With respect to `std::unordered_map` documentation, the `Hash` type
defined in the template is over the `Key` and not `T`, the value. This
hasher is incorrectly named as the `FilterHeader` is the value within this map.
I consider this a bug as opposed to a refactor as the key and value
relationship is implied to be `filter header -> block hash` when it is
the opposite.
Further, the hasher for the key already exists via `BlockHasher`.
ref: https://en.cppreference.com/w/cpp/container/unordered_map.html
Moved CBlockUndo disk read lookups from child index classes to
the base index class.
The goal is for child index classes to synchronize only through
events, without directly accessing the chain database.
This change will enable future parallel synchronization mechanisms,
reduce database access (when batched), and contribute toward the
goal of running indexes in a separate process (with no chain
database access).
Besides that, this commit also documents how NextSyncBlock() behaves.
It is not immediately clear this function could return the first
block after the fork point during a reorg.
Move ReadBlock code from CoinStatsIndex::CustomRewind to BaseIndex::Rewind
Move ReadUndo code from CoinStatsIndex::ReverseBlock to BaseIndex::Rewind
This commit does change behavior slightly. Since the new CustomRemove
methods only take a single block at a time instead of a range of
disconnected blocks, when they call CopyHeightIndexToHashIndex they will
now do an index seek for each removed block instead of only seeking once
to the height of the earliest removed block. Seeking instead of scanning
is a little worse for performance if there is a >1 block reorg, but
probably not noticeable unless the reorg is very deep.
This removes unused includes, primitives/block found manually, and the
others by iwyu:
blockfilter.h should remove these lines:
- #include <serialize.h> // lines 16-16
- #include <undo.h> // lines 18-18
Replace CommitInternal method with CustomCommit and use interfaces::Chain
instead of CChainState to generate block locator.
This commit does not change behavior in any way, except in the
(m_best_block_index == nullptr) case, which was added recently in
https://github.com/bitcoin/bitcoin/pull/24117 as part of an ongoing attempt to
prevent index corruption if bitcoind is interrupted during startup. New
behavior in that case should be slightly better than the old behavior (skipping
the entire custom+base commit now vs only skipping the base commit previously)
and this might avoid more cases of corruption.
Replace Rewind method with CustomRewind and pass block hashes and
heights instead of CBlockIndex* pointers
This commit does not change behavior in any way.
Replace overriden index Init() methods that use the best block
CBlockIndex* pointer with pure CustomInit() callbacks that are passed
the block hash and height.
This gets rid of more CBlockIndex* pointer uses so indexes can work
outside the bitcoin-node process. It also simplifies the initialization
call sequence so index implementations are not responsible for
initializing the base class.
There is a slight change in behavior here since now the best block
pointer is loaded and checked before the custom index init functions are
called instead of while they are called.
e734228d8585c0870c71ce8ba8c037f8cf8b249a Update GCSFilter benchmarks (Calvin Kim)
aee9a8140b3a58b744766f9e89572f1d953a808b Add GCSFilterDecodeSkipCheck benchmark (Patrick Strateman)
299023c1d9962628d158fac0306f8531506a0123 Add GCSFilterDecode and GCSBlockFilterGetHash benchmarks. (Patrick Strateman)
b0a53d50d9142bed51a8372eeb848816bfa94da8 Make sanity check in GCSFilter constructor optional (Patrick Strateman)
Pull request description:
This PR picks up the abandoned #19280
BlockFilterIndex was depending on `GolombRiceDecode()` during the filter decode to sanity check that the filter wasn't corrupt. However, we can check for corruption by ensuring that the encoded blockfilter's hash matches up with the one stored in the index database.
Benchmarks that were added in #19280 showed that checking the hash is much faster.
The benchmarks were changed to nanobench and the relevant benchmarks were like below, showing a clear win for the hash check method.
```
| ns/elem | elem/s | err% | ins/elem | bra/elem | miss% | total | benchmark
|--------------------:|--------------------:|--------:|----------------:|---------------:|--------:|----------:|:----------
| 531.40 | 1,881,819.43 | 0.3% | 3,527.01 | 411.00 | 0.2% | 0.01 | `DecodeCheckedGCSFilter`
| 258,220.50 | 3,872.66 | 0.1% | 2,990,092.00 | 586,706.00 | 1.7% | 0.01 | `DecodeGCSFilter`
| 13,036.77 | 76,706.09 | 0.3% | 64,238.24 | 513.04 | 0.2% | 0.01 | `BlockFilterGetHash`
```
ACKs for top commit:
mzumsande:
Code Review ACK e734228d8585c0870c71ce8ba8c037f8cf8b249a
theStack:
Code-review ACK e734228d8585c0870c71ce8ba8c037f8cf8b249a
stickies-v:
ACK e734228d8585c0870c71ce8ba8c037f8cf8b249a
ryanofsky:
Code review ACK e734228d8585c0870c71ce8ba8c037f8cf8b249a, with caveat that I mostly paid attention to the main code, not the changes to the benchmark. Only changes since last review were changes to the benchmark code.
Tree-SHA512: 02b86eab7b554e1a57a15b17a4d6d71faa91b556c637b0da29f0c9ee76597a110be8e3b4d0c158d4cab04af0623de18b764837be0ec2a72afcfe1ad9c78a83c6
281fd1a4a032cded7f9ea9857e3e99fc793c714b Replace KeyIDHasher with SaltedSipHasher (Andrew Chow)
210b693db66e7c5b618014b5a287aee15af00045 Add generic SaltedSipHasher (Andrew Chow)
95e61c1cf2a91d041c8025306ba36f0ea2806894 Move Hashers to util/hasher.{cpp/h} (Andrew Chow)
Pull request description:
There are existing `SaltedOutPointHasher` and `SaltedTxidHasher` classes used for `std::unordered_map` and `std::unordered_set` that could be useful in other places in the codebase. So we these to their own `saltedhash.{cpp/h}` file. An existing `KeyIDHasher` is moved there too. Additionally, `ScriptIDHasher`, `SaltedPubkeyHasher`, and `SaltedScriptHasher` are added so that they can be used in future work.
`KeyIDHasher` and `ScriptIDHasher` are not salted so that equality comparisons of maps and sets keyed by `CKeyID` and `CScriptID` will actually work.
Split from #19602 (and a few other PRs/branches I have).
ACKs for top commit:
laanwj:
Code review ACK 281fd1a4a032cded7f9ea9857e3e99fc793c714b
jonatack:
ACK 281fd1a4a032cded7f9ea9857e3e99fc793c714b, code review, debug build and ran bitcoind after rebasing to master @ dff0f6f753ea
fjahr:
utACK 281fd1a4a032cded7f9ea9857e3e99fc793c714b
Tree-SHA512: bb03b231ccf3c9ecefc997b8da9c3770af4819f9be5b0a72997a103864e84046a2ac39b8eadf0dc9247bdccd53f86f433642e3a098882e6748341a9e7736271b
Move the hashers that we use for hash tables to a common place.
Moved hashers:
- SaltedTxidHasher
- SaltedOutpointHasher
- FilterHeaderHasher
- SignatureCacheHasher
- BlockHasher
Cache block filter headers at heights of multiples of 1000 in memory.
Block filter headers at height 1000x are checkpointed, and will be the
most frequently requested. Cache them in memory to avoid costly disk
reads.
