fix(hot): incremental splitter; promote constants to BlockNumberList#67
Open
rswanson wants to merge 2 commits into
Open
fix(hot): incremental splitter; promote constants to BlockNumberList#67rswanson wants to merge 2 commits into
rswanson wants to merge 2 commits into
Conversation
…rList Addresses review feedback on #66. - Promote shard sizing constants from `ShardedKey` to `BlockNumberList`, where they belong: they are properties of the roaring treemap encoding, not the key: - `BlockNumberList::MAX_ENCODED_BYTES` (was `ShardedKey::MAX_SHARD_BYTES`) - `BlockNumberList::SAFE_INDICES_PER_SHARD = 67` — the count guaranteed to fit in `MAX_ENCODED_BYTES` under any u64 distribution, derived from the worst case of one index per `RoaringBitmap` (~22 B/index) plus the 8-byte treemap header. - Document the roaring serialisation overhead on `SAFE_INDICES_PER_SHARD` (cookie, container descriptors, offset table, payload) so the bound is reproducible. - Add characterisation tests in `signet-storage-types` covering dense runs, sparse-within-a-bitmap (~10 B/index), and worst-case across bitmaps (~22 B/index). The `safe_indices_per_shard_is_tight` test fails if the bound becomes loose, prompting future re-derivation. - Replace the binary-search splitter with an exact single-pass incremental builder: push, check, roll back on overflow. Avoids the `mid == 1` footgun in the previous implementation and does one `BlockNumberList` rebuild per shard boundary instead of `log2(n)` full re-encodings per shard. - Add a count-based fast path using `SAFE_INDICES_PER_SHARD` so the common case skips `serialized_size()` entirely. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Self-review cleanup of the previous commit. - Use `BlockNumberList::remove(idx)` + `clear()` to roll back the overflowing index in place, instead of rebuilding the prefix from a re-sliced `Vec<u64>`. Drops one allocation + one full re-encoding per shard boundary. - Pass `last_shard.iter()` directly into `chunk_by_encoded_size` instead of materialising a `Vec<u64>` first. Drops the per-write N-sized allocation on the slow path. - Collapse `shard_start` + working `shard` to a single `prev_in_shard` flag — they tracked overlapping information, and the previous slow path serialised the same data twice (once into `shard`, once into the rebuilt `prefix`). - Add `BlockNumberList::remove` so the splitter doesn't need to reach through the tuple-struct boundary. - Drop two redundant comments (`// Pre-sorted, ...: push cannot fail.` restated the `expect` message; `// Emit the final (open) shard.` restated the code). - Fix `let mut last_shard = last_shard;` no-op rebind by moving `mut` into the destructuring. - Name the conformance test's previously-bare `30` as `SHARDS_TO_FORCE: usize = 30`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Addresses the four asks from the review of #66 plus the binary-search
critique posted as a follow-up.
Changes
1. Promote constants from
ShardedKeytoBlockNumberList(review #4)The byte budget and the count bound are properties of the roaring
encoding, not of the key — moved them where they belong:
BlockNumberList::MAX_ENCODED_BYTES(wasShardedKey::MAX_SHARD_BYTES)BlockNumberList::SAFE_INDICES_PER_SHARD = 672. Characterise the roaring treemap encoding (review #1, #2, #3)
SAFE_INDICES_PER_SHARDis documented with the per-component overhead(treemap header,
RoaringBitmapcookie, container descriptors + offsettable, container payload) and derived from the worst case of one index
per
RoaringBitmap(~22 B/index).Five characterisation tests added to
signet-storage-types::int_list:dense_run_encodes_compactlysparse_within_single_bitmap_costs_ten_bytes_per_indexsparse_across_bitmaps_costs_twenty_two_bytes_per_indexsafe_indices_per_shard_fits_worst_casesafe_indices_per_shard_is_tightThe last two fail loudly if the constant drifts out of sync with the
encoding (e.g. on a roaring upgrade), forcing a re-derivation.
3. Replace binary search with exact incremental splitter
Per the Option A follow-up: the previous
max_prefix_fittingdidlog2(n)probes per shard, each probe rebuilding the fullBlockNumberListfrom scratch. It also had aif mid == 1 { break; }branch that silently returned
best = 1without verifying that oneelement actually fit.
The new
chunk_by_encoded_sizewalks indices once through a workingBlockNumberList, checkingserialized_size()after each push, androlling back the overflowing index into the next shard. One full
rebuild per shard boundary, no probing, no
mid == 1footgun.4. Count-based fast path
append_to_sharded_historycheckslen() <= SAFE_INDICES_PER_SHARDbefore calling
serialized_size(), so the common case (smallappends) skips the size check entirely.
Out of scope
The layering observation (move the budget to a backend trait so
non-MDBX backends don't waste work splitting) is deferred to a separate
issue per its own author. This PR keeps
MAX_ENCODED_BYTESon theencoding type as a single shared default; a future PR can introduce a
HotKvWrite::max_value_bytes()override without churning the splitter.Test plan
cargo +nightly fmt -- --checkcargo clippy --workspace --all-targets --all-features -- -D warningscargo clippy --workspace --all-targets --no-default-features -- -D warningsRUSTDOCFLAGS="-D warnings" cargo doc --workspace --no-depscargo t -p signet-storage-types— five new characterisation tests passcargo t -p signet-hot --all-features— 103 tests pass, incl.mem_conformancewhich runs the newtest_history_shard_fits_in_dupsort_limitregressioncargo t -p signet-hot-mdbx --all-features— 49 tests pass🤖 Generated with Claude Code