Eventual-Inc/Daft - CodSpeed

Daft

Blog Docs Changelog

Performance History

Latest Results

feat(checkpoint): idempotent write_deltalake + cross-sink helpers * write_deltalake(checkpoint=daft.IdempotentCommit(store, key)) produces exactly one Delta commit per logical commit, tagged with daft.idempotence-key in custom_metadata. Retries with the same key short-circuit via Delta history walk. * Shared helpers in daft/dataframe/_checkpoint_commit.py used by both the new delta path and the existing iceberg path. * Iceberg refactor: route through shared helpers; partitioning struct column included on partitioned tables. Co-Authored-By: ChengHui Chen <chenghui.cch@alibaba-inc.com> Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

rohit/feature/df-2042-consolidated-checkpoint-rework

31 minutes ago

chris/dashboard-fixes

3 hours ago

removed comments

euan/to-struct

3 hours ago

feat: distributed tests

euan/forward-asof

3 hours ago

fix(dashbaord): stop status animations and highlight failed operators for queries

chris/dashboard-fixes

4 hours ago

euan/forward-asof

4 hours ago

style(ext): move extension helper imports to module level Hoist json and get_loaded_extension_paths to top-level imports in flotilla.py. Keep deferred `import daft` in _load_extensions_from_env to avoid circular imports on Ray workers. Co-authored-by: Cursor <cursoragent@cursor.com>

desmond/native-extensions-ray-support

5 hours ago

perf: push COUNT(*) through Project nodes and support filtered count pushdown Three improvements to the PushDownAggregation optimizer rule: 1. Walk through Project nodes for COUNT(*, All). Row counts are invariant under projection, so intervening Project nodes (e.g., from with_column casts) no longer block count pushdown. This makes SELECT COUNT(*) read only parquet footer metadata instead of scanning a column. 2. Handle Alias-wrapped count expressions. SQL COUNT(*) produces Alias(Count(col), "count") but the rule previously only matched bare Count expressions. The Alias is now unwrapped for pattern matching and stripped before pushing into scan (the executor matches on Expr::Agg(AggExpr::Count(..)) directly). 3. Implement SupportsPushdownFilters for GlobScanOperator on parquet files. This enables the strict_pushdown path in PushDownAggregation to combine filter + count pushdown. A new stream_parquet_filtered_count execution path reads only filter columns, applies the predicate, and counts surviving rows without materializing other columns. Benchmarked on ClickBench (c6a.4xlarge, single parquet, same codebase): - Q1 (SELECT COUNT(*)): 106ms -> 11ms (9.4x, metadata-only read) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

desmond/count-pushdown-through-projects

5 hours ago

Latest Branches

0%

feat(checkpoint): idempotent write_deltalake + cross-sink helpers#6932

1 hour ago

0746f2a

rohit/feature/df-2042-consolidated-checkpoint-rework

0%

fix(dashbaord): highlight failed operators for queries#6930

3 hours ago

9747c80

chris/dashboard-fixes

0%

fix: handle broadcast of scalars when calling to_struct#6931

4 hours ago

1ed42e0

euan/to-struct

© 2026 CodSpeed Technology

Home Terms Privacy Docs