bench: isolate Partial aggregation gains without Final buckets (do not merge) by Rachelint · Pull Request #26081 · apache/datafusion

Rachelint · 2026-10-06T11:21:04Z

Which issue does this PR close?

None. Benchmark-only ablation for #25959; do not merge.

Rationale for this change

Measure whether the new Partial working-set policy retains its benefit when Final uses ordinary hash aggregation. The earlier requests on #25959 failed during checkout because the runner only fetched that PR history. This draft makes both experimental refs reachable.

What changes are included in this PR?

126109f50158b9dbb4e9fd13ce46d9520268b789: old Partial policy from cf6819d, with Final bucketing disabled.
70ac2eed7ad82bd0915ad526cd735bdcacbf7458: new Partial policy from efd384f, with Final bucketing disabled.
The merge commit preserves the new variant tree and includes the baseline in its ancestry for the runner.
The only ablation change forces the Final constructor bucket threshold to zero; Partial still reads the experimental configuration. Original perf: route final aggregate buckets by column #25959 is unchanged.

What is the testing strategy for this PR?

Formatting and focused diff inspection passed. Two partitioned ClickBench comparisons: old versus new Partial with Final bucketing disabled on both sides; then complete efd384f versus new Partial without Final buckets. Shared threshold: 262144. No full test suite rerun for this experimental switch.

Are there any user-facing changes?

Experiment only, not intended for merge.

…, off by default)

…threshold while its groups do not recur

…ting buckets of unique groups

…f the threshold

…e its input holds them

Rachelint · 2026-10-06T11:21:21Z

run benchmark clickbench_partitioned

baseline:
ref: 126109f
changed:
ref: 70ac2ee
env:
DATAFUSION_EXECUTION_HASH_AGGREGATE_BUCKET_THRESHOLD: "262144"

Rachelint · 2026-10-06T11:21:42Z

run benchmark clickbench_partitioned

baseline:
ref: efd384f
changed:
ref: 70ac2ee
env:
DATAFUSION_EXECUTION_HASH_AGGREGATE_BUCKET_THRESHOLD: "262144"

adriangbot · 2026-10-06T11:26:09Z

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c6015170237-3094-qxtrm 6.12.94+ #1 SMP Fri Aug 21 08:00:16 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 70ac2ee (70ac2ee) to 126109f diff

Run configuration

run benchmark clickbench_partitioned
env:
  DATAFUSION_EXECUTION_HASH_AGGREGATE_BUCKET_THRESHOLD: "262144"
baseline:
  ref: "126109f50158b9dbb4e9fd13ce46d9520268b789"
changed:
  ref: "70ac2eed7ad82bd0915ad526cd735bdcacbf7458"

Results will be posted here when complete

File an issue against this benchmark runner

adriangbot · 2026-10-06T11:35:20Z

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c6015176764-3095-pkdmg 6.12.94+ #1 SMP Fri Aug 21 08:00:16 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 70ac2ee (70ac2ee) to efd384f diff

Run configuration

run benchmark clickbench_partitioned
env:
  DATAFUSION_EXECUTION_HASH_AGGREGATE_BUCKET_THRESHOLD: "262144"
baseline:
  ref: "efd384f9fe0362c7cd9811f860837e6d7e425d11"
changed:
  ref: "70ac2eed7ad82bd0915ad526cd735bdcacbf7458"

Results will be posted here when complete

File an issue against this benchmark runner

adriangbot · 2026-10-06T11:37:58Z

🤖 Benchmark completed (GKE) | trigger

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

Comparing 70ac2ee (70ac2ee) to 126109f diff

Run configuration

run benchmark clickbench_partitioned
env:
  DATAFUSION_EXECUTION_HASH_AGGREGATE_BUCKET_THRESHOLD: "262144"
baseline:
  ref: "126109f50158b9dbb4e9fd13ce46d9520268b789"
changed:
  ref: "70ac2eed7ad82bd0915ad526cd735bdcacbf7458"

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 codex_ablation-partial-new-no-final-buckets
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃       HEAD ┃ codex_ablation-partial-new-no-final-buckets ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0  │    1.24 ms │                                     1.24 ms │     no change │
│ QQuery 1  │   11.71 ms │                                    11.69 ms │     no change │
│ QQuery 2  │   35.48 ms │                                    35.47 ms │     no change │
│ QQuery 3  │   30.75 ms │                                    31.02 ms │     no change │
│ QQuery 4  │  247.58 ms │                                   220.41 ms │ +1.12x faster │
│ QQuery 5  │  283.10 ms │                                   258.61 ms │ +1.09x faster │
│ QQuery 6  │    1.29 ms │                                     1.30 ms │     no change │
│ QQuery 7  │   12.81 ms │                                    12.97 ms │     no change │
│ QQuery 8  │  348.13 ms │                                   317.40 ms │ +1.10x faster │
│ QQuery 9  │  478.06 ms │                                   475.84 ms │     no change │
│ QQuery 10 │   64.12 ms │                                    65.06 ms │     no change │
│ QQuery 11 │   74.45 ms │                                    76.25 ms │     no change │
│ QQuery 12 │  266.86 ms │                                   249.83 ms │ +1.07x faster │
│ QQuery 13 │  365.23 ms │                                   365.10 ms │     no change │
│ QQuery 14 │  285.18 ms │                                   273.87 ms │     no change │
│ QQuery 15 │  291.29 ms │                                   251.62 ms │ +1.16x faster │
│ QQuery 16 │  636.93 ms │                                   573.54 ms │ +1.11x faster │
│ QQuery 17 │  639.05 ms │                                   570.46 ms │ +1.12x faster │
│ QQuery 18 │ 1242.48 ms │                                  1079.02 ms │ +1.15x faster │
│ QQuery 19 │   27.21 ms │                                    27.58 ms │     no change │
│ QQuery 20 │  523.05 ms │                                   521.63 ms │     no change │
│ QQuery 21 │  520.61 ms │                                   523.55 ms │     no change │
│ QQuery 22 │ 1000.47 ms │                                   982.71 ms │     no change │
│ QQuery 23 │ 3127.98 ms │                                  3123.42 ms │     no change │
│ QQuery 24 │   40.80 ms │                                    41.81 ms │     no change │
│ QQuery 25 │  104.09 ms │                                   104.44 ms │     no change │
│ QQuery 26 │   41.07 ms │                                    41.04 ms │     no change │
│ QQuery 27 │  514.54 ms │                                   513.75 ms │     no change │
│ QQuery 28 │ 2904.86 ms │                                  2864.90 ms │     no change │
│ QQuery 29 │   41.36 ms │                                    40.71 ms │     no change │
│ QQuery 30 │  316.29 ms │                                   288.07 ms │ +1.10x faster │
│ QQuery 31 │  283.18 ms │                                   273.67 ms │     no change │
│ QQuery 32 │ 1057.52 ms │                                  1050.60 ms │     no change │
│ QQuery 33 │ 1505.99 ms │                                  1359.73 ms │ +1.11x faster │
│ QQuery 34 │ 1516.01 ms │                                  1330.57 ms │ +1.14x faster │
│ QQuery 35 │  282.28 ms │                                   249.50 ms │ +1.13x faster │
│ QQuery 36 │   64.82 ms │                                    61.89 ms │     no change │
│ QQuery 37 │   35.25 ms │                                    34.59 ms │     no change │
│ QQuery 38 │   42.75 ms │                                    44.88 ms │     no change │
│ QQuery 39 │  135.87 ms │                                   140.62 ms │     no change │
│ QQuery 40 │   14.15 ms │                                    15.01 ms │  1.06x slower │
│ QQuery 41 │   13.29 ms │                                    14.13 ms │  1.06x slower │
│ QQuery 42 │   13.26 ms │                                    13.46 ms │     no change │
└───────────┴────────────┴─────────────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                          ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                          │ 19442.42ms │
│ Total Time (codex_ablation-partial-new-no-final-buckets)   │ 18532.95ms │
│ Average Time (HEAD)                                        │   452.15ms │
│ Average Time (codex_ablation-partial-new-no-final-buckets) │   431.00ms │
│ Queries Faster                                             │         12 │
│ Queries Slower                                             │          2 │
│ Queries with No Change                                     │         29 │
│ Queries with Failure                                       │          0 │
└────────────────────────────────────────────────────────────┴────────────┘

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

Comparing HEAD and codex_ablation-partial-new-no-final-buckets
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃                                  HEAD ┃ codex_ablation-partial-new-no-final-buckets ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0  │          1.24 / 4.10 ±5.61 / 15.32 ms │                1.24 / 3.98 ±5.41 / 14.81 ms │     no change │
│ QQuery 1  │        11.71 / 12.13 ±0.26 / 12.48 ms │              11.69 / 11.93 ±0.24 / 12.30 ms │     no change │
│ QQuery 2  │        35.48 / 35.71 ±0.22 / 35.99 ms │              35.47 / 35.88 ±0.25 / 36.08 ms │     no change │
│ QQuery 3  │        30.75 / 31.34 ±0.75 / 32.78 ms │              31.02 / 31.29 ±0.26 / 31.74 ms │     no change │
│ QQuery 4  │     247.58 / 252.04 ±2.91 / 255.49 ms │           220.41 / 225.67 ±3.78 / 229.96 ms │ +1.12x faster │
│ QQuery 5  │     283.10 / 289.93 ±6.93 / 303.24 ms │           258.61 / 263.07 ±4.02 / 269.86 ms │ +1.10x faster │
│ QQuery 6  │           1.29 / 1.44 ±0.23 / 1.89 ms │                 1.30 / 1.45 ±0.22 / 1.89 ms │     no change │
│ QQuery 7  │        12.81 / 16.09 ±6.30 / 28.69 ms │              12.97 / 13.06 ±0.08 / 13.21 ms │ +1.23x faster │
│ QQuery 8  │     348.13 / 357.70 ±8.08 / 369.33 ms │           317.40 / 320.85 ±2.52 / 325.27 ms │ +1.11x faster │
│ QQuery 9  │     478.06 / 484.51 ±7.31 / 498.41 ms │           475.84 / 482.66 ±8.58 / 499.19 ms │     no change │
│ QQuery 10 │        64.12 / 67.15 ±4.94 / 76.98 ms │             65.06 / 71.64 ±12.19 / 96.00 ms │  1.07x slower │
│ QQuery 11 │        74.45 / 74.80 ±0.34 / 75.26 ms │              76.25 / 77.24 ±1.14 / 79.38 ms │     no change │
│ QQuery 12 │     266.86 / 281.22 ±9.34 / 296.10 ms │           249.83 / 251.88 ±1.75 / 254.82 ms │ +1.12x faster │
│ QQuery 13 │     365.23 / 371.87 ±5.59 / 379.23 ms │           365.10 / 369.45 ±3.74 / 375.77 ms │     no change │
│ QQuery 14 │    285.18 / 299.88 ±11.48 / 314.03 ms │           273.87 / 284.17 ±9.14 / 297.95 ms │ +1.06x faster │
│ QQuery 15 │     291.29 / 299.55 ±4.80 / 304.99 ms │           251.62 / 256.42 ±4.75 / 263.28 ms │ +1.17x faster │
│ QQuery 16 │    636.93 / 650.98 ±10.75 / 669.54 ms │          573.54 / 584.33 ±10.53 / 602.27 ms │ +1.11x faster │
│ QQuery 17 │    639.05 / 648.69 ±12.56 / 672.57 ms │          570.46 / 580.03 ±12.14 / 603.74 ms │ +1.12x faster │
│ QQuery 18 │  1242.48 / 1248.72 ±3.95 / 1252.77 ms │       1079.02 / 1094.66 ±11.89 / 1109.90 ms │ +1.14x faster │
│ QQuery 19 │       27.21 / 34.29 ±13.92 / 62.12 ms │              27.58 / 29.92 ±4.33 / 38.58 ms │ +1.15x faster │
│ QQuery 20 │     523.05 / 529.96 ±4.39 / 535.03 ms │           521.63 / 527.44 ±6.29 / 538.82 ms │     no change │
│ QQuery 21 │     520.61 / 527.42 ±5.56 / 535.62 ms │          523.55 / 543.87 ±17.88 / 571.68 ms │     no change │
│ QQuery 22 │ 1000.47 / 1015.20 ±11.74 / 1034.93 ms │        982.71 / 1012.79 ±24.65 / 1055.00 ms │     no change │
│ QQuery 23 │ 3127.98 / 3228.65 ±64.50 / 3322.32 ms │      3123.42 / 3261.08 ±102.02 / 3408.87 ms │     no change │
│ QQuery 24 │    40.80 / 148.67 ±156.84 / 451.64 ms │            41.81 / 88.99 ±40.68 / 135.79 ms │ +1.67x faster │
│ QQuery 25 │    104.09 / 127.10 ±41.15 / 209.31 ms │          104.44 / 153.05 ±59.40 / 248.21 ms │  1.20x slower │
│ QQuery 26 │        41.07 / 41.63 ±0.36 / 42.14 ms │              41.04 / 41.79 ±0.75 / 43.19 ms │     no change │
│ QQuery 27 │    514.54 / 528.34 ±10.01 / 540.14 ms │          513.75 / 526.15 ±10.67 / 544.73 ms │     no change │
│ QQuery 28 │ 2904.86 / 2996.52 ±85.41 / 3141.47 ms │        2864.90 / 2872.78 ±5.55 / 2882.27 ms │     no change │
│ QQuery 29 │      41.36 / 59.64 ±28.68 / 115.64 ms │              40.71 / 41.26 ±0.53 / 42.20 ms │ +1.45x faster │
│ QQuery 30 │    316.29 / 341.16 ±30.51 / 400.23 ms │          288.07 / 302.42 ±11.29 / 322.37 ms │ +1.13x faster │
│ QQuery 31 │    283.18 / 305.26 ±27.22 / 358.00 ms │          273.67 / 290.39 ±10.58 / 302.89 ms │     no change │
│ QQuery 32 │ 1057.52 / 1165.56 ±92.33 / 1280.00 ms │       1050.60 / 1143.05 ±50.02 / 1193.54 ms │     no change │
│ QQuery 33 │ 1505.99 / 1541.67 ±35.67 / 1596.26 ms │       1359.73 / 1386.79 ±26.71 / 1431.97 ms │ +1.11x faster │
│ QQuery 34 │ 1516.01 / 1533.76 ±12.31 / 1554.51 ms │       1330.57 / 1370.07 ±41.13 / 1441.90 ms │ +1.12x faster │
│ QQuery 35 │    282.28 / 321.63 ±44.87 / 391.88 ms │          249.50 / 283.94 ±27.31 / 327.90 ms │ +1.13x faster │
│ QQuery 36 │        64.82 / 71.45 ±7.36 / 84.58 ms │              61.89 / 68.75 ±4.39 / 73.88 ms │     no change │
│ QQuery 37 │        35.25 / 37.49 ±2.08 / 41.17 ms │              34.59 / 38.46 ±3.43 / 43.43 ms │     no change │
│ QQuery 38 │        42.75 / 47.48 ±3.25 / 51.19 ms │              44.88 / 48.48 ±3.70 / 54.59 ms │     no change │
│ QQuery 39 │     135.87 / 150.64 ±9.90 / 166.17 ms │           140.62 / 143.02 ±2.24 / 146.02 ms │ +1.05x faster │
│ QQuery 40 │        14.15 / 16.79 ±4.04 / 24.78 ms │              15.01 / 16.68 ±2.50 / 21.63 ms │     no change │
│ QQuery 41 │        13.29 / 15.70 ±4.05 / 23.79 ms │              14.13 / 14.88 ±0.54 / 15.68 ms │ +1.06x faster │
│ QQuery 42 │        13.26 / 13.98 ±0.86 / 15.59 ms │              13.46 / 15.08 ±2.77 / 20.61 ms │  1.08x slower │
└───────────┴───────────────────────────────────────┴─────────────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                          ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                          │ 20227.83ms │
│ Total Time (codex_ablation-partial-new-no-final-buckets)   │ 19180.80ms │
│ Average Time (HEAD)                                        │   470.41ms │
│ Average Time (codex_ablation-partial-new-no-final-buckets) │   446.07ms │
│ Queries Faster                                             │         19 │
│ Queries Slower                                             │          3 │
│ Queries with No Change                                     │         21 │
│ Queries with Failure                                       │          0 │
└────────────────────────────────────────────────────────────┴────────────┘

Resource Usage

clickbench_partitioned — base (merge-base)

Metric	Value
Wall time	105.0s
Peak memory	17.9 GiB
Avg memory	5.3 GiB
CPU user	992.1s
CPU sys	92.0s
Peak spill	0 B

clickbench_partitioned — branch

Metric	Value
Wall time	100.0s
Peak memory	16.6 GiB
Avg memory	5.3 GiB
CPU user	946.8s
CPU sys	86.7s
Peak spill	0 B

File an issue against this benchmark runner

adriangbot · 2026-10-06T11:48:30Z

🤖 Benchmark completed (GKE) | trigger

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

Comparing 70ac2ee (70ac2ee) to efd384f diff

Run configuration

run benchmark clickbench_partitioned
env:
  DATAFUSION_EXECUTION_HASH_AGGREGATE_BUCKET_THRESHOLD: "262144"
baseline:
  ref: "efd384f9fe0362c7cd9811f860837e6d7e425d11"
changed:
  ref: "70ac2eed7ad82bd0915ad526cd735bdcacbf7458"

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 codex_ablation-partial-new-no-final-buckets
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃       HEAD ┃ codex_ablation-partial-new-no-final-buckets ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0  │    1.29 ms │                                     1.30 ms │     no change │
│ QQuery 1  │   11.82 ms │                                    12.22 ms │     no change │
│ QQuery 2  │   36.66 ms │                                    36.23 ms │     no change │
│ QQuery 3  │   31.83 ms │                                    31.83 ms │     no change │
│ QQuery 4  │  167.28 ms │                                   231.74 ms │  1.39x slower │
│ QQuery 5  │  255.32 ms │                                   264.38 ms │     no change │
│ QQuery 6  │    1.37 ms │                                     1.38 ms │     no change │
│ QQuery 7  │   13.56 ms │                                    13.48 ms │     no change │
│ QQuery 8  │  271.57 ms │                                   328.74 ms │  1.21x slower │
│ QQuery 9  │  484.58 ms │                                   501.15 ms │     no change │
│ QQuery 10 │   69.93 ms │                                    67.13 ms │     no change │
│ QQuery 11 │   81.28 ms │                                    77.78 ms │     no change │
│ QQuery 12 │  244.42 ms │                                   250.00 ms │     no change │
│ QQuery 13 │  405.39 ms │                                   367.28 ms │ +1.10x faster │
│ QQuery 14 │  271.48 ms │                                   279.50 ms │     no change │
│ QQuery 15 │  182.97 ms │                                   262.63 ms │  1.44x slower │
│ QQuery 16 │  525.74 ms │                                   587.55 ms │  1.12x slower │
│ QQuery 17 │  452.84 ms │                                   595.37 ms │  1.31x slower │
│ QQuery 18 │ 1000.32 ms │                                  1124.62 ms │  1.12x slower │
│ QQuery 19 │   28.77 ms │                                    28.16 ms │     no change │
│ QQuery 20 │  534.38 ms │                                   522.68 ms │     no change │
│ QQuery 21 │  527.03 ms │                                   520.52 ms │     no change │
│ QQuery 22 │ 1049.83 ms │                                  1027.92 ms │     no change │
│ QQuery 23 │ 3223.42 ms │                                  3227.33 ms │     no change │
│ QQuery 24 │   41.65 ms │                                    41.80 ms │     no change │
│ QQuery 25 │  108.98 ms │                                   107.34 ms │     no change │
│ QQuery 26 │   42.07 ms │                                    41.84 ms │     no change │
│ QQuery 27 │  526.53 ms │                                   526.47 ms │     no change │
│ QQuery 28 │ 2960.26 ms │                                  2944.65 ms │     no change │
│ QQuery 29 │   44.69 ms │                                    41.69 ms │ +1.07x faster │
│ QQuery 30 │  264.53 ms │                                   295.69 ms │  1.12x slower │
│ QQuery 31 │  255.60 ms │                                   280.72 ms │  1.10x slower │
│ QQuery 32 │  719.59 ms │                                  1211.07 ms │  1.68x slower │
│ QQuery 33 │ 1386.92 ms │                                  1373.45 ms │     no change │
│ QQuery 34 │ 1407.92 ms │                                  1435.41 ms │     no change │
│ QQuery 35 │  211.04 ms │                                   268.14 ms │  1.27x slower │
│ QQuery 36 │   62.91 ms │                                    65.13 ms │     no change │
│ QQuery 37 │   36.36 ms │                                    36.10 ms │     no change │
│ QQuery 38 │   41.77 ms │                                    42.53 ms │     no change │
│ QQuery 39 │  140.49 ms │                                   145.55 ms │     no change │
│ QQuery 40 │   15.56 ms │                                    15.23 ms │     no change │
│ QQuery 41 │   14.73 ms │                                    14.68 ms │     no change │
│ QQuery 42 │   14.26 ms │                                    14.16 ms │     no change │
└───────────┴────────────┴─────────────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                          ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                          │ 18168.93ms │
│ Total Time (codex_ablation-partial-new-no-final-buckets)   │ 19262.54ms │
│ Average Time (HEAD)                                        │   422.53ms │
│ Average Time (codex_ablation-partial-new-no-final-buckets) │   447.97ms │
│ Queries Faster                                             │          2 │
│ Queries Slower                                             │         10 │
│ Queries with No Change                                     │         31 │
│ Queries with Failure                                       │          0 │
└────────────────────────────────────────────────────────────┴────────────┘

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

Comparing HEAD and codex_ablation-partial-new-no-final-buckets
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃                                   HEAD ┃ codex_ablation-partial-new-no-final-buckets ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0  │           1.29 / 4.31 ±5.85 / 16.00 ms │                1.30 / 4.26 ±5.73 / 15.73 ms │     no change │
│ QQuery 1  │         11.82 / 12.34 ±0.30 / 12.75 ms │              12.22 / 12.39 ±0.18 / 12.68 ms │     no change │
│ QQuery 2  │         36.66 / 36.95 ±0.17 / 37.12 ms │              36.23 / 36.79 ±0.51 / 37.67 ms │     no change │
│ QQuery 3  │         31.83 / 32.48 ±0.80 / 34.03 ms │              31.83 / 32.38 ±0.41 / 32.97 ms │     no change │
│ QQuery 4  │      167.28 / 168.92 ±1.10 / 170.54 ms │           231.74 / 235.44 ±3.65 / 241.26 ms │  1.39x slower │
│ QQuery 5  │      255.32 / 264.32 ±8.59 / 279.45 ms │           264.38 / 271.48 ±4.19 / 275.91 ms │     no change │
│ QQuery 6  │            1.37 / 1.53 ±0.23 / 1.98 ms │                 1.38 / 1.52 ±0.23 / 1.97 ms │     no change │
│ QQuery 7  │         13.56 / 13.65 ±0.06 / 13.72 ms │              13.48 / 13.69 ±0.12 / 13.83 ms │     no change │
│ QQuery 8  │      271.57 / 274.77 ±3.72 / 281.27 ms │           328.74 / 335.63 ±4.78 / 341.30 ms │  1.22x slower │
│ QQuery 9  │     484.58 / 508.82 ±14.86 / 524.98 ms │           501.15 / 508.09 ±8.43 / 523.75 ms │     no change │
│ QQuery 10 │         69.93 / 71.57 ±1.12 / 73.40 ms │              67.13 / 69.78 ±2.64 / 74.79 ms │     no change │
│ QQuery 11 │         81.28 / 82.25 ±0.84 / 83.78 ms │              77.78 / 79.19 ±1.77 / 82.62 ms │     no change │
│ QQuery 12 │      244.42 / 256.69 ±9.40 / 269.50 ms │           250.00 / 254.85 ±3.24 / 258.86 ms │     no change │
│ QQuery 13 │      405.39 / 412.60 ±5.89 / 417.52 ms │           367.28 / 381.04 ±8.20 / 388.28 ms │ +1.08x faster │
│ QQuery 14 │      271.48 / 276.45 ±4.04 / 283.40 ms │           279.50 / 287.41 ±8.44 / 301.82 ms │     no change │
│ QQuery 15 │      182.97 / 189.72 ±8.69 / 206.90 ms │           262.63 / 268.75 ±6.42 / 280.88 ms │  1.42x slower │
│ QQuery 16 │     525.74 / 537.51 ±17.27 / 571.36 ms │           587.55 / 596.52 ±7.77 / 606.86 ms │  1.11x slower │
│ QQuery 17 │      452.84 / 460.40 ±7.66 / 472.21 ms │           595.37 / 602.52 ±4.46 / 608.74 ms │  1.31x slower │
│ QQuery 18 │  1000.32 / 1021.31 ±12.32 / 1035.44 ms │       1124.62 / 1149.41 ±23.16 / 1188.48 ms │  1.13x slower │
│ QQuery 19 │         28.77 / 32.74 ±6.36 / 45.39 ms │              28.16 / 28.55 ±0.43 / 29.38 ms │ +1.15x faster │
│ QQuery 20 │     534.38 / 546.50 ±11.60 / 562.06 ms │           522.68 / 528.40 ±4.77 / 535.10 ms │     no change │
│ QQuery 21 │      527.03 / 541.42 ±8.81 / 552.56 ms │           520.52 / 526.81 ±4.96 / 534.93 ms │     no change │
│ QQuery 22 │  1049.83 / 1064.29 ±12.16 / 1079.54 ms │       1027.92 / 1046.69 ±21.38 / 1079.95 ms │     no change │
│ QQuery 23 │ 3223.42 / 3460.21 ±150.28 / 3633.06 ms │      3227.33 / 3393.10 ±169.68 / 3625.61 ms │     no change │
│ QQuery 24 │       41.65 / 61.61 ±38.34 / 138.29 ms │             41.80 / 53.22 ±14.18 / 77.91 ms │ +1.16x faster │
│ QQuery 25 │    108.98 / 212.04 ±203.04 / 618.12 ms │         107.34 / 209.73 ±194.78 / 599.16 ms │     no change │
│ QQuery 26 │         42.07 / 43.79 ±1.19 / 45.43 ms │              41.84 / 42.67 ±0.94 / 44.30 ms │     no change │
│ QQuery 27 │      526.53 / 536.01 ±7.72 / 545.66 ms │           526.47 / 539.47 ±8.35 / 551.58 ms │     no change │
│ QQuery 28 │  2960.26 / 2987.84 ±14.38 / 3000.92 ms │       2944.65 / 2961.89 ±16.07 / 2988.46 ms │     no change │
│ QQuery 29 │         44.69 / 53.80 ±4.99 / 57.58 ms │            41.69 / 64.13 ±28.85 / 119.55 ms │  1.19x slower │
│ QQuery 30 │     264.53 / 279.60 ±15.65 / 309.70 ms │           295.69 / 305.25 ±8.28 / 317.05 ms │  1.09x slower │
│ QQuery 31 │      255.60 / 267.14 ±8.69 / 280.44 ms │          280.72 / 318.04 ±45.18 / 404.76 ms │  1.19x slower │
│ QQuery 32 │   719.59 / 822.42 ±129.62 / 1072.79 ms │       1211.07 / 1279.09 ±43.54 / 1321.51 ms │  1.56x slower │
│ QQuery 33 │  1386.92 / 1470.88 ±77.68 / 1611.99 ms │       1373.45 / 1422.00 ±30.89 / 1467.37 ms │     no change │
│ QQuery 34 │  1407.92 / 1493.67 ±99.10 / 1684.82 ms │       1435.41 / 1464.09 ±25.30 / 1500.34 ms │     no change │
│ QQuery 35 │     211.04 / 229.18 ±14.85 / 253.98 ms │          268.14 / 283.36 ±13.50 / 308.21 ms │  1.24x slower │
│ QQuery 36 │         62.91 / 65.41 ±1.84 / 68.26 ms │              65.13 / 69.53 ±4.46 / 77.41 ms │  1.06x slower │
│ QQuery 37 │         36.36 / 39.52 ±2.98 / 44.53 ms │              36.10 / 38.30 ±2.90 / 43.99 ms │     no change │
│ QQuery 38 │         41.77 / 46.81 ±3.45 / 51.87 ms │              42.53 / 45.60 ±2.89 / 50.36 ms │     no change │
│ QQuery 39 │      140.49 / 152.39 ±8.42 / 166.18 ms │           145.55 / 156.06 ±8.44 / 165.26 ms │     no change │
│ QQuery 40 │         15.56 / 17.73 ±2.56 / 21.58 ms │              15.23 / 15.76 ±0.67 / 17.04 ms │ +1.13x faster │
│ QQuery 41 │         14.73 / 15.11 ±0.50 / 16.10 ms │              14.68 / 14.85 ±0.14 / 15.05 ms │     no change │
│ QQuery 42 │         14.26 / 15.25 ±1.83 / 18.90 ms │              14.16 / 14.34 ±0.16 / 14.53 ms │ +1.06x faster │
└───────────┴────────────────────────────────────────┴─────────────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                          ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                          │ 19081.94ms │
│ Total Time (codex_ablation-partial-new-no-final-buckets)   │ 19962.06ms │
│ Average Time (HEAD)                                        │   443.77ms │
│ Average Time (codex_ablation-partial-new-no-final-buckets) │   464.23ms │
│ Queries Faster                                             │          5 │
│ Queries Slower                                             │         12 │
│ Queries with No Change                                     │         26 │
│ Queries with Failure                                       │          0 │
└────────────────────────────────────────────────────────────┴────────────┘

Resource Usage

clickbench_partitioned — base (merge-base)

Metric	Value
Wall time	100.0s
Peak memory	16.2 GiB
Avg memory	5.4 GiB
CPU user	912.7s
CPU sys	91.8s
Peak spill	0 B

clickbench_partitioned — branch

Metric	Value
Wall time	105.0s
Peak memory	17.7 GiB
Avg memory	5.4 GiB
CPU user	982.0s
CPU sys	90.8s
Peak spill	0 B

File an issue against this benchmark runner

jayzhan211 and others added 16 commits September 25, 2026 09:39

feat: aggregate large final hash tables as hash buckets (experimental…

4daca18

…, off by default)

feat: flush a partial hash aggregation table that reaches the bucket …

0ba8308

…threshold while its groups do not recur

perf: reuse one hash table across aggregation buckets and skip compac…

efa8dde

…ting buckets of unique groups

refactor: move the aggregation bucket driver into its own module

f91b6ba

perf: move a final table of unique groups into buckets at a quarter o…

2cf0fe9

…f the threshold

perf: let the table of an aggregation bucket keep its group keys wher…

f9df89e

…e its input holds them

perf: route final aggregate buckets by column

b964422

perf: benchmark 32 aggregate buckets without compaction

313a3a9

perf: scatter StringView columns into aggregate buckets

085c221

perf: retain gather when scattering StringView bucket ranges

cf6819d

perf: increase final aggregation fanout to 128 buckets

33aad7a

perf: increase final aggregation fanout to 256 buckets

df55545

perf: bound partial aggregation working sets

efd384f

bench: disable final bucketing for partial policy baseline

126109f

bench: isolate partial working-set optimization without final buckets

70ac2ee

bench: make both ablation refs reachable by benchmark runner

2a661dc

github-actions Bot added documentation Improvements or additions to documentation physical-expr Changes to the physical-expr crates sqllogictest SQL Logic Tests (.slt) common Related to common crate physical-plan Changes to the physical-plan crate labels Oct 6, 2026

Navigation Menu

Repository navigation

bench: isolate Partial aggregation gains without Final buckets (do not merge) - #26081

bench: isolate Partial aggregation gains without Final buckets (do not merge)#26081
Rachelint wants to merge 16 commits into
apache:mainfrom
Rachelint:codex/ablation-partial-new-no-final-buckets

Rachelint commented Oct 6, 2026

Uh oh!

Rachelint commented Oct 6, 2026

Uh oh!

Rachelint commented Oct 6, 2026

Uh oh!

adriangbot commented Oct 6, 2026

Uh oh!

adriangbot commented Oct 6, 2026

Uh oh!

adriangbot commented Oct 6, 2026

Uh oh!

adriangbot commented Oct 6, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

Rachelint commented Oct 6, 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!

Rachelint commented Oct 6, 2026

Uh oh!

Rachelint commented Oct 6, 2026

Uh oh!

adriangbot commented Oct 6, 2026

Uh oh!

adriangbot commented Oct 6, 2026

Uh oh!

adriangbot commented Oct 6, 2026

Uh oh!

adriangbot commented Oct 6, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants