Repository navigation
array_agg is very slow #20465
Description
Describe the bug
Running a simple benchmark (5000 groups, 5000 int64 values per group), array_agg in DataFusion is roughly 30x slower than DuckDB and 60x slower than ClickHouse.
To Reproduce
No response
Expected behavior
No response
Additional context
No response
Activity
neilconway commentedon Feb 21, 2026
take
chore: Add end-to-end benchmark for array_agg, code cleanup (#20496)
alamb commentedon Feb 24, 2026
See also
alamb commentedon Feb 24, 2026
array_agg, Polars is 4 times faster for equal query #17446Let's continue the discussion there
array_agg, Polars is 4 times faster for equal query #17446neilconway commentedon Feb 24, 2026
@alamb Thanks, Andrew! I missed that this issue was previously reported and worked on. At least my approach isn't radically different than the prior consensus 😅
alamb commentedon Feb 27, 2026
Even better is that since this keeps coming up it is a signal of how important it is
perf: Optimize `array_agg()` using `GroupsAccumulator` (#20504)
chore: Add end-to-end benchmark for array_agg, code cleanup (apache#2…
perf: Optimize `array_agg()` using `GroupsAccumulator` (apache#20504)