nimbus-eth2/beacon_chain/consensus_object_pools/consensus_manager.nim

469 lines
19 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
chronicles, chronos, web3/[primitives, engine_api_types],
../spec/datatypes/base,
../consensus_object_pools/[blockchain_dag, block_quarantine, attestation_pool],
../el/el_manager,
../beacon_clock,
./common_tools
from ../el/engine_api_conversions import asBlockHash
from ../spec/beaconstate import
get_expected_withdrawals, has_eth1_withdrawal_credential
from ../spec/datatypes/capella import Withdrawal
from ../spec/eth2_apis/dynamic_fee_recipients import
DynamicFeeRecipientsStore, getDynamicFeeRecipient
from ../validators/keystore_management import
KeymanagerHost, getPerValidatorDefaultFeeRecipient, getSuggestedFeeRecipient,
getSuggestedGasLimit
from ../validators/action_tracker import ActionTracker, getNextProposalSlot
logScope: topics = "cman"
type
ConsensusManager* = object
expectedSlot: Slot
expectedBlockReceived: Future[bool].Raising([CancelledError])
# Validated & Verified
# ----------------------------------------------------------------
dag*: ChainDAGRef
attestationPool*: ref AttestationPool
# Missing info
# ----------------------------------------------------------------
quarantine*: ref Quarantine
# Execution layer integration
# ----------------------------------------------------------------
elManager*: ELManager
# Allow determination of whether there's an upcoming proposal
# ----------------------------------------------------------------
actionTracker*: ActionTracker
# Allow determination of preferred fee recipient during proposals
# ----------------------------------------------------------------
dynamicFeeRecipientsStore: ref DynamicFeeRecipientsStore
validatorsDir: string
defaultFeeRecipient: Opt[Eth1Address]
defaultGasLimit: uint64
# Tracking last proposal forkchoiceUpdated payload information
# ----------------------------------------------------------------
optimisticHead: tuple[bid: BlockId, execution_block_hash: Eth2Digest]
# Initialization
# ------------------------------------------------------------------------------
func new*(T: type ConsensusManager,
dag: ChainDAGRef,
attestationPool: ref AttestationPool,
quarantine: ref Quarantine,
elManager: ELManager,
actionTracker: ActionTracker,
dynamicFeeRecipientsStore: ref DynamicFeeRecipientsStore,
validatorsDir: string,
defaultFeeRecipient: Opt[Eth1Address],
defaultGasLimit: uint64
): ref ConsensusManager =
(ref ConsensusManager)(
dag: dag,
attestationPool: attestationPool,
quarantine: quarantine,
elManager: elManager,
actionTracker: actionTracker,
dynamicFeeRecipientsStore: dynamicFeeRecipientsStore,
validatorsDir: validatorsDir,
defaultFeeRecipient: defaultFeeRecipient,
defaultGasLimit: defaultGasLimit
)
# Consensus Management
# -----------------------------------------------------------------------------------
proc checkExpectedBlock(self: var ConsensusManager) =
if self.expectedBlockReceived == nil:
return
if self.dag.head.slot < self.expectedSlot:
return
self.expectedBlockReceived.complete(true)
self.expectedBlockReceived = nil # Don't keep completed futures around!
proc expectBlock*(self: var ConsensusManager, expectedSlot: Slot): Future[bool]
{.async: (raises: [CancelledError], raw: true).} =
## Return a future that will complete when a head is selected whose slot is
## equal or greater than the given slot, or a new expectation is created
if self.expectedBlockReceived != nil:
# Reset the old future to not leave it hanging.. an alternative would be to
# cancel it, but it doesn't make any practical difference for now
self.expectedBlockReceived.complete(false)
let fut = newFuture[bool]("ConsensusManager.expectBlock")
self.expectedSlot = expectedSlot
self.expectedBlockReceived = fut
# It might happen that by the time we're expecting a block, it might have
# already been processed!
self.checkExpectedBlock()
return fut
func shouldSyncOptimistically*(
optimisticSlot, dagSlot, wallSlot: Slot): bool =
## Determine whether an optimistic execution block hash should be reported
## to the EL client instead of the current head as determined by fork choice.
# Check whether optimistic head is sufficiently ahead of DAG
const minProgress = 8 * SLOTS_PER_EPOCH # Set arbitrarily
if optimisticSlot < dagSlot or optimisticSlot - dagSlot < minProgress:
return false
# Check whether optimistic head has synced sufficiently close to wall slot
const maxAge = 2 * SLOTS_PER_EPOCH # Set arbitrarily
if optimisticSlot < max(wallSlot, maxAge.Slot) - maxAge:
return false
true
func shouldSyncOptimistically*(self: ConsensusManager, wallSlot: Slot): bool =
if self.optimisticHead.execution_block_hash.isZero:
return false
shouldSyncOptimistically(
optimisticSlot = self.optimisticHead.bid.slot,
dagSlot = getStateField(self.dag.headState, slot),
wallSlot = wallSlot)
func optimisticHead*(self: ConsensusManager): BlockId =
self.optimisticHead.bid
func optimisticExecutionBlockHash*(self: ConsensusManager): Eth2Digest =
self.optimisticHead.execution_block_hash
func setOptimisticHead*(
self: var ConsensusManager,
bid: BlockId, execution_block_hash: Eth2Digest) =
self.optimisticHead = (bid: bid, execution_block_hash: execution_block_hash)
proc updateExecutionClientHead*(
self: ref ConsensusManager,
newHead: BeaconHead
): Future[Opt[void]] {.async: (raises: [CancelledError]).} =
let headExecutionBlockHash =
self.dag.loadExecutionBlockHash(newHead.blck).valueOr:
# `BlockRef` are only created for blocks that have passed
# execution block hash validation, either explicitly in
# `block_processor.storeBlock`, or implicitly, e.g., through
# checkpoint sync. With checkpoint sync, the checkpoint block
# is initially not available, so if there is a reorg to it,
# this may be triggered. Such a reorg could happen if the first
# imported chain is completely invalid (after the checkpoint block)
# and is subsequently pruned, in which case checkpoint block is head.
# Because execution block hash validation has already passed,
# we can treat this as `SYNCING`.
warn "Failed to load head execution block hash", head = newHead.blck
return Opt[void].ok()
if headExecutionBlockHash.isZero:
# Blocks without execution payloads can't be optimistic.
self.dag.markBlockVerified(newHead.blck)
return Opt[void].ok()
template callForkchoiceUpdated(attributes: untyped): auto =
await self.elManager.forkchoiceUpdated(
headBlockHash = headExecutionBlockHash,
safeBlockHash = newHead.safeExecutionBlockHash,
finalizedBlockHash = newHead.finalizedExecutionBlockHash,
payloadAttributes = Opt.none attributes)
# Can't use dag.head here because it hasn't been updated yet
let
consensusFork =
self.dag.cfg.consensusForkAtEpoch(newHead.blck.bid.slot.epoch)
(payloadExecutionStatus, _) = withConsensusFork(consensusFork):
when consensusFork >= ConsensusFork.Bellatrix:
callForkchoiceUpdated(consensusFork.PayloadAttributes)
else:
callForkchoiceUpdated(PayloadAttributesV1)
case payloadExecutionStatus
of PayloadExecutionStatus.valid:
self.dag.markBlockVerified(newHead.blck)
of PayloadExecutionStatus.invalid, PayloadExecutionStatus.invalid_block_hash:
self.attestationPool[].forkChoice.mark_root_invalid(newHead.blck.root)
self.quarantine[].addUnviable(newHead.blck.root)
return Opt.none(void)
of PayloadExecutionStatus.accepted, PayloadExecutionStatus.syncing:
# Don't do anything. Either newHead.blck.executionValid was already false,
# in which case it'd be superfluous to set it to false again, or the block
# was marked as `VALID` in the `newPayload` path already, in which case it
# is fine to keep it as valid here. Conceptually, were this to be lines of
# code, it'd be something like
# if newHead.blck.executionValid:
# do nothing because of latter case
# else:
# do nothing because it's a no-op
# So, either way, do nothing.
discard
return Opt[void].ok()
func getKnownValidatorsForBlsChangeTracking(
self: ConsensusManager, newHead: BlockRef): seq[ValidatorIndex] =
# Ensure that large nodes won't be overloaded by a nice-to-have, but
# inessential cosmetic feature.
const MAX_CHECKED_INDICES = 64
if newHead.bid.slot.epoch >= self.dag.cfg.CAPELLA_FORK_EPOCH:
var res = newSeqOfCap[ValidatorIndex](min(
len(self.actionTracker.knownValidators), MAX_CHECKED_INDICES))
for vi in self.actionTracker.knownValidators.keys():
res.add vi
if res.len >= MAX_CHECKED_INDICES:
break
res
else:
# It is not possible for any BLS to execution changes, for any validator,
# to have been yet processed.
# https://github.com/nim-lang/Nim/issues/19802
(static(@[]))
proc updateHead*(self: var ConsensusManager, newHead: BlockRef) =
## Trigger fork choice and update the DAG with the new head block
## This does not automatically prune the DAG after finalization
## `pruneFinalized` must be called for pruning.
# Store the new head in the chain DAG - this may cause epochs to be
# justified and finalized
self.dag.updateHead(
newHead, self.quarantine[],
self.getKnownValidatorsForBlsChangeTracking(newHead))
self.checkExpectedBlock()
proc updateHead*(self: var ConsensusManager, wallSlot: Slot) =
## Trigger fork choice and update the DAG with the new head block
## This does not automatically prune the DAG after finalization
## `pruneFinalized` must be called for pruning.
# Grab the new head according to our latest attestation data
let
newHead = self.attestationPool[].selectOptimisticHead(
wallSlot.start_beacon_time).valueOr:
warn "Head selection failed, using previous head",
head = shortLog(self.dag.head), wallSlot
return
executionBlockHash = self.dag.loadExecutionBlockHash(newHead.blck)
if executionBlockHash.isSome and executionBlockHash.unsafeGet.isZero:
# Blocks without execution payloads can't be optimistic.
self.dag.markBlockVerified(newHead.blck)
self.updateHead(newHead.blck)
func isSynced(dag: ChainDAGRef, wallSlot: Slot): bool =
# This is a tweaked version of the beacon_validators isSynced. TODO, refactor
# that one so this becomes the default version, with the same information to
# work with. For the head slot, use the DAG head regardless of what head the
# proposer forkchoiceUpdated is using, because by the beacon_validators might
# be ready to actually propose, it's going to do so from the DAG head. Given
# the defaultSyncHorizon, it will start triggering in time so that potential
# discrepancies between the head here, and the head the DAG has (which might
# not yet be updated) won't be visible.
if dag.head.slot + defaultSyncHorizon < wallSlot:
false
else:
dag.head.executionValid
proc checkNextProposer(
dag: ChainDAGRef, actionTracker: ActionTracker,
dynamicFeeRecipientsStore: ref DynamicFeeRecipientsStore,
wallSlot: Slot):
Opt[(ValidatorIndex, ValidatorPubKey)] =
let nextWallSlot = wallSlot + 1
# Avoid long rewinds during syncing, when it's not going to propose. Though
# this is preparing for a proposal on `nextWallSlot`, it can't possibly yet
# be on said slot, so still check just `wallSlot`.
if not dag.isSynced(wallSlot):
return Opt.none((ValidatorIndex, ValidatorPubKey))
let proposer = ? dag.getProposer(dag.head, nextWallSlot)
if actionTracker.getNextProposalSlot(wallSlot) != nextWallSlot and
dynamicFeeRecipientsStore[].getDynamicFeeRecipient(
proposer, nextWallSlot.epoch).isNone:
return Opt.none((ValidatorIndex, ValidatorPubKey))
let proposerKey = dag.validatorKey(proposer).get().toPubKey
Opt.some((proposer, proposerKey))
proc checkNextProposer*(self: ref ConsensusManager, wallSlot: Slot):
Opt[(ValidatorIndex, ValidatorPubKey)] =
self.dag.checkNextProposer(
self.actionTracker, self.dynamicFeeRecipientsStore, wallSlot)
proc getFeeRecipient*(
self: ConsensusManager, pubkey: ValidatorPubKey,
validatorIdx: Opt[ValidatorIndex], epoch: Epoch): Eth1Address =
let validator =
if validatorIdx.isSome():
withState(self.dag.headState):
if validatorIdx.get < forkyState.data.validators.lenu64:
Opt.some forkyState.data.validators.item(validatorIdx.get)
else:
Opt.none Validator
else:
Opt.none Validator
getFeeRecipient(self.dynamicFeeRecipientsStore, pubkey, validatorIdx,
validator, self.defaultFeeRecipient, self.validatorsDir,
epoch)
proc getGasLimit*(self: ConsensusManager, pubkey: ValidatorPubKey): uint64 =
getGasLimit(self.validatorsDir, self.defaultGasLimit, pubkey)
proc runProposalForkchoiceUpdated*(
self: ref ConsensusManager, wallSlot: Slot): Future[Opt[void]] {.async: (raises: [CancelledError]).} =
let
nextWallSlot = wallSlot + 1
(validatorIndex, nextProposer) = self.checkNextProposer(wallSlot).valueOr:
return err()
debug "runProposalForkchoiceUpdated: expected to be proposing next slot",
nextWallSlot, validatorIndex, nextProposer
# In Capella and later, computing correct withdrawals would mean creating a
# proposal state. Instead, only do that at proposal time.
if nextWallSlot.is_epoch:
debug "runProposalForkchoiceUpdated: not running early fcU for epoch-aligned proposal slot",
nextWallSlot, validatorIndex, nextProposer
return err()
# Approximately lines up with validator_duties version. Used optimistically/
# opportunistically, so mismatches are fine if not too frequent.
let
timestamp = withState(self.dag.headState):
compute_timestamp_at_slot(forkyState.data, nextWallSlot)
# If the current head block still forms the basis of the eventual proposal
# state, then its `get_randao_mix` will remain unchanged as well, as it is
# constant until the next block.
randomData = withState(self.dag.headState):
get_randao_mix(forkyState.data, get_current_epoch(forkyState.data)).data
feeRecipient = self[].getFeeRecipient(
nextProposer, Opt.some(validatorIndex), nextWallSlot.epoch)
beaconHead = self.attestationPool[].getBeaconHead(self.dag.head)
headBlockHash = ? self.dag.loadExecutionBlockHash(beaconHead.blck)
if headBlockHash.isZero:
return err()
let safeBlockHash = beaconHead.safeExecutionBlockHash
withState(self.dag.headState):
template callForkchoiceUpdated(fcPayloadAttributes: auto) =
let (status, _) = await self.elManager.forkchoiceUpdated(
headBlockHash, safeBlockHash,
beaconHead.finalizedExecutionBlockHash,
payloadAttributes = Opt.some fcPayloadAttributes)
debug "Fork-choice updated for proposal", status
static: doAssert high(ConsensusFork) == ConsensusFork.Electra
when consensusFork >= ConsensusFork.Deneb:
# https://github.com/ethereum/execution-apis/blob/90a46e9137c89d58e818e62fa33a0347bba50085/src/engine/prague.md
# does not define any new forkchoiceUpdated, so reuse V3 from Dencun
callForkchoiceUpdated(PayloadAttributesV3(
timestamp: Quantity timestamp,
prevRandao: FixedBytes[32] randomData,
suggestedFeeRecipient: feeRecipient,
withdrawals:
toEngineWithdrawals get_expected_withdrawals(forkyState.data),
parentBeaconBlockRoot: beaconHead.blck.bid.root.asBlockHash))
elif consensusFork >= ConsensusFork.Capella:
callForkchoiceUpdated(PayloadAttributesV2(
timestamp: Quantity timestamp,
prevRandao: FixedBytes[32] randomData,
suggestedFeeRecipient: feeRecipient,
withdrawals:
toEngineWithdrawals get_expected_withdrawals(forkyState.data)))
else:
callForkchoiceUpdated(PayloadAttributesV1(
timestamp: Quantity timestamp,
prevRandao: FixedBytes[32] randomData,
suggestedFeeRecipient: feeRecipient))
ok()
proc updateHeadWithExecution*(
self: ref ConsensusManager, initialNewHead: BeaconHead,
getBeaconTimeFn: GetBeaconTimeFn) {.async: (raises: [CancelledError]).} =
## Trigger fork choice and update the DAG with the new head block
## This does not automatically prune the DAG after finalization
## `pruneFinalized` must be called for pruning.
# Grab the new head according to our latest attestation data
try:
# Ensure dag.updateHead has most current information
var
attempts = 0
newHead = initialNewHead
while (await self.updateExecutionClientHead(newHead)).isErr:
# This proc is called on every new block; guarantee timely return
inc attempts
const maxAttempts = 5
if attempts >= maxAttempts:
warn "updateHeadWithExecution: too many attempts to recover from invalid payload",
attempts, maxAttempts, newHead, initialNewHead
break
# Select new head for next attempt
let
wallTime = getBeaconTimeFn()
nextHead = self.attestationPool[].selectOptimisticHead(wallTime).valueOr:
warn "Head selection failed after invalid block, using previous head",
newHead, wallSlot = wallTime.slotOrZero
break
warn "updateHeadWithExecution: attempting to recover from invalid payload",
attempts, maxAttempts, newHead, initialNewHead, nextHead
newHead = nextHead
# Store the new head in the chain DAG - this may cause epochs to be
# justified and finalized
self.dag.updateHead(
newHead.blck, self.quarantine[],
self[].getKnownValidatorsForBlsChangeTracking(newHead.blck))
# If this node should propose next slot, start preparing payload. Both
# fcUs are useful: the updateExecutionClientHead(newHead) call updates
# the head state (including optimistic status) that self.dagUpdateHead
# needs while runProposalForkchoiceUpdated requires RANDAO information
# from the head state corresponding to the `newHead` block, which only
# self.dag.updateHead(...) sets up.
discard await self.runProposalForkchoiceUpdated(getBeaconTimeFn().slotOrZero)
self[].checkExpectedBlock()
except CatchableError as exc:
debug "updateHeadWithExecution error",
error = exc.msg
proc pruneStateCachesAndForkChoice*(self: var ConsensusManager) =
## Prune unneeded and invalidated data after finalization
## - the DAG state checkpoints
## - the DAG EpochRef
## - the attestation pool/fork choice
# Cleanup DAG & fork choice if we have a finalized head
if self.dag.needStateCachesAndForkChoicePruning():
self.dag.pruneStateCachesDAG()
self.attestationPool[].prune()