Eventual-Inc
Daft
Blog
Docs
Changelog
Blog
Docs
Changelog
Overview
Branches
Benchmarks
Runs
Performance History
Latest Results
test(flotilla): gate cancel test on Ray runner These flotilla-specific cancellation behaviors don't exist on the native runner, so the test only meaningfully runs under `DAFT_RUNNER=ray`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
colin/test-distributed-limit-cancel-regression
1 minute ago
test: cover generic file casts to image
fix/file-media-cast-from-unknown
18 minutes ago
test: clarify uuidv7 bit coverage
everettVT/uuid-param-generator
24 minutes ago
fixes
chris/shuffle-write-stats
52 minutes ago
fix: error strings
euan/window-firstlastval-agg
53 minutes ago
fix: cast generic files to specific media types
fix/file-media-cast-from-unknown
1 hour ago
fix: err msg
euan/window-firstlastval-agg
2 hours ago
perf: enable per-column compression configuration (#6884) ## Changes Made Adds per-column `column_compression` to the parquet writer so columns holding already-compressed payloads (JPEG/PNG/MP3/etc.) can opt out of codec work. ## Why? I was like, why are we compressing JPEG? We pay to compress JPEG with SNAPPY, then we pay to decompress SNAPPY. The simplest solution was allowing customers to specify per-column compression. Benchmarked snappy vs none on 10K JPEGs (`bytes` column, 70 MB output) with S3 moved out of the timed path, a discarded warmup, and 3 alternating iterations per mode in isolated subprocesses: | Metric | none | snappy | speedup | |---|---|---|---| | Wall time | 0.22s | 0.54s | **+142%** | | CPU user | 0.15s | 0.44s | **+205%** | | Output size | 70.66 MB | 70.15 MB | −0.7% | Snappy spends CPU compressing entropy-coded bytes that don't shrink. Skipping it is ~2.4× faster with no meaningful size penalty. This is a double benefit because we save on decompression as well. ## Related Issues n/a
main
2 hours ago
Latest Branches
CodSpeed Performance Gauge
0%
fix: cast generic file arrays to media-specific types
#6848
44 minutes ago
ff0f166
fix/file-media-cast-from-unknown
CodSpeed Performance Gauge
0%
refactor: parameterize uuid generation
#6961
58 minutes ago
d42823b
everettVT/uuid-param-generator
CodSpeed Performance Gauge
0%
feat(shuffle): track rows/bytes/partitions on shuffle write sinks
#6976
1 hour ago
e14e653
chris/shuffle-write-stats
© 2026 CodSpeed Technology
Home
Terms
Privacy
Docs