Add `streaming_groupby` for stateful streaming aggregation - #21924

Merged

rapids-bot[bot] merged 83 commits into

NVIDIA:release/26.06from

PointKernel:streaming_groupby

May 20, 2026

Commits on May 19, 2026

Add streaming groupby
PointKernel
committed
646e452
Update streaming groupby to match normal groupby behavior
PointKernel
committed
9b8eca0
Const everything
PointKernel
committed
4bcb63f
Minor cleanups
PointKernel
committed
4d9be13
Add proper support for complex aggs
PointKernel
committed
6ed0435
Test cleanups
PointKernel
committed
4fc5369
All fixed
PointKernel
committed
f4695ca
A final cleanup
PointKernel
committed
0c185d6
Address review comments
PointKernel
committed
841148d
this is good to go
PointKernel
committed
83370d8
Example cleanups
PointKernel
committed
328fb12
Minor cleanup
PointKernel
committed
85346eb
Minor build fix with GCC14
PointKernel
committed
7b4c7a7
Use persistent hash table for streaming groupby
PointKernel
committed
454668c
cleanups
PointKernel
committed
14f81eb
improvement
PointKernel
committed
4599992
cleanups
PointKernel
committed
1ee32d3
Disable dynamic growth
PointKernel
committed
aa6c1ae
cleanups
PointKernel
committed
c300f7a
Fix mr and stream usage
PointKernel
committed
ab88165
mr cleanup
PointKernel
committed
7707bb6
Fix nullable bug
PointKernel
committed
ae1ddfb
Remove detail usage in tests
PointKernel
committed
a05a32b
Revert example changes
PointKernel
committed
3299ae1
Add streaming groupby cardinality benchmark
PointKernel
committed
d6eccb9
Expose input size for groupby cardinality benchmarks
PointKernel
committed
1441ff9
Fix a unit test error
PointKernel
committed
9d5c597
Fix the ALL NULL mask ops
PointKernel
committed
f1d455e
Add string col support
PointKernel
committed
c3282ef
All tests passed with n-table comparator
PointKernel
committed
85b3110
Cut memory usage
PointKernel
committed
00c9e52
Plan D
PointKernel
committed
03597a7
Cleanups
PointKernel
committed
94736b2
Util cleanups
PointKernel
committed
2b6d135
Naming cleanups
PointKernel
committed
52178bb
Remove release
PointKernel
committed
f41fcd7
Fix tests
PointKernel
committed
3ea5db6
Fix all tests
PointKernel
committed
e1ed93c
More fixes
PointKernel
committed
41aba07
Add nested support back
PointKernel
committed
37ee31e
Cleanup comments
PointKernel
committed
5f202af
Minor naming + use table select
PointKernel
committed
99347c6
Clean up set creation logic
PointKernel
committed
591eb76
Renaming
PointKernel
committed
4143742
clean up common util
PointKernel
committed
560eab6
Cleanups + renaming + fix minor AI stupidity
PointKernel
committed
0b7eed9
Minor cleanup to add gather_agg_res to avoid duplicate
PointKernel
committed
7323d23
Use STL algos
PointKernel
committed
d111582
Split files to reduce build time
PointKernel
committed
22ea99a
Update max_groups constraints
PointKernel
committed
de3d106
Update to improve build time
PointKernel
committed
e0e8baa
remove unused file
PointKernel
committed
c4faed0
Use current mr for nosync policy
PointKernel
committed
cbd708d
Update tests
PointKernel
committed
160c960
Use flat agg request
PointKernel
committed
035e8e9
Bump libcudf CUDA 13 wheel size limit to 350M
PointKernel
committed
720dd5b
Use encoding capacity
PointKernel
committed
4081eb1
Formatting
PointKernel
committed
d9265d5
Migrate to dense buffer design
PointKernel
committed
a0007a5
Formatting
PointKernel
committed
6432075
Document streaming_groupby intent and use distinct consistently
PointKernel
committed
2e03526
Formatting
PointKernel
committed
cbdc728
Reduce memory usage + cleanups
PointKernel
committed
d8726b5
Formatting
PointKernel
committed
2cfd96b
Minor cleanups
PointKernel
committed
c3bfe21
minor fix
PointKernel
committed
aa71768
Minor comment update
PointKernel
committed
8c0d1fb
Rename max_groups/distinct_count to max_distinct_keys/distinct_keys
PointKernel
committed
97081a7
Fix a naming issue
PointKernel
committed
ce523da
Use naive store instead of atomic
PointKernel
committed
ab20d11
cleanups
PointKernel
committed
8229215
Minor updates
PointKernel
committed
270c33c
Formatting
PointKernel
committed
2cd760a
Use parenthesis form for nullable strings_column_wrapper construction
PointKernel
committed
baeb5dd
Skip n_table_comparator on the first batch in streaming_groupby
PointKernel
committed
49eaecb
Fix the first batch key equal
PointKernel
committed
7b339f3
Use tranform_reduce to avoid count_if pass
PointKernel
committed
18112a0
Fuse insert + compact via thrust::copy_if in streaming_groupby
PointKernel
committed
21ed3ba
Split streaming_groupby insert path into 4 TUs for parallel build
PointKernel
committed
56d8977
Reject decimal128 MIN/MAX/SUM in streaming_groupby
PointKernel
committed
ec6e8b2
Address CodeRabbit review feedback
PointKernel
committed
22d82d4
Address devavret review feedback
PointKernel
committed
251d8db
Bump libcudf CUDA 12 wheel size limit to 700M
PointKernel
committed
8bf5ceb