Eventual-Inc/Daft - CodSpeed

Daft

Blog Docs Changelog

Performance History

Latest Results

perf(parquet): skip column parallelism for small row groups Add MIN_RG_BYTES_FOR_COL_PARALLELISM threshold (16 MiB uncompressed) to fall back to decode_single_rg for small row groups where per-column reader overhead (metadata clones, buffer setup, hconcat) exceeds the benefit of parallel decode. Applied to both local streaming (Path 2) and local bulk (Path 1) read paths. The CodSpeed benchmark file (1024 rows, 16 cols, ~880KB uncompressed) now takes the single-reader fast path instead of spawning 16 column tasks. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

desmond/intra-rg-col-parallelism

10 minutes ago

merge from main

rchowell/ext-decouple

13 minutes ago

Merge branch 'main' into desmond/field-name-arc-str

desmond/field-name-arc-str

16 minutes ago

remove allowdeadcode

universalmind303/remove-builders

30 minutes ago

fix(dashboard): request notification permission before enabling notifications (#6362) ## Summary - The dashboard notification toggles ("On Query Start" / "On Query End") checked `Notification.permission === "granted"` before firing notifications, but never called `Notification.requestPermission()`. Since the browser default is `"default"`, notifications were silently skipped. - Now `Notification.requestPermission()` is called when the user enables either toggle. If permission is denied (or the browser doesn't support it), the toggle stays off. ## Test plan - [x] Enable "On Query End" in the dashboard notification settings - [x] Verify browser permission prompt appears - [x] Grant permission and run a query — verify macOS notification appears when query finishes - [x] Deny permission and verify the toggle stays off 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

main

45 minutes ago

slade/clean-up-repartition

51 minutes ago

refactor(arrow2): goodbye arrow2 (#6363) ## Changes Made <img width="497" height="469" alt="image" src="https://github.com/user-attachments/assets/3fd455e7-0801-4994-94fb-fe48eeb1e142" /> ## Related Issues

main

59 minutes ago

perf(parquet): eliminate redundant file opens in column-parallel decode Column-parallel decode opened the file independently for each column task, adding 16x open() syscall overhead on a 16-column file. Read the file into a bytes::Bytes buffer once and share it across column tasks via cheap Bytes::clone() (atomic refcount, zero-copy). Each column reader gets its own independent cursor over the shared buffer. This fixes the CodSpeed regression in test_show[1 Small File] where the per-column file opens added ~2.6ms overhead on a small 1024-row file. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

desmond/intra-rg-col-parallelism

1 hour ago

Active Branches

perf(parquet): add intra-row-group column parallelism to arrow-rs reader

last run

10 minutes ago

0%

refactor: decouple extension ABI from arrow-rs with Arrow C Data Interfaces

last run

13 minutes ago

0%

refactor(schema): change Field.name from String to Arc<str>

last run

16 minutes ago

+1%

© 2026 CodSpeed Technology

Home Terms Privacy Docs