MergeBatcher: extract-first seal - #793
Merged
frankmcsherry merged 1 commit intoJul 12, 2026
Merged
Conversation
seal was merge-then-extract: every chain was drained into one before the frontier split, so updates parked beyond 'upper' (e.g. delay/enter_at bands, future iterations) were recopied on EVERY seal. Instrumenting seal showed drained-over-shipped ratios of 35-38x on the enter_at-band arrangements of an SCC benchmark (10.3M rows drained per seal stream to ship 293k), against 1.0-1.8x for operators without future-dated residue. seal now extracts each chain in place and merges only the shipped parts — delta-sized, and the merge is where cross-chain consolidation happens (the ship/keep split is by time, so no equal (data, time) group can straddle it). Kept chains re-enter the ladder as-is, sorted by weight with the geometric invariant restored, so residue is copied only at doubling events. Measured on the SCC suite: DDIR-corgi -15%, DDIR-vec -11%, compiled native (whose extract still copies rows; only the drain removal applies) -2.5%. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
frankmcsherry
added a commit
that referenced
this pull request
Jul 13, 2026
The extract-first seal (#793) folded the returned frontier during the per-chain extract, before cross-chain consolidation, so a kept time whose updates cancel across two un-merged chains was wrongly reported — the batcher held a capability that never discharges, stalling progress. Restore the merge-then-extract seal: all chains merge into one (consolidating equal (data,time), dropping zeros) before extract folds the frontier, so the frontier reflects post-consolidation distinct times. Keeps regression tests (frontier_is_post_consolidation) so the property is guarded if a future seal revisits the extract-first idea. Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Merged
frankmcsherry
added a commit
that referenced
this pull request
Jul 15, 2026
Rewrite the release-plz-generated changelogs into the house style: categorized Added/Changed/Removed sections plus a closing summary of the breaking surface. Add the omitted timely 0.31 bump, drop the reverted extract-first seal pair (#793/#796), de-duplicate the HRTB entries, and flag the new Chunk and JoinTactic/ReduceTactic surface as experimental. Co-Authored-By: Claude Opus 4.8 (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.
seal was merge-then-extract: every chain was drained into one before the frontier split, so updates parked beyond 'upper' (e.g. delay/enter_at bands, future iterations) were recopied on EVERY seal. Instrumenting seal showed drained-over-shipped ratios of 35-38x on the enter_at-band arrangements of an SCC benchmark (10.3M rows drained per seal stream to ship 293k), against 1.0-1.8x for operators without future-dated residue.
seal now extracts each chain in place and merges only the shipped parts — delta-sized, and the merge is where cross-chain consolidation happens (the ship/keep split is by time, so no equal (data, time) group can straddle it). Kept chains re-enter the ladder as-is, sorted by weight with the geometric invariant restored, so residue is copied only at doubling events.
Measured on the SCC suite: DDIR-corgi -15%, DDIR-vec -11%, compiled native (whose extract still copies rows; only the drain removal applies) -2.5%.