nimbus-eth2/beacon_chain/sync/request_manager.nim

414 lines
15 KiB
Nim

# beacon_chain
# Copyright (c) 2018-2024 Status Research & Development GmbH
# Licensed and distributed under either of
# * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT).
# * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0).
# at your option. This file may not be copied, modified, or distributed except according to those terms.
{.push raises: [].}
import std/[sequtils, strutils]
import chronos, chronicles
import
../spec/datatypes/[phase0, deneb],
../spec/[forks, network],
../networking/eth2_network,
../consensus_object_pools/block_quarantine,
../consensus_object_pools/blob_quarantine,
"."/sync_protocol, "."/sync_manager,
../gossip_processing/block_processor
from ../beacon_clock import GetBeaconTimeFn
export block_quarantine, sync_manager
logScope:
topics = "requman"
const
SYNC_MAX_REQUESTED_BLOCKS* = 32 # Spec allows up to MAX_REQUEST_BLOCKS.
## Maximum number of blocks which will be requested in each
## `beaconBlocksByRoot` invocation.
PARALLEL_REQUESTS* = 2
## Number of peers we using to resolve our request.
BLOB_GOSSIP_WAIT_TIME_NS* = 2 * 1_000_000_000
## How long to wait for blobs to arrive over gossip before fetching.
POLL_INTERVAL = 1.seconds
type
BlockVerifierFn* = proc(
signedBlock: ForkedSignedBeaconBlock,
maybeFinalized: bool
): Future[Result[void, VerifierError]] {.async: (raises: [CancelledError]).}
BlockLoaderFn* = proc(
blockRoot: Eth2Digest
): Opt[ForkedTrustedSignedBeaconBlock] {.gcsafe, raises: [].}
BlobLoaderFn* = proc(
blobId: BlobIdentifier): Opt[ref BlobSidecar] {.gcsafe, raises: [].}
InhibitFn* = proc: bool {.gcsafe, raises: [].}
RequestManager* = object
network*: Eth2Node
getBeaconTime: GetBeaconTimeFn
inhibit: InhibitFn
quarantine: ref Quarantine
blobQuarantine: ref BlobQuarantine
blockVerifier: BlockVerifierFn
blockLoader: BlockLoaderFn
blobLoader: BlobLoaderFn
blockLoopFuture: Future[void].Raising([CancelledError])
blobLoopFuture: Future[void].Raising([CancelledError])
func shortLog*(x: seq[Eth2Digest]): string =
"[" & x.mapIt(shortLog(it)).join(", ") & "]"
func shortLog*(x: seq[FetchRecord]): string =
"[" & x.mapIt(shortLog(it.root)).join(", ") & "]"
proc init*(T: type RequestManager, network: Eth2Node,
denebEpoch: Epoch,
getBeaconTime: GetBeaconTimeFn,
inhibit: InhibitFn,
quarantine: ref Quarantine,
blobQuarantine: ref BlobQuarantine,
blockVerifier: BlockVerifierFn,
blockLoader: BlockLoaderFn = nil,
blobLoader: BlobLoaderFn = nil): RequestManager =
RequestManager(
network: network,
getBeaconTime: getBeaconTime,
inhibit: inhibit,
quarantine: quarantine,
blobQuarantine: blobQuarantine,
blockVerifier: blockVerifier,
blockLoader: blockLoader,
blobLoader: blobLoader)
proc checkResponse(roots: openArray[Eth2Digest],
blocks: openArray[ref ForkedSignedBeaconBlock]): bool =
## This procedure checks peer's response.
var checks = @roots
if len(blocks) > len(roots):
return false
for blk in blocks:
let res = checks.find(blk[].root)
if res == -1:
return false
else:
checks.del(res)
true
proc checkResponse(idList: seq[BlobIdentifier],
blobs: openArray[ref BlobSidecar]): bool =
if len(blobs) > len(idList):
return false
for blob in blobs:
let block_root = hash_tree_root(blob.signed_block_header.message)
var found = false
for id in idList:
if id.block_root == block_root and id.index == blob.index:
found = true
break
if not found:
return false
blob[].verify_blob_sidecar_inclusion_proof().isOkOr:
return false
true
proc requestBlocksByRoot(rman: RequestManager, items: seq[Eth2Digest]) {.async: (raises: [CancelledError]).} =
var peer: Peer
try:
peer = await rman.network.peerPool.acquire()
debug "Requesting blocks by root", peer = peer, blocks = shortLog(items),
peer_score = peer.getScore()
let blocks = (await beaconBlocksByRoot_v2(peer, BlockRootsList items))
if blocks.isOk:
let ublocks = blocks.get()
if checkResponse(items, ublocks.asSeq()):
var
gotGoodBlock = false
gotUnviableBlock = false
for b in ublocks:
let ver = await rman.blockVerifier(b[], false)
if ver.isErr():
case ver.error()
of VerifierError.MissingParent:
# Ignoring because the order of the blocks that
# we requested may be different from the order in which we need
# these blocks to apply.
discard
of VerifierError.Duplicate:
# Ignoring because these errors could occur due to the
# concurrent/parallel requests we made.
discard
of VerifierError.UnviableFork:
# If they're working a different fork, we'll want to descore them
# but also process the other blocks (in case we can register the
# other blocks as unviable)
gotUnviableBlock = true
of VerifierError.Invalid:
# We stop processing blocks because peer is either sending us
# junk or working a different fork
notice "Received invalid block",
peer = peer, blocks = shortLog(items),
peer_score = peer.getScore()
peer.updateScore(PeerScoreBadValues)
return # Stop processing this junk...
else:
gotGoodBlock = true
if gotUnviableBlock:
notice "Received blocks from an unviable fork",
peer = peer, blocks = shortLog(items),
peer_score = peer.getScore()
peer.updateScore(PeerScoreUnviableFork)
elif gotGoodBlock:
debug "Request manager got good block",
peer = peer, blocks = shortLog(items), ublocks = len(ublocks)
# We reward peer only if it returns something.
peer.updateScore(PeerScoreGoodValues)
else:
debug "Mismatching response to blocks by root",
peer = peer, blocks = shortLog(items), ublocks = len(ublocks)
peer.updateScore(PeerScoreBadResponse)
else:
debug "Blocks by root request failed",
peer = peer, blocks = shortLog(items), err = blocks.error()
peer.updateScore(PeerScoreNoValues)
finally:
if not(isNil(peer)):
rman.network.peerPool.release(peer)
proc fetchBlobsFromNetwork(self: RequestManager,
idList: seq[BlobIdentifier])
{.async: (raises: [CancelledError]).} =
var peer: Peer
try:
peer = await self.network.peerPool.acquire()
debug "Requesting blobs by root", peer = peer, blobs = shortLog(idList),
peer_score = peer.getScore()
let blobs = await blobSidecarsByRoot(peer, BlobIdentifierList idList)
if blobs.isOk:
let ublobs = blobs.get()
if not checkResponse(idList, ublobs.asSeq()):
debug "Mismatched response to blobs by root",
peer = peer, blobs = shortLog(idList), ublobs = len(ublobs)
peer.updateScore(PeerScoreBadResponse)
return
for b in ublobs:
self.blobQuarantine[].put(b)
var curRoot: Eth2Digest
for b in ublobs:
let block_root = hash_tree_root(b.signed_block_header.message)
if block_root != curRoot:
curRoot = block_root
if (let o = self.quarantine[].popBlobless(curRoot); o.isSome):
let b = o.unsafeGet()
discard await self.blockVerifier(b, false)
# TODO:
# If appropriate, return a VerifierError.InvalidBlob from
# verification, check for it here, and penalize the peer accordingly
else:
debug "Blobs by root request failed",
peer = peer, blobs = shortLog(idList), err = blobs.error()
peer.updateScore(PeerScoreNoValues)
finally:
if not(isNil(peer)):
self.network.peerPool.release(peer)
proc requestManagerBlockLoop(
rman: RequestManager) {.async: (raises: [CancelledError]).} =
while true:
# TODO This polling could be replaced with an AsyncEvent that is fired
# from the quarantine when there's work to do
await sleepAsync(POLL_INTERVAL)
if rman.inhibit():
continue
let missingBlockRoots =
rman.quarantine[].checkMissing(SYNC_MAX_REQUESTED_BLOCKS).mapIt(it.root)
if missingBlockRoots.len == 0:
continue
# TODO This logic can be removed if the database schema is extended
# to store non-canonical heads on top of the canonical head!
# If that is done, the database no longer contains extra blocks
# that have not yet been assigned a `BlockRef`
var blockRoots: seq[Eth2Digest]
if rman.blockLoader == nil:
blockRoots = missingBlockRoots
else:
var verifiers:
seq[Future[Result[void, VerifierError]].Raising([CancelledError])]
for blockRoot in missingBlockRoots:
let blck = rman.blockLoader(blockRoot).valueOr:
blockRoots.add blockRoot
continue
debug "Loaded orphaned block from storage", blockRoot
verifiers.add rman.blockVerifier(
blck.asSigned(), maybeFinalized = false)
try:
await allFutures(verifiers)
except CancelledError as exc:
var futs = newSeqOfCap[Future[void].Raising([])](verifiers.len)
for verifier in verifiers:
futs.add verifier.cancelAndWait()
await noCancel allFutures(futs)
raise exc
if blockRoots.len == 0:
continue
debug "Requesting detected missing blocks", blocks = shortLog(blockRoots)
let start = SyncMoment.now(0)
var workers:
array[PARALLEL_REQUESTS, Future[void].Raising([CancelledError])]
for i in 0 ..< PARALLEL_REQUESTS:
workers[i] = rman.requestBlocksByRoot(blockRoots)
await allFutures(workers)
let finish = SyncMoment.now(uint64(len(blockRoots)))
debug "Request manager block tick", blocks = shortLog(blockRoots),
sync_speed = speed(start, finish)
proc getMissingBlobs(rman: RequestManager): seq[BlobIdentifier] =
let
wallTime = rman.getBeaconTime()
wallSlot = wallTime.slotOrZero()
delay = wallTime - wallSlot.start_beacon_time()
waitDur = TimeDiff(nanoseconds: BLOB_GOSSIP_WAIT_TIME_NS)
var
fetches: seq[BlobIdentifier]
ready: seq[Eth2Digest]
for blobless in rman.quarantine[].peekBlobless():
withBlck(blobless):
when consensusFork >= ConsensusFork.Deneb:
# give blobs a chance to arrive over gossip
if forkyBlck.message.slot == wallSlot and delay < waitDur:
debug "Not handling missing blobs early in slot"
continue
if not rman.blobQuarantine[].hasBlobs(forkyBlck):
let missing = rman.blobQuarantine[].blobFetchRecord(forkyBlck)
if len(missing.indices) == 0:
warn "quarantine missing blobs, but missing indices is empty",
blk=blobless.root,
commitments=len(forkyBlck.message.body.blob_kzg_commitments)
for idx in missing.indices:
let id = BlobIdentifier(block_root: blobless.root, index: idx)
if id notin fetches:
fetches.add(id)
else:
# this is a programming error should it occur.
warn "missing blob handler found blobless block with all blobs",
blk=blobless.root,
commitments=len(forkyBlck.message.body.blob_kzg_commitments)
ready.add(blobless.root)
for root in ready:
let blobless = rman.quarantine[].popBlobless(root).valueOr:
continue
discard rman.blockVerifier(blobless, false)
fetches
proc requestManagerBlobLoop(
rman: RequestManager) {.async: (raises: [CancelledError]).} =
while true:
# TODO This polling could be replaced with an AsyncEvent that is fired
# from the quarantine when there's work to do
await sleepAsync(POLL_INTERVAL)
if rman.inhibit():
continue
let missingBlobIds = rman.getMissingBlobs()
if missingBlobIds.len == 0:
continue
# TODO This logic can be removed if the database schema is extended
# to store non-canonical heads on top of the canonical head!
# If that is done, the database no longer contains extra blocks
# that have not yet been assigned a `BlockRef`
var blobIds: seq[BlobIdentifier]
if rman.blobLoader == nil:
blobIds = missingBlobIds
else:
var
blockRoots: seq[Eth2Digest]
curRoot: Eth2Digest
for blobId in missingBlobIds:
if blobId.block_root != curRoot:
curRoot = blobId.block_root
blockRoots.add curRoot
let blob_sidecar = rman.blobLoader(blobId).valueOr:
blobIds.add blobId
if blockRoots.len > 0 and blockRoots[^1] == curRoot:
# A blob is missing, remove from list of fully available blocks
discard blockRoots.pop()
continue
debug "Loaded orphaned blob from storage", blobId
rman.blobQuarantine[].put(blob_sidecar)
var verifiers = newSeqOfCap[
Future[Result[void, VerifierError]]
.Raising([CancelledError])](blockRoots.len)
for blockRoot in blockRoots:
let blck = rman.quarantine[].popBlobless(blockRoot).valueOr:
continue
verifiers.add rman.blockVerifier(blck, maybeFinalized = false)
try:
await allFutures(verifiers)
except CancelledError as exc:
var futs = newSeqOfCap[Future[void].Raising([])](verifiers.len)
for verifier in verifiers:
futs.add verifier.cancelAndWait()
await noCancel allFutures(futs)
raise exc
if blobIds.len > 0:
debug "Requesting detected missing blobs", blobs = shortLog(blobIds)
let start = SyncMoment.now(0)
var workers:
array[PARALLEL_REQUESTS, Future[void].Raising([CancelledError])]
for i in 0 ..< PARALLEL_REQUESTS:
workers[i] = rman.fetchBlobsFromNetwork(blobIds)
await allFutures(workers)
let finish = SyncMoment.now(uint64(len(blobIds)))
debug "Request manager blob tick",
blobs_count = len(blobIds),
sync_speed = speed(start, finish)
proc start*(rman: var RequestManager) =
## Start Request Manager's loops.
rman.blockLoopFuture = rman.requestManagerBlockLoop()
rman.blobLoopFuture = rman.requestManagerBlobLoop()
proc stop*(rman: RequestManager) =
## Stop Request Manager's loop.
if not(isNil(rman.blockLoopFuture)):
rman.blockLoopFuture.cancelSoon()
if not(isNil(rman.blobLoopFuture)):
rman.blobLoopFuture.cancelSoon()