d.queue.PendingNodeData() > 0 {
select {
case d.stateWakeCh <- true:
default:
@@ -1348,7 +1348,6 @@ func (d *Downloader) fetchNodeData() error {
// Log a message to the user and return
d.syncStatsLock.Lock()
defer d.syncStatsLock.Unlock()
-
d.syncStatsStateDone += uint64(delivered)
glog.V(logger.Info).Infof("imported %d state entries in %v: processed %d in total", delivered, time.Since(start), d.syncStatsStateDone)
})
@@ -1391,12 +1390,6 @@ func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliv
case <-d.cancelCh:
return errCancel
- case <-d.hashCh:
- // Out of bounds eth/61 hashes received, ignore them
-
- case <-d.blockCh:
- // Out of bounds eth/61 blocks received, ignore them
-
case packet := <-deliveryCh:
// If the peer was previously banned and failed to deliver it's pack
// in a reasonable time frame, ignore it's message.
@@ -1415,7 +1408,6 @@ func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliv
peer.Promote()
setIdle(peer)
glog.V(logger.Detail).Infof("%s: delivered %s %s(s)", peer, packet.Stats(), strings.ToLower(kind))
- go d.process()
case errInvalidChain:
// The hash chain is invalid (blocks are not ordered properly), abort
@@ -1441,7 +1433,6 @@ func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliv
peer.Demote()
setIdle(peer)
glog.V(logger.Detail).Infof("%s: %s delivery partially failed: %v", peer, strings.ToLower(kind), err)
- go d.process()
}
}
// Blocks assembled, try to update the progress
@@ -1508,7 +1499,6 @@ func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliv
}
if progress {
progressed = true
- go d.process()
}
if request == nil {
continue
@@ -1540,51 +1530,23 @@ func (d *Downloader) fetchParts(errCancel error, deliveryCh chan dataPack, deliv
if !progressed && !throttled && !running && len(idles) == total && pending() > 0 {
return errPeersUnavailable
}
+
+ case <-d.hashCh:
+ case <-d.blockCh:
+ // Ignore eth/61 packets because this is eth/62+.
+ // These can arrive as a late delivery from a previous sync.
}
}
}
// process takes fetch results from the queue and tries to import them into the
-// chain. The type of import operation will depend on the result contents:
-// -
-//
-// The algorithmic flow is as follows:
-// - The `processing` flag is swapped to 1 to ensure singleton access
-// - The current `cancel` channel is retrieved to detect sync abortions
-// - Blocks are iteratively taken from the cache and inserted into the chain
-// - When the cache becomes empty, insertion stops
-// - The `processing` flag is swapped back to 0
-// - A post-exit check is made whether new blocks became available
-// - This step is important: it handles a potential race condition between
-// checking for no more work, and releasing the processing "mutex". In
-// between these state changes, a block may have arrived, but a processing
-// attempt denied, so we need to re-enter to ensure the block isn't left
-// to idle in the cache.
-func (d *Downloader) process() {
- // Make sure only one goroutine is ever allowed to process blocks at once
- if !atomic.CompareAndSwapInt32(&d.processing, 0, 1) {
- return
- }
- // If the processor just exited, but there are freshly pending items, try to
- // reenter. This is needed because the goroutine spinned up for processing
- // the fresh results might have been rejected entry to to this present thread
- // not yet releasing the `processing` state.
- defer func() {
- if atomic.LoadInt32(&d.interrupt) == 0 && d.queue.GetHeadResult() != nil {
- d.process()
- }
- }()
- // Release the lock upon exit (note, before checking for reentry!)
- // the import statistics to zero.
- defer atomic.StoreInt32(&d.processing, 0)
-
- // Repeat the processing as long as there are results to process
+// chain. The type of import operation will depend on the result contents.
+func (d *Downloader) process() error {
+ pivot := d.queue.FastSyncPivot()
for {
- // Fetch the next batch of results
- pivot := d.queue.FastSyncPivot() // Fetch pivot before results to prevent reset race
- results := d.queue.TakeResults()
+ results := d.queue.WaitResults()
if len(results) == 0 {
- return
+ return nil // queue empty
}
if d.chainInsertHook != nil {
d.chainInsertHook(results)
@@ -1597,7 +1559,7 @@ func (d *Downloader) process() {
for len(results) != 0 {
// Check for any termination requests
if atomic.LoadInt32(&d.interrupt) == 1 {
- return
+ return errCancelProcessing
}
// Retrieve the a batch of results to import
var (
@@ -1633,8 +1595,7 @@ func (d *Downloader) process() {
}
if err != nil {
glog.V(logger.Debug).Infof("Result #%d [%x…] processing failed: %v", results[index].Header.Number, results[index].Header.Hash().Bytes()[:4], err)
- d.cancel()
- return
+ return err
}
// Shift the results to the next batch
results = results[items:]
@@ -1685,19 +1646,16 @@ func (d *Downloader) deliver(id string, destCh chan dataPack, packet dataPack, i
dropMeter.Mark(int64(packet.Items()))
}
}()
- // Make sure the downloader is active
- if atomic.LoadInt32(&d.synchronising) == 0 {
- return errNoSyncActive
- }
// Deliver or abort if the sync is canceled while queuing
d.cancelLock.RLock()
cancel := d.cancelCh
d.cancelLock.RUnlock()
-
+ if cancel == nil {
+ return errNoSyncActive
+ }
select {
case destCh <- packet:
return nil
-
case <-cancel:
return errNoSyncActive
}
generated by cgit (git 2.34.1) at 2024-11-16 09:25:28 +0800