perf(sqlite): write search_index rows eight per INSERT (+9%) - #951
Open
angela-helios wants to merge 2 commits into
Open
perf(sqlite): write search_index rows eight per INSERT (+9%)#951angela-helios wants to merge 2 commits into
angela-helios wants to merge 2 commits into
Conversation
A resource writes ~10-30 search_index rows, each via a single-row prepared INSERT stepped once per row. The B-tree maintenance is the irreducible cost, but entering and leaving the statement per row is not: batching eight rows per INSERT (with a single-row remainder) amortizes it. Measured on the real 31 GB bulk-submit manifest, identical 7-minute windows, on top of the FTS-scan fix and the write-path pragma defaults: 382/s -> 417/s (+9%), index-row INSERT cost 1.11 -> 0.85 ms per entry, identical index-row counts per resource. File fan-out at 2 was also measured on the optimized write path and came out neutral-to-negative (388/s vs 417/s, no lock errors): with extraction and fetch already cheap, a second in-flight file only adds write-lock handoffs. SQLite's default file concurrency of 1 stands.
smunini
previously approved these changes
Sep 5, 2026
insert_sql_rows8 was inserted between to_sql_params' doc comment and its signature, so the new function absorbed those docs and to_sql_params was left undocumented, failing the workspace `-D missing-docs` lint. Move the new function above the block and give each its own docs. Claude-Session: https://claude.ai/code/session_016kNZJTXf86QoMqcdWqKCbR
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
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.
Third item in the SQLite bulk-ingest series (#949 FTS scan, #950 write-path defaults), and the multi-row INSERT lever from #945/#947.
A resource writes ~10–30
search_indexrows, each via a single-row prepared INSERT stepped once per row. The B-tree maintenance is the irreducible cost, but entering and leaving the statement per row is not: rows are now written eight per INSERT (8 × 24 positional parameters, single-row remainder), amortizing the per-statement overhead. Same rows, same order, same values — the row-count audit per resource is identical.Measured
Real 31 GB manifest, identical 7-minute windows, single worker, on top of #949 + #950:
Also measured and rejected while here: file fan-out ×2 on the optimized write path — 388/s vs 417/s, no lock errors. With extraction and fetch already cheap, a second in-flight file only adds write-lock handoffs; SQLite's default file concurrency of 1 stands (consistent with #946's clamp).
Running total
Stacked local measurement (#944 + #949 + #950 + this): today's
mainstarts at ~110/s and decays; the combined branches hold 417/s stable — a projected ~38 minutes for the full 954,288-entry import, single worker, zero configuration. The 51-minute full-run verification before the last two increments is on #947.Full sqlite persistence suite green (344).