Performance optimizations to the NOMT storage engine while preserving
correctness (all triecompare cross-validation tests pass at 10K+ scale):
- Pool SHA256 hashers via sync.Pool in HashInternal and HashStem
- Replace allStems map with sorted slice + O(N+M) merge (in-place fast
path for incremental updates avoids allocation entirely)
- Add UpdateSorted to db.DB, skipping redundant sort of pre-sorted ops
- Simplify canonicalRoot to use pre-sorted allStems directly
- Optimize StemSharedBits with byte-level XOR + bits.LeadingZeros8
- Replace stemLess loops with bytes.Compare in all locations
- Eliminate per-stem map alloc in groupAndHashStems (use [256]bool dirty)
- Use stack-allocated [248]bool for downBits in BuildInternalTree
- Remove unused stemPathCmp function
BenchmarkHash/10000/nomt: 9.8ms → 8.2ms (-16%)
BenchmarkBlockWorkload/nomt: 7.7ms → 6.6ms (-14%)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add trie/triecompare/ package with realistic state generation and cross-
validation tests proving NOMT produces identical roots as bintrie at scale
(10K+ accounts, PowerLaw/Uniform/Exponential distributions, multi-block).
Fix a subtle bug in groupAndHashStems: sort.Slice was used instead of
sort.SliceStable, causing non-deterministic results when the same account
is mutated twice in a single block (duplicate stem+suffix entries need
last-writer-wins ordering preserved).
Tests: 5 correctness tests + 4 benchmarks + storage footprint comparison.
All pass with race detector clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Two root causes of hash mismatch fixed:
1. Canonical root computation: Hash() now uses BuildInternalTree(skip=0)
over all known stems instead of the page tree's depth-7 internal root.
The page tree is still updated for persistent storage, but the canonical
root bypasses its depth-7 worker split that added extra wrapping levels.
2. Code chunk grouping: UpdateContractCode now matches bintrie's group-based
key derivation exactly — computing the stem key only at group boundaries
and using groupOffset as the suffix, instead of computing a separate
GetBinaryTreeKey per chunk (which produced different stems).
Cross-validation tests (compat_test.go) assert strict equality:
- TestSingleAccountRootMatch
- TestMultiAccountRootMatch
- TestStorageRootMatch
- TestCodeChunkRootMatch
- TestMixedOpsRootMatch
- 4 BuildInternalTree vs bintrie diagnostic tests
48 tests passing, race-detector clean.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add stem-level flat state storage and the groupAndHashStems pipeline
that bridges NomtTrie's per-slot updates to NOMT's page-tree:
- stemValueDBKey/stemValueDBPrefix: ethdb key format (0x03||stem||suffix)
- loadStemValues: prefix-iteration loader for all 256 slots of a stem
- writeStemValues: batch writer with nil=delete semantics
- groupAndHashStems: sort→group→load→merge→hash→write pipeline
producing sorted []core.StemKeyValue for the page tree
Cross-validated against bintrie's StemNode.Hash algorithm.
11 new tests, all passing with -race.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add stem-aware key encoding wrappers delegating to bintrie for identical
SHA256 key derivation. Add packBasicData/packStorageValue matching
bintrie's exact encoding layout. Stub trie.go with stemUpdate type
pending Phase E implementation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Wire the NOMT binary merkle trie engine into geth's triedb/state
framework. This adds two new packages:
- triedb/nomtdb: backend implementing triedb.backend interface, manages
flat state persistence in ethdb and delegates trie ops to nomt/db
- trie/nomttrie: NomtTrie implementing state.Trie, accumulates LeafOps
during block execution and flushes to NOMT engine on Hash()/Commit()
Key design choices:
- Single flat keyspace: accounts use keccak256(addr), storage uses
keccak256(keccak256(addr) || keccak256(slot)) as 256-bit trie paths
- OpenStorageTrie returns the account trie itself (no separate tries)
- Flat state (account/storage values) stored in ethdb with prefixed keys
- NOMT trie stores only hashes; reads delegate to ethdb flat state
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
modify keyToPath
add comment
add credit
clean up
more clean up
moree clean up
chore
moreee clean up
delete more unused methods
# Conflicts:
# trie/bintrie/expired_node.go
# trie/bintrie/expired_node_test.go
GetStorage and DeleteStorage used GetBinaryTreeKey to compute the tree
key, while UpdateStorage used GetBinaryTreeKeyStorageSlot. The latter
applies storage slot remapping (header offset for slots <64, main
storage prefix for the rest), so reads and deletes were targeting
different tree locations than writes.
Replace GetBinaryTreeKey with GetBinaryTreeKeyStorageSlot in both
GetStorage and DeleteStorage to match UpdateStorage. Add a regression
test that verifies the write→read→delete→read round-trip for main
storage slots.
The `decodeRef` function used `size > hashLen` to reject oversized
embedded nodes, but this incorrectly allowed nodes of exactly 32 bytes
through. The encoding side (hasher.go, stacktrie.go) consistently uses
`len(enc) < 32` to decide whether to embed a node inline, meaning nodes
of 32+ bytes are always hash-referenced. The error message itself
already stated `want size < 32`, confirming the intended threshold.
Changed `size > hashLen` to `size >= hashLen` in `decodeRef` to align
the decoding validation with the encoding logic, the Yellow Paper spec,
and the surrounding comments.
The upstream libray has removed the assembly-based implementation of
keccak. We need to maintain our own library to avoid a peformance
regression.
---------
Co-authored-by: lightclient <lightclient@protonmail.com>
The `Witness` method was not implemented for the binary tree, which
caused `debug_excutionWitness` to panic. This PR fixes that.
Note that the `TransitionTrie` version isn't implemented, and that's on
purpose: more thought must be given to what should go in the global
witness.
Based on [EIP-7864](https://eips.ethereum.org/EIPS/eip-7864), the tree
index should be 32 bytes instead of 31 bytes.
```
def get_tree_key(address: Address32, tree_index: int, sub_index: int):
# Assumes STEM_SUBTREE_WIDTH = 256
return tree_hash(address + tree_index.to_bytes(32, "little"))[:31] + bytes(
[sub_index]
)
```
This PR optimizes memory allocation in StateTrie.PrefetchAccount() and
StateTrie.PrefetchStorage() by preallocating slice capacity when the
final size is known.
This pull request introduces a mechanism to compress trienode history by
storing only the node diffs between consecutive versions.
- For full nodes, only the modified children are recorded in the history;
- For short nodes, only the modified value is stored;
If the node type has changed, or if the node is newly created or
deleted, the entire node value is stored instead.
To mitigate the overhead of reassembling nodes from diffs during history
reads, checkpoints are introduced by periodically storing full node values.
The current checkpoint interval is set to every 16 mutations, though
this parameter may be made configurable in the future.
In order to reduce the amount of code that is embedded into the keeper
binary, I am removing all the verkle code that uses go-verkle and
go-ipa. This will be followed by further PRs that are more like stubs to
replace code when the keeper build is detected.
I'm keeping the binary tree of course. This means that you will still
see `isVerkle` variables all over the codebase, but they will be renamed
when code is touched (i.e. this is not an invitation for 30+ AI slop
PRs).
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This is broken off of #31730 to only focus on testing networks that
start with verkle at genesis.
The PR has seen a lot of work since its creation, and it now targets
creating and re-executing tests for a binary tree testnet without the
transition (so it starts at genesis). The transition tree has been moved
to its own package. It also replaces verkle with the binary tree for
this specific application.
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This change addresses critical issues in the state object duplication
process specific to Verkle trie implementations. Without these
modifications, updates to state objects fail to propagate correctly
through the trie structure after a statedb copy operation, leading to
inaccuracies in the computation of the state root hash.
---------
Co-authored-by: Guillaume Ballet <3272758+gballet@users.noreply.github.com>
- Correct the error message in TestOneElementProof to expect 'v' instead
of 'k'.
- The trie is updated with key "k" and value "v"; on mismatch the
expected value must be 'v'.
- Aligns the message with the actual test logic and other similar checks
in this file, reducing confusion during test failures. No behavioral
changes.
- Adds `NodeIteratorWithPrefix()` method to support iterating only nodes
within a specific key prefix
- Adds `NodeIteratorWithRange()` method to support iterating only nodes
within a specific key range
Current `NodeIterator` always traverses the entire remaining trie from a
start position. For non-ethereum applications using the trie implementation,
there's no way to limit iteration to just a subtree with a specific prefix.
**Usage:**
```go
// Only iterate nodes with prefix "key1"
iter, err := trie.NodeIteratorWithPrefix([]byte("key1"))
```
Testing: Comprehensive test suite covering edge cases and boundary conditions.
Closes#32484
---------
Co-authored-by: gballet <guillaume.ballet@gmail.com>
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This PR is the first step in the trienode history series.
It introduces the `nodeWithOrigin` struct in the path database, which tracks
the original values of dirty nodes to support trienode history construction.
Note, the original value is always empty in this PR, so it won't break the
existing journal for encoding and decoding. The compatibility of journal
should be handled in the following PR.
Implement the binary tree as specified in [eip-7864](https://eips.ethereum.org/EIPS/eip-7864).
This will gradually replace verkle trees in the codebase. This is only
running the tests and will not be executed in production, but will help
me rebase some of my work, so that it doesn't bitrot as much.
---------
Signed-off-by: Guillaume Ballet
Co-authored-by: Parithosh Jayanthi <parithosh.jayanthi@ethereum.org>
Co-authored-by: rjl493456442 <garyrong0905@gmail.com>
This pull introduces a `Prefetch` operation in the trie to prefetch trie
nodes in parallel. It is used by the `triePrefetcher` to accelerate state
loading and improve overall chain processing performance.
This add some of the changes that were missing from #31634. It
introduces the `TransitionTrie`, which is a façade pattern between the
current MPT trie and the overlay tree.
---------
Signed-off-by: Guillaume Ballet <3272758+gballet@users.noreply.github.com>
Co-authored-by: rjl493456442 <garyrong0905@gmail.com>
These changes made in the PR should be highlighted here
The trie tracer is split into two distinct structs: opTracer and prevalueTracer.
The former is specific to MPT, while the latter is generic and applicable to all
trie implementations.
The original values of dirty nodes are tracked in a NodeSet. This serves
as the foundation for both full archive node implementations and the state live
tracer.
This pull request optimizes trie hashing by reducing memory allocation
overhead. Specifically:
- define a fullNodeEncoder pool to reuse encoders and avoid memory
allocations.
- simplify the encoding logic for shortNode and fullNode by getting rid
of the Go interfaces.
The optimization tried to defer allocating the cache map until it was used for the
first time. It's a relic from earlier times, when tries were copied often. This seems
unnecessary now, so we can just create the map when the trie is created.
---------
Co-authored-by: Felix Lange <fjl@twurst.com>
As the preimage will only be stored if `t.preimages != nil`, so no need
to save them into local cache if not enabled. This will reduce the memory
wasted to copy the bytes
---------
Signed-off-by: jsvisa <delweng@gmail.com>
In this pull request, snapshot generation in pathdb has been ported from
the legacy state snapshot implementation. Additionally, when running in
path mode, legacy state snapshot data is now managed by the pathdb
based snapshot logic.
Note: Existing snapshot data will be re-generated, regardless of whether
it was previously fully constructed.
This PR creates a global hasher pool that can be used by all packages.
It also removes a bunch of the package local pools.
It also updates a few locations to use available hashers or the global
hashing pool to reduce allocations all over the codebase.
This change should reduce global allocation count by ~1%
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This pull request introduces a SyncKeyValue function to the
ethdb.KeyValueStore
interface, providing the ability to forcibly flush all previous writes
to disk.
This functionality is critical for go-ethereum, which internally uses
two independent
database engines: a key-value store (such as Pebble, LevelDB, or
memoryDB for
testing) and a flat-file–based freezer. To ensure write-order
consistency between
these engines, the key-value store must be explicitly synced before
writing to the
freezer and vice versa.
Fixes
- https://github.com/ethereum/go-ethereum/issues/31405
- https://github.com/ethereum/go-ethereum/issues/29819
This PR adds checking for an edgecase which theoretically can happen in
the range-prover. Right now, we check that a key does not overwrite a
previous one by checking that the key is increasing. However, if keys
are of different lengths, it is possible to create a key which is
increasing _and_ overwrites the previous key. Example: `0xaabbcc`
followed by `0xaabbccdd`.
This can not happen in go-ethereum, which always uses fixed-size paths
for accounts and storage slot paths in the trie, but it might happen if
the range prover is used without guaranteed fixed-size keys.
This PR also adds some testcases for the errors that are expected.
This pull request removes the node copy operation to reduce memory
allocation. Key Changes as below:
**(a) Use `decodeNodeUnsafe` for decoding nodes retrieved from the trie
node reader**
In the current implementation of the MPT, once a trie node blob is
retrieved, it is passed to `decodeNode` for decoding. However,
`decodeNode` assumes the supplied byte slice might be mutated later, so
it performs a deep copy internally before parsing the node.
Given that the node reader is implemented by the path database and the
hash database, both of which guarantee the immutability of the returned
byte slice. By restricting the node reader interface to explicitly
guarantee that the returned byte slice will not be modified, we can
safely replace `decodeNode` with `decodeNodeUnsafe`. This eliminates the
need for a redundant byte copy during each node resolution.
**(b) Modify the trie in place**
In the current implementation of the MPT, a copy of a trie node is
created before any modifications are made. These modifications include:
- Node resolution: Converting the value from a hash to the actual node.
- Node hashing: Tagging the hash into its cache.
- Node commit: Replacing the children with its hash.
- Structural changes: For example, adding a new child to a fullNode or
replacing a child of a shortNode.
This mechanism ensures that modifications only affect the live tree,
leaving all previously created copies unaffected.
Unfortunately, this property leads to a huge memory allocation
requirement. For example, if we want to modify the fullNode for n times,
the node will be copied for n times.
In this pull request, all the trie modifications are made in place. In
order to make sure all previously created copies are unaffected, the
`Copy` function now will deep-copy all the live nodes rather than the
root node itself.
With this change, while the `Copy` function becomes more expensive, it's
totally acceptable as it's not a frequently used one. For the normal
trie operations (Get, GetNode, Hash, Commit, Insert, Delete), the node
copy is not required anymore.
This PR removes the assumption of the stacktrie and trie to have the
same ordering. This was hit by the fuzzers on oss-fuzz
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This fixes an error where executing `evm run --dump ...` omits preimages
from the dump (because the statedb used for execution is a copy of
another instance).
