Eventual-Inc/Daft - CodSpeed

Daft

Blog Docs Changelog

Performance History

Latest Results

fix: add tenacity retries to Google Sheets upload in benchmarking (#6713) ## Summary - Google Sheets API occasionally returns transient 503 errors during benchmark result uploads - Adds `tenacity` retry decorator to `upload_to_google_sheets` with exponential backoff (2s–60s, up to 5 attempts) - Retries only on transient HTTP status codes: 429, 500, 502, 503, 504

main

56 minutes ago

chenghuichen:agg-ext

1 hour ago

fix: add tenacity retries to Google Sheets upload in benchmarking Google Sheets API occasionally returns transient 503 errors, causing benchmark runs to fail at the upload step. Wrap upload_to_google_sheets with tenacity retry logic for retryable status codes (429, 500, 502, 503, 504) with exponential backoff.

jeev/benchmarking-gspread-retries

2 hours ago

feat: Union Type (#6497) ## Changes Made This PR introduces the Union type and array to daft-core which allows storing values of multiple different child types in a single column/array. This PR adds support for reading from pyarrow unionarrays and basic operations like slicing, concat, filter and take as well as accessing type ids, child fields and union mode from the daft union type. ## Related Issues This work is part of [Discussion #6411 ](https://github.com/Eventual-Inc/Daft/discussions/6411) to add geospatial support to Daft.

main

2 hours ago

fix: skip schema pruning on Source node when can_absorb_select is false (#6501) ## Summary - Skips projection pushdown into materialized scans (`ScanState::Tasks`) - Adds regression tests for custom `DataSource` joins that previously panicked - Fixes `DataSource.get_tasks` test to use async API Fixes #6500 ## Problem Two issues with projection pushdown on Source nodes: 1. **Materialized scans**: When `ScanState::Tasks` is already materialized, pushing projections down would incorrectly modify already-built scan tasks. 2. **Custom DataSource joins**: When a custom `DataSource` applies `pushdowns.columns` and the query involves **join + select/agg**, the pushed-down column set could previously be incomplete. This was fixed upstream in the column requirement computation, and the regression tests now verify it stays correct. ## Fix In `push_down_projection.rs`: 1. **Guard against materialized scans** — early return `Transformed::no` when `ScanState::Tasks(_)` to avoid modifying already-built scan tasks 2. **Keep standard schema pruning** — always prune Source schema + set column pushdowns + retry optimization (matching main behavior), which correctly computes required columns including join keys ## Tests - Added `tests/io/test_datasource_join.py` with three regression tests: - `test_join_select_does_not_panic` — join + select - `test_join_groupby_agg_does_not_panic` — join + groupby/agg - `test_multi_join_agg_does_not_panic` — three-way join + aggregation - Added `test_projection_does_not_pushdown_into_materialized_scan` Rust unit test --------- Co-authored-by: Ofer Helman <ofer.helman@vastdata.com> Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

main

2 hours ago

chenghuichen:agg-ext

3 hours ago

feat: implement e, pi, factorial, and hypot math functions (#6682) ## Summary - Adds 4 math functions from the missing list in #3793: - **`e()`** — Returns Euler's number (2.718...), zero-argument constant - **`pi()`** — Returns pi (3.14159...), zero-argument constant - **`factorial(expr)`** — Returns factorial of a non-negative integer, returns NaN for negative or non-integer inputs - **`hypot(a, b)`** — Returns sqrt(a² + b²), follows the existing `atan2` binary function pattern - All 4 are registered in the Rust FUNCTION_REGISTRY (auto-available in SQL) - Python wrappers and exports added Resolves part of #3793 ## Test plan - [x] 8 new tests covering all 4 functions (constants, edge cases, bad input, None handling) - [x] All 330 tests in `test_numeric.py` pass with no regressions - [x] `ruff check` and `ruff format` clean - [x] `cargo fmt` clean

main

3 hours ago

feat: support catalog-qualified identifiers in create_table, drop_table (#6680) ## Summary - `create_table`, `create_table_if_not_exists`, and `drop_table` now resolve catalog-qualified identifiers (e.g. `my_cat.my_schema.my_table`) by routing to the matched catalog with the prefix stripped - Replicates the existing resolution pattern from `get_table` (Rule 3: if `identifier[0]` matches an attached catalog, strip it and route there) - Python-only change (no Rust changes) Resolves #4566 ## Test plan - [x] 3 new tests in `tests/catalog/test_catalog.py` covering `create_table`, `create_table_if_not_exists`, and `drop_table` with catalog-qualified identifiers - [x] Full `tests/catalog/` suite passes with no regressions

main

3 hours ago

Latest Branches

0%

feat: introduce AggFn trait and two-stage UDAF aggregation pipeline#6704

2 hours ago

8c98957

chenghuichen:agg-ext

-1%

fix: add tenacity retries to Google Sheets upload in benchmarking#6713

2 hours ago

a0800a1

jeev/benchmarking-gspread-retries

0%

fix(docs): replace deprecated .struct.get() with bracket indexing in examples#6709

4 hours ago

426e855

everettVT/peaceful-kilby

© 2026 CodSpeed Technology

Home Terms Privacy Docs