nimbus-eth2/beacon_chain/gossip_processing/gossip_to_consensus.nim

398 lines
14 KiB
Nim

# beacon_chain
# Copyright (c) 2018-2021 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.
import
std/math,
stew/results,
chronicles, chronos, metrics,
../spec/[crypto, datatypes, digest],
../consensus_object_pools/[block_clearance, blockchain_dag, attestation_pool],
./consensus_manager,
../beacon_node_types,
../beacon_clock, ../conf, ../ssz/sszdump
# Gossip Queue Manager
# ------------------------------------------------------------------------------
# The queue manager moves blocks from "Gossip validated" to "Consensus verified"
declareHistogram beacon_store_block_duration_seconds,
"storeBlock() duration", buckets = [0.25, 0.5, 1, 2, 4, 8, Inf]
type
SyncBlock* = object
blk*: SignedBeaconBlock
resfut*: Future[Result[void, BlockError]]
BlockEntry* = object
# Exported for "test_sync_manager"
v*: SyncBlock
AttestationEntry = object
v: Attestation
attesting_indices: seq[ValidatorIndex]
AggregateEntry* = AttestationEntry
VerifQueueManager* = object
## This manages the queues of blocks and attestations.
## Blocks and attestations are enqueued in a gossip-validated state
##
## from:
## - Gossip (when synced)
## - SyncManager (during sync)
## - RequestManager (missing ancestor blocks)
##
## are then consensus-verified and added to:
## - the blockchain DAG
## - database
## - attestation pool
## - fork choice
##
## The queue manager doesn't manage exits (voluntary, attester slashing or proposer slashing)
## as don't need extra verification and can be added to the exit pool as soon as they are gossip-validated.
# Config
# ----------------------------------------------------------------
dumpEnabled: bool
dumpDirInvalid: string
dumpDirIncoming: string
# Clock
# ----------------------------------------------------------------
getWallTime: GetWallTimeFn
# Producers
# ----------------------------------------------------------------
blocksQueue*: AsyncQueue[BlockEntry] # Exported for "test_sync_manager"
attestationsQueue: AsyncQueue[AttestationEntry]
aggregatesQueue: AsyncQueue[AggregateEntry]
# Consumer
# ----------------------------------------------------------------
consensusManager: ref ConsensusManager
## Blockchain DAG, AttestationPool and Quarantine
{.push raises: [Defect].}
# Initialization
# ------------------------------------------------------------------------------
proc new*(T: type VerifQueueManager,
conf: BeaconNodeConf,
consensusManager: ref ConsensusManager,
getWallTime: GetWallTimeFn): ref VerifQueueManager =
(ref VerifQueueManager)(
dumpEnabled: conf.dumpEnabled,
dumpDirInvalid: conf.dumpDirInvalid,
dumpDirIncoming: conf.dumpDirIncoming,
getWallTime: getWallTime,
blocksQueue: newAsyncQueue[BlockEntry](1),
# limit to the max number of aggregates we expect to see in one slot
aggregatesQueue: newAsyncQueue[AggregateEntry](
(TARGET_AGGREGATORS_PER_COMMITTEE * MAX_COMMITTEES_PER_SLOT).int),
# This queue is a bit harder to bound reasonably - we want to get a good
# spread of votes across committees - ideally at least TARGET_COMMITTEE_SIZE
# per committee - assuming randomness in vote arrival, this limit should
# cover that but of course, when votes arrive depends on a number of
# factors that are not entire random
attestationsQueue: newAsyncQueue[AttestationEntry](
(TARGET_COMMITTEE_SIZE * MAX_COMMITTEES_PER_SLOT).int),
consensusManager: consensusManager
)
# Sync callbacks
# ------------------------------------------------------------------------------
proc done*(blk: SyncBlock) =
## Send signal to [Sync/Request]Manager that the block ``blk`` has passed
## verification successfully.
if blk.resfut != nil:
blk.resfut.complete(Result[void, BlockError].ok())
proc fail*(blk: SyncBlock, error: BlockError) =
## Send signal to [Sync/Request]Manager that the block ``blk`` has NOT passed
## verification with specific ``error``.
if blk.resfut != nil:
blk.resfut.complete(Result[void, BlockError].err(error))
proc complete*(blk: SyncBlock, res: Result[void, BlockError]) =
## Send signal to [Sync/Request]Manager about result ``res`` of block ``blk``
## verification.
if blk.resfut != nil:
blk.resfut.complete(res)
# Enqueue
# ------------------------------------------------------------------------------
{.pop.}
proc addBlock*(self: var VerifQueueManager, syncBlock: SyncBlock) {.raises: [Exception].} =
## Enqueue a Gossip-validated block for consensus verification
# Backpressure:
# If no item can be enqueued because buffer is full,
# we suspend here.
# Producers:
# - Gossip (when synced)
# - SyncManager (during sync)
# - RequestManager (missing ancestor blocks)
# TODO: solve the signature requiring raise: [Exception]
# even when push/pop is used
asyncSpawn(
try:
self.blocksQueue.addLast(BlockEntry(v: syncBlock))
except Exception as e:
# Chronos can in theory raise an untyped exception in `internalCheckComplete`
# which asyncSpawn doesn't like.
raiseAssert e.msg
)
{.push raises: [Defect].}
proc addAttestation*(self: var VerifQueueManager, att: Attestation, att_indices: seq[ValidatorIndex]) =
## Enqueue a Gossip-validated attestation for consensus verification
# Backpressure:
# no handling
# Producer:
# - Gossip (when synced)
while self.attestationsQueue.full():
try:
notice "Queue full, dropping oldest attestation",
dropped = shortLog(self.attestationsQueue[0].v)
discard self.attestationsQueue.popFirstNoWait()
except AsyncQueueEmptyError as exc:
raiseAssert "If queue is full, we have at least one item! " & exc.msg
try:
self.attestationsQueue.addLastNoWait(
AttestationEntry(v: att, attesting_indices: att_indices))
except AsyncQueueFullError as exc:
raiseAssert "We just checked that queue is not full! " & exc.msg
proc addAggregate*(self: var VerifQueueManager, agg: SignedAggregateAndProof, att_indices: seq[ValidatorIndex]) =
## Enqueue a Gossip-validated aggregate attestation for consensus verification
# Backpressure:
# no handling
# Producer:
# - Gossip (when synced)
while self.aggregatesQueue.full():
try:
notice "Queue full, dropping oldest aggregate",
dropped = shortLog(self.aggregatesQueue[0].v)
discard self.aggregatesQueue.popFirstNoWait()
except AsyncQueueEmptyError as exc:
raiseAssert "We just checked that queue is not full! " & exc.msg
try:
self.aggregatesQueue.addLastNoWait(AggregateEntry(
v: agg.message.aggregate,
attesting_indices: att_indices))
except AsyncQueueFullError as exc:
raiseAssert "We just checked that queue is not full! " & exc.msg
# Storage
# ------------------------------------------------------------------------------
proc dumpBlock*[T](
self: VerifQueueManager, signedBlock: SignedBeaconBlock,
res: Result[T, (ValidationResult, BlockError)]) =
if self.dumpEnabled and res.isErr:
case res.error[1]
of Invalid:
dump(
self.dumpDirInvalid, signedBlock)
of MissingParent:
dump(
self.dumpDirIncoming, signedBlock)
else:
discard
proc storeBlock(
self: var VerifQueueManager, signedBlock: SignedBeaconBlock,
wallSlot: Slot): Result[void, BlockError] =
let
start = Moment.now()
attestationPool = self.consensusManager.attestationPool
let blck = self.consensusManager.chainDag.addRawBlock(self.consensusManager.quarantine, signedBlock) do (
blckRef: BlockRef, trustedBlock: TrustedSignedBeaconBlock,
epochRef: EpochRef, state: HashedBeaconState):
# Callback add to fork choice if valid
attestationPool[].addForkChoice(
epochRef, blckRef, trustedBlock.message, wallSlot)
self.dumpBlock(signedBlock, blck)
# There can be a scenario where we receive a block we already received.
# However this block was before the last finalized epoch and so its parent
# was pruned from the ForkChoice.
if blck.isErr:
return err(blck.error[1])
let duration = (Moment.now() - start).toFloatSeconds()
beacon_store_block_duration_seconds.observe(duration)
ok()
# Event Loop
# ------------------------------------------------------------------------------
proc processAttestation(
self: var VerifQueueManager, entry: AttestationEntry) =
logScope:
signature = shortLog(entry.v.signature)
let
wallTime = self.getWallTime()
(afterGenesis, wallSlot) = wallTime.toSlot()
if not afterGenesis:
error "Processing attestation before genesis, clock turned back?"
quit 1
trace "Processing attestation"
self.consensusManager.attestationPool[].addAttestation(
entry.v, entry.attesting_indices, wallSlot)
proc processAggregate(
self: var VerifQueueManager, entry: AggregateEntry) =
logScope:
signature = shortLog(entry.v.signature)
let
wallTime = self.getWallTime()
(afterGenesis, wallSlot) = wallTime.toSlot()
if not afterGenesis:
error "Processing aggregate before genesis, clock turned back?"
quit 1
trace "Processing aggregate"
self.consensusManager.attestationPool[].addAttestation(
entry.v, entry.attesting_indices, wallSlot)
proc processBlock(self: var VerifQueueManager, entry: BlockEntry) =
logScope:
blockRoot = shortLog(entry.v.blk.root)
let
wallTime = self.getWallTime()
(afterGenesis, wallSlot) = wallTime.toSlot()
if not afterGenesis:
error "Processing block before genesis, clock turned back?"
quit 1
let
start = now(chronos.Moment)
res = self.storeBlock(entry.v.blk, wallSlot)
storeDone = now(chronos.Moment)
if res.isOk():
# Eagerly update head in case the new block gets selected
self.consensusManager[].updateHead(wallSlot) # This also eagerly prunes the blocks DAG to prevent processing forks.
# self.consensusManager.pruneStateCachesDAG() # Amortized pruning, we don't prune states & fork choice here but in `onSlotEnd`()
let updateDone = now(chronos.Moment)
let storeBlockDuration = storeDone - start
let updateHeadDuration = updateDone - storeDone
let overallDuration = updateDone - start
let storeSpeed =
block:
let secs = float(chronos.seconds(1).nanoseconds)
if not(overallDuration.isZero()):
let v = secs / float(overallDuration.nanoseconds)
round(v * 10_000) / 10_000
else:
0.0
debug "Block processed",
local_head_slot = self.consensusManager.chainDag.head.slot,
store_speed = storeSpeed,
block_slot = entry.v.blk.message.slot,
store_block_duration = $storeBlockDuration,
update_head_duration = $updateHeadDuration,
overall_duration = $overallDuration
if entry.v.resFut != nil:
entry.v.resFut.complete(Result[void, BlockError].ok())
elif res.error() in {BlockError.Duplicate, BlockError.Old}:
# These are harmless / valid outcomes - for the purpose of scoring peers,
# they are ok
if entry.v.resFut != nil:
entry.v.resFut.complete(Result[void, BlockError].ok())
else:
if entry.v.resFut != nil:
entry.v.resFut.complete(Result[void, BlockError].err(res.error()))
{.pop.} # Chronos: Error: can raise an unlisted exception: ref Exception
proc runQueueProcessingLoop*(self: ref VerifQueueManager) {.async.} =
# Blocks in eth2 arrive on a schedule for every slot:
#
# * Block arrives at time 0
# * Attestations arrives at time 4
# * Aggregate arrives at time 8
var
blockFut = self[].blocksQueue.popFirst()
aggregateFut = self[].aggregatesQueue.popFirst()
attestationFut = self[].attestationsQueue.popFirst()
while true:
# Cooperative concurrency: one idle calculation step per loop - because
# we run both networking and CPU-heavy things like block processing
# on the same thread, we need to make sure that there is steady progress
# on the networking side or we get long lockups that lead to timeouts.
const
# We cap waiting for an idle slot in case there's a lot of network traffic
# taking up all CPU - we don't want to _completely_ stop processing blocks
# in this case (attestations will get dropped) - doing so also allows us
# to benefit from more batching / larger network reads when under load.
idleTimeout = 10.milliseconds
# Attestation processing is fairly quick and therefore done in batches to
# avoid some of the `Future` overhead
attestationBatch = 16
discard await idleAsync().withTimeout(idleTimeout)
# Avoid one more `await` when there's work to do
if not (blockFut.finished or aggregateFut.finished or attestationFut.finished):
trace "Waiting for processing work"
await blockFut or aggregateFut or attestationFut
# Only run one task per idle iteration, in priority order: blocks are needed
# for all other processing - then come aggregates which are cheap to
# process but might have a big impact on fork choice - last come
# attestations which individually have the smallest effect on chain progress
if blockFut.finished:
self[].processBlock(blockFut.read())
blockFut = self[].blocksQueue.popFirst()
elif aggregateFut.finished:
# aggregates will be dropped under heavy load on producer side
self[].processAggregate(aggregateFut.read())
for i in 0..<attestationBatch: # process a few at a time - this is fairly fast
if self[].aggregatesQueue.empty():
break
self[].processAggregate(self[].aggregatesQueue.popFirstNoWait())
aggregateFut = self[].aggregatesQueue.popFirst()
elif attestationFut.finished:
# attestations will be dropped under heavy load on producer side
self[].processAttestation(attestationFut.read())
for i in 0..<attestationBatch: # process a few at a time - this is fairly fast
if self[].attestationsQueue.empty():
break
self[].processAttestation(self[].attestationsQueue.popFirstNoWait())
attestationFut = self[].attestationsQueue.popFirst()