* focus on performance improvement in many aspects.
1. Do BlockBody verification concurrently;
2. Do calculation of intermediate root concurrently;
3. Preload accounts before processing blocks;
4. Make the snapshot layers configurable.
5. Reuse some object to reduce GC.
add
* rlp: improve decoder stream implementation (#22858)
This commit makes various cleanup changes to rlp.Stream.
* rlp: shrink Stream struct
This removes a lot of unused padding space in Stream by reordering the
fields. The size of Stream changes from 120 bytes to 88 bytes. Stream
instances are internally cached and reused using sync.Pool, so this does
not improve performance.
* rlp: simplify list stack
The list stack kept track of the size of the current list context as
well as the current offset into it. The size had to be stored in the
stack in order to subtract it from the remaining bytes of any enclosing
list in ListEnd. It seems that this can be implemented in a simpler
way: just subtract the size from the enclosing list context in List instead.
* rlp: use atomic.Value for type cache (#22902)
All encoding/decoding operations read the type cache to find the
writer/decoder function responsible for a type. When analyzing CPU
profiles of geth during sync, I found that the use of sync.RWMutex in
cache lookups appears in the profiles. It seems we are running into
CPU cache contention problems when package rlp is heavily used
on all CPU cores during sync.
This change makes it use atomic.Value + a writer lock instead of
sync.RWMutex. In the common case where the typeinfo entry is present in
the cache, we simply fetch the map and lookup the type.
* rlp: optimize byte array handling (#22924)
This change improves the performance of encoding/decoding [N]byte.
name old time/op new time/op delta
DecodeByteArrayStruct-8 336ns ± 0% 246ns ± 0% -26.98% (p=0.000 n=9+10)
EncodeByteArrayStruct-8 225ns ± 1% 148ns ± 1% -34.12% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
DecodeByteArrayStruct-8 120B ± 0% 48B ± 0% -60.00% (p=0.000 n=10+10)
EncodeByteArrayStruct-8 0.00B 0.00B ~ (all equal)
* rlp: optimize big.Int decoding for size <= 32 bytes (#22927)
This change grows the static integer buffer in Stream to 32 bytes,
making it possible to decode 256bit integers without allocating a
temporary buffer.
In the recent commit 088da24, Stream struct size decreased from 120
bytes down to 88 bytes. This commit grows the struct to 112 bytes again,
but the size change will not degrade performance because Stream
instances are internally cached in sync.Pool.
name old time/op new time/op delta
DecodeBigInts-8 12.2µs ± 0% 8.6µs ± 4% -29.58% (p=0.000 n=9+10)
name old speed new speed delta
DecodeBigInts-8 230MB/s ± 0% 326MB/s ± 4% +42.04% (p=0.000 n=9+10)
* eth/protocols/eth, les: avoid Raw() when decoding HashOrNumber (#22841)
Getting the raw value is not necessary to decode this type, and
decoding it directly from the stream is faster.
* fix testcase
* debug no lazy
* fix can not repair
* address comments
Co-authored-by: Felix Lange <fjl@twurst.com>
This commit splits the eth package, separating the handling of eth and snap protocols. It also includes the capability to run snap sync (https://github.com/ethereum/devp2p/blob/master/caps/snap.md) , but does not enable it by default.
Co-authored-by: Marius van der Wijden <m.vanderwijden@live.de>
Co-authored-by: Martin Holst Swende <martin@swende.se>
* trie: update tests to check commit integrity
* trie: polish committer
* trie: fix typo
* trie: remove hasvalue notion
According to the benchmarks, type assertion between the pointer and
interface is extremely fast.
BenchmarkIntmethod-12 1000000000 1.91 ns/op
BenchmarkInterface-12 1000000000 2.13 ns/op
BenchmarkTypeSwitch-12 1000000000 1.81 ns/op
BenchmarkTypeAssertion-12 2000000000 1.78 ns/op
So the overhead for asserting whether the shortnode has "valuenode"
child is super tiny. No necessary to have another field.
* trie: linter nitpicks
Co-authored-by: Martin Holst Swende <martin@swende.se>
The current trie memory database/cache that we do pruning on stores
trie nodes as binary rlp encoded blobs, and also stores the node
relationships/references for GC purposes. However, most of the trie
nodes (everything apart from a value node) is in essence just a
collection of references.
This PR switches out the RLP encoded trie blobs with the
collapsed-but-not-serialized trie nodes. This permits most of the
references to be recovered from within the node data structure,
avoiding the need to track them a second time (expensive memory wise).
* trie: make fullnode children hash calculation concurrently
* trie: thread out only on topmost fullnode
* trie: clean up full node children hash calculation
* trie: minor code fixups
* ethdb: add Putter interface and Has method
* ethdb: improve docs and add IdealBatchSize
* ethdb: remove memory batch lock
Batches are not safe for concurrent use.
* core: use ethdb.Putter for Write* functions
This covers the easy cases.
* core/state: simplify StateSync
* trie: optimize local node check
* ethdb: add ValueSize to Batch
* core: optimize HasHeader check
This avoids one random database read get the block number. For many uses
of HasHeader, the expectation is that it's actually there. Using Has
avoids a load + decode of the value.
* core: write fast sync block data in batches
Collect writes into batches up to the ideal size instead of issuing many
small, concurrent writes.
* eth/downloader: commit larger state batches
Collect nodes into a batch up to the ideal size instead of committing
whenever a node is received.
* core: optimize HasBlock check
This avoids a random database read to get the number.
* core: use numberCache in HasHeader
numberCache has higher capacity, increasing the odds of finding the
header without a database lookup.
* core: write imported block data using a batch
Restore batch writes of state and add blocks, tx entries, receipts to
the same batch. The change also simplifies the miner.
This commit also removes posting of logs when a forked block is imported.
* core: fix DB write error handling
* ethdb: use RLock for Has
* core: fix HasBlock comment
* ethdb: remove Set
Set deadlocks immediately and isn't part of the Database interface.
* trie: add Err to Iterator
This is useful for testing because the underlying NodeIterator doesn't
need to be kept in a separate variable just to get the error.
* trie: add LeafKey to iterator, panic when not at leaf
LeafKey is useful for callers that can't interpret Path.
* trie: retry failed seek/peek in iterator Next
Instead of failing iteration irrecoverably, make it so Next retries the
pending seek or peek every time.
Smaller changes in this commit make this easier to test:
* The iterator previously returned from Next on encountering a hash
node. This caused it to visit the same path twice.
* Path returned nibbles with terminator symbol for valueNode attached
to fullNode, but removed it for valueNode attached to shortNode. Now
the terminator is always present. This makes Path unique to each node
and simplifies Leaf.
* trie: add Path to MissingNodeError
The light client trie iterator needs to know the path of the node that's
missing so it can retrieve a proof for it. NodeIterator.Path is not
sufficient because it is updated when the node is resolved and actually
visited by the iterator.
Also remove unused fields. They were added a long time ago before we
knew which fields would be needed for the light client.
The 'step' method is split into two parts, 'peek' and 'push'. peek
returns the next state but doesn't make it current.
The end of iteration was previously tracked by setting 'trie' to nil.
End of iteration is now tracked using the 'iteratorEnd' error, which is
slightly cleaner and requires less code.
Make it so each iterator has exactly one public constructor:
- NodeIterators can be created through a method.
- Iterators can be created through NewIterator on any NodeIterator.
The key was constructed from nibbles, which isn't possible for all
nodes. Remove the only use of Key in LightTrie by always retrying with
the original key that was looked up.
Commit 40cdcf1183 broke the optimisation which kept nodes resolved
during Get in the trie. The decoder assigned cache generation 0
unconditionally, causing resolved nodes to get flushed on Commit.
This commit fixes it and adds two tests.
* trie: store nodes as pointers
This avoids memory copies when unwrapping node interface values.
name old time/op new time/op delta
Get 388ns ± 8% 215ns ± 2% -44.56% (p=0.000 n=15+15)
GetDB 363ns ± 3% 202ns ± 2% -44.21% (p=0.000 n=15+15)
UpdateBE 1.57µs ± 2% 1.29µs ± 3% -17.80% (p=0.000 n=13+15)
UpdateLE 1.92µs ± 2% 1.61µs ± 2% -16.25% (p=0.000 n=14+14)
HashBE 2.16µs ± 6% 2.18µs ± 6% ~ (p=0.436 n=15+15)
HashLE 7.43µs ± 3% 7.21µs ± 3% -2.96% (p=0.000 n=15+13)
* trie: close temporary databases in GetDB benchmark
* trie: don't keep []byte from DB load around
Nodes decoded from a DB load kept hashes and values as sub-slices of
the DB value. This can be a problem because loading from leveldb often
returns []byte with a cap that's larger than necessary, increasing
memory usage.
* trie: unload old cached nodes
* trie, core/state: use cache unloading for account trie
* trie: use explicit private flags (fixes Go 1.5 reflection issue).
* trie: fixup cachegen overflow at request of nick
* core/state: rename journal size constant
Delete crashed if a fullNode contained a valueNode directly. This bug is
very unlikely to occur with SecureTrie, but can happen with regular
tries. This commit also introduces a randomised test which triggers all
trie operations, which should prevent such bugs in the future.
Credit for finding this bug goes to Github user @rjl493456442.
Fixed database merge strategy to use the correct database. Due to a copy
paste fail when doing type evaluation the same database was being
iterated (chain), all others were ignored.
Removed state prefixing because {H(code): code} is stored in the same
database as the rest of the state.
