nimbus-eth1/fluffy/tools/portal_bridge/portal_bridge_beacon.nim

471 lines
16 KiB
Nim

# Fluffy
# Copyright (c) 2023-2024 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/os,
chronos,
chronicles,
chronicles/topics_registry,
stew/byteutils,
eth/async_utils,
json_rpc/clients/httpclient,
beacon_chain/spec/eth2_apis/rest_beacon_client,
../../network/beacon/beacon_content,
../../rpc/portal_rpc_client,
../eth_data_exporter/cl_data_exporter,
./[portal_bridge_conf, portal_bridge_common]
const
largeRequestsTimeout = 120.seconds # For downloading large items such as states.
restRequestsTimeout = 30.seconds
# TODO: From nimbus_binary_common, but we don't want to import that.
proc sleepAsync(t: TimeDiff): Future[void] =
sleepAsync(nanoseconds(if t.nanoseconds < 0: 0'i64 else: t.nanoseconds))
proc gossipLCBootstrapUpdate(
restClient: RestClientRef,
portalRpcClient: RpcClient,
trustedBlockRoot: Eth2Digest,
cfg: RuntimeConfig,
forkDigests: ref ForkDigests,
): Future[Result[void, string]] {.async.} =
var bootstrap =
try:
info "Downloading LC bootstrap"
awaitWithTimeout(
restClient.getLightClientBootstrap(trustedBlockRoot, cfg, forkDigests),
restRequestsTimeout,
):
return err("Attempt to download LC bootstrap timed out")
except CatchableError as exc:
return err("Unable to download LC bootstrap: " & exc.msg)
withForkyObject(bootstrap):
when lcDataFork > LightClientDataFork.None:
let
slot = forkyObject.header.beacon.slot
contentKey = encode(bootstrapContentKey(trustedBlockRoot))
forkDigest = forkDigestAtEpoch(forkDigests[], epoch(slot), cfg)
content = encodeBootstrapForked(forkDigest, bootstrap)
proc GossipRpcAndClose(): Future[Result[void, string]] {.async.} =
try:
let
contentKeyHex = contentKey.asSeq().toHex()
peers = await portalRpcClient.portal_beaconRandomGossip(
contentKeyHex, content.toHex()
)
info "Beacon LC bootstrap gossiped", peers, contentKey = contentKeyHex
return ok()
except CatchableError as e:
return err("JSON-RPC error: " & $e.msg)
let res = await GossipRpcAndClose()
if res.isOk():
return ok()
else:
return err(res.error)
else:
return err("No LC bootstraps pre Altair")
proc gossipLCUpdates(
restClient: RestClientRef,
portalRpcClient: RpcClient,
startPeriod: uint64,
count: uint64,
cfg: RuntimeConfig,
forkDigests: ref ForkDigests,
): Future[Result[void, string]] {.async.} =
var updates =
try:
info "Downloading LC updates", count
awaitWithTimeout(
restClient.getLightClientUpdatesByRange(
SyncCommitteePeriod(startPeriod), count, cfg, forkDigests
),
restRequestsTimeout,
):
return err("Attempt to download LC updates timed out")
except CatchableError as exc:
return err("Unable to download LC updates: " & exc.msg)
if updates.len() > 0:
withForkyObject(updates[0]):
when lcDataFork > LightClientDataFork.None:
let
slot = forkyObject.attested_header.beacon.slot
period = slot.sync_committee_period
contentKey = encode(updateContentKey(period.uint64, count))
forkDigest = forkDigestAtEpoch(forkDigests[], epoch(slot), cfg)
content = encodeLightClientUpdatesForked(forkDigest, updates)
proc GossipRpcAndClose(): Future[Result[void, string]] {.async.} =
try:
let
contentKeyHex = contentKey.asSeq().toHex()
peers = await portalRpcClient.portal_beaconRandomGossip(
contentKeyHex, content.toHex()
)
info "Beacon LC update gossiped",
peers, contentKey = contentKeyHex, period, count
return ok()
except CatchableError as e:
return err("JSON-RPC error: " & $e.msg)
let res = await GossipRpcAndClose()
if res.isOk():
return ok()
else:
return err(res.error)
else:
return err("No LC updates pre Altair")
else:
# TODO:
# currently only error if no updates at all found. This might be due
# to selecting future period or too old period.
# Might want to error here in case count != updates.len or might not want to
# error at all and perhaps return the updates.len.
return err("No updates downloaded")
proc gossipLCFinalityUpdate(
restClient: RestClientRef,
portalRpcClient: RpcClient,
cfg: RuntimeConfig,
forkDigests: ref ForkDigests,
): Future[Result[(Slot, Eth2Digest), string]] {.async.} =
var update =
try:
info "Downloading LC finality update"
awaitWithTimeout(
restClient.getLightClientFinalityUpdate(cfg, forkDigests), restRequestsTimeout
):
return err("Attempt to download LC finality update timed out")
except CatchableError as exc:
return err("Unable to download LC finality update: " & exc.msg)
withForkyObject(update):
when lcDataFork > LightClientDataFork.None:
let
finalizedSlot = forkyObject.finalized_header.beacon.slot
blockRoot = hash_tree_root(forkyObject.finalized_header.beacon)
contentKey = encode(finalityUpdateContentKey(finalizedSlot.uint64))
forkDigest = forkDigestAtEpoch(
forkDigests[], epoch(forkyObject.attested_header.beacon.slot), cfg
)
content = encodeFinalityUpdateForked(forkDigest, update)
proc GossipRpcAndClose(): Future[Result[void, string]] {.async.} =
try:
let
contentKeyHex = contentKey.asSeq().toHex()
peers = await portalRpcClient.portal_beaconRandomGossip(
contentKeyHex, content.toHex()
)
info "Beacon LC finality update gossiped",
peers, contentKey = contentKeyHex, finalizedSlot
return ok()
except CatchableError as e:
return err("JSON-RPC error: " & $e.msg)
let res = await GossipRpcAndClose()
if res.isOk():
return ok((finalizedSlot, blockRoot))
else:
return err(res.error)
else:
return err("No LC updates pre Altair")
proc gossipLCOptimisticUpdate(
restClient: RestClientRef,
portalRpcClient: RpcClient,
cfg: RuntimeConfig,
forkDigests: ref ForkDigests,
): Future[Result[Slot, string]] {.async.} =
var update =
try:
info "Downloading LC optimistic update"
awaitWithTimeout(
restClient.getLightClientOptimisticUpdate(cfg, forkDigests), restRequestsTimeout
):
return err("Attempt to download LC optimistic update timed out")
except CatchableError as exc:
return err("Unable to download LC optimistic update: " & exc.msg)
withForkyObject(update):
when lcDataFork > LightClientDataFork.None:
let
slot = forkyObject.signature_slot
contentKey = encode(optimisticUpdateContentKey(slot.uint64))
forkDigest = forkDigestAtEpoch(
forkDigests[], epoch(forkyObject.attested_header.beacon.slot), cfg
)
content = encodeOptimisticUpdateForked(forkDigest, update)
proc GossipRpcAndClose(): Future[Result[void, string]] {.async.} =
try:
let
contentKeyHex = contentKey.asSeq().toHex()
peers = await portalRpcClient.portal_beaconRandomGossip(
contentKeyHex, content.toHex()
)
info "Beacon LC optimistic update gossiped",
peers, contentKey = contentKeyHex, slot
return ok()
except CatchableError as e:
return err("JSON-RPC error: " & $e.msg)
let res = await GossipRpcAndClose()
if res.isOk():
return ok(slot)
else:
return err(res.error)
else:
return err("No LC updates pre Altair")
proc gossipHistoricalSummaries(
restClient: RestClientRef,
portalRpcClient: RpcClient,
cfg: RuntimeConfig,
forkDigests: ref ForkDigests,
): Future[Result[void, string]] {.async.} =
let state =
try:
notice "Downloading beacon state"
awaitWithTimeout(
restClient.getStateV2(StateIdent.init(StateIdentType.Finalized), cfg),
largeRequestsTimeout,
):
return err("Attempt to download beacon state timed out")
except CatchableError as exc:
return err("Unable to download beacon state: " & exc.msg)
if state == nil:
return err("No beacon state found")
withState(state[]):
when consensusFork >= ConsensusFork.Capella:
let
historical_summaries = forkyState.data.historical_summaries
proof = ?buildProof(state[])
epoch = forkyState.data.slot.epoch()
forkDigest = forkDigestAtEpoch(forkDigests[], epoch, cfg)
summariesWithProof = HistoricalSummariesWithProof(
epoch: epoch, historical_summaries: historical_summaries, proof: proof
)
contentKey = encode(historicalSummariesContentKey(epoch.uint64))
content = encodeSsz(summariesWithProof, forkDigest)
try:
let peers = await portalRpcClient.portal_beaconRandomGossip(
contentKey.asSeq().toHex(), content.toHex()
)
info "Beacon historical_summaries gossiped", peers, epoch
return ok()
except CatchableError as e:
return err("JSON-RPC error: " & $e.msg)
else:
return err("No historical_summaries pre Capella")
proc runBeacon*(config: PortalBridgeConf) {.raises: [CatchableError].} =
notice "Launching Fluffy beacon chain bridge", cmdParams = commandLineParams()
let
(cfg, forkDigests, beaconClock) = getBeaconData()
getBeaconTime = beaconClock.getBeaconTimeFn()
portalRpcClient = newRpcClientConnect(config.portalRpcUrl)
restClient = RestClientRef.new(config.restUrl).valueOr:
fatal "Cannot connect to server", error = $error
quit QuitFailure
proc backfill(
beaconRestClient: RestClientRef,
portalRpcClient: RpcClient,
backfillAmount: uint64,
trustedBlockRoot: Option[TrustedDigest],
) {.async.} =
# TODO:
# It can get tricky when we need to bootstrap the beacon network with
# a portal_bridge:
# - Either a very recent bootstrap needs to be taken so that no updates are
# required for the nodes to sync.
# - Or the bridge needs to be tuned together with the selected bootstrap to
# provide the right amount of backfill updates.
# - Or the above point could get automatically implemented here based on the
# provided trusted-block-root
# Bootstrap backfill, currently just one bootstrap selected by
# trusted-block-root, could become a selected list, or some other way.
if trustedBlockRoot.isSome():
let res = await gossipLCBootstrapUpdate(
beaconRestClient, portalRpcClient, trustedBlockRoot.get(), cfg, forkDigests
)
if res.isErr():
warn "Error gossiping LC bootstrap", error = res.error
await portalRpcClient.close()
# Add some seconds delay to allow the bootstrap to be gossiped around.
# Without the bootstrap, following updates will not get accepted.
await sleepAsync(5.seconds)
# Updates backfill, selected by backfillAmount
# Might want to alter this to default backfill to the
# `MIN_EPOCHS_FOR_BLOCK_REQUESTS`.
# TODO: This can be up to 128, but our JSON-RPC requests fail with a value
# higher than 16. TBI
const updatesPerRequest = 16
let
wallSlot = getBeaconTime().slotOrZero()
currentPeriod = wallSlot div (SLOTS_PER_EPOCH * EPOCHS_PER_SYNC_COMMITTEE_PERIOD)
requestAmount = backfillAmount div updatesPerRequest
leftOver = backfillAmount mod updatesPerRequest
for i in 0 ..< requestAmount:
let res = await gossipLCUpdates(
beaconRestClient,
portalRpcClient,
currentPeriod - updatesPerRequest * (i + 1) + 1,
updatesPerRequest,
cfg,
forkDigests,
)
if res.isErr():
warn "Error gossiping LC updates", error = res.error
await portalRpcClient.close()
if leftOver > 0:
let res = await gossipLCUpdates(
beaconRestClient,
portalRpcClient,
currentPeriod - updatesPerRequest * requestAmount - leftOver + 1,
leftOver,
cfg,
forkDigests,
)
if res.isErr():
warn "Error gossiping LC updates", error = res.error
await portalRpcClient.close()
var
lastOptimisticUpdateSlot = Slot(0)
lastFinalityUpdateEpoch = epoch(lastOptimisticUpdateSlot)
lastUpdatePeriod = sync_committee_period(lastOptimisticUpdateSlot)
proc onSlotGossip(wallTime: BeaconTime, lastSlot: Slot) {.async.} =
let
wallSlot = wallTime.slotOrZero()
wallEpoch = epoch(wallSlot)
wallPeriod = sync_committee_period(wallSlot)
notice "Slot start info",
slot = wallSlot,
epoch = wallEpoch,
period = wallPeriod,
lastOptimisticUpdateSlot,
lastFinalityUpdateEpoch,
lastUpdatePeriod,
slotsTillNextEpoch = SLOTS_PER_EPOCH - (wallSlot mod SLOTS_PER_EPOCH),
slotsTillNextPeriod =
SLOTS_PER_SYNC_COMMITTEE_PERIOD - (wallSlot mod SLOTS_PER_SYNC_COMMITTEE_PERIOD)
if wallSlot > lastOptimisticUpdateSlot + 1:
# TODO: If this turns out to be too tricky to not gossip old updates,
# then an alternative could be to verify in the gossip calls if the actual
# slot number received is the correct one, before gossiping into Portal.
# And/or look into possibly using eth/v1/events for
# light_client_finality_update and light_client_optimistic_update if that
# is something that works.
# Or basically `lightClientOptimisticUpdateSlotOffset`
await sleepAsync((SECONDS_PER_SLOT div INTERVALS_PER_SLOT).int.seconds)
let res =
await gossipLCOptimisticUpdate(restClient, portalRpcClient, cfg, forkDigests)
if res.isErr():
warn "Error gossiping LC optimistic update", error = res.error
else:
if wallEpoch > lastFinalityUpdateEpoch + 2 and wallSlot > start_slot(wallEpoch):
let res =
await gossipLCFinalityUpdate(restClient, portalRpcClient, cfg, forkDigests)
if res.isErr():
warn "Error gossiping LC finality update", error = res.error
else:
let (slot, blockRoot) = res.value()
lastFinalityUpdateEpoch = epoch(slot)
let res = await gossipLCBootstrapUpdate(
restClient, portalRpcClient, blockRoot, cfg, forkDigests
)
if res.isErr():
warn "Error gossiping LC bootstrap", error = res.error
let res2 = await gossipHistoricalSummaries(
restClient, portalRpcClient, cfg, forkDigests
)
if res2.isErr():
warn "Error gossiping historical summaries", error = res.error
if wallPeriod > lastUpdatePeriod and wallSlot > start_slot(wallEpoch):
# TODO: Need to delay timing here also with one slot?
let res = await gossipLCUpdates(
restClient,
portalRpcClient,
sync_committee_period(wallSlot).uint64,
1,
cfg,
forkDigests,
)
if res.isErr():
warn "Error gossiping LC update", error = res.error
else:
lastUpdatePeriod = wallPeriod
lastOptimisticUpdateSlot = res.get()
proc runOnSlotLoop() {.async.} =
var
curSlot = getBeaconTime().slotOrZero()
nextSlot = curSlot + 1
timeToNextSlot = nextSlot.start_beacon_time() - getBeaconTime()
while true:
await sleepAsync(timeToNextSlot)
let
wallTime = getBeaconTime()
wallSlot = wallTime.slotOrZero()
await onSlotGossip(wallTime, curSlot)
curSlot = wallSlot
nextSlot = wallSlot + 1
timeToNextSlot = nextSlot.start_beacon_time() - getBeaconTime()
waitFor backfill(
restClient, portalRpcClient, config.backfillAmount, config.trustedBlockRoot
)
asyncSpawn runOnSlotLoop()
while true:
poll()