This PR changes the database access of the base part of filter rows that
are stored in groups of 32 adjacent maps for improved database storage
size and data access efficiency.
Before this grouped storage was introduced, filter rows were not cached
because the access pattern of either the index rendering or the search
does not really benefit from caching. Also no mutex was necessary for
filter row access. Storing adjacent rows in groups complicated the
situation as a search typically required reading all or most of adjacent
rows of a group, so in order to implement the single row read operation
without having to read the entire group up to 32 times, a cache for the
base row groups was added. This also introduced data race issues for
concurrenct read/write in the same group which was avoided by locking
the `indexLock` mutex. Unfortunately this also led to slowed down or
temporarily blocked search operations when indexing was in progress.
This PR returns to the original concept of uncached, no-mutex filter map
access by increasing read efficiency in a better way; similiarly to
write operations that already operate on groups of filter maps, now
`getFilterMapRow` is also replaced by `getFilterMapRows` that accepts a
single `rowIndex` and a list of `mapIndices`. It slightly complicates
`singleMatcherInstance.getMatchesForLayer` which now has to collect
groups of map indices accessed in the same row, but in exchange it
guarantees maximum read efficiency while avoiding read/write mutex
interference.
Note: a follow-up refactoring is WIP that further changes the database
access scheme by prodiving an immutable index view to the matcher, makes
the whole indexer more straightforward with no callbacks, and entirely
removes the concept of matcher syncing with `validBlocks` and the
resulting multiple retry logic in `eth/filters/filter.go`. This might
take a bit longer to finish though and in the meantime this change could
hopefully already solve the blocked request issues.
This implements a backing store for chain history based on era1 files.
The new store is integrated with the freezer. Queries for blocks and receipts
below the current freezer tail are handled by the era store.
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
Co-authored-by: Felix Lange <fjl@twurst.com>
Co-authored-by: lightclient <lightclient@protonmail.com>
Some tests involving transactions near the txMaxSize limit were flaky.
This was due to ECDSA signatures occasionally having leading zeros,
which are omitted during RLP encoding — making the final transaction
size 1 byte smaller than expected.
To address this, a new helper function pricedDataTransactionWithFixedSignature
was added. It ensures both r and s are exactly 32 bytes (i.e., no leading zeros),
producing transactions with deterministic size.
This PR implements eth/69. This protocol version drops the bloom filter
from receipts messages, reducing the amount of data needed for a sync
by ~530GB (2.3B txs * 256 byte) uncompressed. Compressed this will
be reduced to ~100GB
The new version also changes the Status message and introduces the
BlockRangeUpdate message to relay information about the available history
range.
---------
Co-authored-by: Felix Lange <fjl@twurst.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.
Adding values to the witness introduces a new class of issues for
computing gas: if there is not enough gas to cover adding an item to the
witness, then the item should not be added to the witness.
The problem happens when several items are added together, and that
process runs out of gas. The witness gas computation needs a way to
signal that not enough gas was provided. These values can not be
hardcoded, however, as they are context dependent, i.e. two calls to the
same function with the same parameters can give two different results.
The approach is to return both the gas that was actually consumed, and
the gas that was necessary. If the values don't match, then a witness
update OOG'd. The caller should then charge the `consumed` value
(remaining gas will be 0) and error out.
Why not return a boolean instead of the wanted value? Because when
several items are touched, we want to distinguish which item lacked gas.
---------
Signed-off-by: Guillaume Ballet <3272758+gballet@users.noreply.github.com>
This adds a metric called `chain/mgasps`, which records how many million
gas per second are being used during block insertion.
The value is calculated as `usedGas * 1000 / elapsed`, and it's updated
in the `insertStats.report` method. Also cleaned up the log output to
reuse the same value instead of recalculating it.
Useful for monitoring block processing throughput.
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This PR introduces an allocation-free version of the
Transaction.EffectiveGasTip method to improve performance by reducing
memory allocations.
## Changes
- Added a new `EffectiveGasTipInto` method that accepts a destination
parameter to avoid memory allocations
- Refactored the existing `EffectiveGasTip` method to use the new
allocation-free implementation
- Updated related methods (`EffectiveGasTipValue`, `EffectiveGasTipCmp`,
`EffectiveGasTipIntCmp`) to use the allocation-free approach
- Added tests and benchmarks to verify correctness and measure
performance improvements
## Motivation
In high-transaction-volume environments, the `EffectiveGasTip` method is
called frequently. Reducing memory allocations in this method decreases
garbage collection pressure and improves overall system performance.
## Benchmark Results
As-Is
BenchmarkEffectiveGasTip/Original-10 42089140 27.45 ns/op 8 B/op 1
allocs/op
To-Be
BenchmarkEffectiveGasTip/IntoMethod-10 72353263 16.73 ns/op 0 B/op 0
allocs/op
## Summary of Improvements
- **Performance**: ~39% faster execution (27.45 ns/op → 16.73 ns/op)
- **Memory**: Eliminated all allocations (8 B/op → 0 B/op)
- **Allocation count**: Reduced from 1 to 0 allocations per operation
This optimization follows the same pattern successfully applied to other
methods in the codebase, maintaining API compatibility while improving
performance.
## Safety & Compatibility
This optimization has no side effects or adverse impacts because:
- It maintains functional equivalence as confirmed by comprehensive
tests
- It preserves API compatibility with existing callers
- It follows clear memory ownership patterns with the destination
parameter
- It maintains thread safety by only modifying the caller-provided
destination parameter
This optimization follows the same pattern successfully applied to other
methods in the codebase, providing better performance without
compromising stability or correctness.
---------
Co-authored-by: lightclient <lightclient@protonmail.com>
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 enhances the block prefetcher by executing transactions
in parallel to warm the cache alongside the main block processor.
Unlike the original prefetcher, which only executes the next block and
is limited to chain syncing, the new implementation can be applied to any
block. This makes it useful not only during chain sync but also for regular
block insertion after the initial sync.
---------
Co-authored-by: Marius van der Wijden <m.vanderwijden@live.de>
This PR fixes an issue that could lead to data corruption.
Writing the state history may fail due to insufficient disk space or
other potential errors. With this change, the entire state insertion
will be aborted instead of silently ignoring the error.
Without this fix, state transitions would continue while the associated
state history is lost. After a restart, the resulting gap would be detected,
making recovery impossible.
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
updates the log entries in `core/filtermaps/indexer.go` to remove double
quotes around keys like "first block" and "last block", changing them to
`firstblock` and `lastblock`. This brings them in line with the general
logging style used across the codebase, where log keys are unquoted
single words.
For example, the log:
` INFO [...] "first block"=..., "last block"=...`
Is now rendered as:
` INFO [...] firstblock=..., lastblock=...`
This change improves readability and maintains consistency with logs
such as:
` INFO [...] number=2 sealhash=... uncles=0 txs=0 ...`
No functional behavior is changed — this is purely a formatting cleanup
for better developer experience.
Fixes https://github.com/ethereum/go-ethereum/issues/31732.
This logic was removed in the recent refactoring in the txindexer to
handle history cutoff (#31393). It was first introduced in this PR:
https://github.com/ethereum/go-ethereum/pull/28908.
I have tested it and it works as an alternative to #31745.
This PR packs 3 changes to the flow of fetching txs from the API:
- It caches the indexer tail after each run is over to avoid hitting the
db all the time as was done originally in #28908.
- Changes `backend.GetTransaction`. It doesn't return an error anymore
when tx indexer is in progress. It shifts the responsibility to the
caller to check the progress. The reason is that in most cases we anyway
check the txpool for the tx. If it was indeed a pending tx we can avoid
the indexer progress check.
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This PR fixes an initialization bug that in some cases caused the map
renderer to leave the last, partially rendered map as is and resume
rendering from the next map. At initialization we check whether the
existing rendered maps are consistent with the current chain view and
revert them if necessary. Until now this happened through an ugly hacky
solution, a "limited" chain view that was supposed to trigger a rollback
of some maps in the renderer logic if necessary. This whole setup worked
under assumptions that just weren't true any more. As a result it always
tried to revert the last map but also it did not shorten the indexed
range, only set `headIndexed` to false which indicated to the renderer
logic that the last map is fully populated (which it wasn't).
Now an explicit rollback of any unusable (reorged) maps happens at
startup, which also means that no hacky chain view is necessary, as soon
as the new `FilterMaps` is returned, the indexed range and view are
consistent with each other.
In the first commit an extra check is also added to `writeFinishedMaps`
so that if there is ever again a bug that would result in a gapped index
then it will not break the db with writing the incomplete data. Instead
it will return an indexing error which causes the indexer to revert to
unindexed mode and print an error log instantly. Hopefully this will not
ever happen in the future, but in order to test this safeguard check I
manually triggered the bug with only the first commit enabled, which
caused an indexing error as expected. With the second commit added (the
actual fix) the same operation succeeded without any issues.
Note that the database version is also bumped in this PR in order to
enforce a full reindexing as any existing database might be potentially
broken.
Fixes https://github.com/ethereum/go-ethereum/issues/31729
This PR fixes the out-of-range block number logic of `getBlockLvPointer`
which sometimes caused searches to fail if the head was updated in the
wrong moment. This logic ensures that querying the pointer of a future
block returns the pointer after the last fully indexed block (instead of
failing) and therefore an async range update will not cause the search
to fail. Earier this behaviour only worked when `headIndexed` was true
and `headDelimiter` pointed to the end of the indexed range. Now it also
works for an unfinished index.
This logic is also moved from `FilterMaps.getBlockLvPointer` to
`FilterMapsMatcherBackend.GetBlockLvPointer` because it is only required
by the search anyways. `FilterMaps.getBlockLvPointer` now only returns a
pointer for existing blocks, consistently with how it is used in the
indexer/renderer.
Note that this unhandled case has been present in the code for a long
time but went unnoticed because either one of two previously fixed bugs
did prevent it from being triggered; the incorrectly positive
`tempRange.headIndexed` (fixed in
https://github.com/ethereum/go-ethereum/pull/31680), though caused other
problems, prevented this one from being triggered as with a positive
`headIndexed` no database read was triggered in `getBlockLvPointer`.
Also, the unnecessary `indexLock` in `synced()` (fixed in
https://github.com/ethereum/go-ethereum/pull/31708) usually did prevent
the search seeing the temp range and therefore avoided noticeable
issues.
This changes the filtermaps to only pull up the raw receipts, not the
derived receipts which saves a lot of allocations.
During normal execution this will reduce the allocations of the whole
geth node by ~15%.
This PR applies the config overrides to the new config as well,
otherwise they will not be applied to defined configs, making
shadowforks impossible.
To test:
```
> ./build/bin/geth --override.prague 123 --dev --datadir /tmp/geth
INFO [04-28|21:20:47.009] - Prague: @123
> ./build/bin/geth --override.prague 321 --dev --datadir /tmp/geth
INFO [04-28|21:23:59.760] - Prague: @321
``
TruncatePending shows up bright red on our nodes, because it computes
the length of a map multiple times.
I don't know why this is so expensive, but around 20% of our time is
spent on this, which is super weird.
```
//PR: BenchmarkTruncatePending-24 17498 69397 ns/op 32872 B/op 3 allocs/op
//Master: BenchmarkTruncatePending-24 9960 123954 ns/op 32872 B/op 3 allocs/op
```
```
benchmark old ns/op new ns/op delta
BenchmarkTruncatePending-24 123954 69397 -44.01%
benchmark old allocs new allocs delta
BenchmarkTruncatePending-24 3 3 +0.00%
benchmark old bytes new bytes delta
BenchmarkTruncatePending-24 32872 32872 +0.00%
```
This simple PR is a 44% improvement over the old state
```
OUTINE ======================== github.com/ethereum/go-ethereum/core/txpool/legacypool.(*LegacyPool).truncatePending in github.com/ethereum/go-ethereum/core/txpool/legacypool/legacypool.go
1.96s 18.02s (flat, cum) 19.57% of Total
. . 1495:func (pool *LegacyPool) truncatePending() {
. . 1496: pending := uint64(0)
60ms 2.99s 1497: for _, list := range pool.pending {
250ms 5.48s 1498: pending += uint64(list.Len())
. . 1499: }
. . 1500: if pending <= pool.config.GlobalSlots {
. . 1501: return
. . 1502: }
. . 1503:
. . 1504: pendingBeforeCap := pending
. . 1505: // Assemble a spam order to penalize large transactors first
. 510ms 1506: spammers := prque.New[int64, common.Address](nil)
140ms 2.50s 1507: for addr, list := range pool.pending {
. . 1508: // Only evict transactions from high rollers
50ms 5.08s 1509: if uint64(list.Len()) > pool.config.AccountSlots {
. . 1510: spammers.Push(addr, int64(list.Len()))
. . 1511: }
. . 1512: }
. . 1513: // Gradually drop transactions from offenders
. . 1514: offenders := []common.Address{}
```
```go
// Benchmarks the speed of batch transaction insertion in case of multiple accounts.
func BenchmarkTruncatePending(b *testing.B) {
// Generate a batch of transactions to enqueue into the pool
pool, _ := setupPool()
defer pool.Close()
b.ReportAllocs()
batches := make(types.Transactions, 4096+1024+1)
for i := range len(batches) {
key, _ := crypto.GenerateKey()
account := crypto.PubkeyToAddress(key.PublicKey)
pool.currentState.AddBalance(account, uint256.NewInt(1000000), tracing.BalanceChangeUnspecified)
tx := transaction(uint64(0), 100000, key)
batches[i] = tx
}
for _, tx := range batches {
pool.addRemotesSync([]*types.Transaction{tx})
}
b.ResetTimer()
// benchmark truncating the pending
for range b.N {
pool.truncatePending()
}
}
```
This PR fixes a deadlock situation is deleteTailEpoch that might arise
when
range delete is running in iterator based fallback mode (either using
leveldb
database or the hashdb state storage scheme).
In this case a stopCb callback is called periodically that does check
events,
including matcher sync requests, in which case it tries to acquire
indexLock
for read access, while deleteTailEpoch already held it for write access.
This pull request removes the indexLock acquiring in
`FilterMapsMatcherBackend.synced`
as this function is only called in the indexLoop.
Fixes https://github.com/ethereum/go-ethereum/issues/31700
This PR adds the `AuthorizationList` field to the `CallMsg` interface to support `eth_call`
and `eth_estimateGas` of set-code transactions.
---------
Co-authored-by: Sina Mahmoodi <itz.s1na@gmail.com>
This PR ensures that caching a slice or a slice of slices will never
affect the original version by always cloning a slice fetched from cache
if it is not used in a guaranteed read only way.
This PR changes the chain view update mechanism of the log filter.
Previously the head updates were all wired through the indexer, even in
unindexed mode. This was both a bit weird and also unsafe as the
indexer's chain view was updates asynchronously with some delay, making
some log related tests flaky. Also, the reorg safety of the indexed
search was integrated with unindexed search in a weird way, relying on
`syncRange.ValidBlocks` in the unindexed case too, with a special
condition added to only consider the head of the valid range but not the
tail in the unindexed case.
In this PR the current chain view is directly accessible through the
filter backend and unindexed search is also chain view based, making it
inherently safe. The matcher sync mechanism is now only used for indexed
search as originally intended, removing a few ugly special conditions.
The PR is currently based on top of
https://github.com/ethereum/go-ethereum/pull/31642
Together they fix https://github.com/ethereum/go-ethereum/issues/31518
and replace https://github.com/ethereum/go-ethereum/pull/31542
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This PR makes `filtermaps.ChainView` thread safe because it is used
concurrently both by the indexer and multiple matcher threads. Even
though it represents an immutable view of the chain, adding a mutex lock
to the `blockHash` function is necessary because it does so by extending
its list of non-canonical hashes if the underlying blockchain is
changed.
The unsafe concurrency did cause a panic once after running the unit
tests for several hours and it could also happen during live operation.
This PR makes the conditions for using a map rendering snapshot stricter
so that whenever a reorg happens, only a snapshot of a common ancestor
block can be used. The issue fixed in
https://github.com/ethereum/go-ethereum/pull/31642 originated from using
a snapshot that wasn't a common ancestor. For example in the following
reorg scenario: `A->B`, then `A->B2`, then `A->B2->C2`, then `A->B->C`
the last reorg triggered a render from snapshot `B` saved earlier. Now
this is possible under certain conditions but extra care is needed, for
example if block `B` crosses a map boundary then it should not be
allowed. With the latest fix the checks are sufficient but I realized I
would just feel safer if we disallowed this rare and risky scenario
altogether and just render from snapshot `A` after the last reorg in the
example above. The performance difference if a few milliseconds and it
occurs rarely (about once a day on Holesky, probably much more rare on
Mainnet).
Note that this PR only makes the snapshot conditions stricter and
`TestIndexerRandomRange` does check that snapshots are still used
whenever it's obviously possible (adding blocks after the current head
without a reorg) so this change can be considered safe. Also I am
running the unit tests and the fuzzer and everything seems to be fine.
This pull request improves error handling for local transaction submissions.
Specifically, if a transaction fails with a temporary error but might be
accepted later, the error will not be returned to the user; instead, the
transaction will be tracked locally for resubmission.
However, if the transaction fails with a permanent error (e.g., invalid
transaction or insufficient balance), the error will be propagated to the user.
These errors returned in the legacyPool are regarded as temporary failure:
- `ErrOutOfOrderTxFromDelegated`
- `txpool.ErrInflightTxLimitReached`
- `ErrAuthorityReserved`
- `txpool.ErrUnderpriced`
- `ErrTxPoolOverflow`
- `ErrFutureReplacePending`
Notably, InsufficientBalance is also treated as a permanent error, as
it’s highly unlikely that users will transfer funds into the sender account
after submitting the transaction. Otherwise, users may be confused—seeing
their transaction submitted but unaware that the sender lacks sufficient funds—and
continue waiting for it to be included.
---------
Co-authored-by: lightclient <lightclient@protonmail.com>
closes#31310
This has been requested a few times in the past and I think it is a nice
quality-of-life improvement for users. At a predetermined interval,
there will now be a "Fork ready" log when a future fork is scheduled,
but not yet active.
It can only possibly print after block import, which kinda avoids the
scenario where the client isn't progressing or is syncing and the user
thinks it's "ready" because it sees a ready log.
New output:
```console
INFO [03-08|21:32:57.472] Imported new potential chain segment number=7 hash=aa24ee..f09e62 blocks=1 txs=0 mgas=0.000 elapsed="874.916µs" mgasps=0.000 snapdiffs=973.00B triediffs=7.05KiB triedirty=0.00B
INFO [03-08|21:32:57.473] Ready for fork activation fork=Prague date="18 Mar 25 19:29 CET" remaining=237h57m0s timestamp=1,742,322,597
INFO [03-08|21:32:57.475] Chain head was updated number=7 hash=aa24ee..f09e62 root=19b0de..8d32f2 elapsed="129.125µs"
```
Easiest way to verify this behavior is to apply this patch and run `geth
--dev --dev.period=12`
```patch
diff --git a/params/config.go b/params/config.go
index 9c7719d901..030c4f80e7 100644
--- a/params/config.go
+++ b/params/config.go
@@ -174,7 +174,7 @@ var (
ShanghaiTime: newUint64(0),
CancunTime: newUint64(0),
TerminalTotalDifficulty: big.NewInt(0),
- PragueTime: newUint64(0),
+ PragueTime: newUint64(uint64(time.Now().Add(time.Hour * 300).Unix())),
BlobScheduleConfig: &BlobScheduleConfig{
Cancun: DefaultCancunBlobConfig,
Prague: DefaultPragueBlobConfig,
```
This is an attempt at fixing #31601. I think what happens is the startup
logic will try to get the full block body (it's `bc.loadLastState`) and
fail because genesis block has been pruned from the freezer. This will
cause it to keep repeating the reset logic, causing a deadlock.
This can happen when due to an unsuccessful sync we don't have the state
for the head (or any other state) fully, and try to redo the snap sync.
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This fixes an issue where running geth with `--history.chain postmerge`
would not work on an empty database.
```
ERROR[04-16|23:11:12.913] Chain history database is pruned to unknown block tail=0
Fatal: Failed to register the Ethereum service: unexpected database tail
```
This PR fixes a bug in the map renderer that sometimes used an obsolete
block log value pointer to initialize the iterator for rendering from a
snapshot. This bug was triggered by chain reorgs and sometimes caused
indexing errors and invalid search results. A few other conditions are
also made safer that were not reported to cause issues yet but could
potentially be unsafe in some corner cases. A new unit test is also
added that reproduced the bug but passes with the new fixes.
Fixes https://github.com/ethereum/go-ethereum/issues/31593
Might also fix https://github.com/ethereum/go-ethereum/issues/31589
though this issue has not been reproduced yet, but it appears to be
related to a log index database corruption around a specific block,
similarly to the other issue.
Note that running this branch resets and regenerates the log index
database. For this purpose a `Version` field has been added to
`rawdb.FilterMapsRange` which will also make this easier in the future
if a breaking database change is needed or the existing one is
considered potentially broken due to a bug, like in this case.
