vortex-data/vortex - CodSpeed

vortex

Blog Docs Changelog

Performance History

Latest Results

Keep normal-sized dictionary regions as separate DuckDB assignments Only coalesce tiny dictionary regions below 8192 rows, allowing workers to balance dictionary expressions without cutting their cached value domains. Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

ji/duckdb-v2-scan-driver

5 hours ago

fix(bench): retry the post-ingest database write across RDS restarts post-ingest.py connected to the benchmarks database once, so four ingest jobs failed during the two RDS restarts on 2026-10-28 (an engine upgrade refused connections for 33 seconds, an OS update for about 3 minutes). The write -- connect, with a fresh IAM token, and ingest -- now runs again after 5, 10, 20, 40, 60, 60 and 60 seconds while it fails with a lost or refused connection. A statement timeout, a validation error or a contract violation is not retried. Retrying the whole write is safe: the commit row is an idempotent upsert, and the fact rows land in one transaction, so a connection lost mid-write leaves nothing half-written. The worst case, about 5.5 minutes, fits the 10-minute commit-metadata job. Tests cover a retry that succeeds, retries that run out, an error that is not retried, and a schedule long enough for an OS-update restart. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Will Manning <will@willmanning.io>

wm/post-ingest-retries

7 hours ago

Keep optional dictionary partition metadata from rejecting skipped chunks Fall back to the driver's normal partitioning when a lazy chunk's metadata cannot be initialized. Selection and pruning may avoid executing that chunk. Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

ji/duckdb-v2-scan-driver

7 hours ago

Choose sparse dictionary access from predicate and row demand Preserve selective predicate demand at scan leaves while retaining vectorized evaluation for dense scan masks. Memoize referenced string dictionary predicates and reuse packed whole results after bulk evaluation; keep numeric comparisons on their vectorized path. Read dictionary codes before referenced value chunks for sparse random access, preserve duplicate/null code order, and keep partial reads out of whole-value caches. Fetch flat value segments alongside codes. Validation: 1,890 affected library tests passed (2 S3 skips), workspace all-target/all-feature Clippy with warnings denied, pinned formatting, layout doctests, and five complete Q23 row comparisons against V1. Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

ji/codex/scan-v2-topk-feedback

7 hours ago

Allow intentional single row ranges in dictionary assignment tests Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

ji/duckdb-v2-scan-driver

7 hours ago

Lock file maintenance

renovate/lock-file-maintenance-rust-lock-file-maintenance

7 hours ago

Avoid eager row-count lookup in dictionary boundary traversal Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

ji/duckdb-v2-scan-driver

7 hours ago

Skip dictionary-boundary planning for numeric projections Check the projected dtype before initializing concatenation children. Numeric scans and projections with multiple string columns retain their existing planning cost. Cover this with a lazy unsupported child that must remain unopened. Signed-off-by: Joe Isaacs <joe.isaacs@live.co.uk>

ji/duckdb-v2-scan-driver

7 hours ago

Latest Branches

0%

Bound V2 shared reads and partition DuckDB scan work#10857

5 hours ago

d704b47

ji/duckdb-v2-scan-driver

+12%

fix(bench): retry the post-ingest database write across RDS restarts#11018

7 hours ago

c277378

wm/post-ingest-retries

+10%

Improve Scan V2 TopK feedback and sparse payload retrieval#10964

7 hours ago

83d5434

ji/codex/scan-v2-topk-feedback

© 2026 CodSpeed Technology

Home Terms Privacy Docs