go-ethereum/eth/downloader/statesync.go

482 lines
16 KiB
Go
Raw Normal View History

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
// Copyright 2017 The go-ethereum Authors
// This file is part of the go-ethereum library.
//
// The go-ethereum library is free software: you can redistribute it and/or modify
// 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.
//
// The go-ethereum library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public License
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
package downloader
import (
"fmt"
"hash"
"sync"
"time"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
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/core/state"
"github.com/ethereum/go-ethereum/ethdb"
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/log"
"github.com/ethereum/go-ethereum/trie"
"golang.org/x/crypto/sha3"
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
)
2018-04-04 05:25:02 -05:00
// stateReq represents a batch of state fetch requests grouped together into
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
// a single data retrieval network packet.
type stateReq struct {
items []common.Hash // Hashes of the state items to download
tasks map[common.Hash]*stateTask // Download tasks to track previous attempts
timeout time.Duration // Maximum round trip time for this to complete
timer *time.Timer // Timer to fire when the RTT timeout expires
peer *peerConnection // Peer that we're requesting from
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
response [][]byte // Response data of the peer (nil for timeouts)
dropped bool // Flag whether the peer dropped off early
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
}
// timedOut returns if this request timed out.
func (req *stateReq) timedOut() bool {
return req.response == nil
}
// stateSyncStats is a collection of progress stats to report during a state trie
// sync to RPC requests as well as to display in user logs.
type stateSyncStats struct {
processed uint64 // Number of state entries processed
duplicate uint64 // Number of state entries downloaded twice
unexpected uint64 // Number of non-requested state entries received
pending uint64 // Number of still pending state entries
}
// syncState starts downloading state with the given root hash.
func (d *Downloader) syncState(root common.Hash) *stateSync {
s := newStateSync(d, root)
select {
case d.stateSyncStart <- s:
case <-d.quitCh:
s.err = errCancelStateFetch
close(s.done)
}
return s
}
// stateFetcher manages the active state sync and accepts requests
// on its behalf.
func (d *Downloader) stateFetcher() {
for {
select {
case s := <-d.stateSyncStart:
for next := s; next != nil; {
next = d.runStateSync(next)
}
case <-d.stateCh:
// Ignore state responses while no sync is running.
case <-d.quitCh:
return
}
}
}
// runStateSync runs a state synchronisation until it completes or another root
// hash is requested to be switched over to.
func (d *Downloader) runStateSync(s *stateSync) *stateSync {
var (
active = make(map[string]*stateReq) // Currently in-flight requests
finished []*stateReq // Completed or failed requests
timeout = make(chan *stateReq) // Timed out active requests
)
defer func() {
// Cancel active request timers on exit. Also set peers to idle so they're
// available for the next sync.
for _, req := range active {
req.timer.Stop()
req.peer.SetNodeDataIdle(len(req.items))
}
}()
// Run the state sync.
go s.run()
defer s.Cancel()
// Listen for peer departure events to cancel assigned tasks
peerDrop := make(chan *peerConnection, 1024)
peerSub := s.d.peers.SubscribePeerDrops(peerDrop)
defer peerSub.Unsubscribe()
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
for {
// Enable sending of the first buffered element if there is one.
var (
deliverReq *stateReq
deliverReqCh chan *stateReq
)
if len(finished) > 0 {
deliverReq = finished[0]
deliverReqCh = s.deliver
}
select {
// The stateSync lifecycle:
case next := <-d.stateSyncStart:
return next
case <-s.done:
return nil
// Send the next finished request to the current sync:
case deliverReqCh <- deliverReq:
// Shift out the first request, but also set the emptied slot to nil for GC
copy(finished, finished[1:])
finished[len(finished)-1] = nil
finished = finished[:len(finished)-1]
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
// Handle incoming state packs:
case pack := <-d.stateCh:
2018-04-04 05:25:02 -05:00
// Discard any data not requested (or previously timed out)
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
req := active[pack.PeerId()]
if req == nil {
log.Debug("Unrequested node data", "peer", pack.PeerId(), "len", pack.Items())
continue
}
// Finalize the request and queue up for processing
req.timer.Stop()
req.response = pack.(*statePack).states
finished = append(finished, req)
delete(active, pack.PeerId())
// Handle dropped peer connections:
case p := <-peerDrop:
// Skip if no request is currently pending
req := active[p.id]
if req == nil {
continue
}
// Finalize the request and queue up for processing
req.timer.Stop()
req.dropped = true
finished = append(finished, req)
delete(active, p.id)
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
// Handle timed-out requests:
case req := <-timeout:
// If the peer is already requesting something else, ignore the stale timeout.
// This can happen when the timeout and the delivery happens simultaneously,
// causing both pathways to trigger.
if active[req.peer.id] != req {
continue
}
// Move the timed out data back into the download queue
finished = append(finished, req)
delete(active, req.peer.id)
// Track outgoing state requests:
case req := <-d.trackStateReq:
// If an active request already exists for this peer, we have a problem. In
// theory the trie node schedule must never assign two requests to the same
2018-04-04 05:25:02 -05:00
// peer. In practice however, a peer might receive a request, disconnect and
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
// immediately reconnect before the previous times out. In this case the first
// request is never honored, alas we must not silently overwrite it, as that
// causes valid requests to go missing and sync to get stuck.
if old := active[req.peer.id]; old != nil {
log.Warn("Busy peer assigned new state fetch", "peer", old.peer.id)
// Make sure the previous one doesn't get siletly lost
old.timer.Stop()
old.dropped = true
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
finished = append(finished, old)
}
// Start a timer to notify the sync loop if the peer stalled.
req.timer = time.AfterFunc(req.timeout, func() {
select {
case timeout <- req:
case <-s.done:
// Prevent leaking of timer goroutines in the unlikely case where a
// timer is fired just before exiting runStateSync.
}
})
active[req.peer.id] = req
}
}
}
// stateSync schedules requests for downloading a particular state trie defined
// by a given state root.
type stateSync struct {
d *Downloader // Downloader instance to access and manage current peerset
sched *trie.Sync // State trie sync scheduler defining the tasks
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
keccak hash.Hash // Keccak256 hasher to verify deliveries with
tasks map[common.Hash]*stateTask // Set of tasks currently queued for retrieval
numUncommitted int
bytesUncommitted int
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
deliver chan *stateReq // Delivery channel multiplexing peer responses
cancel chan struct{} // Channel to signal a termination request
cancelOnce sync.Once // Ensures cancel only ever gets called once
done chan struct{} // Channel to signal termination completion
err error // Any error hit during sync (set before completion)
}
2018-04-04 05:25:02 -05:00
// stateTask represents a single trie node download task, containing a set of
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
// peers already attempted retrieval from to detect stalled syncs and abort.
type stateTask struct {
attempts map[string]struct{}
}
// newStateSync creates a new state trie download scheduler. This method does not
// yet start the sync. The user needs to call run to initiate.
func newStateSync(d *Downloader, root common.Hash) *stateSync {
return &stateSync{
d: d,
sched: state.NewStateSync(root, d.stateDB),
keccak: sha3.NewLegacyKeccak256(),
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
tasks: make(map[common.Hash]*stateTask),
deliver: make(chan *stateReq),
cancel: make(chan struct{}),
done: make(chan struct{}),
}
}
// run starts the task assignment and response processing loop, blocking until
// it finishes, and finally notifying any goroutines waiting for the loop to
// finish.
func (s *stateSync) run() {
s.err = s.loop()
close(s.done)
}
// Wait blocks until the sync is done or canceled.
func (s *stateSync) Wait() error {
<-s.done
return s.err
}
// Cancel cancels the sync and waits until it has shut down.
func (s *stateSync) Cancel() error {
s.cancelOnce.Do(func() { close(s.cancel) })
return s.Wait()
}
// loop is the main event loop of a state trie sync. It it responsible for the
// assignment of new tasks to peers (including sending it to them) as well as
// for the processing of inbound data. Note, that the loop does not directly
// receive data from peers, rather those are buffered up in the downloader and
// pushed here async. The reason is to decouple processing from data receipt
// and timeouts.
func (s *stateSync) loop() (err error) {
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
// Listen for new peer events to assign tasks to them
newPeer := make(chan *peerConnection, 1024)
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
peerSub := s.d.peers.SubscribeNewPeers(newPeer)
defer peerSub.Unsubscribe()
defer func() {
cerr := s.commit(true)
if err == nil {
err = cerr
}
}()
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
// Keep assigning new tasks until the sync completes or aborts
for s.sched.Pending() > 0 {
if err = s.commit(false); err != nil {
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
return err
}
s.assignTasks()
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
// Tasks assigned, wait for something to happen
select {
case <-newPeer:
// New peer arrived, try to assign it download tasks
case <-s.cancel:
return errCancelStateFetch
case <-s.d.cancelCh:
return errCancelStateFetch
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
case req := <-s.deliver:
// Response, disconnect or timeout triggered, drop the peer if stalling
log.Trace("Received node data response", "peer", req.peer.id, "count", len(req.response), "dropped", req.dropped, "timeout", !req.dropped && req.timedOut())
if len(req.items) <= 2 && !req.dropped && req.timedOut() {
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
// 2 items are the minimum requested, if even that times out, we've no use of
// this peer at the moment.
log.Warn("Stalling state sync, dropping peer", "peer", req.peer.id)
s.d.dropPeer(req.peer.id)
}
// Process all the received blobs and check for stale delivery
delivered, err := s.process(req)
if err != nil {
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
log.Warn("Node data write error", "err", err)
return err
}
req.peer.SetNodeDataIdle(delivered)
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
}
}
return nil
}
func (s *stateSync) commit(force bool) error {
if !force && s.bytesUncommitted < ethdb.IdealBatchSize {
return nil
}
start := time.Now()
b := s.d.stateDB.NewBatch()
if written, err := s.sched.Commit(b); written == 0 || err != nil {
return err
}
if err := b.Write(); err != nil {
return fmt.Errorf("DB write error: %v", err)
}
s.updateStats(s.numUncommitted, 0, 0, time.Since(start))
s.numUncommitted = 0
s.bytesUncommitted = 0
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
return nil
}
2018-04-04 05:25:02 -05:00
// assignTasks attempts to assign new tasks to all idle peers, either from the
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
// batch currently being retried, or fetching new data from the trie sync itself.
func (s *stateSync) assignTasks() {
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
// Iterate over all idle peers and try to assign them state fetches
peers, _ := s.d.peers.NodeDataIdlePeers()
for _, p := range peers {
// Assign a batch of fetches proportional to the estimated latency/bandwidth
cap := p.NodeDataCapacity(s.d.requestRTT())
req := &stateReq{peer: p, timeout: s.d.requestTTL()}
s.fillTasks(cap, req)
// If the peer was assigned tasks to fetch, send the network request
if len(req.items) > 0 {
req.peer.log.Trace("Requesting new batch of data", "type", "state", "count", len(req.items))
select {
case s.d.trackStateReq <- req:
req.peer.FetchNodeData(req.items)
case <-s.cancel:
case <-s.d.cancelCh:
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
}
}
}
}
// fillTasks fills the given request object with a maximum of n state download
// tasks to send to the remote peer.
func (s *stateSync) fillTasks(n int, req *stateReq) {
// Refill available tasks from the scheduler.
if len(s.tasks) < n {
new := s.sched.Missing(n - len(s.tasks))
for _, hash := range new {
s.tasks[hash] = &stateTask{make(map[string]struct{})}
}
}
// Find tasks that haven't been tried with the request's peer.
req.items = make([]common.Hash, 0, n)
req.tasks = make(map[common.Hash]*stateTask, n)
for hash, t := range s.tasks {
// Stop when we've gathered enough requests
if len(req.items) == n {
break
}
// Skip any requests we've already tried from this peer
if _, ok := t.attempts[req.peer.id]; ok {
continue
}
// Assign the request to this peer
t.attempts[req.peer.id] = struct{}{}
req.items = append(req.items, hash)
req.tasks[hash] = t
delete(s.tasks, hash)
}
}
// process iterates over a batch of delivered state data, injecting each item
// into a running state sync, re-queuing any items that were requested but not
// delivered. Returns whether the peer actually managed to deliver anything of
// value, and any error that occurred.
func (s *stateSync) process(req *stateReq) (int, error) {
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
// Collect processing stats and update progress if valid data was received
duplicate, unexpected, successful := 0, 0, 0
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
defer func(start time.Time) {
if duplicate > 0 || unexpected > 0 {
s.updateStats(0, duplicate, unexpected, time.Since(start))
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
}
}(time.Now())
// Iterate over all the delivered data and inject one-by-one into the trie
for _, blob := range req.response {
_, hash, err := s.processNodeData(blob)
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
switch err {
case nil:
s.numUncommitted++
s.bytesUncommitted += len(blob)
successful++
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
case trie.ErrNotRequested:
unexpected++
case trie.ErrAlreadyProcessed:
duplicate++
default:
return successful, fmt.Errorf("invalid state node %s: %v", hash.TerminalString(), err)
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
}
if _, ok := req.tasks[hash]; ok {
delete(req.tasks, hash)
}
}
// Put unfulfilled tasks back into the retry queue
npeers := s.d.peers.Len()
for hash, task := range req.tasks {
// If the node did deliver something, missing items may be due to a protocol
// limit or a previous timeout + delayed delivery. Both cases should permit
// the node to retry the missing items (to avoid single-peer stalls).
if len(req.response) > 0 || req.timedOut() {
delete(task.attempts, req.peer.id)
}
// If we've requested the node too many times already, it may be a malicious
// sync where nobody has the right data. Abort.
if len(task.attempts) >= npeers {
return successful, fmt.Errorf("state node %s failed with all peers (%d tries, %d peers)", hash.TerminalString(), len(task.attempts), npeers)
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
}
// Missing item, place into the retry queue.
s.tasks[hash] = task
}
return successful, nil
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
}
// processNodeData tries to inject a trie node data blob delivered from a remote
// peer into the state trie, returning whether anything useful was written or any
// error occurred.
func (s *stateSync) processNodeData(blob []byte) (bool, common.Hash, error) {
res := trie.SyncResult{Data: blob}
s.keccak.Reset()
s.keccak.Write(blob)
s.keccak.Sum(res.Hash[:0])
committed, _, err := s.sched.Process([]trie.SyncResult{res})
return committed, res.Hash, err
}
// updateStats bumps the various state sync progress counters and displays a log
// message for the user to see.
func (s *stateSync) updateStats(written, duplicate, unexpected int, duration time.Duration) {
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
s.d.syncStatsLock.Lock()
defer s.d.syncStatsLock.Unlock()
s.d.syncStatsState.pending = uint64(s.sched.Pending())
s.d.syncStatsState.processed += uint64(written)
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
s.d.syncStatsState.duplicate += uint64(duplicate)
s.d.syncStatsState.unexpected += uint64(unexpected)
if written > 0 || duplicate > 0 || unexpected > 0 {
log.Info("Imported new state entries", "count", written, "elapsed", common.PrettyDuration(duration), "processed", s.d.syncStatsState.processed, "pending", s.d.syncStatsState.pending, "retry", len(s.tasks), "duplicate", s.d.syncStatsState.duplicate, "unexpected", s.d.syncStatsState.unexpected)
}
if written > 0 {
rawdb.WriteFastTrieProgress(s.d.stateDB, s.d.syncStatsState.processed)
}
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
}