nimbus-eth1/nimbus/sync/misc/best_pivot.nim

456 lines
15 KiB
Nim
Raw Normal View History

# Nimbus
# Copyright (c) 2018-2021 Status Research & Development GmbH
# Licensed and distributed under either of
# * MIT license (license terms in the root directory or at
# https://opensource.org/licenses/MIT).
# * Apache v2 license (license terms in the root directory or at
# https://www.apache.org/licenses/LICENSE-2.0).
# at your option. This file may not be copied, modified, or distributed
# except according to those terms.
## Negotiate a pivot header base on the *best header* values. Buddies that
## cannot provide a minimal block number will be disconnected.
##
## Ackn: nim-eth/eth/p2p/blockchain_sync.nim: `startSyncWithPeer()`
import
std/[hashes, options, sets],
chronicles,
chronos,
eth/[common, p2p],
stew/byteutils,
".."/[protocol, sync_desc, types]
{.push raises:[Defect].}
logScope:
topics = "best-pivot"
const
extraTraceMessages = false or true
## Additional trace commands
pivotMinPeersToStartSync* = 2
## Wait for consensus of at least this number of peers before syncing.
pivotFailCountMax* = 3
## Stop after a peer fails too often while negotiating. This happens if
## a peer responses repeatedly with useless data.
type
BestPivotCtxRef* = ref object of RootRef
## Data shared by all peers.
rng: ref HmacDrbgContext ## Random generator
untrusted: seq[Peer] ## Clean up list
trusted: HashSet[Peer] ## Peers ready for delivery
relaxedMode: bool ## Not using strictly `trusted` set
BestPivotWorkerRef* = ref object of RootRef
## Data for this peer only
global: BestPivotCtxRef ## Common data
header: Option[BlockHeader] ## Pivot header (if any)
ctrl: BuddyCtrlRef ## Control and state settings
peer: Peer ## network peer
comFailCount: int ## Beware of repeated network errors
# ------------------------------------------------------------------------------
# Private helpers
# ------------------------------------------------------------------------------
proc hash(peer: Peer): Hash =
## Mixin `HashSet[Peer]` handler
hash(cast[pointer](peer))
template safeTransport(
bp: BestPivotWorkerRef;
info: static[string];
code: untyped) =
try:
code
except TransportError as e:
error info & ", stop", peer=bp.peer, error=($e.name), msg=e.msg
bp.ctrl.stopped = true
proc rand(r: ref HmacDrbgContext; maxVal: uint64): uint64 =
# github.com/nim-lang/Nim/tree/version-1-6/lib/pure/random.nim#L216
const
randMax = high(uint64)
if 0 < maxVal:
if maxVal == randMax:
var x: uint64
r[].generate(x)
return x
while true:
var x: uint64
r[].generate(x)
# avoid `mod` bias, so `x <= n*maxVal <= randMax` for some integer `n`
if x <= randMax - (randMax mod maxVal):
# uint -> int
return x mod (maxVal + 1)
proc rand(r: ref HmacDrbgContext; maxVal: int): int =
if 0 < maxVal: # somehow making sense of `maxVal = -1`
return cast[int](r.rand(maxVal.uint64))
# ------------------------------------------------------------------------------
# Private functions
# ------------------------------------------------------------------------------
proc getRandomTrustedPeer(bp: BestPivotWorkerRef): Result[Peer,void] =
## Return random entry from `trusted` peer different from this peer set if
## there are enough
##
## Ackn: nim-eth/eth/p2p/blockchain_sync.nim: `randomTrustedPeer()`
let
nPeers = bp.global.trusted.len
offInx = if bp.peer in bp.global.trusted: 2 else: 1
if 0 < nPeers:
var (walkInx, stopInx) = (0, bp.global.rng.rand(nPeers - offInx))
for p in bp.global.trusted:
if p == bp.peer:
continue
if walkInx == stopInx:
return ok(p)
walkInx.inc
err()
proc getBestHeader(
bp: BestPivotWorkerRef;
): Future[Result[BlockHeader,void]]
{.async.} =
## Get best block number from best block hash.
##
## Ackn: nim-eth/eth/p2p/blockchain_sync.nim: `getBestBlockNumber()`
let
peer = bp.peer
startHash = peer.state(eth).bestBlockHash
reqLen = 1u
hdrReq = BlocksRequest(
startBlock: HashOrNum(
isHash: true,
hash: startHash),
maxResults: reqLen,
skip: 0,
reverse: true)
trace trEthSendSendingGetBlockHeaders, peer,
startBlock=startHash.data.toHex, reqLen
var hdrResp: Option[blockHeadersObj]
bp.safeTransport("Error fetching block header"):
hdrResp = await peer.getBlockHeaders(hdrReq)
if bp.ctrl.stopped:
return err()
if hdrResp.isNone:
bp.comFailCount.inc
trace trEthRecvReceivedBlockHeaders, peer, reqLen,
response="n/a", comFailCount=bp.comFailCount
if pivotFailCountMax < bp.comFailCount:
bp.ctrl.zombie = true
return err()
let hdrRespLen = hdrResp.get.headers.len
if hdrRespLen == 1:
let
header = hdrResp.get.headers[0]
blockNumber = header.blockNumber
trace trEthRecvReceivedBlockHeaders, peer, hdrRespLen, blockNumber
bp.comFailCount = 0 # reset fail count
return ok(header)
trace trEthRecvReceivedBlockHeaders, peer, reqLen, hdrRespLen
return err()
proc agreesOnChain(
bp: BestPivotWorkerRef;
other: Peer
): Future[Result[void,bool]]
{.async.} =
## Returns `true` if one of the peers `bp.peer` or `other` acknowledges
## existence of the best block of the other peer. The values returned mean
## * ok() -- `peer` is trusted
## * err(true) -- `peer` is untrusted
## * err(false) -- `other` is dead
##
## Ackn: nim-eth/eth/p2p/blockchain_sync.nim: `peersAgreeOnChain()`
let
peer = bp.peer
var
start = peer
fetch = other
swapped = false
# Make sure that `fetch` has not the smaller difficulty.
if fetch.state(eth).bestDifficulty < start.state(eth).bestDifficulty:
swap(fetch, start)
swapped = true
let
startHash = start.state(eth).bestBlockHash
hdrReq = BlocksRequest(
startBlock: HashOrNum(
isHash: true,
hash: startHash),
maxResults: 1,
skip: 0,
reverse: true)
trace trEthSendSendingGetBlockHeaders, peer, start, fetch,
startBlock=startHash.data.toHex, hdrReqLen=1, swapped
var hdrResp: Option[blockHeadersObj]
bp.safeTransport("Error fetching block header"):
hdrResp = await fetch.getBlockHeaders(hdrReq)
if bp.ctrl.stopped:
if swapped:
return err(true)
# No need to terminate `peer` if it was the `other`, failing nevertheless
bp.ctrl.stopped = false
return err(false)
if hdrResp.isSome:
let hdrRespLen = hdrResp.get.headers.len
if 0 < hdrRespLen:
let blockNumber = hdrResp.get.headers[0].blockNumber
trace trEthRecvReceivedBlockHeaders, peer, start, fetch,
hdrRespLen, blockNumber
return ok()
trace trEthRecvReceivedBlockHeaders, peer, start, fetch,
blockNumber="n/a", swapped
return err(true)
# ------------------------------------------------------------------------------
# Public functions, constructor
# ------------------------------------------------------------------------------
proc init*(
T: type BestPivotCtxRef; ## Global data descriptor type
rng: ref HmacDrbgContext; ## Random generator
): T =
## Global constructor, shared data
T(rng: rng)
proc init*(
T: type BestPivotWorkerRef; ## Global data descriptor type
ctx: BestPivotCtxRef; ## Global data descriptor
ctrl: BuddyCtrlRef; ## Control and state settings
peer: Peer; ## For fetching data from network
): T =
## Buddy/local constructor
T(global: ctx,
header: none(BlockHeader),
ctrl: ctrl,
peer: peer)
proc clear*(bp: BestPivotWorkerRef) =
## Reset descriptor
bp.global.untrusted.add bp.peer
bp.header = none(BlockHeader)
# ------------------------------------------------------------------------------
# Public functions
# ------------------------------------------------------------------------------
proc pivotRelaxedMode*(ctx: BestPivotCtxRef; enable = false) =
## Controls relaxed mode. In relaxed mode, the *best header* is fetched
## from the network and used as pivot if its block number is large enough.
## Otherwise, the default is to find at least `pivotMinPeersToStartSync`
## peers (this one included) that agree on a minimum pivot.
ctx.relaxedMode = enable
Prep for full sync after snap (#1253) * Split fetch accounts into sub-modules details: There will be separated modules for accounts snapshot, storage snapshot, and healing for either. * Allow to rebase pivot before negotiated header why: Peers seem to have not too many snapshots available. By setting back the pivot block header slightly, the chances might be higher to find more peers to serve this pivot. Experiment on mainnet showed that setting back too much (tested with 1024), the chances to find matching snapshot peers seem to decrease. * Add accounts healing * Update variable/field naming in `worker_desc` for readability * Handle leaf nodes in accounts healing why: There is no need to fetch accounts when they had been added by the healing process. On the flip side, these accounts must be checked for storage data and the batch queue updated, accordingly. * Reorganising accounts hash ranges batch queue why: The aim is to formally cover as many accounts as possible for different pivot state root environments. Formerly, this was tried by starting the accounts batch queue at a random value for each pivot (and wrapping around.) Now, each pivot environment starts with an interval set mutually disjunct from any interval set retrieved with other pivot state roots. also: Stop fishing for more pivots in `worker` if 100% download is reached * Reorganise/update accounts healing why: Error handling was wrong and the (math. complexity of) whole process could be better managed. details: Much of the algorithm is now documented at the top of the file `heal_accounts.nim`
2022-10-08 17:20:50 +00:00
proc pivotHeader*(bp: BestPivotWorkerRef): Result[BlockHeader,void] =
## Returns cached block header if available and the buddy `peer` is trusted.
## In relaxed mode (see `pivotRelaxedMode()`), also lesser trusted pivots
## are returned.
if bp.header.isSome and
bp.peer notin bp.global.untrusted:
if pivotMinPeersToStartSync <= bp.global.trusted.len and
bp.peer in bp.global.trusted:
return ok(bp.header.unsafeGet)
if bp.global.relaxedMode:
when extraTraceMessages:
trace "Returning not fully trusted pivot", peer=bp.peer,
trusted=bp.global.trusted.len, untrusted=bp.global.untrusted.len
return ok(bp.header.unsafeGet)
err()
proc pivotHeader*(
bp: BestPivotWorkerRef; ## Worker peer
relaxedMode: bool; ## One time relaxed mode flag
): Result[BlockHeader,void] =
## Variant of `pivotHeader()` with `relaxedMode` flag as function argument.
if bp.header.isSome and
bp.peer notin bp.global.untrusted:
if pivotMinPeersToStartSync <= bp.global.trusted.len and
bp.peer in bp.global.trusted:
return ok(bp.header.unsafeGet)
if relaxedMode:
when extraTraceMessages:
trace "Returning not fully trusted pivot", peer=bp.peer,
trusted=bp.global.trusted.len, untrusted=bp.global.untrusted.len
return ok(bp.header.unsafeGet)
err()
Prep for full sync after snap (#1253) * Split fetch accounts into sub-modules details: There will be separated modules for accounts snapshot, storage snapshot, and healing for either. * Allow to rebase pivot before negotiated header why: Peers seem to have not too many snapshots available. By setting back the pivot block header slightly, the chances might be higher to find more peers to serve this pivot. Experiment on mainnet showed that setting back too much (tested with 1024), the chances to find matching snapshot peers seem to decrease. * Add accounts healing * Update variable/field naming in `worker_desc` for readability * Handle leaf nodes in accounts healing why: There is no need to fetch accounts when they had been added by the healing process. On the flip side, these accounts must be checked for storage data and the batch queue updated, accordingly. * Reorganising accounts hash ranges batch queue why: The aim is to formally cover as many accounts as possible for different pivot state root environments. Formerly, this was tried by starting the accounts batch queue at a random value for each pivot (and wrapping around.) Now, each pivot environment starts with an interval set mutually disjunct from any interval set retrieved with other pivot state roots. also: Stop fishing for more pivots in `worker` if 100% download is reached * Reorganise/update accounts healing why: Error handling was wrong and the (math. complexity of) whole process could be better managed. details: Much of the algorithm is now documented at the top of the file `heal_accounts.nim`
2022-10-08 17:20:50 +00:00
proc pivotNegotiate*(
bp: BestPivotWorkerRef; ## Worker peer
minBlockNumber: Option[BlockNumber]; ## Minimum block number to expect
Prep for full sync after snap (#1253) * Split fetch accounts into sub-modules details: There will be separated modules for accounts snapshot, storage snapshot, and healing for either. * Allow to rebase pivot before negotiated header why: Peers seem to have not too many snapshots available. By setting back the pivot block header slightly, the chances might be higher to find more peers to serve this pivot. Experiment on mainnet showed that setting back too much (tested with 1024), the chances to find matching snapshot peers seem to decrease. * Add accounts healing * Update variable/field naming in `worker_desc` for readability * Handle leaf nodes in accounts healing why: There is no need to fetch accounts when they had been added by the healing process. On the flip side, these accounts must be checked for storage data and the batch queue updated, accordingly. * Reorganising accounts hash ranges batch queue why: The aim is to formally cover as many accounts as possible for different pivot state root environments. Formerly, this was tried by starting the accounts batch queue at a random value for each pivot (and wrapping around.) Now, each pivot environment starts with an interval set mutually disjunct from any interval set retrieved with other pivot state roots. also: Stop fishing for more pivots in `worker` if 100% download is reached * Reorganise/update accounts healing why: Error handling was wrong and the (math. complexity of) whole process could be better managed. details: Much of the algorithm is now documented at the top of the file `heal_accounts.nim`
2022-10-08 17:20:50 +00:00
): Future[bool]
{.async.} =
## Negotiate best header pivot. This function must be run in *single mode* at
## the beginning of a running worker peer. If the function returns `true`,
## the current `buddy` can be used for syncing and the function
## `bestPivotHeader()` will succeed returning a `BlockHeader`.
##
## In relaxed mode (see `pivotRelaxedMode()`), negotiation stopps when there
## is a *best header*. It caches the best header and returns `true` it the
## block number is large enough.
##
## Ackn: nim-eth/eth/p2p/blockchain_sync.nim: `startSyncWithPeer()`
##
let peer = bp.peer
# Delayed clean up batch list
if 0 < bp.global.untrusted.len:
when extraTraceMessages:
trace "Removing untrusted peers", peer, trusted=bp.global.trusted.len,
untrusted=bp.global.untrusted.len, runState=bp.ctrl.state
bp.global.trusted = bp.global.trusted - bp.global.untrusted.toHashSet
bp.global.untrusted.setLen(0)
if bp.header.isNone:
when extraTraceMessages:
# Only log for the first time (if any)
trace "Pivot initialisation", peer,
trusted=bp.global.trusted.len, runState=bp.ctrl.state
let rc = await bp.getBestHeader()
# Beware of peer terminating the session right after communicating
if rc.isErr or bp.ctrl.stopped:
return false
let
bestNumber = rc.value.blockNumber
minNumber = minBlockNumber.get(otherwise = 0.toBlockNumber)
if bestNumber < minNumber:
bp.ctrl.zombie = true
trace "Useless peer, best number too low", peer,
trusted=bp.global.trusted.len, runState=bp.ctrl.state,
minNumber, bestNumber
return false
bp.header = some(rc.value)
# No further negotiation if in relaxed mode
if bp.global.relaxedMode:
return true
if pivotMinPeersToStartSync <= bp.global.trusted.len:
# We have enough trusted peers. Validate new peer against trusted
let rc = bp.getRandomTrustedPeer()
if rc.isOK:
let rx = await bp.agreesOnChain(rc.value)
if rx.isOk:
bp.global.trusted.incl peer
when extraTraceMessages:
let bestHeader =
if bp.header.isSome: "#" & $bp.header.get.blockNumber
else: "nil"
trace "Accepting peer", peer, trusted=bp.global.trusted.len,
untrusted=bp.global.untrusted.len, runState=bp.ctrl.state,
bestHeader
return true
if not rx.error:
# Other peer is dead
bp.global.trusted.excl rc.value
return false
# If there are no trusted peers yet, assume this very peer is trusted,
# but do not finish initialisation until there are more peers.
if bp.global.trusted.len == 0:
bp.global.trusted.incl peer
when extraTraceMessages:
let bestHeader =
if bp.header.isSome: "#" & $bp.header.get.blockNumber
else: "nil"
trace "Assume initial trusted peer", peer,
trusted=bp.global.trusted.len, runState=bp.ctrl.state, bestHeader
return false
if bp.global.trusted.len == 1 and bp.peer in bp.global.trusted:
# Ignore degenerate case, note that `trusted.len < minPeersToStartSync`
return false
# At this point we have some "trusted" candidates, but they are not
# "trusted" enough. We evaluate `peer` against all other candidates. If
# one of the candidates disagrees, we swap it for `peer`. If all candidates
# agree, we add `peer` to trusted set. The peers in the set will become
# "fully trusted" (and sync will start) when the set is big enough
var
agreeScore = 0
otherPeer: Peer
deadPeers: HashSet[Peer]
when extraTraceMessages:
trace "Trust scoring peer", peer,
trusted=bp.global.trusted.len, runState=bp.ctrl.state
for p in bp.global.trusted:
if peer == p:
inc agreeScore
else:
let rc = await bp.agreesOnChain(p)
if rc.isOk:
inc agreeScore
elif bp.ctrl.stopped:
# Beware of terminated session
return false
elif rc.error:
otherPeer = p
else:
# `Other` peer is dead
deadPeers.incl p
# Normalise
if 0 < deadPeers.len:
bp.global.trusted = bp.global.trusted - deadPeers
if bp.global.trusted.len == 0 or
bp.global.trusted.len == 1 and bp.peer in bp.global.trusted:
return false
# Check for the number of peers that disagree
case bp.global.trusted.len - agreeScore:
of 0:
bp.global.trusted.incl peer # best possible outcome
when extraTraceMessages:
trace "Agreeable trust score for peer", peer,
trusted=bp.global.trusted.len, runState=bp.ctrl.state
of 1:
bp.global.trusted.excl otherPeer
bp.global.trusted.incl peer
when extraTraceMessages:
trace "Other peer no longer trusted", peer,
otherPeer, trusted=bp.global.trusted.len, runState=bp.ctrl.state
else:
when extraTraceMessages:
trace "Peer not trusted", peer,
trusted=bp.global.trusted.len, runState=bp.ctrl.state
discard
# Evaluate status, finally
if pivotMinPeersToStartSync <= bp.global.trusted.len:
when extraTraceMessages:
let bestHeader =
if bp.header.isSome: "#" & $bp.header.get.blockNumber
else: "nil"
trace "Peer trusted now", peer,
trusted=bp.global.trusted.len, runState=bp.ctrl.state, bestHeader
return true
# ------------------------------------------------------------------------------
# End
# ------------------------------------------------------------------------------