2015-07-06 19:54:22 -05:00
|
|
|
// Copyright 2015 The go-ethereum Authors
|
2015-07-22 11:48:40 -05:00
|
|
|
// This file is part of the go-ethereum library.
|
2015-07-06 19:54:22 -05:00
|
|
|
//
|
2015-07-23 11:35:11 -05:00
|
|
|
// The go-ethereum library is free software: you can redistribute it and/or modify
|
2015-07-06 19:54:22 -05:00
|
|
|
// it under the terms of the GNU Lesser General Public License as published by
|
|
|
|
// the Free Software Foundation, either version 3 of the License, or
|
|
|
|
// (at your option) any later version.
|
|
|
|
//
|
2015-07-22 11:48:40 -05:00
|
|
|
// The go-ethereum library is distributed in the hope that it will be useful,
|
2015-07-06 19:54:22 -05:00
|
|
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
2015-07-22 11:48:40 -05:00
|
|
|
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
2015-07-06 19:54:22 -05:00
|
|
|
// GNU Lesser General Public License for more details.
|
|
|
|
//
|
|
|
|
// You should have received a copy of the GNU Lesser General Public License
|
2015-07-22 11:48:40 -05:00
|
|
|
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
|
2015-07-06 19:54:22 -05:00
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
// Contains the active peer-set of the downloader, maintaining both failures
|
|
|
|
// as well as reputation metrics to prioritize the block retrievals.
|
|
|
|
|
2015-04-12 05:38:25 -05:00
|
|
|
package downloader
|
|
|
|
|
|
|
|
import (
|
2015-04-12 06:33:42 -05:00
|
|
|
"errors"
|
2016-07-25 07:14:14 -05:00
|
|
|
"math/big"
|
2015-04-12 05:38:25 -05:00
|
|
|
"sync"
|
2015-06-03 06:39:21 -05:00
|
|
|
"time"
|
2015-04-12 05:38:25 -05:00
|
|
|
|
|
|
|
"github.com/ethereum/go-ethereum/common"
|
2021-02-18 10:54:29 -06:00
|
|
|
"github.com/ethereum/go-ethereum/eth/protocols/eth"
|
eth/downloader: separate state sync from queue (#14460)
* eth/downloader: separate state sync from queue
Scheduling of state node downloads hogged the downloader queue lock when
new requests were scheduled. This caused timeouts for other requests.
With this change, state sync is fully independent of all other downloads
and doesn't involve the queue at all.
State sync is started and checked on in processContent. This is slightly
awkward because processContent doesn't have a select loop. Instead, the
queue is closed by an auxiliary goroutine when state sync fails. We
tried several alternatives to this but settled on the current approach
because it's the least amount of change overall.
Handling of the pivot block has changed slightly: the queue previously
prevented import of pivot block receipts before the state of the pivot
block was available. In this commit, the receipt will be imported before
the state. This causes an annoyance where the pivot block is committed
as fast block head even when state downloads fail. Stay tuned for more
updates in this area ;)
* eth/downloader: remove cancelTimeout channel
* eth/downloader: retry state requests on timeout
* eth/downloader: improve comment
* eth/downloader: mark peers idle when state sync is done
* eth/downloader: move pivot block splitting to processContent
This change also ensures that pivot block receipts aren't imported
before the pivot block itself.
* eth/downloader: limit state node retries
* eth/downloader: improve state node error handling and retry check
* eth/downloader: remove maxStateNodeRetries
It fails the sync too much.
* eth/downloader: remove last use of cancelCh in statesync.go
Fixes TestDeliverHeadersHang*Fast and (hopefully)
the weird cancellation behaviour at the end of fast sync.
* eth/downloader: fix leak in runStateSync
* eth/downloader: don't run processFullSyncContent in LightSync mode
* eth/downloader: improve comments
* eth/downloader: fix vet, megacheck
* eth/downloader: remove unrequested tasks anyway
* eth/downloader, trie: various polishes around duplicate items
This commit explicitly tracks duplicate and unexpected state
delieveries done against a trie Sync structure, also adding
there to import info logs.
The commit moves the db batch used to commit trie changes one
level deeper so its flushed after every node insertion. This
is needed to avoid a lot of duplicate retrievals caused by
inconsistencies between Sync internals and database. A better
approach is to track not-yet-written states in trie.Sync and
flush on commit, but I'm focuing on correctness first now.
The commit fixes a regression around pivot block fail count.
The counter previously was reset to 1 if and only if a sync
cycle progressed (inserted at least 1 entry to the database).
The current code reset it already if a node was delivered,
which is not stong enough, because unless it ends up written
to disk, an attacker can just loop and attack ad infinitum.
The commit also fixes a regression around state deliveries
and timeouts. The old downloader tracked if a delivery is
stale (none of the deliveries were requestedt), in which
case it didn't mark the node idle and did not send further
requests, since it signals a past timeout. The current code
did mark it idle even on stale deliveries, which eventually
caused two requests to be in flight at the same time, making
the deliveries always stale and mass duplicating retrievals
between multiple peers.
* eth/downloader: fix state request leak
This commit fixes the hang seen sometimes while doing the state
sync. The cause of the hang was a rare combination of events:
request state data from peer, peer drops and reconnects almost
immediately. This caused a new download task to be assigned to
the peer, overwriting the old one still waiting for a timeout,
which in turned leaked the requests out, never to be retried.
The fix is to ensure that a task assignment moves any pending
one back into the retry queue.
The commit also fixes a regression with peer dropping due to
stalls. The current code considered a peer stalling if they
timed out delivering 1 item. However, the downloader never
requests only one, the minimum is 2 (attempt to fine tune
estimated latency/bandwidth). The fix is simply to drop if
a timeout is detected at 2 items.
Apart from the above bugfixes, the commit contains some code
polishes I made while debugging the hang.
* core, eth, trie: support batched trie sync db writes
* trie: rename SyncMemCache to syncMemBatch
2017-06-22 07:26:03 -05:00
|
|
|
"github.com/ethereum/go-ethereum/event"
|
2017-02-24 10:23:03 -06:00
|
|
|
"github.com/ethereum/go-ethereum/log"
|
2021-05-19 07:09:03 -05:00
|
|
|
"github.com/ethereum/go-ethereum/p2p/msgrate"
|
2015-04-12 05:38:25 -05:00
|
|
|
)
|
|
|
|
|
2015-10-29 11:37:26 -05:00
|
|
|
const (
|
2021-05-19 07:09:03 -05:00
|
|
|
maxLackingHashes = 4096 // Maximum number of entries allowed on the list or lacking items
|
2015-10-29 11:37:26 -05:00
|
|
|
)
|
2015-11-04 04:18:48 -06:00
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
var (
|
|
|
|
errAlreadyRegistered = errors.New("peer is already registered")
|
|
|
|
errNotRegistered = errors.New("peer is not registered")
|
|
|
|
)
|
2015-04-13 09:38:32 -05:00
|
|
|
|
2017-07-05 04:42:37 -05:00
|
|
|
// peerConnection represents an active peer from which hashes and blocks are retrieved.
|
2017-06-28 07:25:08 -05:00
|
|
|
type peerConnection struct {
|
2016-07-25 07:14:14 -05:00
|
|
|
id string // Unique identifier of the peer
|
2015-05-11 06:26:20 -05:00
|
|
|
|
2021-05-19 07:09:03 -05:00
|
|
|
rates *msgrate.Tracker // Tracker to hone in on the number of items retrievable per second
|
2015-10-29 11:37:26 -05:00
|
|
|
lacking map[common.Hash]struct{} // Set of hashes not to request (didn't have previously)
|
2015-05-11 06:26:20 -05:00
|
|
|
|
2017-06-28 07:25:08 -05:00
|
|
|
peer Peer
|
2015-09-28 11:27:31 -05:00
|
|
|
|
2020-12-14 03:27:15 -06:00
|
|
|
version uint // Eth protocol version number to switch strategies
|
2017-03-02 07:06:16 -06:00
|
|
|
log log.Logger // Contextual logger to add extra infos to peer logs
|
2015-10-29 11:37:26 -05:00
|
|
|
lock sync.RWMutex
|
2015-04-18 13:25:55 -05:00
|
|
|
}
|
|
|
|
|
2017-07-05 04:42:37 -05:00
|
|
|
// LightPeer encapsulates the methods required to synchronise with a remote light peer.
|
2017-07-03 09:17:12 -05:00
|
|
|
type LightPeer interface {
|
2017-06-28 07:25:08 -05:00
|
|
|
Head() (common.Hash, *big.Int)
|
2021-11-26 05:26:03 -06:00
|
|
|
RequestHeadersByHash(common.Hash, int, int, bool, chan *eth.Response) (*eth.Request, error)
|
|
|
|
RequestHeadersByNumber(uint64, int, int, bool, chan *eth.Response) (*eth.Request, error)
|
2017-07-03 09:17:12 -05:00
|
|
|
}
|
|
|
|
|
2017-07-05 04:42:37 -05:00
|
|
|
// Peer encapsulates the methods required to synchronise with a remote full peer.
|
2017-07-03 09:17:12 -05:00
|
|
|
type Peer interface {
|
|
|
|
LightPeer
|
2021-11-26 05:26:03 -06:00
|
|
|
RequestBodies([]common.Hash, chan *eth.Response) (*eth.Request, error)
|
|
|
|
RequestReceipts([]common.Hash, chan *eth.Response) (*eth.Request, error)
|
2017-06-28 07:25:08 -05:00
|
|
|
}
|
|
|
|
|
2017-07-03 09:17:12 -05:00
|
|
|
// lightPeerWrapper wraps a LightPeer struct, stubbing out the Peer-only methods.
|
|
|
|
type lightPeerWrapper struct {
|
|
|
|
peer LightPeer
|
|
|
|
}
|
2017-02-24 10:23:03 -06:00
|
|
|
|
2017-07-03 09:17:12 -05:00
|
|
|
func (w *lightPeerWrapper) Head() (common.Hash, *big.Int) { return w.peer.Head() }
|
2021-11-26 05:26:03 -06:00
|
|
|
func (w *lightPeerWrapper) RequestHeadersByHash(h common.Hash, amount int, skip int, reverse bool, sink chan *eth.Response) (*eth.Request, error) {
|
|
|
|
return w.peer.RequestHeadersByHash(h, amount, skip, reverse, sink)
|
2017-07-03 09:17:12 -05:00
|
|
|
}
|
2021-11-26 05:26:03 -06:00
|
|
|
func (w *lightPeerWrapper) RequestHeadersByNumber(i uint64, amount int, skip int, reverse bool, sink chan *eth.Response) (*eth.Request, error) {
|
|
|
|
return w.peer.RequestHeadersByNumber(i, amount, skip, reverse, sink)
|
2017-07-03 09:17:12 -05:00
|
|
|
}
|
2021-11-26 05:26:03 -06:00
|
|
|
func (w *lightPeerWrapper) RequestBodies([]common.Hash, chan *eth.Response) (*eth.Request, error) {
|
2017-07-03 09:17:12 -05:00
|
|
|
panic("RequestBodies not supported in light client mode sync")
|
|
|
|
}
|
2021-11-26 05:26:03 -06:00
|
|
|
func (w *lightPeerWrapper) RequestReceipts([]common.Hash, chan *eth.Response) (*eth.Request, error) {
|
2017-07-03 09:17:12 -05:00
|
|
|
panic("RequestReceipts not supported in light client mode sync")
|
|
|
|
}
|
|
|
|
|
2017-07-05 04:42:37 -05:00
|
|
|
// newPeerConnection creates a new downloader peer.
|
2020-12-14 03:27:15 -06:00
|
|
|
func newPeerConnection(id string, version uint, peer Peer, logger log.Logger) *peerConnection {
|
2017-06-28 07:25:08 -05:00
|
|
|
return &peerConnection{
|
2015-10-29 11:37:26 -05:00
|
|
|
id: id,
|
|
|
|
lacking: make(map[common.Hash]struct{}),
|
2020-07-24 02:46:26 -05:00
|
|
|
peer: peer,
|
2015-08-14 13:25:41 -05:00
|
|
|
version: version,
|
2017-03-02 07:06:16 -06:00
|
|
|
log: logger,
|
2015-04-13 09:38:32 -05:00
|
|
|
}
|
2015-05-11 06:26:20 -05:00
|
|
|
}
|
2015-04-13 09:38:32 -05:00
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
// Reset clears the internal state of a peer entity.
|
2017-06-28 07:25:08 -05:00
|
|
|
func (p *peerConnection) Reset() {
|
2015-10-29 11:37:26 -05:00
|
|
|
p.lock.Lock()
|
|
|
|
defer p.lock.Unlock()
|
|
|
|
|
2015-11-04 04:18:48 -06:00
|
|
|
p.lacking = make(map[common.Hash]struct{})
|
2015-04-13 09:38:32 -05:00
|
|
|
}
|
|
|
|
|
2021-11-26 05:26:03 -06:00
|
|
|
// UpdateHeaderRate updates the peer's estimated header retrieval throughput with
|
|
|
|
// the current measurement.
|
|
|
|
func (p *peerConnection) UpdateHeaderRate(delivered int, elapsed time.Duration) {
|
|
|
|
p.rates.Update(eth.BlockHeadersMsg, elapsed, delivered)
|
2015-09-28 11:27:31 -05:00
|
|
|
}
|
|
|
|
|
2021-11-26 05:26:03 -06:00
|
|
|
// UpdateBodyRate updates the peer's estimated body retrieval throughput with the
|
|
|
|
// current measurement.
|
|
|
|
func (p *peerConnection) UpdateBodyRate(delivered int, elapsed time.Duration) {
|
|
|
|
p.rates.Update(eth.BlockBodiesMsg, elapsed, delivered)
|
2015-10-05 11:37:56 -05:00
|
|
|
}
|
|
|
|
|
2021-11-26 05:26:03 -06:00
|
|
|
// UpdateReceiptRate updates the peer's estimated receipt retrieval throughput
|
|
|
|
// with the current measurement.
|
|
|
|
func (p *peerConnection) UpdateReceiptRate(delivered int, elapsed time.Duration) {
|
|
|
|
p.rates.Update(eth.ReceiptsMsg, elapsed, delivered)
|
2016-02-25 10:36:42 -06:00
|
|
|
}
|
|
|
|
|
2021-11-26 05:26:03 -06:00
|
|
|
// HeaderCapacity retrieves the peer's header download allowance based on its
|
2016-02-25 10:36:42 -06:00
|
|
|
// previously discovered throughput.
|
2017-06-28 07:25:08 -05:00
|
|
|
func (p *peerConnection) HeaderCapacity(targetRTT time.Duration) int {
|
2021-05-27 11:43:55 -05:00
|
|
|
cap := p.rates.Capacity(eth.BlockHeadersMsg, targetRTT)
|
2021-05-19 07:09:03 -05:00
|
|
|
if cap > MaxHeaderFetch {
|
|
|
|
cap = MaxHeaderFetch
|
|
|
|
}
|
|
|
|
return cap
|
2016-02-25 10:36:42 -06:00
|
|
|
}
|
|
|
|
|
2021-11-26 05:26:03 -06:00
|
|
|
// BodyCapacity retrieves the peer's body download allowance based on its
|
2015-10-29 11:37:26 -05:00
|
|
|
// previously discovered throughput.
|
2021-11-26 05:26:03 -06:00
|
|
|
func (p *peerConnection) BodyCapacity(targetRTT time.Duration) int {
|
2021-05-27 11:43:55 -05:00
|
|
|
cap := p.rates.Capacity(eth.BlockBodiesMsg, targetRTT)
|
2021-05-19 07:09:03 -05:00
|
|
|
if cap > MaxBlockFetch {
|
|
|
|
cap = MaxBlockFetch
|
|
|
|
}
|
|
|
|
return cap
|
2015-06-03 06:39:21 -05:00
|
|
|
}
|
|
|
|
|
2015-10-29 11:37:26 -05:00
|
|
|
// ReceiptCapacity retrieves the peers receipt download allowance based on its
|
|
|
|
// previously discovered throughput.
|
2017-06-28 07:25:08 -05:00
|
|
|
func (p *peerConnection) ReceiptCapacity(targetRTT time.Duration) int {
|
2021-05-27 11:43:55 -05:00
|
|
|
cap := p.rates.Capacity(eth.ReceiptsMsg, targetRTT)
|
2021-05-19 07:09:03 -05:00
|
|
|
if cap > MaxReceiptFetch {
|
|
|
|
cap = MaxReceiptFetch
|
|
|
|
}
|
|
|
|
return cap
|
2015-05-11 06:26:20 -05:00
|
|
|
}
|
2015-04-12 05:38:25 -05:00
|
|
|
|
2015-11-04 04:18:48 -06:00
|
|
|
// MarkLacking appends a new entity to the set of items (blocks, receipts, states)
|
|
|
|
// that a peer is known not to have (i.e. have been requested before). If the
|
|
|
|
// set reaches its maximum allowed capacity, items are randomly dropped off.
|
2017-06-28 07:25:08 -05:00
|
|
|
func (p *peerConnection) MarkLacking(hash common.Hash) {
|
2015-10-29 11:37:26 -05:00
|
|
|
p.lock.Lock()
|
|
|
|
defer p.lock.Unlock()
|
2015-11-04 04:18:48 -06:00
|
|
|
|
|
|
|
for len(p.lacking) >= maxLackingHashes {
|
2017-01-06 08:52:03 -06:00
|
|
|
for drop := range p.lacking {
|
2015-11-04 04:18:48 -06:00
|
|
|
delete(p.lacking, drop)
|
|
|
|
break
|
|
|
|
}
|
|
|
|
}
|
|
|
|
p.lacking[hash] = struct{}{}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Lacks retrieves whether the hash of a blockchain item is on the peers lacking
|
|
|
|
// list (i.e. whether we know that the peer does not have it).
|
2017-06-28 07:25:08 -05:00
|
|
|
func (p *peerConnection) Lacks(hash common.Hash) bool {
|
2015-10-29 11:37:26 -05:00
|
|
|
p.lock.RLock()
|
|
|
|
defer p.lock.RUnlock()
|
2015-11-04 04:18:48 -06:00
|
|
|
|
|
|
|
_, ok := p.lacking[hash]
|
|
|
|
return ok
|
|
|
|
}
|
|
|
|
|
2021-11-26 05:26:03 -06:00
|
|
|
// peeringEvent is sent on the peer event feed when a remote peer connects or
|
|
|
|
// disconnects.
|
|
|
|
type peeringEvent struct {
|
|
|
|
peer *peerConnection
|
|
|
|
join bool
|
|
|
|
}
|
|
|
|
|
2016-02-25 10:36:42 -06:00
|
|
|
// peerSet represents the collection of active peer participating in the chain
|
2015-05-11 06:26:20 -05:00
|
|
|
// download procedure.
|
|
|
|
type peerSet struct {
|
2021-11-26 05:26:03 -06:00
|
|
|
peers map[string]*peerConnection
|
|
|
|
rates *msgrate.Trackers // Set of rate trackers to give the sync a common beat
|
|
|
|
events event.Feed // Feed to publish peer lifecycle events on
|
2021-05-19 07:09:03 -05:00
|
|
|
|
|
|
|
lock sync.RWMutex
|
2015-04-12 05:38:25 -05:00
|
|
|
}
|
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
// newPeerSet creates a new peer set top track the active download sources.
|
|
|
|
func newPeerSet() *peerSet {
|
|
|
|
return &peerSet{
|
2017-06-28 07:25:08 -05:00
|
|
|
peers: make(map[string]*peerConnection),
|
2021-05-19 07:09:03 -05:00
|
|
|
rates: msgrate.NewTrackers(log.New("proto", "eth")),
|
2015-04-18 11:54:57 -05:00
|
|
|
}
|
2015-04-12 05:38:25 -05:00
|
|
|
}
|
|
|
|
|
2021-11-26 05:26:03 -06:00
|
|
|
// SubscribeEvents subscribes to peer arrival and departure events.
|
|
|
|
func (ps *peerSet) SubscribeEvents(ch chan<- *peeringEvent) event.Subscription {
|
|
|
|
return ps.events.Subscribe(ch)
|
2017-09-12 06:39:34 -05:00
|
|
|
}
|
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
// Reset iterates over the current peer set, and resets each of the known peers
|
|
|
|
// to prepare for a next batch of block retrieval.
|
|
|
|
func (ps *peerSet) Reset() {
|
|
|
|
ps.lock.RLock()
|
|
|
|
defer ps.lock.RUnlock()
|
2015-04-12 05:38:25 -05:00
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
for _, peer := range ps.peers {
|
|
|
|
peer.Reset()
|
2015-04-12 06:33:42 -05:00
|
|
|
}
|
2015-05-11 06:26:20 -05:00
|
|
|
}
|
2015-04-12 06:33:42 -05:00
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
// Register injects a new peer into the working set, or returns an error if the
|
|
|
|
// peer is already known.
|
2015-10-29 11:37:26 -05:00
|
|
|
//
|
|
|
|
// The method also sets the starting throughput values of the new peer to the
|
2016-02-25 10:36:42 -06:00
|
|
|
// average of all existing peers, to give it a realistic chance of being used
|
2015-10-29 11:37:26 -05:00
|
|
|
// for data retrievals.
|
2017-06-28 07:25:08 -05:00
|
|
|
func (ps *peerSet) Register(p *peerConnection) error {
|
2016-06-01 10:07:25 -05:00
|
|
|
// Register the new peer with some meaningful defaults
|
2015-05-11 06:26:20 -05:00
|
|
|
ps.lock.Lock()
|
|
|
|
if _, ok := ps.peers[p.id]; ok {
|
eth/downloader: separate state sync from queue (#14460)
* eth/downloader: separate state sync from queue
Scheduling of state node downloads hogged the downloader queue lock when
new requests were scheduled. This caused timeouts for other requests.
With this change, state sync is fully independent of all other downloads
and doesn't involve the queue at all.
State sync is started and checked on in processContent. This is slightly
awkward because processContent doesn't have a select loop. Instead, the
queue is closed by an auxiliary goroutine when state sync fails. We
tried several alternatives to this but settled on the current approach
because it's the least amount of change overall.
Handling of the pivot block has changed slightly: the queue previously
prevented import of pivot block receipts before the state of the pivot
block was available. In this commit, the receipt will be imported before
the state. This causes an annoyance where the pivot block is committed
as fast block head even when state downloads fail. Stay tuned for more
updates in this area ;)
* eth/downloader: remove cancelTimeout channel
* eth/downloader: retry state requests on timeout
* eth/downloader: improve comment
* eth/downloader: mark peers idle when state sync is done
* eth/downloader: move pivot block splitting to processContent
This change also ensures that pivot block receipts aren't imported
before the pivot block itself.
* eth/downloader: limit state node retries
* eth/downloader: improve state node error handling and retry check
* eth/downloader: remove maxStateNodeRetries
It fails the sync too much.
* eth/downloader: remove last use of cancelCh in statesync.go
Fixes TestDeliverHeadersHang*Fast and (hopefully)
the weird cancellation behaviour at the end of fast sync.
* eth/downloader: fix leak in runStateSync
* eth/downloader: don't run processFullSyncContent in LightSync mode
* eth/downloader: improve comments
* eth/downloader: fix vet, megacheck
* eth/downloader: remove unrequested tasks anyway
* eth/downloader, trie: various polishes around duplicate items
This commit explicitly tracks duplicate and unexpected state
delieveries done against a trie Sync structure, also adding
there to import info logs.
The commit moves the db batch used to commit trie changes one
level deeper so its flushed after every node insertion. This
is needed to avoid a lot of duplicate retrievals caused by
inconsistencies between Sync internals and database. A better
approach is to track not-yet-written states in trie.Sync and
flush on commit, but I'm focuing on correctness first now.
The commit fixes a regression around pivot block fail count.
The counter previously was reset to 1 if and only if a sync
cycle progressed (inserted at least 1 entry to the database).
The current code reset it already if a node was delivered,
which is not stong enough, because unless it ends up written
to disk, an attacker can just loop and attack ad infinitum.
The commit also fixes a regression around state deliveries
and timeouts. The old downloader tracked if a delivery is
stale (none of the deliveries were requestedt), in which
case it didn't mark the node idle and did not send further
requests, since it signals a past timeout. The current code
did mark it idle even on stale deliveries, which eventually
caused two requests to be in flight at the same time, making
the deliveries always stale and mass duplicating retrievals
between multiple peers.
* eth/downloader: fix state request leak
This commit fixes the hang seen sometimes while doing the state
sync. The cause of the hang was a rare combination of events:
request state data from peer, peer drops and reconnects almost
immediately. This caused a new download task to be assigned to
the peer, overwriting the old one still waiting for a timeout,
which in turned leaked the requests out, never to be retried.
The fix is to ensure that a task assignment moves any pending
one back into the retry queue.
The commit also fixes a regression with peer dropping due to
stalls. The current code considered a peer stalling if they
timed out delivering 1 item. However, the downloader never
requests only one, the minimum is 2 (attempt to fine tune
estimated latency/bandwidth). The fix is simply to drop if
a timeout is detected at 2 items.
Apart from the above bugfixes, the commit contains some code
polishes I made while debugging the hang.
* core, eth, trie: support batched trie sync db writes
* trie: rename SyncMemCache to syncMemBatch
2017-06-22 07:26:03 -05:00
|
|
|
ps.lock.Unlock()
|
2015-05-11 06:26:20 -05:00
|
|
|
return errAlreadyRegistered
|
2015-05-06 07:32:53 -05:00
|
|
|
}
|
2021-05-19 07:09:03 -05:00
|
|
|
p.rates = msgrate.NewTracker(ps.rates.MeanCapacities(), ps.rates.MedianRoundTrip())
|
|
|
|
if err := ps.rates.Track(p.id, p.rates); err != nil {
|
|
|
|
return err
|
2015-10-29 11:37:26 -05:00
|
|
|
}
|
2015-05-11 06:26:20 -05:00
|
|
|
ps.peers[p.id] = p
|
eth/downloader: separate state sync from queue (#14460)
* eth/downloader: separate state sync from queue
Scheduling of state node downloads hogged the downloader queue lock when
new requests were scheduled. This caused timeouts for other requests.
With this change, state sync is fully independent of all other downloads
and doesn't involve the queue at all.
State sync is started and checked on in processContent. This is slightly
awkward because processContent doesn't have a select loop. Instead, the
queue is closed by an auxiliary goroutine when state sync fails. We
tried several alternatives to this but settled on the current approach
because it's the least amount of change overall.
Handling of the pivot block has changed slightly: the queue previously
prevented import of pivot block receipts before the state of the pivot
block was available. In this commit, the receipt will be imported before
the state. This causes an annoyance where the pivot block is committed
as fast block head even when state downloads fail. Stay tuned for more
updates in this area ;)
* eth/downloader: remove cancelTimeout channel
* eth/downloader: retry state requests on timeout
* eth/downloader: improve comment
* eth/downloader: mark peers idle when state sync is done
* eth/downloader: move pivot block splitting to processContent
This change also ensures that pivot block receipts aren't imported
before the pivot block itself.
* eth/downloader: limit state node retries
* eth/downloader: improve state node error handling and retry check
* eth/downloader: remove maxStateNodeRetries
It fails the sync too much.
* eth/downloader: remove last use of cancelCh in statesync.go
Fixes TestDeliverHeadersHang*Fast and (hopefully)
the weird cancellation behaviour at the end of fast sync.
* eth/downloader: fix leak in runStateSync
* eth/downloader: don't run processFullSyncContent in LightSync mode
* eth/downloader: improve comments
* eth/downloader: fix vet, megacheck
* eth/downloader: remove unrequested tasks anyway
* eth/downloader, trie: various polishes around duplicate items
This commit explicitly tracks duplicate and unexpected state
delieveries done against a trie Sync structure, also adding
there to import info logs.
The commit moves the db batch used to commit trie changes one
level deeper so its flushed after every node insertion. This
is needed to avoid a lot of duplicate retrievals caused by
inconsistencies between Sync internals and database. A better
approach is to track not-yet-written states in trie.Sync and
flush on commit, but I'm focuing on correctness first now.
The commit fixes a regression around pivot block fail count.
The counter previously was reset to 1 if and only if a sync
cycle progressed (inserted at least 1 entry to the database).
The current code reset it already if a node was delivered,
which is not stong enough, because unless it ends up written
to disk, an attacker can just loop and attack ad infinitum.
The commit also fixes a regression around state deliveries
and timeouts. The old downloader tracked if a delivery is
stale (none of the deliveries were requestedt), in which
case it didn't mark the node idle and did not send further
requests, since it signals a past timeout. The current code
did mark it idle even on stale deliveries, which eventually
caused two requests to be in flight at the same time, making
the deliveries always stale and mass duplicating retrievals
between multiple peers.
* eth/downloader: fix state request leak
This commit fixes the hang seen sometimes while doing the state
sync. The cause of the hang was a rare combination of events:
request state data from peer, peer drops and reconnects almost
immediately. This caused a new download task to be assigned to
the peer, overwriting the old one still waiting for a timeout,
which in turned leaked the requests out, never to be retried.
The fix is to ensure that a task assignment moves any pending
one back into the retry queue.
The commit also fixes a regression with peer dropping due to
stalls. The current code considered a peer stalling if they
timed out delivering 1 item. However, the downloader never
requests only one, the minimum is 2 (attempt to fine tune
estimated latency/bandwidth). The fix is simply to drop if
a timeout is detected at 2 items.
Apart from the above bugfixes, the commit contains some code
polishes I made while debugging the hang.
* core, eth, trie: support batched trie sync db writes
* trie: rename SyncMemCache to syncMemBatch
2017-06-22 07:26:03 -05:00
|
|
|
ps.lock.Unlock()
|
|
|
|
|
2021-11-26 05:26:03 -06:00
|
|
|
ps.events.Send(&peeringEvent{peer: p, join: true})
|
2015-05-11 06:26:20 -05:00
|
|
|
return nil
|
|
|
|
}
|
2015-04-12 06:33:42 -05:00
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
// Unregister removes a remote peer from the active set, disabling any further
|
|
|
|
// actions to/from that particular entity.
|
|
|
|
func (ps *peerSet) Unregister(id string) error {
|
|
|
|
ps.lock.Lock()
|
2017-09-12 06:39:34 -05:00
|
|
|
p, ok := ps.peers[id]
|
|
|
|
if !ok {
|
2020-06-15 07:46:27 -05:00
|
|
|
ps.lock.Unlock()
|
2015-05-11 06:26:20 -05:00
|
|
|
return errNotRegistered
|
|
|
|
}
|
|
|
|
delete(ps.peers, id)
|
2021-05-19 07:09:03 -05:00
|
|
|
ps.rates.Untrack(id)
|
2017-09-12 06:39:34 -05:00
|
|
|
ps.lock.Unlock()
|
|
|
|
|
2021-11-26 05:26:03 -06:00
|
|
|
ps.events.Send(&peeringEvent{peer: p, join: false})
|
2015-04-12 06:33:42 -05:00
|
|
|
return nil
|
2015-04-12 05:38:25 -05:00
|
|
|
}
|
2015-04-15 19:16:33 -05:00
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
// Peer retrieves the registered peer with the given id.
|
2017-06-28 07:25:08 -05:00
|
|
|
func (ps *peerSet) Peer(id string) *peerConnection {
|
2015-05-11 06:26:20 -05:00
|
|
|
ps.lock.RLock()
|
|
|
|
defer ps.lock.RUnlock()
|
|
|
|
|
|
|
|
return ps.peers[id]
|
|
|
|
}
|
|
|
|
|
2015-05-11 09:06:42 -05:00
|
|
|
// Len returns if the current number of peers in the set.
|
|
|
|
func (ps *peerSet) Len() int {
|
2015-05-11 06:26:20 -05:00
|
|
|
ps.lock.RLock()
|
|
|
|
defer ps.lock.RUnlock()
|
2015-04-15 19:16:33 -05:00
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
return len(ps.peers)
|
2015-04-15 19:16:33 -05:00
|
|
|
}
|
|
|
|
|
2015-05-11 06:26:20 -05:00
|
|
|
// AllPeers retrieves a flat list of all the peers within the set.
|
2017-06-28 07:25:08 -05:00
|
|
|
func (ps *peerSet) AllPeers() []*peerConnection {
|
2015-05-11 06:26:20 -05:00
|
|
|
ps.lock.RLock()
|
|
|
|
defer ps.lock.RUnlock()
|
2015-04-15 19:16:33 -05:00
|
|
|
|
2017-06-28 07:25:08 -05:00
|
|
|
list := make([]*peerConnection, 0, len(ps.peers))
|
2015-05-11 06:26:20 -05:00
|
|
|
for _, p := range ps.peers {
|
|
|
|
list = append(list, p)
|
2015-04-15 19:16:33 -05:00
|
|
|
}
|
2015-05-11 06:26:20 -05:00
|
|
|
return list
|
2015-04-15 19:16:33 -05:00
|
|
|
}
|
2015-04-18 13:25:55 -05:00
|
|
|
|
2021-05-27 11:43:55 -05:00
|
|
|
// peerCapacitySort implements sort.Interface.
|
|
|
|
// It sorts peer connections by capacity (descending).
|
|
|
|
type peerCapacitySort struct {
|
2020-07-24 02:46:26 -05:00
|
|
|
p []*peerConnection
|
2021-05-27 11:43:55 -05:00
|
|
|
tp []int
|
2020-07-24 02:46:26 -05:00
|
|
|
}
|
|
|
|
|
2021-05-27 11:43:55 -05:00
|
|
|
func (ps *peerCapacitySort) Len() int {
|
2020-07-24 02:46:26 -05:00
|
|
|
return len(ps.p)
|
|
|
|
}
|
|
|
|
|
2021-05-27 11:43:55 -05:00
|
|
|
func (ps *peerCapacitySort) Less(i, j int) bool {
|
2020-07-24 02:46:26 -05:00
|
|
|
return ps.tp[i] > ps.tp[j]
|
|
|
|
}
|
|
|
|
|
2021-05-27 11:43:55 -05:00
|
|
|
func (ps *peerCapacitySort) Swap(i, j int) {
|
2020-07-24 02:46:26 -05:00
|
|
|
ps.p[i], ps.p[j] = ps.p[j], ps.p[i]
|
|
|
|
ps.tp[i], ps.tp[j] = ps.tp[j], ps.tp[i]
|
|
|
|
}
|