docs(design): close-out disposition for #405 (re-measured on main) - #634
Merged
Merged
Conversation
…d on main) Records the final disposition of the three ranked items in commandprompt#405, re-measured against current main so the issue can be closed: - Item 1 (position-list intersection): MEASURED again on main 977bb44. Two queries returning the same 2,000 surviving rows show 25x the payload-decode cost (28.9 ms vs 726.7 ms) when survivors are spread one-per-vector, with 1,998,000 rows removed by filter -- decode tracks surviving VECTORS not positions. We prune vectors, not positions; Abadi's full bit-string intersection stays retracted (a trade, not a free win, PR commandprompt#601); the worthwhile subset shipped as the fold-path payload deferral (commandprompt#617). - Item 2 (branch predication): N/A -- the tree builds no packed selection vector to predicate (the selection loops were deleted in commandprompt#200). Recorded. - Item 3 (vector length): reframed. COLUMNAR_NATIVE_VECTOR_LENGTH is vestigial (write-only catalog stamp); the execution stride is a literal and the runtime knob is chunk_group_row_limit. Not a single tunable constant. Recorded. Docs only (design note); no code, no behaviour change. Recommends CLOSE. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TgAk1gqeME7DHpJw8xxybu
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Docs-only close-out for #405 (materialization strategies from Abadi + X100). Appends a final disposition section to
design/ISSUE_405_LATE_MATERIALIZATION.md, re-measured on currentmain(977bb44) so the issue can be reasonably closed. No code, no behaviour change.What it records
Item 1 (position-list intersection before materialization) — measured again on main. Two queries returning the same 2,000 surviving rows on a 2,000,000-row fixture (pg18n non-assert, bloom off, serial):
Same survivor count, 25× the payload-decode cost, ~2M rows materialised to return 2,000 → decode tracks surviving vectors, not positions. We prune vectors, not positions; Abadi's bit-string intersection stays retracted (a measured trade, PR #601); the worthwhile subset shipped as the fold-path payload deferral (#617,
test/native_fold_deferral.sh).Item 2 (branch predication) — N/A: the tree builds no packed selection vector to predicate (those loops were deleted in #200). Recorded.
Item 3 (vector length vs column count) — reframed:
COLUMNAR_NATIVE_VECTOR_LENGTHis vestigial (write-only catalog stamp, 3 sites); the execution stride is a literal and the runtime granularity knob ischunk_group_row_limit. Not a single tunable constant. Recorded.Recommendation: CLOSE #405 — item 1 measured + dispositioned with the useful part shipped; items 2/3 recorded as the issue itself proposed.
Scope
Docs only (a
design/note; not user-facingdocs/, sodocs_styledoes not apply). No code, no CHANGELOG entry.🤖 Generated with Claude Code