nimbus-eth2/beacon_chain/sync/sync_manager.nim

783 lines
28 KiB
Nim
Raw Normal View History

# 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/[strutils, sequtils, algorithm]
import stew/base10, chronos, chronicles
import
../spec/datatypes/[phase0, altair],
2022-06-02 09:39:08 +00:00
../spec/eth2_apis/rest_types,
../spec/[helpers, forks, network],
../networking/[peer_pool, peer_scores, eth2_network],
../gossip_processing/block_processor,
../beacon_clock,
"."/[sync_protocol, sync_queue]
export phase0, altair, merge, chronos, chronicles, results,
helpers, peer_scores, sync_queue, forks, sync_protocol
2020-01-21 18:30:21 +00:00
logScope:
topics = "syncman"
2020-01-21 18:30:21 +00:00
const
SyncWorkersCount* = 10
## Number of sync workers to spawn
StatusUpdateInterval* = chronos.minutes(1)
## Minimum time between two subsequent calls to update peer's status
StatusExpirationTime* = chronos.minutes(2)
## Time time it takes for the peer's status information to expire.
2020-01-21 18:30:21 +00:00
type
SyncWorkerStatus* {.pure.} = enum
Sleeping, WaitingPeer, UpdatingStatus, Requesting, Downloading, Queueing,
Processing
SyncManagerFlag* {.pure.} = enum
NoMonitor
SyncWorker*[A, B] = object
future: Future[void].Raising([CancelledError])
status: SyncWorkerStatus
2020-04-20 14:59:18 +00:00
SyncManager*[A, B] = ref object
pool: PeerPool[A, B]
DENEB_FORK_EPOCH: Epoch
MIN_EPOCHS_FOR_BLOB_SIDECARS_REQUESTS: uint64
2020-04-20 14:59:18 +00:00
responseTimeout: chronos.Duration
maxHeadAge: uint64
getLocalHeadSlot: GetSlotCallback
getLocalWallSlot: GetSlotCallback
getSafeSlot: GetSlotCallback
getFirstSlot: GetSlotCallback
getLastSlot: GetSlotCallback
progressPivot: Slot
workers: array[SyncWorkersCount, SyncWorker[A, B]]
notInSyncEvent: AsyncEvent
rangeAge: uint64
chunkSize: uint64
queue: SyncQueue[A]
syncFut: Future[void]
blockVerifier: BlockVerifier
inProgress*: bool
insSyncSpeed*: float
avgSyncSpeed*: float
syncStatus*: string
direction: SyncQueueKind
ident*: string
flags: set[SyncManagerFlag]
SyncMoment* = object
stamp*: chronos.Moment
slots*: uint64
BeaconBlocksRes =
NetRes[List[ref ForkedSignedBeaconBlock, Limit MAX_REQUEST_BLOCKS]]
BlobSidecarsRes = NetRes[List[ref BlobSidecar, Limit(MAX_REQUEST_BLOB_SIDECARS)]]
2020-04-20 14:59:18 +00:00
proc now*(sm: typedesc[SyncMoment], slots: uint64): SyncMoment {.inline.} =
SyncMoment(stamp: now(chronos.Moment), slots: slots)
proc speed*(start, finish: SyncMoment): float {.inline.} =
## Returns number of slots per second.
if finish.slots <= start.slots or finish.stamp <= start.stamp:
0.0 # replays for example
else:
let
slots = float(finish.slots - start.slots)
dur = toFloatSeconds(finish.stamp - start.stamp)
slots / dur
proc initQueue[A, B](man: SyncManager[A, B]) =
case man.direction
of SyncQueueKind.Forward:
man.queue = SyncQueue.init(A, man.direction, man.getFirstSlot(),
man.getLastSlot(), man.chunkSize,
man.getSafeSlot, man.blockVerifier,
1, man.ident)
of SyncQueueKind.Backward:
let
firstSlot = man.getFirstSlot()
lastSlot = man.getLastSlot()
startSlot = if firstSlot == lastSlot:
# This case should never be happened in real life because
# there is present check `needsBackfill().
firstSlot
else:
firstSlot - 1'u64
man.queue = SyncQueue.init(A, man.direction, startSlot, lastSlot,
man.chunkSize, man.getSafeSlot,
man.blockVerifier, 1, man.ident)
2020-01-21 18:30:21 +00:00
proc newSyncManager*[A, B](pool: PeerPool[A, B],
denebEpoch: Epoch,
minEpochsForBlobSidecarsRequests: uint64,
direction: SyncQueueKind,
2020-04-20 14:59:18 +00:00
getLocalHeadSlotCb: GetSlotCallback,
getLocalWallSlotCb: GetSlotCallback,
2020-07-06 12:53:48 +00:00
getFinalizedSlotCb: GetSlotCallback,
getBackfillSlotCb: GetSlotCallback,
era: load blocks and states (#3394) * era: load blocks and states Era files contain finalized history and can be thought of as an alternative source for block and state data that allows clients to avoid syncing this information from the P2P network - the P2P network is then used to "top up" the client with the most recent data. They can be freely shared in the community via whatever means (http, torrent, etc) and serve as a permanent cold store of consensus data (and, after the merge, execution data) for history buffs and bean counters alike. This PR gently introduces support for loading blocks and states in two cases: block requests from rest/p2p and frontfilling when doing checkpoint sync. The era files are used as a secondary source if the information is not found in the database - compared to the database, there are a few key differences: * the database stores the block indexed by block root while the era file indexes by slot - the former is used only in rest, while the latter is used both by p2p and rest. * when loading blocks from era files, the root is no longer trivially available - if it is needed, it must either be computed (slow) or cached (messy) - the good news is that for p2p requests, it is not needed * in era files, "framed" snappy encoding is used while in the database we store unframed snappy - for p2p2 requests, the latter requires recompression while the former could avoid it * front-filling is the process of using era files to replace backfilling - in theory this front-filling could happen from any block and front-fills with gaps could also be entertained, but our backfilling algorithm cannot take advantage of this because there's no (simple) way to tell it to "skip" a range. * front-filling, as implemented, is a bit slow (10s to load mainnet): we load the full BeaconState for every era to grab the roots of the blocks - it would be better to partially load the state - as such, it would also be good to be able to partially decompress snappy blobs * lookups from REST via root are served by first looking up a block summary in the database, then using the slot to load the block data from the era file - however, there needs to be an option to create the summary table from era files to fully support historical queries To test this, `ncli_db` has an era file exporter: the files it creates should be placed in an `era` folder next to `db` in the data directory. What's interesting in particular about this setup is that `db` remains as the source of truth for security purposes - it stores the latest synced head root which in turn determines where a node "starts" its consensus participation - the era directory however can be freely shared between nodes / people without any (significant) security implications, assuming the era files are consistent / not broken. There's lots of future improvements to be had: * we can drop the in-memory `BlockRef` index almost entirely - at this point, resident memory usage of Nimbus should drop to a cool 500-600 mb * we could serve era files via REST trivially: this would drop backfill times to whatever time it takes to download the files - unlike the current implementation that downloads block by block, downloading an era at a time almost entirely cuts out request overhead * we can "reasonably" recreate detailed state history from almost any point in time, turning an O(slot) process into O(1) effectively - we'll still need caches and indices to do this with sufficient efficiency for the rest api, but at least it cuts the whole process down to minutes instead of hours, for arbitrary points in time * CI: ignore failures with Nim-1.6 (temporary) * test fixes Co-authored-by: Ștefan Talpalaru <stefantalpalaru@yahoo.com>
2022-03-23 08:58:17 +00:00
getFrontfillSlotCb: GetSlotCallback,
progressPivot: Slot,
blockVerifier: BlockVerifier,
maxHeadAge = uint64(SLOTS_PER_EPOCH * 1),
2020-04-20 14:59:18 +00:00
chunkSize = uint64(SLOTS_PER_EPOCH),
flags: set[SyncManagerFlag] = {},
ident = "main"
2020-04-20 14:59:18 +00:00
): SyncManager[A, B] =
let (getFirstSlot, getLastSlot, getSafeSlot) = case direction
of SyncQueueKind.Forward:
(getLocalHeadSlotCb, getLocalWallSlotCb, getFinalizedSlotCb)
of SyncQueueKind.Backward:
era: load blocks and states (#3394) * era: load blocks and states Era files contain finalized history and can be thought of as an alternative source for block and state data that allows clients to avoid syncing this information from the P2P network - the P2P network is then used to "top up" the client with the most recent data. They can be freely shared in the community via whatever means (http, torrent, etc) and serve as a permanent cold store of consensus data (and, after the merge, execution data) for history buffs and bean counters alike. This PR gently introduces support for loading blocks and states in two cases: block requests from rest/p2p and frontfilling when doing checkpoint sync. The era files are used as a secondary source if the information is not found in the database - compared to the database, there are a few key differences: * the database stores the block indexed by block root while the era file indexes by slot - the former is used only in rest, while the latter is used both by p2p and rest. * when loading blocks from era files, the root is no longer trivially available - if it is needed, it must either be computed (slow) or cached (messy) - the good news is that for p2p requests, it is not needed * in era files, "framed" snappy encoding is used while in the database we store unframed snappy - for p2p2 requests, the latter requires recompression while the former could avoid it * front-filling is the process of using era files to replace backfilling - in theory this front-filling could happen from any block and front-fills with gaps could also be entertained, but our backfilling algorithm cannot take advantage of this because there's no (simple) way to tell it to "skip" a range. * front-filling, as implemented, is a bit slow (10s to load mainnet): we load the full BeaconState for every era to grab the roots of the blocks - it would be better to partially load the state - as such, it would also be good to be able to partially decompress snappy blobs * lookups from REST via root are served by first looking up a block summary in the database, then using the slot to load the block data from the era file - however, there needs to be an option to create the summary table from era files to fully support historical queries To test this, `ncli_db` has an era file exporter: the files it creates should be placed in an `era` folder next to `db` in the data directory. What's interesting in particular about this setup is that `db` remains as the source of truth for security purposes - it stores the latest synced head root which in turn determines where a node "starts" its consensus participation - the era directory however can be freely shared between nodes / people without any (significant) security implications, assuming the era files are consistent / not broken. There's lots of future improvements to be had: * we can drop the in-memory `BlockRef` index almost entirely - at this point, resident memory usage of Nimbus should drop to a cool 500-600 mb * we could serve era files via REST trivially: this would drop backfill times to whatever time it takes to download the files - unlike the current implementation that downloads block by block, downloading an era at a time almost entirely cuts out request overhead * we can "reasonably" recreate detailed state history from almost any point in time, turning an O(slot) process into O(1) effectively - we'll still need caches and indices to do this with sufficient efficiency for the rest api, but at least it cuts the whole process down to minutes instead of hours, for arbitrary points in time * CI: ignore failures with Nim-1.6 (temporary) * test fixes Co-authored-by: Ștefan Talpalaru <stefantalpalaru@yahoo.com>
2022-03-23 08:58:17 +00:00
(getBackfillSlotCb, getFrontfillSlotCb, getBackfillSlotCb)
var res = SyncManager[A, B](
2020-04-20 14:59:18 +00:00
pool: pool,
DENEB_FORK_EPOCH: denebEpoch,
MIN_EPOCHS_FOR_BLOB_SIDECARS_REQUESTS: minEpochsForBlobSidecarsRequests,
2020-04-20 14:59:18 +00:00
getLocalHeadSlot: getLocalHeadSlotCb,
getLocalWallSlot: getLocalWallSlotCb,
getSafeSlot: getSafeSlot,
getFirstSlot: getFirstSlot,
getLastSlot: getLastSlot,
progressPivot: progressPivot,
2020-04-20 14:59:18 +00:00
maxHeadAge: maxHeadAge,
chunkSize: chunkSize,
blockVerifier: blockVerifier,
notInSyncEvent: newAsyncEvent(),
direction: direction,
ident: ident,
flags: flags
2020-04-20 14:59:18 +00:00
)
res.initQueue()
res
2020-04-20 14:59:18 +00:00
proc getBlocks[A, B](man: SyncManager[A, B], peer: A,
req: SyncRequest): Future[BeaconBlocksRes] {.
async: (raises: [CancelledError], raw: true).} =
mixin getScore, `==`
logScope:
peer_score = peer.getScore()
peer_speed = peer.netKbps()
sync_ident = man.ident
direction = man.direction
topics = "syncman"
2020-04-20 14:59:18 +00:00
doAssert(not(req.isEmpty()), "Request must not be empty!")
debug "Requesting blocks from peer", request = req
beaconBlocksByRange_v2(peer, req.slot, req.count, 1'u64)
proc shouldGetBlobs[A, B](man: SyncManager[A, B], e: Epoch): bool =
let wallEpoch = man.getLocalWallSlot().epoch
e >= man.DENEB_FORK_EPOCH and
(wallEpoch < man.MIN_EPOCHS_FOR_BLOB_SIDECARS_REQUESTS or
e >= wallEpoch - man.MIN_EPOCHS_FOR_BLOB_SIDECARS_REQUESTS)
proc getBlobSidecars[A, B](man: SyncManager[A, B], peer: A,
req: SyncRequest): Future[BlobSidecarsRes]
{.async: (raises: [CancelledError], raw: true).} =
mixin getScore, `==`
logScope:
peer_score = peer.getScore()
peer_speed = peer.netKbps()
sync_ident = man.ident
direction = man.direction
topics = "syncman"
doAssert(not(req.isEmpty()), "Request must not be empty!")
debug "Requesting blobs sidecars from peer", request = req
blobSidecarsByRange(peer, req.slot, req.count)
proc remainingSlots(man: SyncManager): uint64 =
let
first = man.getFirstSlot()
last = man.getLastSlot()
if man.direction == SyncQueueKind.Forward:
if last > first:
man.getLastSlot() - man.getFirstSlot()
else:
0'u64
else:
if first > last:
man.getFirstSlot() - man.getLastSlot()
else:
0'u64
func groupBlobs*[T](req: SyncRequest[T],
blocks: seq[ref ForkedSignedBeaconBlock],
blobs: seq[ref BlobSidecar]):
Result[seq[BlobSidecars], string] =
var
grouped = newSeq[BlobSidecars](len(blocks))
blob_cursor = 0
for block_idx, blck in blocks:
withBlck(blck[]):
when consensusFork >= ConsensusFork.Deneb:
template kzgs: untyped = forkyBlck.message.body.blob_kzg_commitments
if kzgs.len == 0:
continue
# Clients MUST include all blob sidecars of each block from which they include blob sidecars.
# The following blob sidecars, where they exist, MUST be sent in consecutive (slot, index) order.
# https://github.com/ethereum/consensus-specs/blob/v1.4.0-beta.5/specs/deneb/p2p-interface.md#blobsidecarsbyrange-v1
let header = forkyBlck.toSignedBeaconBlockHeader()
for blob_idx, kzg_commitment in kzgs:
if blob_cursor >= blobs.len:
return err("BlobSidecar: response too short")
let blob_sidecar = blobs[blob_cursor]
if blob_sidecar.index != BlobIndex blob_idx:
return err("BlobSidecar: unexpected index")
if blob_sidecar.kzg_commitment != kzg_commitment:
return err("BlobSidecar: unexpected kzg_commitment")
if blob_sidecar.signed_block_header != header:
return err("BlobSidecar: unexpected signed_block_header")
grouped[block_idx].add(blob_sidecar)
inc blob_cursor
if blob_cursor != len(blobs):
# we reached end of blocks without consuming all blobs so either
# the peer we got too few blocks in the paired request, or the
# peer is sending us spurious blobs.
Result[seq[BlobSidecars], string].err "invalid block or blob sequence"
else:
Result[seq[BlobSidecars], string].ok grouped
func checkBlobs(blobs: seq[BlobSidecars]): Result[void, string] =
for blob_sidecars in blobs:
for blob_sidecar in blob_sidecars:
? blob_sidecar[].verify_blob_sidecar_inclusion_proof()
ok()
proc syncStep[A, B](man: SyncManager[A, B], index: int, peer: A)
{.async: (raises: [CancelledError]).} =
logScope:
peer_score = peer.getScore()
peer_speed = peer.netKbps()
index = index
sync_ident = man.ident
topics = "syncman"
var
headSlot = man.getLocalHeadSlot()
wallSlot = man.getLocalWallSlot()
peerSlot = peer.getHeadSlot()
block: # Check that peer status is recent and relevant
logScope:
peer = peer
direction = man.direction
debug "Peer's syncing status", wall_clock_slot = wallSlot,
remote_head_slot = peerSlot, local_head_slot = headSlot
let
peerStatusAge = Moment.now() - peer.getStatusLastTime()
needsUpdate =
# Latest status we got is old
peerStatusAge >= StatusExpirationTime or
# The point we need to sync is close to where the peer is
man.getFirstSlot() >= peerSlot
if needsUpdate:
man.workers[index].status = SyncWorkerStatus.UpdatingStatus
# Avoid a stampede of requests, but make them more frequent in case the
# peer is "close" to the slot range of interest
if peerStatusAge < StatusExpirationTime div 2:
await sleepAsync(StatusExpirationTime div 2 - peerStatusAge)
trace "Updating peer's status information", wall_clock_slot = wallSlot,
remote_head_slot = peerSlot, local_head_slot = headSlot
2020-09-24 12:14:29 +00:00
if not(await peer.updateStatus()):
peer.updateScore(PeerScoreNoStatus)
debug "Failed to get remote peer's status, exiting",
peer_head_slot = peerSlot
return
let newPeerSlot = peer.getHeadSlot()
if peerSlot >= newPeerSlot:
peer.updateScore(PeerScoreStaleStatus)
debug "Peer's status information is stale",
wall_clock_slot = wallSlot, remote_old_head_slot = peerSlot,
local_head_slot = headSlot, remote_new_head_slot = newPeerSlot
else:
debug "Peer's status information updated", wall_clock_slot = wallSlot,
remote_old_head_slot = peerSlot, local_head_slot = headSlot,
remote_new_head_slot = newPeerSlot
peer.updateScore(PeerScoreGoodStatus)
peerSlot = newPeerSlot
# Time passed - enough to move slots, if sleep happened
headSlot = man.getLocalHeadSlot()
wallSlot = man.getLocalWallSlot()
if man.remainingSlots() <= man.maxHeadAge:
logScope:
peer = peer
direction = man.direction
case man.direction
of SyncQueueKind.Forward:
info "We are in sync with network", wall_clock_slot = wallSlot,
remote_head_slot = peerSlot, local_head_slot = headSlot
of SyncQueueKind.Backward:
info "Backfill complete", wall_clock_slot = wallSlot,
remote_head_slot = peerSlot, local_head_slot = headSlot
# We clear SyncManager's `notInSyncEvent` so all the workers will become
# sleeping soon.
man.notInSyncEvent.clear()
return
# Find out if the peer potentially can give useful blocks - in the case of
# forward sync, they can be useful if they have blocks newer than our head -
# in the case of backwards sync, they're useful if they have blocks newer than
# the backfill point
if man.getFirstSlot() >= peerSlot:
# This is not very good solution because we should not discriminate and/or
# penalize peers which are in sync process too, but their latest head is
# lower then our latest head. We should keep connections with such peers
# (so this peers are able to get in sync using our data), but we should
# not use this peers for syncing because this peers are useless for us.
# Right now we decreasing peer's score a bit, so it will not be
# disconnected due to low peer's score, but new fresh peers could replace
# peers with low latest head.
debug "Peer's head slot is lower then local head slot", peer = peer,
wall_clock_slot = wallSlot, remote_head_slot = peerSlot,
local_last_slot = man.getLastSlot(),
local_first_slot = man.getFirstSlot(),
direction = man.direction
peer.updateScore(PeerScoreUseless)
return
2020-04-20 14:59:18 +00:00
# Wall clock keeps ticking, so we need to update the queue
man.queue.updateLastSlot(man.getLastSlot())
man.workers[index].status = SyncWorkerStatus.Requesting
let req = man.queue.pop(peerSlot, peer)
if req.isEmpty():
# SyncQueue could return empty request in 2 cases:
# 1. There no more slots in SyncQueue to download (we are synced, but
# our ``notInSyncEvent`` is not yet cleared).
# 2. Current peer's known head slot is too low to satisfy request.
#
# To avoid endless loop we going to wait for RESP_TIMEOUT time here.
# This time is enough for all pending requests to finish and it is also
# enough for main sync loop to clear ``notInSyncEvent``.
debug "Empty request received from queue, exiting", peer = peer,
local_head_slot = headSlot, remote_head_slot = peerSlot,
queue_input_slot = man.queue.inpSlot,
queue_output_slot = man.queue.outSlot,
queue_last_slot = man.queue.finalSlot, direction = man.direction
await sleepAsync(RESP_TIMEOUT_DUR)
return
debug "Creating new request for peer", wall_clock_slot = wallSlot,
remote_head_slot = peerSlot, local_head_slot = headSlot,
request = req
man.workers[index].status = SyncWorkerStatus.Downloading
2020-04-20 14:59:18 +00:00
let blocks = await man.getBlocks(peer, req)
if blocks.isErr():
peer.updateScore(PeerScoreNoValues)
man.queue.push(req)
debug "Failed to receive blocks on request",
request = req, err = blocks.error
return
let blockData = blocks.get().asSeq()
let blockSmap = getShortMap(req, blockData)
debug "Received blocks on request", blocks_count = len(blockData),
blocks_map = blockSmap, request = req
let slots = mapIt(blockData, it[].slot)
if not(checkResponse(req, slots)):
peer.updateScore(PeerScoreBadResponse)
man.queue.push(req)
warn "Received blocks sequence is not in requested range",
blocks_count = len(blockData), blocks_map = blockSmap,
request = req
return
let shouldGetBlobs =
if not man.shouldGetBlobs(req.slot.epoch):
false
else:
var hasBlobs = false
for blck in blockData:
withBlck(blck[]):
when consensusFork >= ConsensusFork.Deneb:
if forkyBlck.message.body.blob_kzg_commitments.len > 0:
hasBlobs = true
break
hasBlobs
func combine(acc: seq[Slot], cur: Slot): seq[Slot] =
var copy = acc
if copy[copy.len-1] != cur:
copy.add(cur)
copy
let blobData =
if shouldGetBlobs:
let blobs = await man.getBlobSidecars(peer, req)
if blobs.isErr():
peer.updateScore(PeerScoreNoValues)
man.queue.push(req)
debug "Failed to receive blobs on request",
request = req, err = blobs.error
return
let blobData = blobs.get().asSeq()
let blobSmap = getShortMap(req, blobData)
debug "Received blobs on request", blobs_count = len(blobData),
blobs_map = blobSmap, request = req
if len(blobData) > 0:
let slots = mapIt(blobData, it[].signed_block_header.message.slot)
let uniqueSlots = foldl(slots, combine(a, b), @[slots[0]])
if not(checkResponse(req, uniqueSlots)):
peer.updateScore(PeerScoreBadResponse)
man.queue.push(req)
warn "Received blobs sequence is not in requested range",
blobs_count = len(blobData), blobs_map = getShortMap(req, blobData),
request = req
return
let groupedBlobs = groupBlobs(req, blockData, blobData)
if groupedBlobs.isErr():
peer.updateScore(PeerScoreNoValues)
man.queue.push(req)
info "Received blobs sequence is inconsistent",
blobs_map = getShortMap(req, blobData), request = req, msg=groupedBlobs.error()
return
if (let checkRes = groupedBlobs.get.checkBlobs(); checkRes.isErr):
peer.updateScore(PeerScoreBadResponse)
man.queue.push(req)
warn "Received blobs sequence is invalid",
blobs_count = len(blobData),
blobs_map = getShortMap(req, blobData),
request = req,
msg = checkRes.error
return
Opt.some(groupedBlobs.get())
else:
Opt.none(seq[BlobSidecars])
if len(blockData) == 0 and man.direction == SyncQueueKind.Backward and
req.contains(man.getSafeSlot()):
# The sync protocol does not distinguish between:
# - All requested slots are empty
# - Peer does not have data available about requested range
#
# However, we include the `backfill` slot in backward sync requests.
# If we receive an empty response to a request covering that slot,
# we know that the response is incomplete and can descore.
peer.updateScore(PeerScoreNoValues)
man.queue.push(req)
debug "Response does not include known-to-exist block", request = req
return
2020-01-21 18:30:21 +00:00
# Scoring will happen in `syncUpdate`.
man.workers[index].status = SyncWorkerStatus.Queueing
let
peerFinalized = peer.getFinalizedEpoch().start_slot()
lastSlot = req.slot + req.count
# The peer claims the block is finalized - our own block processing will
# verify this point down the line
# TODO descore peers that lie
maybeFinalized = lastSlot < peerFinalized
await man.queue.push(req, blockData, blobData, maybeFinalized, proc() =
man.workers[index].status = SyncWorkerStatus.Processing)
proc syncWorker[A, B](man: SyncManager[A, B], index: int) {.async: (raises: [CancelledError]).} =
mixin getKey, getScore, getHeadSlot
logScope:
index = index
sync_ident = man.ident
direction = man.direction
topics = "syncman"
debug "Starting syncing worker"
var peer: A = nil
try:
while true:
man.workers[index].status = SyncWorkerStatus.Sleeping
# This event is going to be set until we are not in sync with network
await man.notInSyncEvent.wait()
man.workers[index].status = SyncWorkerStatus.WaitingPeer
peer = await man.pool.acquire()
await man.syncStep(index, peer)
man.pool.release(peer)
peer = nil
finally:
if not(isNil(peer)):
man.pool.release(peer)
debug "Sync worker stopped"
proc getWorkersStats[A, B](man: SyncManager[A, B]): tuple[map: string,
sleeping: int,
waiting: int,
pending: int] =
var map = newString(len(man.workers))
var sleeping, waiting, pending: int
for i in 0 ..< len(man.workers):
var ch: char
case man.workers[i].status
of SyncWorkerStatus.Sleeping:
ch = 's'
inc(sleeping)
of SyncWorkerStatus.WaitingPeer:
ch = 'w'
inc(waiting)
of SyncWorkerStatus.UpdatingStatus:
ch = 'U'
inc(pending)
of SyncWorkerStatus.Requesting:
ch = 'R'
inc(pending)
of SyncWorkerStatus.Downloading:
ch = 'D'
inc(pending)
of SyncWorkerStatus.Queueing:
ch = 'Q'
inc(pending)
of SyncWorkerStatus.Processing:
ch = 'P'
inc(pending)
map[i] = ch
(map, sleeping, waiting, pending)
2020-01-21 18:30:21 +00:00
proc startWorkers[A, B](man: SyncManager[A, B]) =
# Starting all the synchronization workers.
for i in 0 ..< len(man.workers):
man.workers[i].future = syncWorker[A, B](man, i)
proc toTimeLeftString*(d: Duration): string =
if d == InfiniteDuration:
"--h--m"
else:
var v = d
var res = ""
let ndays = chronos.days(v)
if ndays > 0:
res = res & (if ndays < 10: "0" & $ndays else: $ndays) & "d"
v = v - chronos.days(ndays)
let nhours = chronos.hours(v)
if nhours > 0:
res = res & (if nhours < 10: "0" & $nhours else: $nhours) & "h"
v = v - chronos.hours(nhours)
else:
res = res & "00h"
let nmins = chronos.minutes(v)
if nmins > 0:
res = res & (if nmins < 10: "0" & $nmins else: $nmins) & "m"
v = v - chronos.minutes(nmins)
else:
res = res & "00m"
res
proc syncClose[A, B](man: SyncManager[A, B],
speedTaskFut: Future[void]) {.async.} =
var pending: seq[FutureBase]
if not(speedTaskFut.finished()):
pending.add(speedTaskFut.cancelAndWait())
for worker in man.workers:
doAssert(worker.status in {Sleeping, WaitingPeer})
pending.add(worker.future.cancelAndWait())
await noCancel allFutures(pending)
proc syncLoop[A, B](man: SyncManager[A, B]) {.async.} =
logScope:
sync_ident = man.ident
direction = man.direction
topics = "syncman"
2020-04-20 14:59:18 +00:00
mixin getKey, getScore
var pauseTime = 0
2020-01-21 18:30:21 +00:00
man.startWorkers()
2020-01-21 18:30:21 +00:00
debug "Synchronization loop started"
proc averageSpeedTask() {.async: (raises: [CancelledError]).} =
while true:
# Reset sync speeds between each loss-of-sync event
man.avgSyncSpeed = 0
man.insSyncSpeed = 0
await man.notInSyncEvent.wait()
# Give the node time to connect to peers and get the sync process started
await sleepAsync(seconds(SECONDS_PER_SLOT.int64))
var
stamp = SyncMoment.now(man.queue.progress())
syncCount = 0
while man.inProgress:
await sleepAsync(seconds(SECONDS_PER_SLOT.int64))
let
newStamp = SyncMoment.now(man.queue.progress())
slotsPerSec = speed(stamp, newStamp)
syncCount += 1
man.insSyncSpeed = slotsPerSec
man.avgSyncSpeed =
man.avgSyncSpeed + (slotsPerSec - man.avgSyncSpeed) / float(syncCount)
stamp = newStamp
let averageSpeedTaskFut = averageSpeedTask()
2020-04-20 14:59:18 +00:00
while true:
let wallSlot = man.getLocalWallSlot()
let headSlot = man.getLocalHeadSlot()
2020-04-20 14:59:18 +00:00
let (map, sleeping, waiting, pending) = man.getWorkersStats()
debug "Current syncing state", workers_map = map,
sleeping_workers_count = sleeping,
waiting_workers_count = waiting,
pending_workers_count = pending,
wall_head_slot = wallSlot, local_head_slot = headSlot,
pause_time = $chronos.seconds(pauseTime),
avg_sync_speed = man.avgSyncSpeed, ins_sync_speed = man.insSyncSpeed
let
pivot = man.progressPivot
progress =
case man.queue.kind
of SyncQueueKind.Forward:
if man.queue.outSlot >= pivot:
man.queue.outSlot - pivot
else:
0'u64
of SyncQueueKind.Backward:
if pivot >= man.queue.outSlot:
pivot - man.queue.outSlot
else:
0'u64
total =
case man.queue.kind
of SyncQueueKind.Forward:
if man.queue.finalSlot >= pivot:
man.queue.finalSlot + 1'u64 - pivot
else:
0'u64
of SyncQueueKind.Backward:
if pivot >= man.queue.finalSlot:
pivot + 1'u64 - man.queue.finalSlot
else:
0'u64
remaining = total - progress
done =
if total > 0:
progress.float / total.float
else:
1.0
timeleft =
if man.avgSyncSpeed >= 0.001:
Duration.fromFloatSeconds(remaining.float / man.avgSyncSpeed)
else:
InfiniteDuration
currentSlot = Base10.toString(
if man.queue.kind == SyncQueueKind.Forward:
max(uint64(man.queue.outSlot), 1'u64) - 1'u64
else:
uint64(man.queue.outSlot) + 1'u64
)
2020-04-20 14:59:18 +00:00
# Update status string
man.syncStatus = timeleft.toTimeLeftString() & " (" &
(done * 100).formatBiggestFloat(ffDecimal, 2) & "%) " &
man.avgSyncSpeed.formatBiggestFloat(ffDecimal, 4) &
"slots/s (" & map & ":" & currentSlot & ")"
if man.remainingSlots() <= man.maxHeadAge:
man.notInSyncEvent.clear()
# We are marking SyncManager as not working only when we are in sync and
# all sync workers are in `Sleeping` state.
if pending > 0:
debug "Synchronization loop waits for workers completion",
wall_head_slot = wallSlot, local_head_slot = headSlot,
difference = (wallSlot - headSlot), max_head_age = man.maxHeadAge,
sleeping_workers_count = sleeping,
waiting_workers_count = waiting, pending_workers_count = pending
# We already synced, so we should reset all the pending workers from
# any state they have.
man.queue.clearAndWakeup()
man.inProgress = true
else:
case man.direction
of SyncQueueKind.Forward:
if man.inProgress:
if SyncManagerFlag.NoMonitor in man.flags:
await man.syncClose(averageSpeedTaskFut)
man.inProgress = false
debug "Forward synchronization process finished, exiting",
wall_head_slot = wallSlot, local_head_slot = headSlot,
difference = (wallSlot - headSlot),
max_head_age = man.maxHeadAge
break
else:
man.inProgress = false
debug "Forward synchronization process finished, sleeping",
wall_head_slot = wallSlot, local_head_slot = headSlot,
difference = (wallSlot - headSlot),
max_head_age = man.maxHeadAge
else:
debug "Synchronization loop sleeping", wall_head_slot = wallSlot,
local_head_slot = headSlot,
difference = (wallSlot - headSlot),
max_head_age = man.maxHeadAge
of SyncQueueKind.Backward:
# Backward syncing is going to be executed only once, so we exit loop
# and stop all pending tasks which belongs to this instance (sync
# workers, speed calculation task).
await man.syncClose(averageSpeedTaskFut)
man.inProgress = false
debug "Backward synchronization process finished, exiting",
wall_head_slot = wallSlot, local_head_slot = headSlot,
backfill_slot = man.getLastSlot(),
max_head_age = man.maxHeadAge
break
else:
if not(man.notInSyncEvent.isSet()):
# We get here only if we lost sync for more then `maxHeadAge` period.
if pending == 0:
man.initQueue()
man.notInSyncEvent.fire()
man.inProgress = true
debug "Node lost sync for more then preset period",
period = man.maxHeadAge, wall_head_slot = wallSlot,
local_head_slot = headSlot,
missing_slots = man.remainingSlots(),
progress = float(man.queue.progress())
else:
man.notInSyncEvent.fire()
man.inProgress = true
await sleepAsync(chronos.seconds(2))
proc start*[A, B](man: SyncManager[A, B]) =
## Starts SyncManager's main loop.
man.syncFut = man.syncLoop()