perf: add fully ordered group values implementation for primitive by rluvaton · Pull Request #25931 · apache/datafusion

rluvaton · 2026-10-01T09:55:37Z

Which issue does this PR close?

None

Rationale for this change

Improving performance. there is no need to do unnecessary computation when the group by columns are fully sorted

What changes are included in this PR?

Implemented GroupValues for new struct FullyOrderedGroupValues that avoid keeping hash map and map probing

What is the testing strategy for this PR?

existing tests

Are there any user-facing changes?

no

This allows for optimized implementation given that we know that the input is ordered

rluvaton · 2026-10-01T09:55:50Z

run benchmarks tpch10 h2o_medium clickbench_partitioned

env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "16G"

rluvaton · 2026-10-01T09:57:07Z

run benchmark external_aggr

rluvaton · 2026-10-01T10:08:19Z

show benchmarks queue

rluvaton · 2026-10-01T10:08:55Z

show benchmark queue

adriangbot · 2026-10-01T10:08:57Z

Hi @rluvaton, you asked to view the benchmark queue (#25931 (comment)).

Comment	Repo	PR	User	Benchmarks	Status
#5929052627	apache/datafusion	#25931	rluvaton	["tpch10"]	running
#5929052627	apache/datafusion	#25931	rluvaton	["h2o_medium"]	running
#5929052627	apache/datafusion	#25931	rluvaton	["clickbench_partitioned"]	running
#5929070511	apache/datafusion	#25931	rluvaton	["external_aggr"]	running

File an issue against this benchmark runner

adriangbot · 2026-10-01T10:30:36Z

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16-lssd (12 vCPU / 65 GiB) | Linux bench-c5929052627-2998-2d7g8 6.12.94+ #1 SMP Wed Aug 19 07:47:20 UTC 2026 aarch64 GNU/Linux

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Comparing add-fully-ordered-groups-values (8dec018) to 6924c6a (merge-base) diff

Run configuration

run benchmark h2o_medium
env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "16G"

Results will be posted here when complete

File an issue against this benchmark runner

adriangbot · 2026-10-01T10:30:36Z

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16-lssd (12 vCPU / 65 GiB) | Linux bench-c5929052627-2999-p5pvj 6.12.94+ #1 SMP Wed Aug 19 07:47:20 UTC 2026 aarch64 GNU/Linux

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Comparing add-fully-ordered-groups-values (8dec018) to 6924c6a (merge-base) diff

Run configuration

run benchmark clickbench_partitioned
env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "16G"

Results will be posted here when complete

File an issue against this benchmark runner

adriangbot · 2026-10-01T10:30:37Z

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16-lssd (12 vCPU / 65 GiB) | Linux bench-c5929070511-3000-w9rj4 6.12.94+ #1 SMP Wed Aug 19 07:47:20 UTC 2026 aarch64 GNU/Linux

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Comparing add-fully-ordered-groups-values (8dec018) to 6924c6a (merge-base) diff

Run configuration

run benchmark external_aggr

Results will be posted here when complete

File an issue against this benchmark runner

adriangbot · 2026-10-01T10:30:54Z

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16-lssd (12 vCPU / 65 GiB) | Linux bench-c5929052627-2997-tt5h5 6.12.94+ #1 SMP Wed Aug 19 07:47:20 UTC 2026 aarch64 GNU/Linux

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Comparing add-fully-ordered-groups-values (8dec018) to 6924c6a (merge-base) diff

Run configuration

run benchmark tpch10
env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "16G"

Results will be posted here when complete

File an issue against this benchmark runner

adriangbot · 2026-10-01T10:39:34Z

🤖 Benchmark completed (GKE) | trigger

Instance: c4a-highmem-16-lssd (12 vCPU / 65 GiB)

Comparing add-fully-ordered-groups-values (8dec018) to 6924c6a (merge-base) diff

Run configuration

run benchmark tpch10
env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "16G"

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Details

Comparing HEAD and add-fully-ordered-groups-values
--------------------
Benchmark tpch_sf10.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Query     ┃      HEAD ┃ add-fully-ordered-groups-values ┃    Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ QQuery 1  │ 326.50 ms │                       326.52 ms │ no change │
│ QQuery 2  │  86.92 ms │                        90.45 ms │ no change │
│ QQuery 3  │ 212.62 ms │                       219.05 ms │ no change │
│ QQuery 4  │ 114.18 ms │                       113.00 ms │ no change │
│ QQuery 5  │ 352.57 ms │                       344.77 ms │ no change │
│ QQuery 6  │ 124.29 ms │                       125.30 ms │ no change │
│ QQuery 7  │ 436.59 ms │                       445.43 ms │ no change │
│ QQuery 8  │ 354.86 ms │                       355.50 ms │ no change │
│ QQuery 9  │ 514.25 ms │                       511.11 ms │ no change │
│ QQuery 10 │ 282.08 ms │                       293.92 ms │ no change │
│ QQuery 11 │  61.68 ms │                        62.43 ms │ no change │
│ QQuery 12 │ 148.48 ms │                       151.54 ms │ no change │
│ QQuery 13 │ 301.35 ms │                       297.96 ms │ no change │
│ QQuery 14 │ 168.35 ms │                       167.82 ms │ no change │
│ QQuery 15 │ 298.12 ms │                       298.30 ms │ no change │
│ QQuery 16 │  63.27 ms │                        64.61 ms │ no change │
│ QQuery 17 │ 581.26 ms │                       584.76 ms │ no change │
│ QQuery 18 │ 718.86 ms │                       710.80 ms │ no change │
│ QQuery 19 │ 229.34 ms │                       230.64 ms │ no change │
│ QQuery 20 │ 261.26 ms │                       267.15 ms │ no change │
│ QQuery 21 │ 661.17 ms │                       656.05 ms │ no change │
│ QQuery 22 │  60.01 ms │                        57.11 ms │ no change │
└───────────┴───────────┴─────────────────────────────────┴───────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                              ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                              │ 6358.02ms │
│ Total Time (add-fully-ordered-groups-values)   │ 6374.20ms │
│ Average Time (HEAD)                            │  289.00ms │
│ Average Time (add-fully-ordered-groups-values) │  289.74ms │
│ Queries Faster                                 │         0 │
│ Queries Slower                                 │         0 │
│ Queries with No Change                         │        22 │
│ Queries with Failure                           │         0 │
└────────────────────────────────────────────────┴───────────┘

Distribution per query (min / mean ±stddev / max):

Comparing HEAD and add-fully-ordered-groups-values
--------------------
Benchmark tpch_sf10.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Query     ┃                               HEAD ┃    add-fully-ordered-groups-values ┃    Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ QQuery 1  │  326.50 / 328.90 ±1.46 / 330.53 ms │  326.52 / 328.82 ±1.39 / 330.13 ms │ no change │
│ QQuery 2  │     86.92 / 90.62 ±2.52 / 94.03 ms │     90.45 / 92.60 ±1.55 / 94.74 ms │ no change │
│ QQuery 3  │  212.62 / 215.46 ±2.10 / 218.92 ms │  219.05 / 223.00 ±4.92 / 232.51 ms │ no change │
│ QQuery 4  │  114.18 / 115.09 ±0.96 / 116.91 ms │  113.00 / 115.56 ±1.52 / 117.53 ms │ no change │
│ QQuery 5  │  352.57 / 356.49 ±4.38 / 364.95 ms │  344.77 / 356.00 ±8.70 / 371.52 ms │ no change │
│ QQuery 6  │  124.29 / 124.75 ±0.36 / 125.25 ms │  125.30 / 127.24 ±2.84 / 132.79 ms │ no change │
│ QQuery 7  │ 436.59 / 450.79 ±10.20 / 467.60 ms │ 445.43 / 460.73 ±11.78 / 474.50 ms │ no change │
│ QQuery 8  │  354.86 / 359.66 ±4.78 / 367.58 ms │  355.50 / 364.61 ±8.89 / 379.49 ms │ no change │
│ QQuery 9  │ 514.25 / 526.37 ±16.46 / 558.70 ms │ 511.11 / 529.50 ±20.06 / 568.19 ms │ no change │
│ QQuery 10 │  282.08 / 290.06 ±5.92 / 298.03 ms │  293.92 / 302.17 ±8.85 / 318.86 ms │ no change │
│ QQuery 11 │     61.68 / 63.12 ±0.88 / 64.25 ms │     62.43 / 63.78 ±0.80 / 64.52 ms │ no change │
│ QQuery 12 │  148.48 / 157.06 ±6.89 / 166.05 ms │  151.54 / 156.02 ±4.70 / 162.66 ms │ no change │
│ QQuery 13 │ 301.35 / 319.29 ±19.00 / 353.04 ms │  297.96 / 312.92 ±8.34 / 323.50 ms │ no change │
│ QQuery 14 │  168.35 / 173.50 ±5.38 / 180.64 ms │  167.82 / 172.33 ±3.80 / 178.43 ms │ no change │
│ QQuery 15 │  298.12 / 300.98 ±1.65 / 302.75 ms │  298.30 / 299.70 ±0.85 / 300.66 ms │ no change │
│ QQuery 16 │     63.27 / 65.56 ±1.72 / 68.16 ms │     64.61 / 65.84 ±1.40 / 68.48 ms │ no change │
│ QQuery 17 │  581.26 / 590.00 ±8.04 / 602.45 ms │  584.76 / 592.31 ±7.24 / 604.15 ms │ no change │
│ QQuery 18 │ 718.86 / 737.89 ±20.24 / 776.97 ms │ 710.80 / 745.82 ±20.59 / 773.48 ms │ no change │
│ QQuery 19 │ 229.34 / 245.65 ±13.91 / 262.89 ms │ 230.64 / 244.64 ±18.24 / 278.46 ms │ no change │
│ QQuery 20 │  261.26 / 270.48 ±6.91 / 279.51 ms │  267.15 / 274.16 ±4.89 / 280.55 ms │ no change │
│ QQuery 21 │  661.17 / 670.68 ±5.95 / 677.56 ms │  656.05 / 666.60 ±7.23 / 677.97 ms │ no change │
│ QQuery 22 │     60.01 / 62.48 ±1.72 / 64.91 ms │     57.11 / 60.00 ±2.51 / 64.07 ms │ no change │
└───────────┴────────────────────────────────────┴────────────────────────────────────┴───────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                              ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                              │ 6514.87ms │
│ Total Time (add-fully-ordered-groups-values)   │ 6554.34ms │
│ Average Time (HEAD)                            │  296.13ms │
│ Average Time (add-fully-ordered-groups-values) │  297.92ms │
│ Queries Faster                                 │         0 │
│ Queries Slower                                 │         0 │
│ Queries with No Change                         │        22 │
│ Queries with Failure                           │         0 │
└────────────────────────────────────────────────┴───────────┘

Memory Pool Peaks

Peak MemoryPool reservation per query — what DataFusion's accounting believes it reserved. Recorded only for benchmarks that write a results JSON (the dfbench suites, and the suites bench.sh runs through the Criterion SQL harness since #25644), and only under a memory limit: DATAFUSION_RUNTIME_MEMORY_LIMIT, or one the suite sets itself.

Base: 6924c6a (merge-base) | Changed: add-fully-ordered-groups-values

tpch10 — tpch_sf10

Query	Base	Changed	Change
Query 1	31.4 MiB	30.0 MiB	-4.4%
Query 2	114.3 MiB	112.5 MiB	-1.6%
Query 3	108.7 MiB	110.5 MiB	+1.7%
Query 4	42.5 MiB	42.5 MiB	+0.0%
Query 5	135.3 MiB	135.3 MiB	+0.0%
Query 6	832 B	832 B	+0.0%
Query 7	965.5 MiB	965.4 MiB	-0.0%
Query 8	374.2 MiB	374.2 MiB	+0.0%
Query 9	1.3 GiB	1.3 GiB	+0.0%
Query 10	484.9 MiB	480.1 MiB	-1.0%
Query 11	41.3 MiB	59.1 MiB	+43.0%
Query 12	27.1 MiB	36.2 MiB	+33.3%
Query 13	126.7 MiB	125.0 MiB	-1.4%
Query 14	86.9 MiB	86.9 MiB	+0.0%
Query 15	69.6 MiB	69.6 MiB	+0.0%
Query 16	178.3 MiB	179.7 MiB	+0.8%
Query 17	164.2 MiB	166.5 MiB	+1.4%
Query 18	3.7 GiB	3.7 GiB	+0.0%
Query 19	9.5 MiB	9.5 MiB	+0.0%
Query 20	311.8 MiB	308.1 MiB	-1.2%
Query 21	422.0 MiB	424.6 MiB	+0.6%
Query 22	30.3 MiB	40.2 MiB	+32.7%

Pool accounting vs. process RSS

Max pool peak is the largest reservation any single query in the run reached; peak RSS covers the whole invocation, including data loading and allocator retention, and the two high-water marks need not coincide in time. The gap is therefore an upper bound on what the pool did not account for, not a measurement of it.

Benchmark	Side	Max pool peak	Peak RSS	Gap	RSS / pool
tpch10	base (`6924c6a (merge-base)`)	3.7 GiB	4.8 GiB	1.2 GiB	1.3×
tpch10	changed (`add-fully-ordered-groups-values`)	3.7 GiB	5.3 GiB	1.7 GiB	1.5×

Resource Usage

tpch10 — base (merge-base)

Metric	Value
Wall time	35.0s
Peak memory	4.8 GiB
Avg memory	1.5 GiB
CPU user	332.6s
CPU sys	17.7s
Peak spill	0 B

tpch10 — branch

Metric	Value
Wall time	35.0s
Peak memory	5.3 GiB
Avg memory	1.5 GiB
CPU user	333.3s
CPU sys	18.3s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-10-01T10:41:37Z

🤖 Benchmark completed (GKE) | trigger

Instance: c4a-highmem-16-lssd (12 vCPU / 65 GiB)

Comparing add-fully-ordered-groups-values (8dec018) to 6924c6a (merge-base) diff

Run configuration

run benchmark external_aggr

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Details

Comparing HEAD and add-fully-ordered-groups-values
--------------------
Benchmark external_aggr.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃      HEAD ┃ add-fully-ordered-groups-values ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ Q1(64.0 MB)  │  51.60 ms │                        48.81 ms │ +1.06x faster │
│ Q1(32.0 MB)  │  49.08 ms │                        45.80 ms │ +1.07x faster │
│ Q1(16.0 MB)  │  50.44 ms │                        47.58 ms │ +1.06x faster │
│ Q2(512.0 MB) │ 270.15 ms │                       273.84 ms │     no change │
│ Q2(256.0 MB) │ 265.57 ms │                       271.09 ms │     no change │
│ Q2(128.0 MB) │ 244.82 ms │                       240.63 ms │     no change │
│ Q2(64.0 MB)  │ 237.81 ms │                       233.49 ms │     no change │
│ Q2(32.0 MB)  │ 300.01 ms │                       298.07 ms │     no change │
└──────────────┴───────────┴─────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                              ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                              │ 1469.47ms │
│ Total Time (add-fully-ordered-groups-values)   │ 1459.31ms │
│ Average Time (HEAD)                            │  183.68ms │
│ Average Time (add-fully-ordered-groups-values) │  182.41ms │
│ Queries Faster                                 │         3 │
│ Queries Slower                                 │         0 │
│ Queries with No Change                         │         5 │
│ Queries with Failure                           │         0 │
└────────────────────────────────────────────────┴───────────┘

Distribution per query (min / mean ±stddev / max):

Comparing HEAD and add-fully-ordered-groups-values
--------------------
Benchmark external_aggr.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃                               HEAD ┃    add-fully-ordered-groups-values ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ Q1(64.0 MB)  │     51.60 / 55.25 ±3.37 / 61.40 ms │     48.81 / 50.96 ±2.73 / 56.08 ms │ +1.08x faster │
│ Q1(32.0 MB)  │     49.08 / 51.12 ±1.24 / 52.35 ms │     45.80 / 47.16 ±1.07 / 48.43 ms │ +1.08x faster │
│ Q1(16.0 MB)  │     50.44 / 53.73 ±1.84 / 55.38 ms │     47.58 / 48.31 ±0.78 / 49.48 ms │ +1.11x faster │
│ Q2(512.0 MB) │  270.15 / 281.63 ±8.39 / 296.16 ms │  273.84 / 280.74 ±6.23 / 289.81 ms │     no change │
│ Q2(256.0 MB) │ 265.57 / 283.68 ±12.71 / 299.87 ms │ 271.09 / 287.99 ±13.24 / 302.90 ms │     no change │
│ Q2(128.0 MB) │  244.82 / 250.52 ±4.53 / 255.78 ms │ 240.63 / 251.94 ±12.05 / 272.02 ms │     no change │
│ Q2(64.0 MB)  │  237.81 / 239.40 ±1.14 / 241.37 ms │  233.49 / 237.40 ±2.40 / 240.33 ms │     no change │
│ Q2(32.0 MB)  │  300.01 / 301.62 ±1.33 / 303.43 ms │  298.07 / 301.08 ±3.32 / 307.01 ms │     no change │
└──────────────┴────────────────────────────────────┴────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                              ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                              │ 1516.95ms │
│ Total Time (add-fully-ordered-groups-values)   │ 1505.59ms │
│ Average Time (HEAD)                            │  189.62ms │
│ Average Time (add-fully-ordered-groups-values) │  188.20ms │
│ Queries Faster                                 │         3 │
│ Queries Slower                                 │         0 │
│ Queries with No Change                         │         5 │
│ Queries with Failure                           │         0 │
└────────────────────────────────────────────────┴───────────┘

Memory Pool Peaks

Peak MemoryPool reservation per query — what DataFusion's accounting believes it reserved. Recorded only for benchmarks that write a results JSON (the dfbench suites, and the suites bench.sh runs through the Criterion SQL harness since #25644), and only under a memory limit: DATAFUSION_RUNTIME_MEMORY_LIMIT, or one the suite sets itself.

Base: 6924c6a (merge-base) | Changed: add-fully-ordered-groups-values

external_aggr

Query	Base	Changed	Change
1(64.0 MB)	35.0 MiB	36.8 MiB	+5.0%
1(32.0 MB)	19.5 MiB	18.6 MiB	-4.5%
1(16.0 MB)	10.0 MiB	9.2 MiB	-7.8%
2(512.0 MB)	135.9 MiB	137.8 MiB	+1.5%
2(256.0 MB)	97.8 MiB	97.4 MiB	-0.4%
2(128.0 MB)	49.2 MiB	49.1 MiB	-0.3%
2(64.0 MB)	36.4 MiB	36.4 MiB	-0.0%
2(32.0 MB)	17.5 MiB	17.9 MiB	+2.1%

Pool accounting vs. process RSS

Max pool peak is the largest reservation any single query in the run reached; peak RSS covers the whole invocation, including data loading and allocator retention, and the two high-water marks need not coincide in time. The gap is therefore an upper bound on what the pool did not account for, not a measurement of it.

Benchmark	Side	Max pool peak	Peak RSS	Gap	RSS / pool
external_aggr	base (`6924c6a (merge-base)`)	135.9 MiB	426.1 MiB	290.2 MiB	3.1×
external_aggr	changed (`add-fully-ordered-groups-values`)	137.8 MiB	462.7 MiB	324.8 MiB	3.4×

Resource Usage

external_aggr — base (merge-base)

Metric	Value
Wall time	105.1s
Peak memory	426.1 MiB
Avg memory	29.6 MiB
CPU user	25.9s
CPU sys	3.8s
Peak spill	99.2 MiB

external_aggr — branch

Metric	Value
Wall time	105.0s
Peak memory	462.7 MiB
Avg memory	31.0 MiB
CPU user	25.8s
CPU sys	3.7s
Peak spill	92.4 MiB

File an issue against this benchmark runner

adriangbot · 2026-10-01T10:44:33Z

🤖 Benchmark completed (GKE) | trigger

Instance: c4a-highmem-16-lssd (12 vCPU / 65 GiB)

Comparing add-fully-ordered-groups-values (8dec018) to 6924c6a (merge-base) diff

Run configuration

run benchmark clickbench_partitioned
env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "16G"

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Details

Comparing HEAD and add-fully-ordered-groups-values
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃       HEAD ┃ add-fully-ordered-groups-values ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0  │    1.26 ms │                         1.26 ms │     no change │
│ QQuery 1  │   12.12 ms │                        12.01 ms │     no change │
│ QQuery 2  │   37.46 ms │                        37.52 ms │     no change │
│ QQuery 3  │   32.68 ms │                        32.08 ms │     no change │
│ QQuery 4  │  249.20 ms │                       246.00 ms │     no change │
│ QQuery 5  │  282.43 ms │                       280.10 ms │     no change │
│ QQuery 6  │    1.32 ms │                         1.32 ms │     no change │
│ QQuery 7  │   13.26 ms │                        13.34 ms │     no change │
│ QQuery 8  │  351.39 ms │                       350.63 ms │     no change │
│ QQuery 9  │  489.03 ms │                       484.54 ms │     no change │
│ QQuery 10 │   65.81 ms │                        65.41 ms │     no change │
│ QQuery 11 │   75.41 ms │                        75.59 ms │     no change │
│ QQuery 12 │  273.36 ms │                       275.48 ms │     no change │
│ QQuery 13 │  374.69 ms │                       374.28 ms │     no change │
│ QQuery 14 │  289.37 ms │                       292.06 ms │     no change │
│ QQuery 15 │  298.32 ms │                       305.05 ms │     no change │
│ QQuery 16 │  645.22 ms │                       646.52 ms │     no change │
│ QQuery 17 │  640.42 ms │                       649.44 ms │     no change │
│ QQuery 18 │ 1318.91 ms │                      1323.04 ms │     no change │
│ QQuery 19 │   28.37 ms │                        28.19 ms │     no change │
│ QQuery 20 │  527.75 ms │                       524.79 ms │     no change │
│ QQuery 21 │  519.09 ms │                       515.20 ms │     no change │
│ QQuery 22 │ 1032.30 ms │                      1023.63 ms │     no change │
│ QQuery 23 │ 3175.74 ms │                      3193.65 ms │     no change │
│ QQuery 24 │   40.96 ms │                        40.77 ms │     no change │
│ QQuery 25 │  107.54 ms │                       105.51 ms │     no change │
│ QQuery 26 │   41.54 ms │                        41.05 ms │     no change │
│ QQuery 27 │  524.74 ms │                       527.16 ms │     no change │
│ QQuery 28 │ 2853.83 ms │                      2962.69 ms │     no change │
│ QQuery 29 │   42.75 ms │                        42.75 ms │     no change │
│ QQuery 30 │  321.00 ms │                       318.52 ms │     no change │
│ QQuery 31 │  285.55 ms │                       287.83 ms │     no change │
│ QQuery 32 │ 4159.17 ms │                      4050.78 ms │     no change │
│ QQuery 33 │ 1625.77 ms │                      1617.60 ms │     no change │
│ QQuery 34 │ 1641.08 ms │                      1629.41 ms │     no change │
│ QQuery 35 │  319.47 ms │                       323.93 ms │     no change │
│ QQuery 36 │   68.99 ms │                        70.83 ms │     no change │
│ QQuery 37 │   35.53 ms │                        35.86 ms │     no change │
│ QQuery 38 │   48.44 ms │                        44.76 ms │ +1.08x faster │
│ QQuery 39 │  150.92 ms │                       151.39 ms │     no change │
│ QQuery 40 │   15.59 ms │                        14.77 ms │ +1.06x faster │
│ QQuery 41 │   14.53 ms │                        14.20 ms │     no change │
│ QQuery 42 │   13.84 ms │                        13.89 ms │     no change │
└───────────┴────────────┴─────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                              ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                              │ 23046.16ms │
│ Total Time (add-fully-ordered-groups-values)   │ 23044.80ms │
│ Average Time (HEAD)                            │   535.96ms │
│ Average Time (add-fully-ordered-groups-values) │   535.93ms │
│ Queries Faster                                 │          2 │
│ Queries Slower                                 │          0 │
│ Queries with No Change                         │         41 │
│ Queries with Failure                           │          0 │
└────────────────────────────────────────────────┴────────────┘

Distribution per query (min / mean ±stddev / max):

Comparing HEAD and add-fully-ordered-groups-values
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃                                   HEAD ┃        add-fully-ordered-groups-values ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0  │           1.26 / 4.15 ±5.62 / 15.40 ms │           1.26 / 4.17 ±5.62 / 15.40 ms │     no change │
│ QQuery 1  │         12.12 / 12.30 ±0.12 / 12.47 ms │         12.01 / 12.30 ±0.16 / 12.45 ms │     no change │
│ QQuery 2  │         37.46 / 37.81 ±0.28 / 38.21 ms │         37.52 / 37.73 ±0.25 / 38.18 ms │     no change │
│ QQuery 3  │         32.68 / 33.30 ±0.83 / 34.85 ms │         32.08 / 32.62 ±0.43 / 33.24 ms │     no change │
│ QQuery 4  │      249.20 / 255.48 ±4.14 / 261.60 ms │      246.00 / 255.54 ±5.75 / 262.86 ms │     no change │
│ QQuery 5  │      282.43 / 285.74 ±2.92 / 290.95 ms │      280.10 / 286.46 ±5.07 / 295.67 ms │     no change │
│ QQuery 6  │            1.32 / 1.47 ±0.22 / 1.91 ms │            1.32 / 1.47 ±0.24 / 1.94 ms │     no change │
│ QQuery 7  │         13.26 / 13.74 ±0.30 / 14.20 ms │         13.34 / 13.61 ±0.16 / 13.78 ms │     no change │
│ QQuery 8  │      351.39 / 354.73 ±2.89 / 359.87 ms │      350.63 / 354.24 ±3.50 / 360.79 ms │     no change │
│ QQuery 9  │      489.03 / 494.76 ±4.71 / 500.76 ms │      484.54 / 493.91 ±6.31 / 500.91 ms │     no change │
│ QQuery 10 │         65.81 / 67.26 ±1.74 / 70.57 ms │         65.41 / 66.23 ±0.66 / 67.07 ms │     no change │
│ QQuery 11 │         75.41 / 76.33 ±0.92 / 78.02 ms │         75.59 / 76.67 ±1.38 / 79.38 ms │     no change │
│ QQuery 12 │      273.36 / 280.25 ±5.39 / 286.68 ms │      275.48 / 281.49 ±5.11 / 287.98 ms │     no change │
│ QQuery 13 │      374.69 / 380.53 ±3.40 / 385.32 ms │     374.28 / 388.56 ±15.47 / 417.62 ms │     no change │
│ QQuery 14 │      289.37 / 295.94 ±4.88 / 303.73 ms │      292.06 / 299.29 ±6.15 / 308.12 ms │     no change │
│ QQuery 15 │      298.32 / 307.50 ±6.16 / 315.14 ms │     305.05 / 317.26 ±10.96 / 336.83 ms │     no change │
│ QQuery 16 │      645.22 / 655.92 ±8.87 / 670.27 ms │      646.52 / 655.05 ±5.43 / 663.61 ms │     no change │
│ QQuery 17 │     640.42 / 656.99 ±14.40 / 681.91 ms │     649.44 / 664.76 ±14.24 / 685.81 ms │     no change │
│ QQuery 18 │  1318.91 / 1342.44 ±12.93 / 1355.09 ms │  1323.04 / 1351.24 ±21.62 / 1378.40 ms │     no change │
│ QQuery 19 │         28.37 / 32.59 ±5.62 / 43.52 ms │        28.19 / 37.13 ±11.95 / 58.25 ms │  1.14x slower │
│ QQuery 20 │      527.75 / 536.42 ±7.74 / 549.94 ms │      524.79 / 528.87 ±5.42 / 539.35 ms │     no change │
│ QQuery 21 │     519.09 / 533.96 ±16.48 / 565.93 ms │     515.20 / 530.46 ±12.72 / 550.19 ms │     no change │
│ QQuery 22 │   1032.30 / 1041.66 ±5.29 / 1048.32 ms │  1023.63 / 1087.21 ±53.85 / 1168.72 ms │     no change │
│ QQuery 23 │ 3175.74 / 3352.28 ±101.47 / 3450.58 ms │ 3193.65 / 3443.89 ±291.06 / 3981.25 ms │     no change │
│ QQuery 24 │       40.96 / 83.62 ±39.40 / 141.83 ms │         40.77 / 41.12 ±0.31 / 41.65 ms │ +2.03x faster │
│ QQuery 25 │     107.54 / 151.03 ±81.70 / 314.35 ms │    105.51 / 227.15 ±229.97 / 686.92 ms │  1.50x slower │
│ QQuery 26 │         41.54 / 42.62 ±0.60 / 43.38 ms │         41.05 / 41.84 ±0.57 / 42.81 ms │     no change │
│ QQuery 27 │      524.74 / 529.25 ±2.88 / 532.78 ms │     527.16 / 548.11 ±27.43 / 601.97 ms │     no change │
│ QQuery 28 │  2853.83 / 2888.10 ±18.38 / 2907.17 ms │  2962.69 / 3003.61 ±50.61 / 3098.99 ms │     no change │
│ QQuery 29 │         42.75 / 49.83 ±8.58 / 61.24 ms │       42.75 / 65.62 ±34.93 / 134.15 ms │  1.32x slower │
│ QQuery 30 │      321.00 / 327.58 ±6.03 / 338.19 ms │     318.52 / 330.57 ±17.10 / 363.45 ms │     no change │
│ QQuery 31 │      285.55 / 298.97 ±7.36 / 305.97 ms │      287.83 / 305.75 ±9.27 / 313.51 ms │     no change │
│ QQuery 32 │  4159.17 / 4304.12 ±87.17 / 4404.63 ms │  4050.78 / 4229.73 ±94.71 / 4314.41 ms │     no change │
│ QQuery 33 │  1625.77 / 1702.05 ±92.97 / 1882.42 ms │  1617.60 / 1651.82 ±20.37 / 1675.50 ms │     no change │
│ QQuery 34 │  1641.08 / 1721.52 ±86.45 / 1857.54 ms │  1629.41 / 1707.33 ±79.33 / 1859.34 ms │     no change │
│ QQuery 35 │    319.47 / 456.25 ±228.86 / 912.89 ms │     323.93 / 386.64 ±82.00 / 545.32 ms │ +1.18x faster │
│ QQuery 36 │         68.99 / 72.38 ±2.29 / 75.23 ms │         70.83 / 74.47 ±2.65 / 78.13 ms │     no change │
│ QQuery 37 │         35.53 / 38.31 ±3.52 / 45.20 ms │         35.86 / 40.90 ±4.85 / 48.50 ms │  1.07x slower │
│ QQuery 38 │        48.44 / 63.34 ±17.58 / 87.96 ms │         44.76 / 48.54 ±4.22 / 55.08 ms │ +1.30x faster │
│ QQuery 39 │     150.92 / 165.38 ±13.80 / 191.30 ms │     151.39 / 164.99 ±20.03 / 204.66 ms │     no change │
│ QQuery 40 │         15.59 / 21.57 ±5.22 / 28.83 ms │         14.77 / 16.43 ±3.03 / 22.48 ms │ +1.31x faster │
│ QQuery 41 │         14.53 / 15.07 ±0.36 / 15.61 ms │         14.20 / 15.63 ±2.12 / 19.78 ms │     no change │
│ QQuery 42 │         13.84 / 13.96 ±0.09 / 14.10 ms │         13.89 / 14.09 ±0.11 / 14.19 ms │     no change │
└───────────┴────────────────────────────────────────┴────────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                              ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                              │ 23998.51ms │
│ Total Time (add-fully-ordered-groups-values)   │ 24134.54ms │
│ Average Time (HEAD)                            │   558.10ms │
│ Average Time (add-fully-ordered-groups-values) │   561.27ms │
│ Queries Faster                                 │          4 │
│ Queries Slower                                 │          4 │
│ Queries with No Change                         │         35 │
│ Queries with Failure                           │          0 │
└────────────────────────────────────────────────┴────────────┘

Memory Pool Peaks

Peak MemoryPool reservation per query — what DataFusion's accounting believes it reserved. Recorded only for benchmarks that write a results JSON (the dfbench suites, and the suites bench.sh runs through the Criterion SQL harness since #25644), and only under a memory limit: DATAFUSION_RUNTIME_MEMORY_LIMIT, or one the suite sets itself.

Base: 6924c6a (merge-base) | Changed: add-fully-ordered-groups-values

clickbench_partitioned

Query	Base	Changed	Change
Query 0	0 B	0 B	0.0%
Query 1	104 B	104 B	+0.0%
Query 2	936 B	936 B	+0.0%
Query 3	312 B	312 B	+0.0%
Query 4	775.0 MiB	773.0 MiB	-0.2%
Query 5	1.2 GiB	1.2 GiB	+0.7%
Query 6	0 B	0 B	0.0%
Query 7	60.0 MiB	60.0 MiB	+0.0%
Query 8	946.2 MiB	941.3 MiB	-0.5%
Query 9	593.6 MiB	551.7 MiB	-7.0%
Query 10	120.1 MiB	116.8 MiB	-2.8%
Query 11	121.2 MiB	121.6 MiB	+0.3%
Query 12	1.4 GiB	1.3 GiB	-0.6%
Query 13	1.6 GiB	1.6 GiB	-0.6%
Query 14	1.3 GiB	1.3 GiB	+4.3%
Query 15	1.2 GiB	1.2 GiB	-1.3%
Query 16	3.5 GiB	3.4 GiB	-0.8%
Query 17	3.4 GiB	3.4 GiB	+0.9%
Query 18	7.9 GiB	7.9 GiB	+0.1%
Query 19	0 B	0 B	0.0%
Query 20	104 B	104 B	+0.0%
Query 21	3.7 MiB	3.7 MiB	-0.1%
Query 22	3.4 MiB	3.4 MiB	+0.4%
Query 23	3.9 GiB	3.1 GiB	-21.4%
Query 24	59.1 MiB	57.7 MiB	-2.5%
Query 25	161.5 MiB	160.0 MiB	-1.0%
Query 26	59.0 MiB	59.7 MiB	+1.1%
Query 27	2.2 MiB	2.5 MiB	+9.8%
Query 28	1.8 GiB	1.8 GiB	-0.3%
Query 29	624 B	624 B	+0.0%
Query 30	706.4 MiB	724.3 MiB	+2.5%
Query 31	1.5 GiB	1.5 GiB	-0.1%
Query 32	3.6 GiB	3.6 GiB	+0.2%
Query 33	7.7 GiB	7.1 GiB	-7.9%
Query 34	7.2 GiB	7.4 GiB	+2.0%
Query 35	607.5 MiB	596.7 MiB	-1.8%
Query 36	114.2 MiB	116.0 MiB	+1.5%
Query 37	6.6 MiB	6.6 MiB	+0.0%
Query 38	6.0 MiB	5.5 MiB	-9.0%
Query 39	297.8 MiB	296.9 MiB	-0.3%
Query 40	2.0 MiB	1.8 MiB	-9.2%
Query 41	3.1 MiB	3.1 MiB	+0.0%
Query 42	1.6 MiB	1.7 MiB	+8.7%

Pool accounting vs. process RSS

Max pool peak is the largest reservation any single query in the run reached; peak RSS covers the whole invocation, including data loading and allocator retention, and the two high-water marks need not coincide in time. The gap is therefore an upper bound on what the pool did not account for, not a measurement of it.

Benchmark	Side	Max pool peak	Peak RSS	Gap	RSS / pool
clickbench_partitioned	base (`6924c6a (merge-base)`)	7.9 GiB	18.2 GiB	10.3 GiB	2.3×
clickbench_partitioned	changed (`add-fully-ordered-groups-values`)	7.9 GiB	18.4 GiB	10.4 GiB	2.3×

Resource Usage

clickbench_partitioned — base (merge-base)

Metric	Value
Wall time	125.0s
Peak memory	18.2 GiB
Avg memory	6.0 GiB
CPU user	1170.3s
CPU sys	123.5s
Peak spill	4.2 GiB

clickbench_partitioned — branch

Metric	Value
Wall time	125.0s
Peak memory	18.4 GiB
Avg memory	6.0 GiB
CPU user	1175.5s
CPU sys	127.0s
Peak spill	4.2 GiB

File an issue against this benchmark runner

adriangbot · 2026-10-01T10:45:48Z

🤖 Benchmark completed (GKE) | trigger

Instance: c4a-highmem-16-lssd (12 vCPU / 65 GiB)

Comparing add-fully-ordered-groups-values (8dec018) to 6924c6a (merge-base) diff

Run configuration

run benchmark h2o_medium
env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "16G"

CPU Details (lscpu)

Architecture:                            aarch64
CPU op-mode(s):                          64-bit
Byte Order:                              Little Endian
CPU(s):                                  16
On-line CPU(s) list:                     0-15
Vendor ID:                               ARM
Model name:                              Neoverse-V2
Model:                                   1
Thread(s) per core:                      1
Core(s) per cluster:                     16
Socket(s):                               -
Cluster(s):                              1
Stepping:                                r0p1
BogoMIPS:                                2000.00
Flags:                                   fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 sve asimdfhm dit uscat ilrcpc flagm sb paca pacg dcpodp sve2 sveaes svepmull svebitperm svesha3 svesm4 flagm2 frint svei8mm svebf16 i8mm bf16 dgh rng bti
L1d cache:                               1 MiB (16 instances)
L1i cache:                               1 MiB (16 instances)
L2 cache:                                32 MiB (16 instances)
L3 cache:                                80 MiB (1 instance)
NUMA node(s):                            1
NUMA node0 CPU(s):                       0-15
Vulnerability Gather data sampling:      Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit:             Not affected
Vulnerability L1tf:                      Not affected
Vulnerability Mds:                       Not affected
Vulnerability Meltdown:                  Not affected
Vulnerability Mmio stale data:           Not affected
Vulnerability Reg file data sampling:    Not affected
Vulnerability Retbleed:                  Not affected
Vulnerability Spec rstack overflow:      Not affected
Vulnerability Spec store bypass:         Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1:                Mitigation; __user pointer sanitization
Vulnerability Spectre v2:                Mitigation; CSV2, BHB
Vulnerability Srbds:                     Not affected
Vulnerability Tsa:                       Not affected
Vulnerability Tsx async abort:           Not affected
Vulnerability Vmscape:                   Not affected

Details

Comparing HEAD and add-fully-ordered-groups-values
--------------------
Benchmark h2o.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃        HEAD ┃ add-fully-ordered-groups-values ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1  │  1093.87 ms │                      1016.56 ms │ +1.08x faster │
│ QQuery 2  │  2327.63 ms │                      2253.19 ms │     no change │
│ QQuery 3  │  2021.09 ms │                      1929.00 ms │     no change │
│ QQuery 4  │  1398.09 ms │                      1315.96 ms │ +1.06x faster │
│ QQuery 5  │  1887.53 ms │                      1839.49 ms │     no change │
│ QQuery 6  │  1608.79 ms │                      1532.76 ms │     no change │
│ QQuery 7  │  1855.32 ms │                      1791.01 ms │     no change │
│ QQuery 8  │  3576.58 ms │                      3385.04 ms │ +1.06x faster │
│ QQuery 9  │  7416.86 ms │                      7274.96 ms │     no change │
│ QQuery 10 │ 12366.64 ms │                     12720.57 ms │     no change │
└───────────┴─────────────┴─────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                              ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                              │ 35552.40ms │
│ Total Time (add-fully-ordered-groups-values)   │ 35058.54ms │
│ Average Time (HEAD)                            │  3555.24ms │
│ Average Time (add-fully-ordered-groups-values) │  3505.85ms │
│ Queries Faster                                 │          3 │
│ Queries Slower                                 │          0 │
│ Queries with No Change                         │          7 │
│ Queries with Failure                           │          0 │
└────────────────────────────────────────────────┴────────────┘

Distribution per query (min / mean ±stddev / max):

Comparing HEAD and add-fully-ordered-groups-values
--------------------
Benchmark h2o.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃                                      HEAD ┃           add-fully-ordered-groups-values ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1  │     1093.87 / 1103.02 ±12.26 / 1120.35 ms │      1016.56 / 1019.15 ±2.93 / 1023.25 ms │ +1.08x faster │
│ QQuery 2  │     2327.63 / 2363.46 ±28.55 / 2397.49 ms │     2253.19 / 2271.28 ±12.84 / 2281.77 ms │     no change │
│ QQuery 3  │     2021.09 / 2042.59 ±15.49 / 2056.99 ms │     1929.00 / 1967.55 ±41.50 / 2025.15 ms │     no change │
│ QQuery 4  │      1398.09 / 1400.38 ±2.91 / 1404.50 ms │      1315.96 / 1321.78 ±7.47 / 1332.33 ms │ +1.06x faster │
│ QQuery 5  │     1887.53 / 1903.74 ±16.52 / 1926.42 ms │     1839.49 / 1852.42 ±10.97 / 1866.30 ms │     no change │
│ QQuery 6  │      1608.79 / 1612.55 ±4.16 / 1618.35 ms │     1532.76 / 1552.36 ±14.00 / 1564.57 ms │     no change │
│ QQuery 7  │     1855.32 / 1898.67 ±36.82 / 1945.33 ms │     1791.01 / 1810.60 ±13.85 / 1820.42 ms │     no change │
│ QQuery 8  │     3576.58 / 3623.87 ±45.40 / 3685.12 ms │     3385.04 / 3447.40 ±46.65 / 3497.22 ms │     no change │
│ QQuery 9  │     7416.86 / 7434.77 ±15.13 / 7453.86 ms │     7274.96 / 7342.44 ±55.05 / 7409.80 ms │     no change │
│ QQuery 10 │ 12366.64 / 12725.20 ±256.55 / 12952.42 ms │ 12720.57 / 12886.77 ±202.31 / 13171.56 ms │     no change │
└───────────┴───────────────────────────────────────────┴───────────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                              ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                              │ 36108.27ms │
│ Total Time (add-fully-ordered-groups-values)   │ 35471.74ms │
│ Average Time (HEAD)                            │  3610.83ms │
│ Average Time (add-fully-ordered-groups-values) │  3547.17ms │
│ Queries Faster                                 │          2 │
│ Queries Slower                                 │          0 │
│ Queries with No Change                         │          8 │
│ Queries with Failure                           │          0 │
└────────────────────────────────────────────────┴────────────┘

Memory Pool Peaks

Peak MemoryPool reservation per query — what DataFusion's accounting believes it reserved. Recorded only for benchmarks that write a results JSON (the dfbench suites, and the suites bench.sh runs through the Criterion SQL harness since #25644), and only under a memory limit: DATAFUSION_RUNTIME_MEMORY_LIMIT, or one the suite sets itself.

Base: 6924c6a (merge-base) | Changed: add-fully-ordered-groups-values

h2o_medium — h2o

Query	Base	Changed	Change
Query 1	1.9 MiB	1.9 MiB	+0.0%
Query 2	4.2 GiB	4.2 GiB	-0.0%
Query 3	142.3 MiB	145.3 MiB	+2.1%
Query 4	4.4 MiB	4.0 MiB	-9.9%
Query 5	98.1 MiB	97.4 MiB	-0.7%
Query 6	1.1 GiB	1.0 GiB	-5.5%
Query 7	122.9 MiB	119.8 MiB	-2.6%
Query 8	3.9 GiB	3.9 GiB	+0.0%
Query 9	4.7 GiB	4.7 GiB	+0.2%
Query 10	5.5 GiB	5.5 GiB	+0.1%

Pool accounting vs. process RSS

Max pool peak is the largest reservation any single query in the run reached; peak RSS covers the whole invocation, including data loading and allocator retention, and the two high-water marks need not coincide in time. The gap is therefore an upper bound on what the pool did not account for, not a measurement of it.

Benchmark	Side	Max pool peak	Peak RSS	Gap	RSS / pool
h2o_medium	base (`6924c6a (merge-base)`)	5.5 GiB	12.7 GiB	7.2 GiB	2.3×
h2o_medium	changed (`add-fully-ordered-groups-values`)	5.5 GiB	11.9 GiB	6.5 GiB	2.2×

Resource Usage

h2o_medium — base (merge-base)

Metric	Value
Wall time	110.0s
Peak memory	12.7 GiB
Avg memory	3.8 GiB
CPU user	1148.3s
CPU sys	91.2s
Peak spill	7.3 GiB

h2o_medium — branch

Metric	Value
Wall time	110.0s
Peak memory	11.9 GiB
Avg memory	3.8 GiB
CPU user	1124.8s
CPU sys	92.7s
Peak spill	7.3 GiB

File an issue against this benchmark runner

perf: add fully ordered group values implementation for primitive

eb88c11

This allows for optimized implementation given that we know that the input is ordered

github-actions Bot added the physical-plan Changes to the physical-plan crate label Oct 1, 2026

fix compile

7829192

connonicalize and extract

8dec018

rluvaton added 3 commits October 1, 2026 13:53

cleanup

90651d4

lint

65e2981

fix

274e3b7

Conversation

rluvaton commented Oct 1, 2026

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

What is the testing strategy for this PR?

Are there any user-facing changes?

Uh oh!

rluvaton commented Oct 1, 2026

Uh oh!

rluvaton commented Oct 1, 2026

Uh oh!

rluvaton commented Oct 1, 2026

Uh oh!

rluvaton commented Oct 1, 2026

Uh oh!

adriangbot commented Oct 1, 2026

Uh oh!

adriangbot commented Oct 1, 2026

Uh oh!

adriangbot commented Oct 1, 2026

Uh oh!

adriangbot commented Oct 1, 2026

Uh oh!

adriangbot commented Oct 1, 2026

Uh oh!

adriangbot commented Oct 1, 2026

Uh oh!

adriangbot commented Oct 1, 2026

Uh oh!

adriangbot commented Oct 1, 2026

Uh oh!

adriangbot commented Oct 1, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants