Expose accumulator state to allow prefix scanning

[avantgardnerio]

Summary

Expose state of aggregate streams within BWAG so downstream prefix scanning can take place.

API

// physical-plan/src/windows/bounded_window_agg_exec.rs
pub type FinalizedWindowStateObserver = Arc<
    dyn Fn(usize, &PartitionKey, &[Option<Vec<ScalarValue>>]) -> Result<()>
        + Send + Sync,
>;

impl BoundedWindowAggExec {
    pub fn with_finalized_state_observer(mut self, obs: FinalizedWindowStateObserver) -> Self { … }
}

// physical-expr/src/window/window_expr.rs
impl WindowState {
    /// `Accumulator::state()` if this is an aggregate window function, `None` otherwise.
    pub fn aggregate_state(&mut self) -> Result<Option<Vec<ScalarValue>>> { … }
}

[codecov-commenter]

Codecov Report

❌ Patch coverage is 82.00000% with 108 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.01%. Comparing base (33ad1cc) to head (aa4d196).
⚠️ Report is 18 commits behind head on main.

Files with missing lines	Patch %	Lines
...ysical-plan/src/windows/bounded_window_agg_exec.rs	85.01%	21 Missing and 49 partials ⚠️
datafusion/physical-plan/src/windows/mod.rs	72.72%	13 Missing and 5 partials ⚠️
datafusion/physical-expr/src/window/window_expr.rs	58.97%	13 Missing and 3 partials ⚠️
...zer/src/ensure_requirements/enforce_sorting/mod.rs	84.00%	3 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #24035      +/-   ##
==========================================
+ Coverage   80.99%   81.01%   +0.02%     
==========================================
  Files        1106     1106              
  Lines      383352   384686    +1334     
  Branches   383352   384686    +1334     
==========================================
+ Hits       310488   311665    +1177     
- Misses      54544    54620      +76     
- Partials    18320    18401      +81     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[neilconway]

Thanks for this contribution @avantgardnerio ! I'd find it helpful if you'd elaborate a little bit more about the motivation for this change in the PR description. For example, some intended use-cases, what kind of performance improvement this unlocks, etc.

[avantgardnerio]

@neilconway I'm trying to speed up window functions using parallel prefix scans. I am presently incubating this in Ballista, and this is the minimum API exposure that I need to do it for non-decomposable operations like approx_distinct() (vs others like AVG, STDDEV via Welford / Chan, etc). Though, the work is certainly not limited to Ballista, it has shown improvement in DataFusion as well.

The jury is still out about re-partition cost vs performance benefit, but the signs are hopeful:

And at least from a big-O time perspective (table 1) it should be optimal for some queries (select my_agg() over unbounded preceding...)

[avantgardnerio]

@gene-bordegaray and @JSOD11 you guys might be interested as well.

[alamb]

Thanks @avantgardnerio and @neilconway -- I left some comments

[alamb]

I may not fully understand prefix scanning, but it seems to me like this API will only give you access to the window state for the single last row in each partition.

Don't you potentially need access to the window state for the last N rows in a partition (e.g the HALO rows) 🤔

[alamb]

FWIW claude claims this doesn't get run with windows like

UNBOUNDED PRECEDING → CURRENT ROW

[alamb]

I think it would also be super useful to add some sort of example / test that shows how you intend to use this API (for exmple some simple example for computing a window function in parallel or something 🤔

that way we could see the API in action

[avantgardnerio]

this doesn't get run with windows like

Thanks @alamb ! That was a critical bug that would have defeated the whole point. It is now fixed and asserted in test_finalized_state_observer_fires_on_causal_frame()

[avantgardnerio]

access to the window state for the single last row

Yes, this is exactly what is required.

need access to the window state for the last N rows

No, not for prefix scanning. (answer below)

e.g the HALO rows

The HPC "halo" term is a good fit for bounded preceding/following (surrounding cells, in 1D) but doesn't extend cleanly to "last row of every other partition." Regardless of the name, this PR doesn't take that approach - because although it works for SUM, and decomposes for AVG (sum+count), it fails by the time you get to arbitrary accumulators like approx_distinct.

add some sort of example / test

Which is exactly where (the newly added) test_prefix_scan_across_tasks_matches_single_bwag() comes in. It shows that with only the accumulator state of the very last row of the entire (DF) partition, parallel partitioned prefix-scanned sums produce exactly the same results as a single (DF) partition BWAG. Given the observer exposes Accumulator::state() directly, any function that supports Accumulator::merge_batch - including approx_distinct - can be prefix-scanned the same way, with the downstream consumer merging sketches instead of adding scalars.

Edit: added the qualifier (DF) partition to distinguish between the ambiguously named (SQL window) partition.

[avantgardnerio]

Pseudo code, stripped directly from the test_prefix_scan_across_tasks_matches_single_bwag test, for those who value brevity:

        // Two tasks under range partition on sn:
        let (task1_out, task1_total) = run_running_sum_task(&[1, 1, 2, 2, 3, 3, 4, 4]);
        let (task2_out, task2_total) = run_running_sum_task(&[5, 5, 6, 6, 7, 7, 8, 8]);

        // Local (uncorrected) outputs and totals — first pass.
        assert_eq!(task1_out, vec![1, 2, 4, 6, 9, 12, 16, 20]);
        assert_eq!(task2_out, vec![5, 10, 16, 22, 29, 36, 44, 52]);

        // Prefix scan over per-task totals → carry-in for each task. Task 0's
        // carry-in is 0; task N's carry-in is the sum of tasks [0, N).
        let carry_ins = [0u64, task1_total];

        // Second pass: shift each task's local values by its carry-in.
        let task1_final: Vec<u64> = task1_out.iter().map(|v| v + carry_ins[0]).collect();
        let task2_final: Vec<u64> = task2_out.iter().map(|v| v + carry_ins[1]).collect();
        let parallel_result: Vec<u64> = task1_final.iter().chain(task2_final.iter());

        // Oracle: single BWAG over the full concatenated input.
        let (single_result, single_total) = run_running_sum_task(
            &[1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8]);

        assert_eq!(parallel_result, single_result);
        assert_eq!(single_result,
            vec![1, 2, 4, 6, 9, 12, 16, 20, 25, 30, 36, 42, 49, 56, 64, 72]
        );

[alamb]

Thanks @avantgardnerio

[avantgardnerio]

@timsaucer thank you for the thorough review. These are all legitimate concerns:

1. I was not able to reproduce this, but I added a test anyway, as well as your suggested mitigation
2. Following the belt-and-suspenders theme, I made this case hard-error in a central place
3. I've updated all the construction sites to include the callback during clones
4. It now prevents installation of a callback if any frames are sliding.

I think all of the above stem from this trying to fire per-partition-key, rather than once, at the end, when the motivating case is prefix scanning. What would you think about a getter shape where a downstream node reads Some(state) / None from the BWAG it already holds via input()? Scope is single-group-per-output-partition (matches distributed prefix-scan; wouldn't cover per-key PARTITION BY windows). Concerns 1/2/3 collapse to not-applicable, and 4 is a one-line gate at the write site.

[timsaucer]

Thank you for all the work on this!

[alamb]

Thank you @avantgardnerio and @timsaucer

[alamb]

run benchmark bounded_window

[alamb]

run benchmark window_query_sql

[alamb]

run benchmark h2o_small_window

[alamb]

(just running some benchmarks to make sure this doesn't change performance noticably)

[adriangbot]

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c5257786963-1548-jvh8r 6.12.85+ #1 SMP Wed Jun 17 20:31:55 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 brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark bounded_window

Results will be posted here when complete

---

File an issue against this benchmark runner

[adriangbot]

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c5257789157-1550-mwtd6 6.12.85+ #1 SMP Wed Jun 17 20:31:55 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 brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark h2o_small_window

Results will be posted here when complete

---

File an issue against this benchmark runner

[adriangbot]

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c5257788057-1549-cxjlj 6.12.85+ #1 SMP Wed Jun 17 20:31:55 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 brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark window_query_sql

Results will be posted here when complete

---

File an issue against this benchmark runner

[adriangbot]

🤖 Benchmark completed (GKE) | trigger

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

Comparing brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark bounded_window

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

group                                                       HEAD                                   brent_bwag-finalized-state-observer
-----                                                       ----                                   -----------------------------------
bounded_window_partitions/linear 100 partitions             1.00     83.8±0.09ms        ? ?/sec    1.00     83.9±0.05ms        ? ?/sec
bounded_window_partitions/linear 10000 partitions           1.00    457.1±1.67ms        ? ?/sec    1.00    456.5±5.82ms        ? ?/sec
bounded_window_partitions/linear multi 10000 partitions     1.00    709.8±2.96ms        ? ?/sec    1.02    724.1±7.03ms        ? ?/sec
bounded_window_partitions/linear rows 10000 partitions      1.00    409.9±4.19ms        ? ?/sec    1.04    427.0±5.99ms        ? ?/sec
bounded_window_partitions/linear sparse 32768 partitions    1.00    452.1±4.61ms        ? ?/sec    1.00    450.5±2.64ms        ? ?/sec
bounded_window_partitions/sorted 10000 partitions           1.00     70.7±0.21ms        ? ?/sec    1.00     70.9±0.17ms        ? ?/sec

Resource Usage

bounded_window — base (merge-base)

Metric	Value
Wall time	475.1s
Peak memory	88.9 MiB
Avg memory	9.8 MiB
CPU user	84.6s
CPU sys	0.2s
Peak spill	0 B

bounded_window — branch

Metric	Value
Wall time	470.1s
Peak memory	91.2 MiB
Avg memory	10.8 MiB
CPU user	90.0s
CPU sys	0.1s
Peak spill	0 B

---

File an issue against this benchmark runner

[adriangbot]

🤖 Benchmark completed (GKE) | trigger

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

Comparing brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark h2o_small_window

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 brent_bwag-finalized-state-observer
--------------------
Benchmark h2o_window.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃       HEAD ┃ brent_bwag-finalized-state-observer ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1  │  140.09 ms │                           139.61 ms │     no change │
│ QQuery 2  │ 1958.73 ms │                          1954.55 ms │     no change │
│ QQuery 3  │ 2519.09 ms │                          2825.21 ms │  1.12x slower │
│ QQuery 4  │  486.56 ms │                           468.42 ms │     no change │
│ QQuery 5  │ 1238.25 ms │                          1250.95 ms │     no change │
│ QQuery 6  │ 2849.53 ms │                          2866.78 ms │     no change │
│ QQuery 7  │ 2626.92 ms │                          2639.56 ms │     no change │
│ QQuery 8  │ 9535.86 ms │                          9547.38 ms │     no change │
│ QQuery 9  │  415.93 ms │                           407.97 ms │     no change │
│ QQuery 10 │  544.19 ms │                           538.88 ms │     no change │
│ QQuery 11 │  527.02 ms │                           519.35 ms │     no change │
│ QQuery 12 │ 1073.63 ms │                          1076.98 ms │     no change │
│ QQuery 13 │  308.01 ms │                           319.21 ms │     no change │
│ QQuery 14 │  407.38 ms │                           366.71 ms │ +1.11x faster │
│ QQuery 15 │  354.05 ms │                           342.45 ms │     no change │
│ QQuery 16 │  320.59 ms │                           328.13 ms │     no change │
│ QQuery 17 │  355.69 ms │                           322.25 ms │ +1.10x faster │
│ QQuery 18 │  413.38 ms │                           430.51 ms │     no change │
│ QQuery 19 │  365.71 ms │                           387.11 ms │  1.06x slower │
│ QQuery 20 │  385.90 ms │                           417.78 ms │  1.08x slower │
│ QQuery 21 │  332.06 ms │                           355.50 ms │  1.07x slower │
│ QQuery 22 │  380.59 ms │                           375.47 ms │     no change │
│ QQuery 23 │  357.97 ms │                           356.96 ms │     no change │
│ QQuery 24 │  436.54 ms │                           436.11 ms │     no change │
│ QQuery 25 │  377.74 ms │                           379.31 ms │     no change │
│ QQuery 26 │  384.04 ms │                           404.51 ms │  1.05x slower │
│ QQuery 27 │  365.12 ms │                           356.72 ms │     no change │
│ QQuery 28 │  387.05 ms │                           372.17 ms │     no change │
│ QQuery 29 │  342.05 ms │                           342.96 ms │     no change │
└───────────┴────────────┴─────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 30189.65ms │
│ Total Time (brent_bwag-finalized-state-observer)   │ 30529.51ms │
│ Average Time (HEAD)                                │  1041.02ms │
│ Average Time (brent_bwag-finalized-state-observer) │  1052.74ms │
│ Queries Faster                                     │          2 │
│ Queries Slower                                     │          5 │
│ Queries with No Change                             │         22 │
│ Queries with Failure                               │          0 │
└────────────────────────────────────────────────────┴────────────┘

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

Comparing HEAD and brent_bwag-finalized-state-observer
--------------------
Benchmark h2o_window.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃                                   HEAD ┃   brent_bwag-finalized-state-observer ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1  │      140.09 / 146.83 ±7.65 / 157.52 ms │     139.61 / 145.68 ±4.44 / 150.12 ms │     no change │
│ QQuery 2  │   1958.73 / 1958.94 ±0.24 / 1959.27 ms │  1954.55 / 1957.96 ±2.75 / 1961.28 ms │     no change │
│ QQuery 3  │ 2519.09 / 2662.55 ±128.21 / 2830.31 ms │ 2825.21 / 2858.15 ±29.98 / 2897.74 ms │  1.07x slower │
│ QQuery 4  │     486.56 / 510.58 ±17.37 / 527.07 ms │    468.42 / 497.35 ±30.67 / 539.80 ms │     no change │
│ QQuery 5  │   1238.25 / 1247.87 ±7.43 / 1256.35 ms │  1250.95 / 1253.26 ±2.31 / 1256.42 ms │     no change │
│ QQuery 6  │   2849.53 / 2857.10 ±6.51 / 2865.43 ms │ 2866.78 / 2876.32 ±12.81 / 2894.42 ms │     no change │
│ QQuery 7  │   2626.92 / 2634.73 ±7.87 / 2645.50 ms │ 2639.56 / 2656.07 ±11.88 / 2667.00 ms │     no change │
│ QQuery 8  │  9535.86 / 9563.66 ±19.68 / 9578.80 ms │  9547.38 / 9552.70 ±3.87 / 9556.48 ms │     no change │
│ QQuery 9  │     415.93 / 458.66 ±30.21 / 480.18 ms │    407.97 / 426.44 ±21.31 / 456.30 ms │ +1.08x faster │
│ QQuery 10 │     544.19 / 569.37 ±31.62 / 613.96 ms │     538.88 / 545.62 ±6.98 / 555.24 ms │     no change │
│ QQuery 11 │     527.02 / 547.59 ±16.98 / 568.60 ms │    519.35 / 536.47 ±15.29 / 556.47 ms │     no change │
│ QQuery 12 │   1073.63 / 1079.93 ±4.47 / 1083.46 ms │  1076.98 / 1086.36 ±9.68 / 1099.69 ms │     no change │
│ QQuery 13 │     308.01 / 339.37 ±35.71 / 389.32 ms │    319.21 / 339.67 ±23.38 / 372.39 ms │     no change │
│ QQuery 14 │      407.38 / 411.50 ±2.92 / 413.85 ms │    366.71 / 409.51 ±34.86 / 452.10 ms │     no change │
│ QQuery 15 │      354.05 / 364.14 ±9.44 / 376.75 ms │    342.45 / 360.91 ±16.66 / 382.82 ms │     no change │
│ QQuery 16 │     320.59 / 339.15 ±18.67 / 364.70 ms │     328.13 / 333.43 ±4.01 / 337.84 ms │     no change │
│ QQuery 17 │      355.69 / 362.10 ±8.14 / 373.59 ms │     322.25 / 330.06 ±5.95 / 336.66 ms │ +1.10x faster │
│ QQuery 18 │     413.38 / 433.03 ±14.10 / 445.78 ms │    430.51 / 442.70 ±13.59 / 461.66 ms │     no change │
│ QQuery 19 │      365.71 / 374.83 ±9.26 / 387.53 ms │     387.11 / 396.63 ±7.21 / 404.54 ms │  1.06x slower │
│ QQuery 20 │      385.90 / 388.98 ±3.06 / 393.16 ms │     417.78 / 425.90 ±5.74 / 429.99 ms │  1.09x slower │
│ QQuery 21 │     332.06 / 365.22 ±40.37 / 422.05 ms │    355.50 / 370.67 ±16.20 / 393.12 ms │     no change │
│ QQuery 22 │      380.59 / 385.21 ±4.55 / 391.40 ms │     375.47 / 382.48 ±5.41 / 388.65 ms │     no change │
│ QQuery 23 │     357.97 / 383.55 ±20.12 / 407.14 ms │    356.96 / 366.96 ±13.12 / 385.49 ms │     no change │
│ QQuery 24 │      436.54 / 448.69 ±8.93 / 457.74 ms │    436.11 / 453.22 ±16.53 / 475.56 ms │     no change │
│ QQuery 25 │      377.74 / 385.56 ±5.92 / 392.07 ms │     379.31 / 383.39 ±3.11 / 386.84 ms │     no change │
│ QQuery 26 │     384.04 / 438.01 ±43.99 / 491.79 ms │    404.51 / 446.59 ±50.41 / 517.47 ms │     no change │
│ QQuery 27 │     365.12 / 382.58 ±24.54 / 417.28 ms │    356.72 / 384.99 ±25.50 / 418.52 ms │     no change │
│ QQuery 28 │      387.05 / 391.35 ±5.60 / 399.26 ms │     372.17 / 380.36 ±8.19 / 391.54 ms │     no change │
│ QQuery 29 │     342.05 / 368.91 ±26.88 / 405.64 ms │    342.96 / 357.57 ±13.74 / 375.97 ms │     no change │
└───────────┴────────────────────────────────────────┴───────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 30799.98ms │
│ Total Time (brent_bwag-finalized-state-observer)   │ 30957.42ms │
│ Average Time (HEAD)                                │  1062.07ms │
│ Average Time (brent_bwag-finalized-state-observer) │  1067.50ms │
│ Queries Faster                                     │          2 │
│ Queries Slower                                     │          3 │
│ Queries with No Change                             │         24 │
│ Queries with Failure                               │          0 │
└────────────────────────────────────────────────────┴────────────┘

Resource Usage

h2o_small_window — base (merge-base)

Metric	Value
Wall time	95.0s
Peak memory	4.1 GiB
Avg memory	972.2 MiB
CPU user	449.2s
CPU sys	17.1s
Peak spill	0 B

h2o_small_window — branch

Metric	Value
Wall time	95.0s
Peak memory	4.1 GiB
Avg memory	1.0 GiB
CPU user	464.4s
CPU sys	16.6s
Peak spill	0 B

---

File an issue against this benchmark runner

[alamb]

run benchmark h2o_small_window

[alamb]

run benchmark bounded_window

[adriangbot]

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c5258171918-1551-djbfl 6.12.85+ #1 SMP Wed Jun 17 20:31:55 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 brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark h2o_small_window

Results will be posted here when complete

---

File an issue against this benchmark runner

[adriangbot]

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c5258175342-1552-q6gl5 6.12.85+ #1 SMP Wed Jun 17 20:31:55 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 brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark bounded_window

Results will be posted here when complete

---

File an issue against this benchmark runner

[adriangbot]

🤖 Benchmark completed (GKE) | trigger

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

Comparing brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark bounded_window

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

group                                                       HEAD                                   brent_bwag-finalized-state-observer
-----                                                       ----                                   -----------------------------------
bounded_window_partitions/linear 100 partitions             1.00     84.0±0.72ms        ? ?/sec    1.00     83.9±0.07ms        ? ?/sec
bounded_window_partitions/linear 10000 partitions           1.00    452.0±1.90ms        ? ?/sec    1.02   459.1±15.99ms        ? ?/sec
bounded_window_partitions/linear multi 10000 partitions     1.00    703.7±5.74ms        ? ?/sec    1.00    704.5±4.25ms        ? ?/sec
bounded_window_partitions/linear rows 10000 partitions      1.01    411.6±3.05ms        ? ?/sec    1.00    406.5±1.39ms        ? ?/sec
bounded_window_partitions/linear sparse 32768 partitions    1.02    446.2±8.16ms        ? ?/sec    1.00    436.5±3.50ms        ? ?/sec
bounded_window_partitions/sorted 10000 partitions           1.00     70.6±0.26ms        ? ?/sec    1.01     71.0±0.25ms        ? ?/sec

Resource Usage

bounded_window — base (merge-base)

Metric	Value
Wall time	220.1s
Peak memory	89.3 MiB
Avg memory	23.7 MiB
CPU user	87.3s
CPU sys	0.1s
Peak spill	0 B

bounded_window — branch

Metric	Value
Wall time	220.0s
Peak memory	89.9 MiB
Avg memory	21.4 MiB
CPU user	87.1s
CPU sys	0.1s
Peak spill	0 B

---

File an issue against this benchmark runner

[adriangbot]

🤖 Benchmark completed (GKE) | trigger

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

Comparing brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark h2o_small_window

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 brent_bwag-finalized-state-observer
--------------------
Benchmark h2o_window.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃       HEAD ┃ brent_bwag-finalized-state-observer ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1  │  140.60 ms │                           140.26 ms │     no change │
│ QQuery 2  │ 1971.38 ms │                          1959.88 ms │     no change │
│ QQuery 3  │ 2603.45 ms │                          2494.99 ms │     no change │
│ QQuery 4  │  455.12 ms │                           486.11 ms │  1.07x slower │
│ QQuery 5  │ 1222.26 ms │                          1272.60 ms │     no change │
│ QQuery 6  │ 2828.92 ms │                          2864.17 ms │     no change │
│ QQuery 7  │ 2685.73 ms │                          2659.05 ms │     no change │
│ QQuery 8  │ 9543.31 ms │                          9586.29 ms │     no change │
│ QQuery 9  │  406.62 ms │                           433.07 ms │  1.07x slower │
│ QQuery 10 │  559.17 ms │                           540.33 ms │     no change │
│ QQuery 11 │  526.52 ms │                           524.02 ms │     no change │
│ QQuery 12 │ 1058.94 ms │                          1079.48 ms │     no change │
│ QQuery 13 │  335.76 ms │                           296.53 ms │ +1.13x faster │
│ QQuery 14 │  407.03 ms │                           385.46 ms │ +1.06x faster │
│ QQuery 15 │  368.00 ms │                           345.69 ms │ +1.06x faster │
│ QQuery 16 │  318.85 ms │                           336.23 ms │  1.05x slower │
│ QQuery 17 │  313.77 ms │                           326.09 ms │     no change │
│ QQuery 18 │  404.66 ms │                           417.98 ms │     no change │
│ QQuery 19 │  385.73 ms │                           376.55 ms │     no change │
│ QQuery 20 │  408.55 ms │                           405.82 ms │     no change │
│ QQuery 21 │  346.45 ms │                           334.68 ms │     no change │
│ QQuery 22 │  369.20 ms │                           375.53 ms │     no change │
│ QQuery 23 │  357.87 ms │                           362.95 ms │     no change │
│ QQuery 24 │  450.54 ms │                           405.73 ms │ +1.11x faster │
│ QQuery 25 │  373.97 ms │                           368.18 ms │     no change │
│ QQuery 26 │  387.17 ms │                           417.37 ms │  1.08x slower │
│ QQuery 27 │  337.79 ms │                           365.38 ms │  1.08x slower │
│ QQuery 28 │  386.77 ms │                           369.34 ms │     no change │
│ QQuery 29 │  350.46 ms │                           346.96 ms │     no change │
└───────────┴────────────┴─────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 30304.59ms │
│ Total Time (brent_bwag-finalized-state-observer)   │ 30276.71ms │
│ Average Time (HEAD)                                │  1044.99ms │
│ Average Time (brent_bwag-finalized-state-observer) │  1044.02ms │
│ Queries Faster                                     │          4 │
│ Queries Slower                                     │          5 │
│ Queries with No Change                             │         20 │
│ Queries with Failure                               │          0 │
└────────────────────────────────────────────────────┴────────────┘

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

Comparing HEAD and brent_bwag-finalized-state-observer
--------------------
Benchmark h2o_window.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃                                   HEAD ┃    brent_bwag-finalized-state-observer ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1  │      140.60 / 142.87 ±2.08 / 145.63 ms │      140.26 / 143.76 ±4.38 / 149.93 ms │     no change │
│ QQuery 2  │   1971.38 / 1972.34 ±0.73 / 1973.14 ms │   1959.88 / 1961.85 ±2.51 / 1965.39 ms │     no change │
│ QQuery 3  │ 2603.45 / 2713.65 ±128.36 / 2893.68 ms │ 2494.99 / 2866.00 ±316.48 / 3268.32 ms │  1.06x slower │
│ QQuery 4  │     455.12 / 472.84 ±23.57 / 506.15 ms │     486.11 / 496.80 ±11.38 / 512.57 ms │  1.05x slower │
│ QQuery 5  │   1222.26 / 1233.80 ±8.32 / 1241.55 ms │   1272.60 / 1274.27 ±1.29 / 1275.73 ms │     no change │
│ QQuery 6  │   2828.92 / 2836.76 ±5.64 / 2841.92 ms │   2864.17 / 2866.84 ±2.22 / 2869.60 ms │     no change │
│ QQuery 7  │   2685.73 / 2690.67 ±3.51 / 2693.56 ms │   2659.05 / 2665.69 ±4.87 / 2670.61 ms │     no change │
│ QQuery 8  │  9543.31 / 9578.95 ±25.34 / 9599.95 ms │  9586.29 / 9596.14 ±11.79 / 9612.72 ms │     no change │
│ QQuery 9  │     406.62 / 434.46 ±20.86 / 456.84 ms │     433.07 / 450.31 ±12.69 / 463.25 ms │     no change │
│ QQuery 10 │     559.17 / 584.09 ±29.47 / 625.48 ms │     540.33 / 553.34 ±11.01 / 567.25 ms │ +1.06x faster │
│ QQuery 11 │     526.52 / 546.46 ±14.12 / 557.30 ms │      524.02 / 529.74 ±6.43 / 538.72 ms │     no change │
│ QQuery 12 │  1058.94 / 1077.17 ±14.75 / 1095.07 ms │  1079.48 / 1112.79 ±29.45 / 1151.09 ms │     no change │
│ QQuery 13 │     335.76 / 353.96 ±25.19 / 389.58 ms │     296.53 / 330.45 ±27.78 / 364.57 ms │ +1.07x faster │
│ QQuery 14 │     407.03 / 420.68 ±12.41 / 437.06 ms │      385.46 / 391.11 ±4.07 / 394.87 ms │ +1.08x faster │
│ QQuery 15 │     368.00 / 391.27 ±20.06 / 416.96 ms │      345.69 / 352.99 ±5.41 / 358.63 ms │ +1.11x faster │
│ QQuery 16 │      318.85 / 332.68 ±9.87 / 341.26 ms │      336.23 / 343.81 ±6.47 / 352.04 ms │     no change │
│ QQuery 17 │      313.77 / 314.45 ±0.65 / 315.33 ms │     326.09 / 344.71 ±19.05 / 370.87 ms │  1.10x slower │
│ QQuery 18 │     404.66 / 432.69 ±20.55 / 453.37 ms │     417.98 / 438.66 ±15.00 / 453.10 ms │     no change │
│ QQuery 19 │     385.73 / 401.73 ±16.35 / 424.19 ms │     376.55 / 393.80 ±14.94 / 413.00 ms │     no change │
│ QQuery 20 │      408.55 / 418.41 ±6.99 / 423.96 ms │      405.82 / 415.73 ±9.65 / 428.82 ms │     no change │
│ QQuery 21 │      346.45 / 351.05 ±4.25 / 356.69 ms │     334.68 / 344.94 ±11.26 / 360.61 ms │     no change │
│ QQuery 22 │      369.20 / 372.18 ±3.14 / 376.52 ms │     375.53 / 396.31 ±23.14 / 428.59 ms │  1.06x slower │
│ QQuery 23 │     357.87 / 373.98 ±19.38 / 401.23 ms │      362.95 / 366.08 ±3.26 / 370.57 ms │     no change │
│ QQuery 24 │      450.54 / 457.42 ±9.22 / 470.45 ms │     405.73 / 440.50 ±25.05 / 463.74 ms │     no change │
│ QQuery 25 │     373.97 / 393.10 ±23.51 / 426.22 ms │     368.18 / 385.20 ±12.99 / 399.69 ms │     no change │
│ QQuery 26 │     387.17 / 400.98 ±11.48 / 415.27 ms │     417.37 / 430.77 ±17.47 / 455.45 ms │  1.07x slower │
│ QQuery 27 │      337.79 / 343.57 ±6.37 / 352.44 ms │     365.38 / 381.99 ±16.45 / 404.40 ms │  1.11x slower │
│ QQuery 28 │     386.77 / 410.63 ±19.01 / 433.28 ms │      369.34 / 377.21 ±7.36 / 387.05 ms │ +1.09x faster │
│ QQuery 29 │      350.46 / 358.02 ±7.44 / 368.15 ms │      346.96 / 350.81 ±2.73 / 352.88 ms │     no change │
└───────────┴────────────────────────────────────────┴────────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 30810.84ms │
│ Total Time (brent_bwag-finalized-state-observer)   │ 31002.60ms │
│ Average Time (HEAD)                                │  1062.44ms │
│ Average Time (brent_bwag-finalized-state-observer) │  1069.06ms │
│ Queries Faster                                     │          5 │
│ Queries Slower                                     │          6 │
│ Queries with No Change                             │         18 │
│ Queries with Failure                               │          0 │
└────────────────────────────────────────────────────┴────────────┘

Resource Usage

h2o_small_window — base (merge-base)

Metric	Value
Wall time	95.0s
Peak memory	4.2 GiB
Avg memory	1.0 GiB
CPU user	453.3s
CPU sys	15.2s
Peak spill	0 B

h2o_small_window — branch

Metric	Value
Wall time	95.0s
Peak memory	4.3 GiB
Avg memory	1.0 GiB
CPU user	464.6s
CPU sys	14.7s
Peak spill	0 B

---

File an issue against this benchmark runner

[adriangbot]

🤖 Benchmark completed (GKE) | trigger

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

Comparing brent/bwag-finalized-state-observer (aa4d196) to 33ad1cc (merge-base) diff

Run configuration

run benchmark window_query_sql

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

group                                                             HEAD                                   brent_bwag-finalized-state-observer
-----                                                             ----                                   -----------------------------------
window empty over, aggregate functions                            1.00     14.3±0.85ms        ? ?/sec    1.10     15.8±0.86ms        ? ?/sec
window empty over, built-in functions                             1.00    170.2±0.87ms        ? ?/sec    1.03    175.8±2.54ms        ? ?/sec
window order by, aggregate functions                              1.00    913.1±3.62ms        ? ?/sec    1.00    914.1±4.13ms        ? ?/sec
window order by, built-in functions                               1.00    848.8±3.41ms        ? ?/sec    1.00    845.4±2.66ms        ? ?/sec
window partition and order by, u64_narrow, aggregate functions    1.01    284.0±1.71ms        ? ?/sec    1.00    282.5±0.88ms        ? ?/sec
window partition and order by, u64_narrow, built-in functions     1.00    204.3±1.18ms        ? ?/sec    1.01    205.4±1.18ms        ? ?/sec
window partition and order by, u64_wide, aggregate functions      1.00  965.7±110.85ms        ? ?/sec    1.05  1016.2±116.52ms        ? ?/sec
window partition and order by, u64_wide, built-in functions       1.03  903.0±141.32ms        ? ?/sec    1.00  875.6±138.12ms        ? ?/sec
window partition by, u64_narrow, aggregate functions              1.00     11.2±0.08ms        ? ?/sec    1.00     11.2±0.09ms        ? ?/sec
window partition by, u64_narrow, built-in functions               1.00     41.8±0.38ms        ? ?/sec    1.00     41.8±0.53ms        ? ?/sec
window partition by, u64_wide, aggregate functions                1.00   588.8±90.93ms        ? ?/sec    1.00   589.6±89.49ms        ? ?/sec
window partition by, u64_wide, built-in functions                 1.00   550.8±60.95ms        ? ?/sec    1.05   577.6±58.83ms        ? ?/sec

Resource Usage

window_query_sql — base (merge-base)

Metric	Value
Wall time	1550.3s
Peak memory	1.4 GiB
Avg memory	357.4 MiB
CPU user	4177.1s
CPU sys	34.8s
Peak spill	0 B

window_query_sql — branch

Metric	Value
Wall time	1425.3s
Peak memory	1.4 GiB
Avg memory	400.1 MiB
CPU user	4321.5s
CPU sys	38.1s
Peak spill	0 B

---

File an issue against this benchmark runner