nimbus-eth1/nimbus/db/aristo/aristo_merge.nim
Jacek Sieka 01ca415721
Store keys together with node data (#2849)
Currently, computed hash keys are stored in a separate column family
with respect to the MPT data they're generated from - this has several
disadvantages:

* A lot of space is wasted because the lookup key (`RootedVertexID`) is
repeated in both tables - this is 30% of the `AriKey` content!
* rocksdb must maintain in-memory bloom filters and LRU caches for said
keys, doubling its "minimal efficient cache size"
* An extra disk traversal must be made to check for existence of cached
hash key
* Doubles the amount of files on disk due to each column family being
its own set of files

Here, the two CFs are joined such that both key and data is stored in
`AriVtx`. This means:

* we save ~30% disk space on repeated lookup keys
* we save ~2gb of memory overhead that can be used to cache data instead
of indices
* we can skip storing hash keys for MPT leaf nodes - these are trivial
to compute and waste a lot of space - previously they had to present in
the `AriKey` CF to avoid having to look in two tables on the happy path.
* There is a small increase in write amplification because when a hash
value is updated for a branch node, we must write both key and branch
data - previously we would write only the key
* There's a small shift in CPU usage - instead of performing lookups in
the database, hashes for leaf nodes are (re)-computed on the fly
* We can return to slightly smaller on-disk SST files since there's
fewer of them, which should reduce disk traffic a bit

Internally, there are also other advantages:

* when clearing keys, we no longer have to store a zero hash in memory -
instead, we deduce staleness of the cached key from the presence of an
updated VertexRef - this saves ~1gb of mem overhead during import
* hash key cache becomes dedicated to branch keys since leaf keys are no
longer stored in memory, reducing churn
* key computation is a lot faster thanks to the skipped second disk
traversal - a key computation for mainnet can be completed in 11 hours
instead of ~2 days (!) thanks to better cache usage and less read
amplification - with additional improvements to the on-disk format, we
can probably get rid of the initial full traversal method of seeding the
key cache on first start after import

All in all, this PR reduces the size of a mainnet database from 160gb to
110gb and the peak memory footprint during import by ~1-2gb.
2024-11-20 09:56:27 +01:00

264 lines
9.6 KiB
Nim

# nimbus-eth1
# Copyright (c) 2023-2024 Status Research & Development GmbH
# Licensed under either of
# * Apache License, version 2.0, ([LICENSE-APACHE](LICENSE-APACHE) or
# http://www.apache.org/licenses/LICENSE-2.0)
# * MIT license ([LICENSE-MIT](LICENSE-MIT) or
# http://opensource.org/licenses/MIT)
# at your option. This file may not be copied, modified, or distributed
# except according to those terms.
## Aristo DB -- Patricia Trie builder, raw node insertion
## ======================================================
##
## This module merges `PathID` values as hexary lookup paths into the
## `Patricia Trie`. When changing vertices (aka nodes without Merkle hashes),
## associated (but separated) Merkle hashes will be deleted unless locked.
## Instead of deleting locked hashes error handling is applied.
##
## Also, nodes (vertices plus merkle hashes) can be added which is needed for
## boundary proofing after `snap/1` download. The vertices are split from the
## nodes and stored as-is on the table holding `Patricia Trie` entries. The
## hashes are stored iin a separate table and the vertices are labelled
## `locked`.
{.push raises: [].}
import
std/typetraits,
eth/common,
results,
"."/[aristo_desc, aristo_fetch, aristo_get, aristo_layers, aristo_vid]
proc layersPutLeaf(
db: AristoDbRef, rvid: RootedVertexID, path: NibblesBuf, payload: LeafPayload
): VertexRef =
let vtx = VertexRef(vType: Leaf, pfx: path, lData: payload)
db.layersPutVtx(rvid, vtx)
vtx
proc mergePayloadImpl(
db: AristoDbRef, # Database, top layer
root: VertexID, # MPT state root
path: openArray[byte], # Leaf item to add to the database
leaf: Opt[VertexRef],
payload: LeafPayload, # Payload value
): Result[(VertexRef, VertexRef, VertexRef), AristoError] =
## Merge the argument `(root,path)` key-value-pair into the top level vertex
## table of the database `db`. The `path` argument is used to address the
## leaf vertex with the payload. It is stored or updated on the database
## accordingly.
##
var
path = NibblesBuf.fromBytes(path)
cur = root
(vtx, _) = db.getVtxRc((root, cur)).valueOr:
if error != GetVtxNotFound:
return err(error)
# We're at the root vertex and there is no data - this must be a fresh
# VertexID!
return ok (db.layersPutLeaf((root, cur), path, payload), nil, nil)
vids: ArrayBuf[NibblesBuf.high + 1, VertexID]
vtxs: ArrayBuf[NibblesBuf.high + 1, VertexRef]
template resetKeys() =
# Reset cached hashes of touched verticies
for i in 2..vids.len:
db.layersResKey((root, vids[^i]), vtxs[^i])
while path.len > 0:
# Clear existing merkle keys along the traversal path
vids.add cur
vtxs.add vtx
let n = path.sharedPrefixLen(vtx.pfx)
case vtx.vType
of Leaf:
let res =
if n == vtx.pfx.len:
# Same path - replace the current vertex with a new payload
if vtx.lData == payload:
return err(MergeNoAction)
let leafVtx = if root == VertexID(1):
var payload = payload.dup()
# TODO can we avoid this hack? it feels like the caller should already
# have set an appropriate stoID - this "fixup" feels risky,
# specially from a caching point of view
payload.stoID = vtx.lData.stoID
db.layersPutLeaf((root, cur), path, payload)
else:
db.layersPutLeaf((root, cur), path, payload)
(leafVtx, nil, nil)
else:
# Turn leaf into a branch (or extension) then insert the two leaves
# into the branch
let branch = VertexRef(vType: Branch, pfx: path.slice(0, n))
let other = block: # Copy of existing leaf node, now one level deeper
let local = db.vidFetch()
branch.bVid[vtx.pfx[n]] = local
db.layersPutLeaf((root, local), vtx.pfx.slice(n + 1), vtx.lData)
let leafVtx = block: # Newly inserted leaf node
let local = db.vidFetch()
branch.bVid[path[n]] = local
db.layersPutLeaf((root, local), path.slice(n + 1), payload)
# Put the branch at the vid where the leaf was
db.layersPutVtx((root, cur), branch)
# We need to return vtx here because its pfx member hasn't yet been
# sliced off and is therefore shared with the hike
(leafVtx, vtx, other)
resetKeys()
return ok(res)
of Branch:
if vtx.pfx.len == n:
# The existing branch is a prefix of the new entry
let
nibble = path[vtx.pfx.len]
next = vtx.bVid[nibble]
if next.isValid:
cur = next
path = path.slice(n + 1)
vtx =
if leaf.isSome and leaf[].isValid and leaf[].pfx == path:
leaf[]
else:
(?db.getVtxRc((root, next)))[0]
else:
# There's no vertex at the branch point - insert the payload as a new
# leaf and update the existing branch
let
local = db.vidFetch()
leafVtx = db.layersPutLeaf((root, local), path.slice(n + 1), payload)
brDup = vtx.dup()
brDup.bVid[nibble] = local
db.layersPutVtx((root, cur), brDup)
resetKeys()
return ok((leafVtx, nil, nil))
else:
# Partial path match - we need to split the existing branch at
# the point of divergence, inserting a new branch
let branch = VertexRef(vType: Branch, pfx: path.slice(0, n))
block: # Copy the existing vertex and add it to the new branch
let local = db.vidFetch()
branch.bVid[vtx.pfx[n]] = local
db.layersPutVtx(
(root, local),
VertexRef(vType: Branch, pfx: vtx.pfx.slice(n + 1), bVid: vtx.bVid),
)
let leafVtx = block: # add the new entry
let local = db.vidFetch()
branch.bVid[path[n]] = local
db.layersPutLeaf((root, local), path.slice(n + 1), payload)
db.layersPutVtx((root, cur), branch)
resetKeys()
return ok((leafVtx, nil, nil))
err(MergeHikeFailed)
# ------------------------------------------------------------------------------
# Public functions
# ------------------------------------------------------------------------------
proc mergeAccountRecord*(
db: AristoDbRef; # Database, top layer
accPath: Hash32; # Even nibbled byte path
accRec: AristoAccount; # Account data
): Result[bool,AristoError] =
## Merge the key-value-pair argument `(accKey,accRec)` as an account
## ledger value, i.e. the the sub-tree starting at `VertexID(1)`.
##
## On success, the function returns `true` if the `accRec` argument was
## not on the database already or different from `accRec`, and `false`
## otherwise.
##
let
pyl = LeafPayload(pType: AccountData, account: accRec)
updated = db.mergePayloadImpl(
VertexID(1), accPath.data, db.cachedAccLeaf(accPath), pyl).valueOr:
if error == MergeNoAction:
return ok false
return err(error)
# Update leaf cache both of the merged value and potentially the displaced
# leaf resulting from splitting a leaf into a branch with two leaves
db.layersPutAccLeaf(accPath, updated[0])
if updated[1].isValid:
let otherPath = Hash32(getBytes(
NibblesBuf.fromBytes(accPath.data).replaceSuffix(updated[1].pfx)))
db.layersPutAccLeaf(otherPath, updated[2])
ok true
proc mergeStorageData*(
db: AristoDbRef; # Database, top layer
accPath: Hash32; # Needed for accounts payload
stoPath: Hash32; # Storage data path (aka key)
stoData: UInt256; # Storage data payload value
): Result[void,AristoError] =
## Store the `stoData` data argument on the storage area addressed by
## `(accPath,stoPath)` where `accPath` is the account key (into the MPT)
## and `stoPath` is the slot path of the corresponding storage area.
##
var accHike: Hike
db.fetchAccountHike(accPath,accHike).isOkOr:
return err(MergeStoAccMissing)
let
stoID = accHike.legs[^1].wp.vtx.lData.stoID
# Provide new storage ID when needed
useID =
if stoID.isValid: stoID # Use as is
elif stoID.vid.isValid: (true, stoID.vid) # Re-use previous vid
else: (true, db.vidFetch()) # Create new vid
mixPath = mixUp(accPath, stoPath)
# Call merge
pyl = LeafPayload(pType: StoData, stoData: stoData)
updated = db.mergePayloadImpl(
useID.vid, stoPath.data, db.cachedStoLeaf(mixPath), pyl).valueOr:
if error == MergeNoAction:
assert stoID.isValid # debugging only
return ok()
return err(error)
# Mark account path Merkle keys for update
db.layersResKeys(accHike)
# Update leaf cache both of the merged value and potentially the displaced
# leaf resulting from splitting a leaf into a branch with two leaves
db.layersPutStoLeaf(mixPath, updated[0])
if updated[1].isValid:
let otherPath = Hash32(getBytes(
NibblesBuf.fromBytes(stoPath.data).replaceSuffix(updated[1].pfx)))
db.layersPutStoLeaf(mixUp(accPath, otherPath), updated[2])
if not stoID.isValid:
# Make sure that there is an account that refers to that storage trie
let leaf = accHike.legs[^1].wp.vtx.dup # Dup on modify
leaf.lData.stoID = useID
db.layersPutAccLeaf(accPath, leaf)
db.layersPutVtx((VertexID(1), accHike.legs[^1].wp.vid), leaf)
ok()
# ------------------------------------------------------------------------------
# End
# ------------------------------------------------------------------------------