nimbus-eth2/beacon_chain/consensus_object_pools/attestation_pool.nim

754 lines
29 KiB
Nim
Raw Normal View History

# beacon_chain
# Copyright (c) 2018-2022 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: [Defect].}
import
# Standard libraries
std/[options, tables, sequtils],
# Status libraries
metrics,
chronicles, stew/byteutils,
# Internal
disentangle eth2 types from the ssz library (#2785) * reorganize ssz dependencies This PR continues the work in https://github.com/status-im/nimbus-eth2/pull/2646, https://github.com/status-im/nimbus-eth2/pull/2779 as well as past issues with serialization and type, to disentangle SSZ from eth2 and at the same time simplify imports and exports with a structured approach. The principal idea here is that when a library wants to introduce SSZ support, they do so via 3 files: * `ssz_codecs` which imports and reexports `codecs` - this covers the basic byte conversions and ensures no overloads get lost * `xxx_merkleization` imports and exports `merkleization` to specialize and get access to `hash_tree_root` and friends * `xxx_ssz_serialization` imports and exports `ssz_serialization` to specialize ssz for a specific library Those that need to interact with SSZ always import the `xxx_` versions of the modules and never `ssz` itself so as to keep imports simple and safe. This is similar to how the REST / JSON-RPC serializers are structured in that someone wanting to serialize spec types to REST-JSON will import `eth2_rest_serialization` and nothing else. * split up ssz into a core library that is independendent of eth2 types * rename `bytes_reader` to `codec` to highlight that it contains coding and decoding of bytes and native ssz types * remove tricky List init overload that causes compile issues * get rid of top-level ssz import * reenable merkleization tests * move some "standard" json serializers to spec * remove `ValidatorIndex` serialization for now * remove test_ssz_merkleization * add tests for over/underlong byte sequences * fix broken seq[byte] test - seq[byte] is not an SSZ type There are a few things this PR doesn't solve: * like #2646 this PR is weak on how to handle root and other dontSerialize fields that "sometimes" should be computed - the same problem appears in REST / JSON-RPC etc * Fix a build problem on macOS * Another way to fix the macOS builds Co-authored-by: Zahary Karadjov <zahary@gmail.com>
2021-08-18 18:57:58 +00:00
../spec/[beaconstate, eth2_merkleization, forks, helpers, validator],
../spec/datatypes/[phase0, altair, bellatrix],
"."/[spec_cache, blockchain_dag, block_quarantine],
../fork_choice/fork_choice,
../beacon_clock
export options, tables, phase0, altair, bellatrix, blockchain_dag, fork_choice
const
ATTESTATION_LOOKBACK* =
min(24'u64, SLOTS_PER_EPOCH) + MIN_ATTESTATION_INCLUSION_DELAY
## The number of slots we'll keep track of in terms of "free" attestations
## that potentially could be added to a newly created block
type
OnAttestationCallback* = proc(data: Attestation) {.gcsafe, raises: [Defect].}
Validation* = object
## Validations collect a set of signatures for a distict attestation - in
## eth2, a single bit is used to keep track of which signatures have been
## added to the aggregate meaning that only non-overlapping aggregates may
## be further combined.
aggregation_bits*: CommitteeValidatorsBits
aggregate_signature*: AggregateSignature
AttestationEntry* = object
## Each entry holds the known signatures for a particular, distinct vote
data*: AttestationData
committee_len*: int
singles*: Table[int, CookedSig] ## \
## On the attestation subnets, only attestations with a single vote are
## allowed - these can be collected separately to top up aggregates with -
## here we collect them by mapping index in committee to a vote
aggregates*: seq[Validation]
AttestationTable* = Table[Eth2Digest, AttestationEntry]
## Depending on the world view of the various validators, they may have
## voted on different states - this map keeps track of each vote keyed by
## hash_tree_root(AttestationData)
AttestationPool* = object
## The attestation pool keeps track of all attestations that potentially
## could be added to a block during block production.
## These attestations also contribute to the fork choice, which combines
## "free" attestations with those found in past blocks - these votes
## are tracked separately in the fork choice.
candidates*: array[ATTESTATION_LOOKBACK, AttestationTable] ## \
## We keep one item per slot such that indexing matches slot number
## together with startingSlot
startingSlot*: Slot ## \
## Generally, we keep attestations only until a slot has been finalized -
## after that, they may no longer affect fork choice.
dag*: ChainDAGRef
quarantine*: ref Quarantine
forkChoice*: ForkChoice
nextAttestationEpoch*: seq[tuple[subnet: Epoch, aggregate: Epoch]] ## \
## sequence based on validator indices
onAttestationAdded*: OnAttestationCallback
logScope: topics = "attpool"
declareGauge attestation_pool_block_attestation_packing_time,
"Time it took to create list of attestations for block"
proc init*(T: type AttestationPool, dag: ChainDAGRef,
quarantine: ref Quarantine,
onAttestation: OnAttestationCallback = nil,
proposerBoosting: bool = false): T =
## Initialize an AttestationPool from the dag `headState`
## The `finalized_root` works around the finalized_checkpoint of the genesis block
## holding a zero_root.
let finalizedEpochRef = dag.getFinalizedEpochRef()
var forkChoice = ForkChoice.init(
finalizedEpochRef,
dag.finalizedHead.blck,
proposerBoosting)
# Feed fork choice with unfinalized history - during startup, block pool only
# keeps track of a single history so we just need to follow it
doAssert dag.heads.len == 1, "Init only supports a single history"
var blocks: seq[BlockRef]
var cur = dag.head
# When the chain is finalizing, the votes between the head block and the
# finalized checkpoint should be enough for a stable fork choice - when the
# chain is not finalizing, we want to seed it with as many votes as possible
# since the whole history of each branch might be significant. It is however
# a game of diminishing returns, and we have to weigh it against the time
# it takes to replay that many blocks during startup and thus miss _new_
# votes.
const ForkChoiceHorizon = 256
while cur != dag.finalizedHead.blck:
blocks.add cur
cur = cur.parent
info "Initializing fork choice", unfinalized_blocks = blocks.len
var epochRef = finalizedEpochRef
for i in 0..<blocks.len:
let
blckRef = blocks[blocks.len - i - 1]
status =
if i < (blocks.len - ForkChoiceHorizon) and (i mod 1024 != 0):
# Fork choice needs to know about the full block tree up to the
# finalization point, but doesn't really need to have overly accurate
# justification and finalization points until we get close to head -
# nonetheless, we'll make sure to pass a fresh finalization point now
# and then to make sure the fork choice data structure doesn't grow
# too big - getting an EpochRef can be expensive.
forkChoice.backend.process_block(
blckRef.root, blckRef.parent.root,
epochRef.current_justified_checkpoint,
epochRef.finalized_checkpoint)
else:
epochRef = dag.getEpochRef(blckRef, blckRef.slot.epoch, false).expect(
"Getting an EpochRef should always work for non-finalized blocks")
let blck = dag.getForkedBlock(blckRef.bid).expect(
"Should be able to load initial fork choice blocks")
withBlck(blck):
forkChoice.process_block(
limit by-root requests to non-finalized blocks (#3293) * limit by-root requests to non-finalized blocks Presently, we keep a mapping from block root to `BlockRef` in memory - this has simplified reasoning about the dag, but is not sustainable with the chain growing. We can distinguish between two cases where by-root access is useful: * unfinalized blocks - this is where the beacon chain is operating generally, by validating incoming data as interesting for future fork choice decisions - bounded by the length of the unfinalized period * finalized blocks - historical access in the REST API etc - no bounds, really In this PR, we limit the by-root block index to the first use case: finalized chain data can more efficiently be addressed by slot number. Future work includes: * limiting the `BlockRef` horizon in general - each instance is 40 bytes+overhead which adds up - this needs further refactoring to deal with the tail vs state problem * persisting the finalized slot-to-hash index - this one also keeps growing unbounded (albeit slowly) Anyway, this PR easily shaves ~128mb of memory usage at the time of writing. * No longer honor `BeaconBlocksByRoot` requests outside of the non-finalized period - previously, Nimbus would generously return any block through this libp2p request - per the spec, finalized blocks should be fetched via `BeaconBlocksByRange` instead. * return `Opt[BlockRef]` instead of `nil` when blocks can't be found - this becomes a lot more common now and thus deserves more attention * `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized blocks from now - `finalizedBlocks` covers the other `BlockRef` instances * in backfill, verify that the last backfilled block leads back to genesis, or panic * add backfill timings to log * fix missing check that `BlockRef` block can be fetched with `getForkedBlock` reliably * shortcut doppelganger check when feature is not enabled * in REST/JSON-RPC, fetch blocks without involving `BlockRef` * fix dag.blocks ref
2022-01-21 11:33:16 +00:00
dag, epochRef, blckRef, blck.message,
blckRef.slot.start_beacon_time)
doAssert status.isOk(), "Error in preloading the fork choice: " & $status.error
info "Fork choice initialized",
justified_epoch = getStateField(
dag.headState, current_justified_checkpoint).epoch,
finalized_epoch = getStateField(dag.headState, finalized_checkpoint).epoch,
finalized_root = shortLog(dag.finalizedHead.blck.root)
T(
dag: dag,
quarantine: quarantine,
forkChoice: forkChoice,
onAttestationAdded: onAttestation
)
proc addForkChoiceVotes(
pool: var AttestationPool, slot: Slot,
attesting_indices: openArray[ValidatorIndex], block_root: Eth2Digest,
wallTime: BeaconTime) =
# Add attestation votes to fork choice
if (let v = pool.forkChoice.on_attestation(
pool.dag, slot, block_root, attesting_indices, wallTime);
v.isErr):
# This indicates that the fork choice and the chain dag are out of sync -
# this is most likely the result of a bug, but we'll try to keep going -
# hopefully the fork choice will heal itself over time.
error "Couldn't add attestation to fork choice, bug?", err = v.error()
func candidateIdx(pool: AttestationPool, slot: Slot): Option[int] =
if slot >= pool.startingSlot and
slot < (pool.startingSlot + pool.candidates.lenu64):
some(int(slot mod pool.candidates.lenu64))
else:
none(int)
proc updateCurrent(pool: var AttestationPool, wallSlot: Slot) =
if wallSlot + 1 < pool.candidates.lenu64:
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
return # Genesis
let
newStartingSlot = wallSlot + 1 - pool.candidates.lenu64
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
if newStartingSlot < pool.startingSlot:
error "Current slot older than attestation pool view, clock reset?",
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
startingSlot = pool.startingSlot, newStartingSlot, wallSlot
return
# As time passes we'll clear out any old attestations as they are no longer
# viable to be included in blocks
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
if newStartingSlot - pool.startingSlot >= pool.candidates.lenu64():
# In case many slots passed since the last update, avoid iterating over
# the same indices over and over
pool.candidates = default(type(pool.candidates))
else:
for i in pool.startingSlot..newStartingSlot:
pool.candidates[i.uint64 mod pool.candidates.lenu64] = AttestationTable()
pool.startingSlot = newStartingSlot
func oneIndex(bits: CommitteeValidatorsBits): Option[int] =
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
# Find the index of the set bit, iff one bit is set
var res = none(int)
for idx in 0..<bits.len():
if bits[idx]:
if res.isNone():
res = some(idx)
else: # More than one bit set!
return none(int)
res
func toAttestation(entry: AttestationEntry, validation: Validation): Attestation =
Attestation(
aggregation_bits: validation.aggregation_bits,
data: entry.data,
signature: validation.aggregate_signature.finish().toValidatorSig()
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
)
func updateAggregates(entry: var AttestationEntry) =
# Upgrade the list of aggregates to ensure that there is at least one
# aggregate (assuming there are singles) and all aggregates have all
# singles incorporated
if entry.singles.len() == 0:
return
if entry.aggregates.len() == 0:
# If there are singles, we can create an aggregate from them that will
# represent our best knowledge about the current votes
for index_in_committee, signature in entry.singles:
if entry.aggregates.len() == 0:
# Create aggregate on first iteration..
entry.aggregates.add(
Validation(
aggregation_bits: CommitteeValidatorsBits.init(entry.committee_len),
aggregate_signature: AggregateSignature.init(signature)
))
else:
entry.aggregates[0].aggregate_signature.aggregate(signature)
entry.aggregates[0].aggregation_bits.setBit(index_in_committee)
else:
# There already exist aggregates - we'll try to top them up by adding
# singles to them - for example, it may happen that we're being asked to
# produce a block 4s after creating an aggregate and new information may
# have arrived by then.
# In theory, also aggregates could be combined but finding the best
# combination is hard, so we'll pragmatically use singles only here
var updated = false
for index_in_committee, signature in entry.singles:
for v in entry.aggregates.mitems():
if not v.aggregation_bits[index_in_committee]:
v.aggregation_bits.setBit(index_in_committee)
v.aggregate_signature.aggregate(signature)
updated = true
if updated:
# One or more aggregates were updated - time to remove the ones that are
# pure subsets of the others. This may lead to quadratic behaviour, but
# the number of aggregates for the entry is limited by the number of
# aggregators on the topic which is capped `is_aggregator` and
# TARGET_AGGREGATORS_PER_COMMITTEE
var i = 0
while i < entry.aggregates.len():
var j = 0
while j < entry.aggregates.len():
if i != j and entry.aggregates[i].aggregation_bits.isSubsetOf(
entry.aggregates[j].aggregation_bits):
entry.aggregates[i] = entry.aggregates[j]
entry.aggregates.del(j)
dec i # Rerun checks on the new `i` item
break
else:
inc j
inc i
func covers(entry: AttestationEntry, bits: CommitteeValidatorsBits): bool =
for i in 0..<entry.aggregates.len():
if bits.isSubsetOf(entry.aggregates[i].aggregation_bits):
return true
false
proc addAttestation(entry: var AttestationEntry,
attestation: Attestation,
signature: CookedSig): bool =
logScope:
attestation = shortLog(attestation)
let
singleIndex = oneIndex(attestation.aggregation_bits)
if singleIndex.isSome():
if singleIndex.get() in entry.singles:
trace "Attestation already seen",
singles = entry.singles.len(),
aggregates = entry.aggregates.len()
return false
debug "Attestation resolved",
singles = entry.singles.len(),
aggregates = entry.aggregates.len()
entry.singles[singleIndex.get()] = signature
else:
# More than one vote in this attestation
if entry.covers(attestation.aggregation_bits):
return false
# Since we're adding a new aggregate, we can now remove existing
# aggregates that don't add any new votes
entry.aggregates.keepItIf(
not it.aggregation_bits.isSubsetOf(attestation.aggregation_bits))
entry.aggregates.add(Validation(
aggregation_bits: attestation.aggregation_bits,
aggregate_signature: AggregateSignature.init(signature)))
debug "Aggregate resolved",
singles = entry.singles.len(),
aggregates = entry.aggregates.len()
true
proc addAttestation*(pool: var AttestationPool,
attestation: Attestation,
attesting_indices: openArray[ValidatorIndex],
signature: CookedSig,
wallTime: BeaconTime) =
## Add an attestation to the pool, assuming it's been validated already.
##
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
## Assuming the votes in the attestation have not already been seen, the
## attestation will be added to the fork choice and lazily added to a list of
## attestations for future aggregation and block production.
logScope:
attestation = shortLog(attestation)
doAssert attestation.signature == signature.toValidatorSig(),
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
"Deserialized signature must match the one in the attestation"
updateCurrent(pool, wallTime.slotOrZero)
let candidateIdx = pool.candidateIdx(attestation.data.slot)
if candidateIdx.isNone:
debug "Skipping old attestation for block production",
startingSlot = pool.startingSlot
return
let attestation_data_root = hash_tree_root(attestation.data)
# TODO withValue is an abomination but hard to use anything else too without
# creating an unnecessary AttestationEntry on the hot path and avoiding
# multiple lookups
pool.candidates[candidateIdx.get()].withValue(attestation_data_root, entry) do:
if not addAttestation(entry[], attestation, signature):
return
do:
if not addAttestation(
pool.candidates[candidateIdx.get()].mgetOrPut(
attestation_data_root,
AttestationEntry(
data: attestation.data,
committee_len: attestation.aggregation_bits.len())),
attestation, signature):
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
return
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
pool.addForkChoiceVotes(
attestation.data.slot, attesting_indices,
attestation.data.beacon_block_root, wallTime)
# Send notification about new attestation via callback.
if not(isNil(pool.onAttestationAdded)):
pool.onAttestationAdded(attestation)
func covers*(
pool: var AttestationPool, data: Attestationdata,
bits: CommitteeValidatorsBits): bool =
## Return true iff the given attestation already is fully covered by one of
## the existing aggregates, making it redundant
## the `var` attestation pool is needed to use `withValue`, else Table becomes
## unusably inefficient
let candidateIdx = pool.candidateIdx(data.slot)
if candidateIdx.isNone:
return false
let attestation_data_root = hash_tree_root(data)
pool.candidates[candidateIdx.get()].withValue(attestation_data_root, entry):
if entry[].covers(bits):
return true
false
proc addForkChoice*(pool: var AttestationPool,
epochRef: EpochRef,
blckRef: BlockRef,
blck: ForkyTrustedBeaconBlock,
wallTime: BeaconTime) =
## Add a verified block to the fork choice context
let state = pool.forkChoice.process_block(
pool.dag, epochRef, blckRef, blck, wallTime)
if state.isErr:
# This indicates that the fork choice and the chain dag are out of sync -
# this is most likely the result of a bug, but we'll try to keep going -
# hopefully the fork choice will heal itself over time.
error "Couldn't add block to fork choice, bug?",
blck = shortLog(blck), err = state.error
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
iterator attestations*(pool: AttestationPool, slot: Option[Slot],
committee_index: Option[CommitteeIndex]): Attestation =
let candidateIndices =
if slot.isSome():
let candidateIdx = pool.candidateIdx(slot.get())
if candidateIdx.isSome():
candidateIdx.get() .. candidateIdx.get()
else:
1 .. 0
else:
0 ..< pool.candidates.len()
for candidateIndex in candidateIndices:
for _, entry in pool.candidates[candidateIndex]:
if committee_index.isNone() or entry.data.index == committee_index.get():
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
var singleAttestation = Attestation(
aggregation_bits: CommitteeValidatorsBits.init(entry.committee_len),
data: entry.data)
for index, signature in entry.singles:
singleAttestation.aggregation_bits.setBit(index)
singleAttestation.signature = signature.toValidatorSig()
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
yield singleAttestation
singleAttestation.aggregation_bits.clearBit(index)
for v in entry.aggregates:
yield entry.toAttestation(v)
type
AttestationCacheKey* = (Slot, uint64)
AttestationCache = Table[AttestationCacheKey, CommitteeValidatorsBits] ##\
## Cache for quick lookup during beacon block construction of attestations
## which have already been included, and therefore should be skipped.
func getAttestationCacheKey(ad: AttestationData): AttestationCacheKey =
# The committee is unique per slot and committee index which means we can use
# it as key for a participation cache - this is checked in `check_attestation`
(ad.slot, ad.index)
func add(
attCache: var AttestationCache, data: AttestationData,
aggregation_bits: CommitteeValidatorsBits) =
let key = data.getAttestationCacheKey()
attCache.withValue(key, v) do:
v[].incl(aggregation_bits)
do:
attCache[key] = aggregation_bits
func init(T: type AttestationCache, state: phase0.HashedBeaconState): T =
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
# Load attestations that are scheduled for being given rewards for
for i in 0..<state.data.previous_epoch_attestations.len():
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
result.add(
state.data.previous_epoch_attestations[i].data,
state.data.previous_epoch_attestations[i].aggregation_bits)
for i in 0..<state.data.current_epoch_attestations.len():
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
result.add(
state.data.current_epoch_attestations[i].data,
state.data.current_epoch_attestations[i].aggregation_bits)
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
func init(
T: type AttestationCache,
state: altair.HashedBeaconState | bellatrix.HashedBeaconState,
cache: var StateCache): T =
# Load attestations that are scheduled for being given rewards for
let
prev_epoch = state.data.get_previous_epoch()
cur_epoch = state.data.get_current_epoch()
template update_attestation_pool_cache(
epoch: Epoch, participation_bitmap: untyped) =
let committees_per_slot = get_committee_count_per_slot(
state.data, epoch, cache)
for committee_index in get_committee_indices(committees_per_slot):
for slot in epoch.slots():
let committee = get_beacon_committee(
state.data, slot, committee_index, cache)
var
validator_bits = CommitteeValidatorsBits.init(committee.len)
for index_in_committee, validator_index in committee:
if participation_bitmap[validator_index] != 0:
# If any flag got set, there was an attestation from this validator.
validator_bits[index_in_committee] = true
result.add((slot, committee_index.uint64), validator_bits)
# This treats all types of rewards as equivalent, which isn't ideal
update_attestation_pool_cache(
prev_epoch, state.data.previous_epoch_participation)
update_attestation_pool_cache(
cur_epoch, state.data.current_epoch_participation)
func score(
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
attCache: var AttestationCache, data: AttestationData,
aggregation_bits: CommitteeValidatorsBits): int =
# The score of an attestation is loosely based on how many new votes it brings
# to the state - a more accurate score function would also look at inclusion
# distance and effective balance.
# TODO cache not var, but `withValue` requires it
let
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
key = data.getAttestationCacheKey()
bitsScore = aggregation_bits.countOnes()
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
attCache.withValue(key, value):
doAssert aggregation_bits.len() == value[].len(),
"check_attestation ensures committee length"
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
# How many votes were in the attestation minues the votes that are the same
return bitsScore - aggregation_bits.countOverlap(value[])
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
# Not found in cache - fresh vote meaning all attestations count
bitsScore
proc getAttestationsForBlock*(pool: var AttestationPool,
state: ForkyHashedBeaconState,
cache: var StateCache): seq[Attestation] =
## Retrieve attestations that may be added to a new block at the slot of the
## given state
## https://github.com/ethereum/consensus-specs/blob/v1.1.10/specs/phase0/validator.md#attestations
let newBlockSlot = state.data.slot.uint64
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
if newBlockSlot < MIN_ATTESTATION_INCLUSION_DELAY:
return # Too close to genesis
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
let
# Attestations produced in a particular slot are added to the block
# at the slot where at least MIN_ATTESTATION_INCLUSION_DELAY have passed
maxAttestationSlot = newBlockSlot - MIN_ATTESTATION_INCLUSION_DELAY
startPackingTick = Moment.now()
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
var
candidates: seq[tuple[
score: int, slot: Slot, entry: ptr AttestationEntry, validation: int]]
attCache =
when state is phase0.HashedBeaconState:
AttestationCache.init(state)
elif state is altair.HashedBeaconState or state is bellatrix.HashedBeaconState:
AttestationCache.init(state, cache)
else:
static: doAssert false
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
for i in 0..<ATTESTATION_LOOKBACK:
if i > maxAttestationSlot: # Around genesis..
break
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
let
slot = Slot(maxAttestationSlot - i)
candidateIdx = pool.candidateIdx(slot)
if candidateIdx.isNone():
# Passed the collection horizon - shouldn't happen because it's based on
# ATTESTATION_LOOKBACK
break
for _, entry in pool.candidates[candidateIdx.get()].mpairs():
entry.updateAggregates()
for j in 0..<entry.aggregates.len():
let
attestation = entry.toAttestation(entry.aggregates[j])
# Attestations are checked based on the state that we're adding the
# attestation to - there might have been a fork between when we first
# saw the attestation and the time that we added it
if not check_attestation(
state.data, attestation, {skipBlsValidation}, cache).isOk():
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
continue
let score = attCache.score(
entry.data, entry.aggregates[j].aggregation_bits)
if score == 0:
# 0 score means the attestation would not bring any votes - discard
# it early
# Note; this must be done _after_ `check_attestation` as it relies on
# the committee to match the state that was used to build the cache
continue
# Careful, must not update the attestation table for the pointer to
# remain valid
candidates.add((score, slot, addr entry, j))
# Using a greedy algorithm, select as many attestations as possible that will
# fit in the block.
#
# For each round, we'll look for the best attestation and add it to the result
# then re-score the other candidates.
#
# A possible improvement here would be to use a maximum cover algorithm.
var
prevEpoch = state.data.get_previous_epoch()
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
prevEpochSpace =
when state is altair.HashedBeaconState or state is bellatrix.HashedBeaconState:
MAX_ATTESTATIONS
elif state is phase0.HashedBeaconState:
state.data.previous_epoch_attestations.maxLen -
state.data.previous_epoch_attestations.len()
else:
raiseAssert "invalid HashedBeaconState fork"
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
var res: seq[Attestation]
let totalCandidates = candidates.len()
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
while candidates.len > 0 and res.lenu64() < MAX_ATTESTATIONS:
block:
# Find the candidate with the highest score - slot is used as a
# tie-breaker so that more recent attestations are added first
let
candidate =
# Fast path for when all remaining candidates fit
if candidates.lenu64 < MAX_ATTESTATIONS: candidates.len - 1
else: maxIndex(candidates)
# TODO slot not used; replace with _ when
# https://github.com/nim-lang/Nim/issues/15972 and
# https://github.com/nim-lang/Nim/issues/16217 are
# fixed in Status's Nim.
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
(_, slot, entry, j) = candidates[candidate]
candidates.del(candidate) # careful, `del` reorders candidates
if entry[].data.target.epoch == prevEpoch:
if prevEpochSpace < 1:
continue # No need to rescore since we didn't add the attestation
prevEpochSpace -= 1
res.add(entry[].toAttestation(entry[].aggregates[j]))
# Update cache so that the new votes are taken into account when updating
# the score below
attCache.add(entry[].data, entry[].aggregates[j].aggregation_bits)
block:
# Because we added some votes, it's quite possible that some candidates
# are no longer interesting - update the scores of the existing candidates
for it in candidates.mitems():
it.score = attCache.score(
it.entry[].data,
it.entry[].aggregates[it.validation].aggregation_bits)
candidates.keepItIf:
# Only keep candidates that might add coverage
it.score > 0
let
packingDur = Moment.now() - startPackingTick
debug "Packed attestations for block",
newBlockSlot, packingDur, totalCandidates, attestations = res.len()
attestation_pool_block_attestation_packing_time.set(
packingDur.toFloatSeconds())
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
res
proc getAttestationsForBlock*(pool: var AttestationPool,
state: ForkedHashedBeaconState,
cache: var StateCache): seq[Attestation] =
withState(state):
pool.getAttestationsForBlock(state, cache)
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
func bestValidation(aggregates: openArray[Validation]): (int, int) =
# Look for best validation based on number of votes in the aggregate
doAssert aggregates.len() > 0,
"updateAggregates should have created at least one aggregate"
var
bestIndex = 0
best = aggregates[bestIndex].aggregation_bits.countOnes()
for i in 1..<aggregates.len():
let count = aggregates[i].aggregation_bits.countOnes()
if count > best:
best = count
bestIndex = i
(bestIndex, best)
func getAggregatedAttestation*(pool: var AttestationPool,
slot: Slot,
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
attestation_data_root: Eth2Digest): Option[Attestation] =
let
candidateIdx = pool.candidateIdx(slot)
if candidateIdx.isNone:
return none(Attestation)
pool.candidates[candidateIdx.get].withValue(attestation_data_root, entry):
entry[].updateAggregates()
let (bestIndex, _) = bestValidation(entry[].aggregates)
# Found the right hash, no need to look further
return some(entry[].toAttestation(entry[].aggregates[bestIndex]))
none(Attestation)
func getAggregatedAttestation*(pool: var AttestationPool,
slot: Slot,
index: CommitteeIndex): Option[Attestation] =
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
## Select the attestation that has the most votes going for it in the given
## slot/index
## https://github.com/ethereum/consensus-specs/blob/v1.1.10/specs/phase0/validator.md#construct-aggregate
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
let
candidateIdx = pool.candidateIdx(slot)
if candidateIdx.isNone:
return none(Attestation)
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
var res: Option[Attestation]
for _, entry in pool.candidates[candidateIdx.get].mpairs():
doAssert entry.data.slot == slot
if index != entry.data.index:
continue
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
entry.updateAggregates()
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
let (bestIndex, best) = bestValidation(entry.aggregates)
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
if res.isNone() or best > res.get().aggregation_bits.countOnes():
res = some(entry.toAttestation(entry.aggregates[bestIndex]))
Revamp attestation pool This is a revamp of the attestation pool that cleans up several aspects of attestation processing as the network grows larger and block space becomes more precious. The aim is to better exploit the divide between attestation subnets and aggregations by keeping the two kinds separate until it's time to either produce a block or aggregate. This means we're no longer eagerly combining single-vote attestations, but rather wait until the last moment, and then try to add singles to all aggregates, including those coming from the network. Importantly, the branch improves on poor aggregate quality and poor attestation packing in cases where block space is running out. A basic greed scoring mechanism is used to select attestations for blocks - attestations are added based on how much many new votes they bring to the table. * Collect single-vote attestations separately and store these until it's time to make aggregates * Create aggregates based on single-vote attestations * Select _best_ aggregate rather than _first_ aggregate when on aggregation duty * Top up all aggregates with singles when it's time make the attestation cut, thus improving the chances of grabbing the best aggregates out there * Improve aggregation test coverage * Improve bitseq operations * Simplify aggregate signature creation * Make attestation cache temporary instead of storing it in attestation pool - most of the time, blocks are not being produced, no need to keep the data around * Remove redundant aggregate storage that was used only for RPC * Use tables to avoid some linear seeks when looking up attestation data * Fix long cleanup on large slot jumps * Avoid some pointers * Speed up iterating all attestations for a slot (fixes #2490)
2021-04-12 20:25:09 +00:00
res
limit by-root requests to non-finalized blocks (#3293) * limit by-root requests to non-finalized blocks Presently, we keep a mapping from block root to `BlockRef` in memory - this has simplified reasoning about the dag, but is not sustainable with the chain growing. We can distinguish between two cases where by-root access is useful: * unfinalized blocks - this is where the beacon chain is operating generally, by validating incoming data as interesting for future fork choice decisions - bounded by the length of the unfinalized period * finalized blocks - historical access in the REST API etc - no bounds, really In this PR, we limit the by-root block index to the first use case: finalized chain data can more efficiently be addressed by slot number. Future work includes: * limiting the `BlockRef` horizon in general - each instance is 40 bytes+overhead which adds up - this needs further refactoring to deal with the tail vs state problem * persisting the finalized slot-to-hash index - this one also keeps growing unbounded (albeit slowly) Anyway, this PR easily shaves ~128mb of memory usage at the time of writing. * No longer honor `BeaconBlocksByRoot` requests outside of the non-finalized period - previously, Nimbus would generously return any block through this libp2p request - per the spec, finalized blocks should be fetched via `BeaconBlocksByRange` instead. * return `Opt[BlockRef]` instead of `nil` when blocks can't be found - this becomes a lot more common now and thus deserves more attention * `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized blocks from now - `finalizedBlocks` covers the other `BlockRef` instances * in backfill, verify that the last backfilled block leads back to genesis, or panic * add backfill timings to log * fix missing check that `BlockRef` block can be fetched with `getForkedBlock` reliably * shortcut doppelganger check when feature is not enabled * in REST/JSON-RPC, fetch blocks without involving `BlockRef` * fix dag.blocks ref
2022-01-21 11:33:16 +00:00
proc selectHead*(pool: var AttestationPool, wallTime: BeaconTime): Opt[BlockRef] =
## Trigger fork choice and returns the new head block.
## Can return `nil`
let newHead = pool.forkChoice.get_head(pool.dag, wallTime)
if newHead.isErr:
error "Couldn't select head", err = newHead.error
limit by-root requests to non-finalized blocks (#3293) * limit by-root requests to non-finalized blocks Presently, we keep a mapping from block root to `BlockRef` in memory - this has simplified reasoning about the dag, but is not sustainable with the chain growing. We can distinguish between two cases where by-root access is useful: * unfinalized blocks - this is where the beacon chain is operating generally, by validating incoming data as interesting for future fork choice decisions - bounded by the length of the unfinalized period * finalized blocks - historical access in the REST API etc - no bounds, really In this PR, we limit the by-root block index to the first use case: finalized chain data can more efficiently be addressed by slot number. Future work includes: * limiting the `BlockRef` horizon in general - each instance is 40 bytes+overhead which adds up - this needs further refactoring to deal with the tail vs state problem * persisting the finalized slot-to-hash index - this one also keeps growing unbounded (albeit slowly) Anyway, this PR easily shaves ~128mb of memory usage at the time of writing. * No longer honor `BeaconBlocksByRoot` requests outside of the non-finalized period - previously, Nimbus would generously return any block through this libp2p request - per the spec, finalized blocks should be fetched via `BeaconBlocksByRange` instead. * return `Opt[BlockRef]` instead of `nil` when blocks can't be found - this becomes a lot more common now and thus deserves more attention * `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized blocks from now - `finalizedBlocks` covers the other `BlockRef` instances * in backfill, verify that the last backfilled block leads back to genesis, or panic * add backfill timings to log * fix missing check that `BlockRef` block can be fetched with `getForkedBlock` reliably * shortcut doppelganger check when feature is not enabled * in REST/JSON-RPC, fetch blocks without involving `BlockRef` * fix dag.blocks ref
2022-01-21 11:33:16 +00:00
err()
else:
limit by-root requests to non-finalized blocks (#3293) * limit by-root requests to non-finalized blocks Presently, we keep a mapping from block root to `BlockRef` in memory - this has simplified reasoning about the dag, but is not sustainable with the chain growing. We can distinguish between two cases where by-root access is useful: * unfinalized blocks - this is where the beacon chain is operating generally, by validating incoming data as interesting for future fork choice decisions - bounded by the length of the unfinalized period * finalized blocks - historical access in the REST API etc - no bounds, really In this PR, we limit the by-root block index to the first use case: finalized chain data can more efficiently be addressed by slot number. Future work includes: * limiting the `BlockRef` horizon in general - each instance is 40 bytes+overhead which adds up - this needs further refactoring to deal with the tail vs state problem * persisting the finalized slot-to-hash index - this one also keeps growing unbounded (albeit slowly) Anyway, this PR easily shaves ~128mb of memory usage at the time of writing. * No longer honor `BeaconBlocksByRoot` requests outside of the non-finalized period - previously, Nimbus would generously return any block through this libp2p request - per the spec, finalized blocks should be fetched via `BeaconBlocksByRange` instead. * return `Opt[BlockRef]` instead of `nil` when blocks can't be found - this becomes a lot more common now and thus deserves more attention * `dag.blocks` -> `dag.forkBlocks` - this index only carries unfinalized blocks from now - `finalizedBlocks` covers the other `BlockRef` instances * in backfill, verify that the last backfilled block leads back to genesis, or panic * add backfill timings to log * fix missing check that `BlockRef` block can be fetched with `getForkedBlock` reliably * shortcut doppelganger check when feature is not enabled * in REST/JSON-RPC, fetch blocks without involving `BlockRef` * fix dag.blocks ref
2022-01-21 11:33:16 +00:00
let ret = pool.dag.getBlockRef(newHead.get())
if ret.isErr():
# This should normally not happen, but if the chain dag and fork choice
# get out of sync, we'll need to try to download the selected head - in
# the meantime, return nil to indicate that no new head was chosen
warn "Fork choice selected unknown head, trying to sync", root = newHead.get()
pool.quarantine[].addMissing(newHead.get())
ret
proc prune*(pool: var AttestationPool) =
if (let v = pool.forkChoice.prune(); v.isErr):
# If pruning fails, it's likely the result of a bug - this shouldn't happen
# but we'll keep running hoping that the fork chocie will recover eventually
error "Couldn't prune fork choice, bug?", err = v.error()