This is something interesting I came across during my benchmarks, we
spent ~3.8% of all allocations allocating the header number on the heap.
```
(pprof) list GetHeaderByHash
Total: 38197204475
ROUTINE ======================== github.com/ethereum/go-ethereum/core.(*BlockChain).GetHeaderByHash in github.com/ethereum/go-ethereum/core/blockchain_reader.go
0 5786566117 (flat, cum) 15.15% of Total
. . 79:func (bc *BlockChain) GetHeaderByHash(hash common.Hash) *types.Header {
. 5786566117 80: return bc.hc.GetHeaderByHash(hash)
. . 81:}
. . 82:
. . 83:// GetHeaderByNumber retrieves a block header from the database by number,
. . 84:// caching it (associated with its hash) if found.
. . 85:func (bc *BlockChain) GetHeaderByNumber(number uint64) *types.Header {
ROUTINE ======================== github.com/ethereum/go-ethereum/core.(*HeaderChain).GetHeaderByHash in github.com/ethereum/go-ethereum/core/headerchain.go
0 5786566117 (flat, cum) 15.15% of Total
. . 404:func (hc *HeaderChain) GetHeaderByHash(hash common.Hash) *types.Header {
. 1471264309 405: number := hc.GetBlockNumber(hash)
. . 406: if number == nil {
. . 407: return nil
. . 408: }
. 4315301808 409: return hc.GetHeader(hash, *number)
. . 410:}
. . 411:
. . 412:// HasHeader checks if a block header is present in the database or not.
. . 413:// In theory, if header is present in the database, all relative components
. . 414:// like td and hash->number should be present too.
(pprof) list GetBlockNumber
Total: 38197204475
ROUTINE ======================== github.com/ethereum/go-ethereum/core.(*HeaderChain).GetBlockNumber in github.com/ethereum/go-ethereum/core/headerchain.go
94438817 1471264309 (flat, cum) 3.85% of Total
. . 100:func (hc *HeaderChain) GetBlockNumber(hash common.Hash) *uint64 {
94438817 94438817 101: if cached, ok := hc.numberCache.Get(hash); ok {
. . 102: return &cached
. . 103: }
. 1376270828 104: number := rawdb.ReadHeaderNumber(hc.chainDb, hash)
. . 105: if number != nil {
. 554664 106: hc.numberCache.Add(hash, *number)
. . 107: }
. . 108: return number
. . 109:}
. . 110:
. . 111:type headerWriteResult struct {
(pprof) list ReadHeaderNumber
Total: 38197204475
ROUTINE ======================== github.com/ethereum/go-ethereum/core/rawdb.ReadHeaderNumber in github.com/ethereum/go-ethereum/core/rawdb/accessors_chain.go
204606513 1376270828 (flat, cum) 3.60% of Total
. . 146:func ReadHeaderNumber(db ethdb.KeyValueReader, hash common.Hash) *uint64 {
109577863 1281242178 147: data, _ := db.Get(headerNumberKey(hash))
. . 148: if len(data) != 8 {
. . 149: return nil
. . 150: }
95028650 95028650 151: number := binary.BigEndian.Uint64(data)
. . 152: return &number
. . 153:}
. . 154:
. . 155:// WriteHeaderNumber stores the hash->number mapping.
. . 156:func WriteHeaderNumber(db ethdb.KeyValueWriter, hash common.Hash, number uint64) {
```
Opening this to discuss the idea, I know that rawdb.EmptyNumber is not a
great name for the variable, open to suggestions
This pull request slightly improves the freezer fsync mechanism by scheduling
the Sync operation based on the number of uncommitted items and original
time interval.
Originally, freezer.Sync was triggered every 30 seconds, which worked well during
active chain synchronization. However, once the initial state sync is complete,
the fixed interval causes Sync to be scheduled too frequently.
To address this, the scheduling logic has been improved to consider both the time
interval and the number of uncommitted items. This additional condition helps
avoid unnecessary Sync operations when the chain is idle.
Introduce file-based state journal in path database, fixing
the Pebble restriction when the journal size exceeds 4GB.
---------
Signed-off-by: jsvisa <delweng@gmail.com>
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This pull request refines the filtermap implementation, defining key
APIs for map and
epoch calculations to improve readability.
This pull request doesn't change any logic, it's a pure cleanup.
---------
Co-authored-by: zsfelfoldi <zsfelfoldi@gmail.com>
Reading a single transaction out of a block shouldn't need decoding the
entire body
---------
Co-authored-by: Felix Lange <fjl@twurst.com>
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
As https://github.com/ethereum/go-ethereum/pull/31769 defined a global
hash pool, so we can reuse it, and also remove the unnecessary
KeccakState buffering
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
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>
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>
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 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
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.
This pull request introduces new sync logic for pruning mode. The downloader will now skip
insertion of block bodies and receipts before the configured history cutoff point.
Originally, in snap sync, the header chain and other components (bodies and receipts) were
inserted separately. However, in Proof-of-Stake, this separation is unnecessary since the
sync target is already verified by the CL.
To simplify the process, this pull request modifies `InsertReceiptChain` to insert headers
along with block bodies and receipts together. Besides, `InsertReceiptChain` doesn't have
the notion of reorg, as the common ancestor is always be found before the sync and extra
side chain is truncated at the beginning if they fall in the ancient store. The stale
canonical chain flags will always be rewritten by the new chain. Explicit reorg logic is
no longer required in `InsertReceiptChain`.
This PR adds `rawdb.SafeDeleteRange` and uses it for range deletion in
`core/filtermaps`. This includes deleting the old bloombits database,
resetting the log index database and removing index data for unindexed
tail epochs (which previously weren't properly implemented for the
fallback case).
`SafeDeleteRange` either calls `ethdb.DeleteRange` if the node uses the
new path based state scheme or uses an iterator based fallback method
that safely skips trie nodes in the range if the old hash based state
scheme is used. Note that `ethdb.DeleteRange` also has its own iterator
based fallback implementation in `ethdb/leveldb`. If a path based state
scheme is used and the backing db is pebble (as it is on the majority of
new nodes) then `rawdb.SafeDeleteRange` uses the fast native range
delete.
Also note that `rawdb.SafeDeleteRange` has different semantics from
`ethdb.DeleteRange`, it does not automatically return if the operation
takes a long time. Instead it receives a `stopCallback` that can
interrupt the process if necessary. This is because in the safe mode
potentially a lot of entries are iterated without being deleted (this is
definitely the case when deleting the old bloombits database which has a
single byte prefix) and therefore restarting the process every time a
fixed number of entries have been iterated would result in a quadratic
run time in the number of skipped entries.
When running in safe mode, unindexing an epoch takes about a second,
removing bloombits takes around 10s while resetting a full log index
might take a few minutes. If a range delete operation takes a
significant amount of time then log messages are printed. Also, any
range delete operation can be interrupted by shutdown (tail uinindexing
can also be interrupted by head indexing, similarly to how tail indexing
works). If the last unindexed epoch might have "dirty" index data left
then the indexed map range points to the first valid epoch and
`cleanedEpochsBefore` points to the previous, potentially dirty one. At
startup it is always assumed that the epoch before the first fully
indexed one might be dirty. New tail maps are never rendered and also no
further maps are unindexed before the previous unindexing is properly
cleaned up.
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
Co-authored-by: Felix Lange <fjl@twurst.com>
Instead of reporting all filtermaps stuff in one line, I'm breaking it
down into the three separate kinds of entries here.
```
+-----------------------+-----------------------------+------------+------------+
| DATABASE | CATEGORY | SIZE | ITEMS |
+-----------------------+-----------------------------+------------+------------+
| Key-Value store | Log index filter-map rows | 59.21 GiB | 616077345 |
| Key-Value store | Log index last-block-of-map | 12.35 MiB | 269755 |
| Key-Value store | Log index block-lv | 421.70 MiB | 22109169 |
```
Also added some other changes to make it easier to debug:
- restored bloombits into the inspect output, so we notice if it doesn't
get deleted for some reason
- tracking of unaccounted key examples
This adds a new subcommand 'geth prune-history' that removes the pre-merge history
on supported networks. Geth is not fully ready to work in this mode, please do not run
this command on your production node.
---------
Co-authored-by: Felix Lange <fjl@twurst.com>
In #31384 we unindex TXes prior to the merge block. However when the
node starts up it will try to re-index those back if the config is to index the
whole chain. This change makes the indexer aware of the history cutoff block,
avoiding reindexing in that segment.
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
Co-authored-by: Felix Lange <fjl@twurst.com>
Here we add the notion of prunable tables for the `TruncateTail` operation
in the freezer. TruncateTail for the chain freezer now only truncates the body and
receipts tables, leaving headers and hashes as-is.
This change also requires changing the validation/repair at startup to allow for
tables with different tail. For the header and hash tables, we now require them to start
at number zero.
---------
Co-authored-by: Felix Lange <fjl@twurst.com>
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
1. The metric of preimage/hits are always the same as preimage/total, prefer to replace
the hits with miss instead.
2. For the state/read/accounts metric, follow the same naming of others,
change into singuar.
Currently, when calculating block's bloom, we loop through all the
receipt logs to calculate the hash value. However, normally, after going
through applyTransaction, the receipt's bloom is already calculated
based on the receipt log, so the block's bloom can be calculated by just
ORing these receipt's blooms.
```
goos: darwin
goarch: arm64
pkg: github.com/ethereum/go-ethereum/core/types
cpu: Apple M1 Pro
BenchmarkCreateBloom
BenchmarkCreateBloom/small
BenchmarkCreateBloom/small-10 810922 1481 ns/op 104 B/op 5 allocs/op
BenchmarkCreateBloom/large
BenchmarkCreateBloom/large-10 8173 143764 ns/op 9614 B/op 401 allocs/op
BenchmarkCreateBloom/small-mergebloom
BenchmarkCreateBloom/small-mergebloom-10 5178918 232.0 ns/op 0 B/op 0 allocs/op
BenchmarkCreateBloom/large-mergebloom
BenchmarkCreateBloom/large-mergebloom-10 54110 22207 ns/op 0 B/op 0 allocs/op
```
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
Co-authored-by: Zsolt Felfoldi <zsfelfoldi@gmail.com>
This PR addresses a flaw in the freezer table upgrade path.
In v1.15.0, freezer table v2 was introduced, including an additional
field (`flushOffset`) maintained in the metadata file. To ensure
backward compatibility, an upgrade path was implemented for legacy
freezer tables by setting `flushOffset` to the size of the index file.
However, if the freezer table is opened in read-only mode, this file
write operation is rejected, causing Geth to shut down entirely.
Given that invalid items in the freezer index file can be detected and
truncated, all items in freezer v0 index files are guaranteed to be
complete. Therefore, when operating in read-only mode, it is safe to
use the freezer data without performing an upgrade.
This is a follow-up PR to #29792 to get rid of the data file sync.
**This is a non-backward compatible change, which increments the
database version from 8 to 9**.
We introduce a flushOffset for each freezer table, which tracks the position
of the most recently fsync’d item in the index file. When this offset moves
forward, it indicates that all index entries below it, along with their corresponding
data items, have been properly persisted to disk. The offset can also be moved
backward when truncating from either the head or tail of the file.
Previously, the data file required an explicit fsync after every mutation, which
was highly inefficient. With the introduction of the flush offset, the synchronization
strategy becomes more flexible, allowing the freezer to sync every 30 seconds
instead.
The data items above the flush offset are regarded volatile and callers must ensure
they are recoverable after the unclean shutdown, or explicitly sync the freezer
before any proceeding operations.
---------
Co-authored-by: Felix Lange <fjl@twurst.com>
The total difficulty is the sum of all block difficulties from genesis
to a certain block. This value was used in PoW for deciding which chain
is heavier, and thus which chain to select. Since PoS has a different
fork selection algorithm, all blocks since the merge have a difficulty
of 0, and all total difficulties are the same for the past 2 years.
Whilst the TDs are mostly useless nowadays, there was never really a
reason to mess around removing them since they are so tiny. This
reasoning changes when we go down the path of pruned chain history. In
order to reconstruct any TD, we **must** retrieve all the headers from
chain head to genesis and then iterate all the difficulties to compute
the TD.
In a world where we completely prune past chain segments (bodies,
receipts, headers), it is not possible to reconstruct the TD at all. In
a world where we still keep chain headers and prune only the rest,
reconstructing it possible as long as we process (or download) the chain
forward from genesis, but trying to snap sync the head first and
backfill later hits the same issue, the TD becomes impossible to
calculate until genesis is backfilled.
All in all, the TD is a messy out-of-state, out-of-consensus computed
field that is overall useless nowadays, but code relying on it forces
the client into certain modes of operation and prevents other modes or
other optimizations. This PR completely nukes out the TD from the node.
It doesn't compute it, it doesn't operate on it, it's as if it didn't
even exist.
Caveats:
- Whenever we have APIs that return TD (devp2p handshake, tracer, etc.)
we return a TD of 0.
- For era files, we recompute the TD during export time (fairly quick)
to retain the format content.
- It is not possible to "verify" the merge point (i.e. with TD gone, TTD
is useless). Since we're not verifying PoW any more, just blindly trust
it, not verifying but blindly trusting the many year old merge point
seems just the same trust model.
- Our tests still need to be able to generate pre and post merge blocks,
so they need a new way to split the merge without TTD. The PR introduces
a settable ttdBlock field on the consensus object which is used by tests
as the block where originally the TTD happened. This is not needed for
live nodes, we never want to generate old blocks.
- One merge transition consensus test was disabled. With a
non-operational TD, testing how the client reacts to TTD is useless, it
cannot react.
Questions:
- Should we also drop total terminal difficulty from the genesis json?
It's a number we cannot react on any more, so maybe it would be cleaner
to get rid of even more concepts.
---------
Co-authored-by: Gary Rong <garyrong0905@gmail.com>
This PR modifies how the metrics library handles `Enabled`: previously,
the package `init` decided whether to serve real metrics or just
dummy-types.
This has several drawbacks:
- During pkg init, we need to determine whether metrics are enabled or
not. So we first hacked in a check if certain geth-specific
commandline-flags were enabled. Then we added a similar check for
geth-env-vars. Then we almost added a very elaborate check for
toml-config-file, plus toml parsing.
- Using "real" types and dummy types interchangeably means that
everything is hidden behind interfaces. This has a performance penalty,
and also it just adds a lot of code.
This PR removes the interface stuff, uses concrete types, and allows for
the setting of Enabled to happen later. It is still assumed that
`metrics.Enable()` is invoked early on.
The somewhat 'heavy' operations, such as ticking meters and exp-decay,
now checks the enable-flag to prevent resource leak.
The change may be large, but it's mostly pretty trivial, and from the
last time I gutted the metrics, I ensured that we have fairly good test
coverage.
---------
Co-authored-by: Felix Lange <fjl@twurst.com>
This PR fixes some issues with benchmarks
- [x] Removes log output from a log-test
- [x] Avoids a `nil`-defer in `triedb/pathdb`
- [x] Fixes some crashes re tracers
- [x] Refactors a very resource-expensive benchmark for blobpol.
**NOTE**: this rewrite touches live production code (a little bit), as
it makes the validator-function used by the blobpool configurable.
- [x] Switch some benches over to use pebble over leveldb
- [x] reduce mem overhead in the setup-phase of some tests
- [x] Marks some tests with a long setup-phase to be skipped if `-short`
is specified (where long is on the order of tens of seconds). Ideally,
in my opinion, one should be able to run with `-benchtime 10ms -short`
and sanity-check all tests very quickly.
- [x] Drops some metrics-bechmark which times the speed of `copy`.
---------
Co-authored-by: Sina Mahmoodi <itz.s1na@gmail.com>
This PR adds `DeleteRange` to `ethdb.KeyValueWriter`. While range
deletion using an iterator can be really slow, `DeleteRange` is natively
supported by pebble and apparently runs in O(1) time (typically 20-30ms
in my tests for removing hundreds of millions of keys and gigabytes of
data). For leveldb and memorydb an iterator based fallback is
implemented. Note that since the iterator method can be slow and a
database function should not unexpectedly block for a very long time,
the number of deleted keys is limited at 10000 which should ensure that
it does not block for more than a second. ErrTooManyKeys is returned if
the range has only been partially deleted. In this case the caller can
repeat the call until it finally succeeds.
Changelog: https://golangci-lint.run/product/changelog/#1610
Removes `exportloopref` (no longer needed), replaces it with
`copyloopvar` which is basically the opposite.
Also adds:
- `durationcheck`
- `gocheckcompilerdirectives`
- `reassign`
- `mirror`
- `tenv`
---------
Co-authored-by: Marius van der Wijden <m.vanderwijden@live.de>
This pull request removes the `fsync` of index files in freezer.ModifyAncients function for
performance gain.
Originally, fsync is added after each freezer write operation to ensure
the written data is truly transferred into disk. Unfortunately, it turns
out `fsync` can be relatively slow, especially on
macOS (see https://github.com/ethereum/go-ethereum/issues/28754 for more
information).
In this pull request, fsync for index file is removed as it turns out
index file can be recovered even after a unclean shutdown. But fsync for data file is still kept, as
we have no meaningful way to validate the data correctness after unclean shutdown.
---
**But why do we need the `fsync` in the first place?**
As it's necessary for freezer to survive/recover after the machine crash
(e.g. power failure).
In linux, whenever the file write is performed, the file metadata update
and data update are
not necessarily performed at the same time. Typically, the metadata will
be flushed/journalled
ahead of the file data. Therefore, we make the pessimistic assumption
that the file is first
extended with invalid "garbage" data (normally zero bytes) and that
afterwards the correct
data replaces the garbage.
We have observed that the index file of the freezer often contain
garbage entry with zero value
(filenumber = 0, offset = 0) after a machine power failure. It proves
that the index file is extended
without the data being flushed. And this corruption can destroy the
whole freezer data eventually.
Performing fsync after each write operation can reduce the time window
for data to be transferred
to the disk and ensure the correctness of the data in the disk to the
greatest extent.
---
**How can we maintain this guarantee without relying on fsync?**
Because the items in the index file are strictly in order, we can
leverage this characteristic to
detect the corruption and truncate them when freezer is opened.
Specifically these validation
rules are performed for each index file:
For two consecutive index items:
- If their file numbers are the same, then the offset of the latter one
MUST not be less than that of the former.
- If the file number of the latter one is equal to that of the former
plus one, then the offset of the latter one MUST not be 0.
- If their file numbers are not equal, and the latter's file number is
not equal to the former plus 1, the latter one is valid
And also, for the first non-head item, it must refer to the earliest
data file, or the next file if the
earliest file is not sufficient to place the first item(very special
case, only theoretical possible
in tests)
With these validation rules, we can detect the invalid item in index
file with greatest possibility.
---
But unfortunately, these scenarios are not covered and could still lead
to a freezer corruption if it occurs:
**All items in index file are in zero value**
It's impossible to distinguish if they are truly zero (e.g. all the data
entries maintained in freezer
are zero size) or just the garbage left by OS. In this case, these index
items will be kept by truncating
the entire data file, namely the freezer is corrupted.
However, we can consider that the probability of this situation
occurring is quite low, and even
if it occurs, the freezer can be considered to be close to an empty
state. Rerun the state sync
should be acceptable.
**Index file is integral while relative data file is corrupted**
It might be possible the data file is corrupted whose file size is
extended correctly with garbage
filled (e.g. zero bytes). In this case, it's impossible to detect the
corruption by index validation.
We can either choose to `fsync` the data file, or blindly believe that
if index file is integral then
the data file could be integral with very high chance. In this pull
request, the first option is taken.
