nimbus-eth2/beacon_chain/sync/sync_overseer.nim

502 lines
17 KiB
Nim

# beacon_chain
# Copyright (c) 2018-2024 Status Research & Development GmbH
# Licensed and distributed under either of
# * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
# * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
# at your option. This file may not be copied, modified, or distributed except according to those terms.
{.push raises: [].}
import std/[strutils, sequtils]
import stew/base10, chronos, chronicles, results
import
../consensus_object_pools/blockchain_list,
../spec/eth2_apis/rest_types,
../spec/[helpers, forks, network, forks_light_client, weak_subjectivity],
../networking/[peer_pool, peer_scores, eth2_network],
../gossip_processing/block_processor,
../[beacon_clock, beacon_node],
./[sync_types, sync_manager, sync_queue]
from ../consensus_object_pools/spec_cache import get_attesting_indices
export sync_types
logScope:
topics = "overseer"
const
PARALLEL_REQUESTS = 3
## Number of peers used to obtain the initial block.
BLOCKS_PROCESS_CHUNK_SIZE = 2
## Number of blocks sent to processing (CPU heavy task).
type
BlockDataRes = Result[BlockData, string]
proc init*(t: typedesc[BlockDataChunk],
stateCallback: OnStateUpdated,
data: openArray[BlockData]): BlockDataChunk =
BlockDataChunk(
blocks: @data,
onStateUpdatedCb: stateCallback,
resfut:
Future[Result[void, string]].Raising([CancelledError]).init(
"blockdata.chunk")
)
proc shortLog*(c: BlockDataChunk): string =
let
map =
(c.blocks.mapIt(shortLog(it.blck.root) & ":" & $it.blck.slot)).
join(", ")
futureState = if c.resfut.finished(): "pending" else: "completed"
"[" & map & "]:" & futureState
iterator chunks*(data: openArray[BlockData],
stateCallback: OnStateUpdated,
maxCount: Positive): BlockDataChunk =
for i in countup(0, len(data) - 1, maxCount):
yield BlockDataChunk.init(stateCallback,
data.toOpenArray(i, min(i + maxCount, len(data)) - 1))
proc getLatestBeaconHeader(
overseer: SyncOverseerRef
): Future[BeaconBlockHeader] {.async: (raises: [CancelledError]).} =
let eventKey = overseer.eventQueue.register()
defer:
overseer.eventQueue.unregister(eventKey)
let events =
try:
await overseer.eventQueue.waitEvents(eventKey)
except CancelledError as exc:
raise exc
except AsyncEventQueueFullError:
raiseAssert "AsyncEventQueueFullError should not happen!"
withForkyHeader(events[^1]):
when lcDataFork > LightClientDataFork.None:
forkyHeader.beacon
else:
raiseAssert "Should not happen"
proc getPeerBlock(
overseer: SyncOverseerRef,
slot: Slot,
): Future[BlockDataRes] {.async: (raises: [CancelledError]).} =
let peer = await overseer.pool.acquire()
try:
let
res = (await getSyncBlockData(peer, slot)).valueOr:
return err(error)
blob =
if res.blobs.isSome():
Opt.some(res.blobs.get()[0])
else:
Opt.none(BlobSidecars)
ok(BlockData(blck: res.blocks[0][], blob: blob))
finally:
overseer.pool.release(peer)
proc getBlock(
overseer: SyncOverseerRef,
slot: Slot,
blockHeader: BeaconBlockHeader
): Future[BlockData] {.async: (raises: [CancelledError]).} =
var workers:
array[PARALLEL_REQUESTS, Future[BlockDataRes].Raising([CancelledError])]
while true:
for i in 0 ..< PARALLEL_REQUESTS:
workers[i] = overseer.getPeerBlock(slot)
try:
await allFutures(workers)
except CancelledError as exc:
let pending =
workers.filterIt(not(it.finished())).mapIt(cancelAndWait(it))
await noCancel allFutures(pending)
raise exc
var results: seq[BlockData]
for i in 0 ..< PARALLEL_REQUESTS:
if workers[i].value.isOk:
results.add(workers[i].value.get())
if len(results) > 0:
for item in results:
withBlck(item.blck):
if forkyBlck.message.toBeaconBlockHeader() == blockHeader:
return item
# Wait for 2 seconds before trying one more time.
await sleepAsync(2.seconds)
proc isWithinWeakSubjectivityPeriod(
overseer: SyncOverseerRef, slot: Slot): bool =
let
dag = overseer.consensusManager.dag
currentSlot = overseer.beaconClock.now().slotOrZero()
checkpoint = Checkpoint(
epoch:
getStateField(dag.headState, slot).epoch(),
root:
getStateField(dag.headState, latest_block_header).state_root)
is_within_weak_subjectivity_period(
dag.cfg, currentSlot, dag.headState, checkpoint)
proc isUntrustedBackfillEmpty(clist: ChainListRef): bool =
clist.tail.isNone()
func speed(start, finish: Moment, entities: int): float =
if entities <= 0:
0.0
else:
float(entities) / toFloatSeconds(finish - start)
proc updatePerformance(overseer: SyncOverseerRef, startTick: Moment,
entities: int) =
let dag = overseer.consensusManager.dag
doAssert(overseer.clist.head.isSome() and overseer.clist.tail.isSome())
let
clistHeadSlot = overseer.clist.head.get().slot
clistTailSlot = overseer.clist.tail.get().slot
doAssert(clistHeadSlot >= dag.head.slot)
let slotsPerSec = speed(startTick, Moment.now(), entities)
inc(overseer.avgSpeedCounter)
overseer.avgSpeed = overseer.avgSpeed +
(slotsPerSec - overseer.avgSpeed) / float(overseer.avgSpeedCounter)
let
total = clistHeadSlot - clistTailSlot
progress = dag.head.slot - clistTailSlot
done = float(progress) / float(total)
remaining = total - progress
timeleft =
if overseer.avgSpeed >= 0.001:
Duration.fromFloatSeconds(remaining.float / overseer.avgSpeed)
else:
InfiniteDuration
# Update status string
overseer.statusMsg = Opt.some(
timeleft.toTimeLeftString() & " (" &
(done * 100).formatBiggestFloat(ffDecimal, 2) & "%) " &
overseer.avgSpeed.formatBiggestFloat(ffDecimal, 4) &
"slots/s (" & $dag.head.slot & ")")
proc blockProcessingLoop(overseer: SyncOverseerRef): Future[void] {.
async: (raises: [CancelledError]).} =
let
consensusManager = overseer.consensusManager
dag = consensusManager.dag
attestationPool = consensusManager.attestationPool
validatorMonitor = overseer.validatorMonitor
proc onBlockAdded(
blckRef: BlockRef, blck: ForkedTrustedSignedBeaconBlock, epochRef: EpochRef,
unrealized: FinalityCheckpoints) {.gcsafe, raises: [].} =
let wallTime = overseer.getBeaconTimeFn()
withBlck(blck):
attestationPool[].addForkChoice(
epochRef, blckRef, unrealized, forkyBlck.message, wallTime)
validatorMonitor[].registerBeaconBlock(
MsgSource.sync, wallTime, forkyBlck.message)
for attestation in forkyBlck.message.body.attestations:
for validator_index in
dag.get_attesting_indices(attestation, true):
validatorMonitor[].registerAttestationInBlock(
attestation.data, validator_index, forkyBlck.message.slot)
withState(dag[].clearanceState):
when (consensusFork >= ConsensusFork.Altair) and
(type(forkyBlck) isnot phase0.TrustedSignedBeaconBlock):
for i in forkyBlck.message.body.sync_aggregate.
sync_committee_bits.oneIndices():
validatorMonitor[].registerSyncAggregateInBlock(
forkyBlck.message.slot, forkyBlck.root,
forkyState.data.current_sync_committee.pubkeys.data[i])
block mainLoop:
while true:
let bchunk = await overseer.blocksQueue.popFirst()
block innerLoop:
for bdata in bchunk.blocks:
block:
let res = addBackfillBlockData(dag, bdata, bchunk.onStateUpdatedCb,
onBlockAdded)
if res.isErr():
let msg = "Unable to add block data to database [" &
$res.error & "]"
bchunk.resfut.complete(Result[void, string].err(msg))
break innerLoop
consensusManager.updateHead(overseer.getBeaconTimeFn).isOkOr:
bchunk.resfut.complete(Result[void, string].err(error))
break innerLoop
bchunk.resfut.complete(Result[void, string].ok())
proc verifyBlockProposer(
fork: Fork,
genesis_validators_root: Eth2Digest,
immutableValidators: openArray[ImmutableValidatorData2],
signedBlock: ForkedSignedBeaconBlock
): Result[void, cstring] =
withBlck(signedBlock):
let proposerKey =
immutableValidators.load(forkyBlck.message.proposer_index).valueOr:
return err("Unable to find proposer key")
if not(verify_block_signature(fork, genesis_validators_root,
forkyBlck.message.slot, forkyBlck.message,
proposerKey, forkyBlck.signature)):
return err("Signature verification failed")
ok()
proc rebuildState(overseer: SyncOverseerRef): Future[void] {.
async: (raises: [CancelledError]).} =
overseer.statusMsg = Opt.some("rebuilding state")
let
consensusManager = overseer.consensusManager
dag = consensusManager.dag
batchVerifier = overseer.batchVerifier
clist =
block:
overseer.clist.seekForSlot(dag.head.slot).isOkOr:
fatal "Unable to find slot in backfill data", reason = error,
path = overseer.clist.path
quit 1
overseer.clist
var
blocks: seq[BlockData]
currentEpoch: Epoch = FAR_FUTURE_EPOCH
let handle = clist.handle.get()
overseer.avgSpeed = 0.0
overseer.avgSpeedCounter = 0
# Set minimum slot number from which LC data is collected.
dag.lcDataStore.cache.tailSlot = clist.head.get().slot
block mainLoop:
while true:
let res = getChainFileTail(handle.handle)
if res.isErr():
fatal "Unable to read backfill data", reason = res.error
quit 1
let bres = res.get()
if bres.isNone():
return
let
data = bres.get()
blockEpoch = data.blck.slot.epoch()
if blockEpoch != currentEpoch:
if len(blocks) != 0:
let
startTick = Moment.now()
blocksOnly = blocks.mapIt(it.blck)
proc onStateUpdate(slot: Slot): Result[void, VerifierError] {.
gcsafe, raises: [].} =
if slot != blocksOnly[0].slot:
# We verify signatures only at the beginning of chunk/epoch, in
# such way we could verify whole epoch's proposer signatures in
# one batch.
return ok()
let
fork =
getStateField(dag.clearanceState, fork)
genesis_validators_root =
getStateField(dag.clearanceState, genesis_validators_root)
verifyBlockProposer(batchVerifier[], fork, genesis_validators_root,
dag.db.immutableValidators, blocksOnly).isOkOr:
for signedBlock in blocksOnly:
verifyBlockProposer(fork, genesis_validators_root,
dag.db.immutableValidators,
signedBlock).isOkOr:
fatal "Unable to verify block proposer",
blck = shortLog(signedBlock), reason = error
return err(VerifierError.Invalid)
ok()
for bchunk in blocks.chunks(onStateUpdate, BLOCKS_PROCESS_CHUNK_SIZE):
try:
overseer.blocksQueue.addLastNoWait(bchunk)
except AsyncQueueFullError:
raiseAssert "Should not happen with unbounded AsyncQueue"
let res = await bchunk.resfut
if res.isErr():
fatal "Unable to add block data to database", reason = res.error
quit 1
let updateTick = Moment.now()
debug "Number of blocks injected",
blocks_count = len(blocks),
head = shortLog(dag.head),
finalized = shortLog(getStateField(
dag.headState, finalized_checkpoint)),
store_update_time = updateTick - startTick
overseer.updatePerformance(startTick, len(blocks))
blocks.setLen(0)
currentEpoch = blockEpoch
if data.blck.slot != GENESIS_SLOT:
blocks.add(data)
proc initUntrustedSync(overseer: SyncOverseerRef): Future[void] {.
async: (raises: [CancelledError]).} =
overseer.statusMsg = Opt.some("awaiting light client")
let blockHeader = await overseer.getLatestBeaconHeader()
notice "Received light client block header",
beacon_header = shortLog(blockHeader),
current_slot = overseer.beaconClock.now().slotOrZero()
overseer.statusMsg = Opt.some("retrieving block")
let
blck = await overseer.getBlock(blockHeader.slot, blockHeader)
blobsCount = if blck.blob.isNone(): 0 else: len(blck.blob.get())
notice "Received beacon block", blck = shortLog(blck.blck),
blobs_count = blobsCount
overseer.statusMsg = Opt.some("storing block")
let res = overseer.clist.addBackfillBlockData(blck.blck, blck.blob)
if res.isErr():
warn "Unable to store initial block", reason = res.error
return
overseer.statusMsg = Opt.none(string)
notice "Initial block being stored",
blck = shortLog(blck.blck), blobs_count = blobsCount
proc startBackfillTask(overseer: SyncOverseerRef): Future[void] {.
async: (raises: []).} =
# This procedure performs delayed start of backfilling process.
while overseer.consensusManager.dag.needsBackfill:
if not(overseer.forwardSync.inProgress):
# Only start the backfiller if it's needed _and_ head sync has completed -
# if we lose sync after having synced head, we could stop the backfilller,
# but this should be a fringe case - might as well keep the logic simple
# for now.
overseer.backwardSync.start()
return
try:
await sleepAsync(chronos.seconds(2))
except CancelledError:
return
proc mainLoop*(
overseer: SyncOverseerRef
): Future[void] {.async: (raises: []).} =
let
dag = overseer.consensusManager.dag
clist = overseer.clist
currentSlot = overseer.beaconClock.now().slotOrZero()
if overseer.isWithinWeakSubjectivityPeriod(currentSlot):
# Starting forward sync manager/monitor.
overseer.forwardSync.start()
# Starting backfill/backward sync manager.
if dag.needsBackfill():
asyncSpawn overseer.startBackfillTask()
return
else:
if dag.needsBackfill():
# Checkpoint/Trusted state we have is too old.
error "Trusted node sync started too long time ago"
quit 1
if not(isUntrustedBackfillEmpty(clist)):
let headSlot = clist.head.get().slot
if not(overseer.isWithinWeakSubjectivityPeriod(headSlot)):
# Light forward sync file is too old.
warn "Light client sync was started too long time ago",
current_slot = currentSlot, backfill_data_slot = headSlot
if overseer.config.longRangeSync == LongRangeSyncMode.Lenient:
# Starting forward sync manager/monitor only.
overseer.forwardSync.start()
return
if overseer.config.longRangeSync == LongRangeSyncMode.Light:
let dagHead = dag.finalizedHead
if dagHead.slot < dag.cfg.ALTAIR_FORK_EPOCH.start_slot:
fatal "Light forward syncing requires a post-Altair state",
head_slot = dagHead.slot,
altair_start_slot = dag.cfg.ALTAIR_FORK_EPOCH.start_slot
quit 1
if isUntrustedBackfillEmpty(clist):
overseer.untrustedInProgress = true
try:
await overseer.initUntrustedSync()
except CancelledError:
return
# We need to update pivot slot to enable timeleft calculation.
overseer.untrustedSync.updatePivot(overseer.clist.tail.get().slot)
# Note: We should not start forward sync manager!
overseer.untrustedSync.start()
# Waiting until untrusted backfilling will not be complete
try:
await overseer.untrustedSync.join()
except CancelledError:
return
notice "Start state rebuilding process"
# We spawn block processing loop to keep async world happy, otherwise
# it could be single cpu heavy procedure call.
let blockProcessingFut = overseer.blockProcessingLoop()
try:
await overseer.rebuildState()
except CancelledError:
await cancelAndWait(blockProcessingFut)
return
clist.clear().isOkOr:
warn "Unable to remove backfill data file",
path = clist.path.chainFilePath(), reason = error
quit 1
overseer.untrustedInProgress = false
# When we finished state rebuilding process - we could start forward
# SyncManager which could perform finish sync.
overseer.forwardSync.start()
proc start*(overseer: SyncOverseerRef) =
overseer.loopFuture = overseer.mainLoop()
proc stop*(overseer: SyncOverseerRef) {.async: (raises: []).} =
doAssert(not(isNil(overseer.loopFuture)),
"SyncOverseer was not started yet")
if not(overseer.loopFuture.finished()):
await cancelAndWait(overseer.loopFuture)