Skip to content

Read promoted Parquet dictionaries natively - #26977

Closed
Rich-T-kid wants to merge 3 commits into
apache:mainfrom
Rich-T-kid:codex/dictionary-wide-offset-reader
Closed

Read promoted Parquet dictionaries natively#26977
Rich-T-kid wants to merge 3 commits into
apache:mainfrom
Rich-T-kid:codex/dictionary-wide-offset-reader

Conversation

@Rich-T-kid

@Rich-T-kid Rich-T-kid commented Oct 31, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Which issue does this PR close?

Closes #26976. Part of #24111. Depends on #26973.

Why is this needed?

After #26973 excludes columns with plain fallback pages, eligible Parquet columns can still take an avoidable conversion path. Wide-offset dictionaries are read as plain values, and binary columns inferred as strings are dictionary-encoded row by row.

This PR keeps the physical dictionary value type in the reader schema. parquet-rs decodes the keys directly, and the output projection casts the much smaller dictionary values once.

Commits

There is no Arrow patch or dependency override.

Performance

The hosted runs compare #26973 (e66caad9) with this PR (0b6467ff) using released Arrow 60.1.0 and dictionary reads on both sides.

  • ClickBench: 23.03 s to 19.61 s (-14.8%). Q10 is 3.03x faster, Q11 2.78x, and Q23 2.01x.
  • TPC-H SF1: all 22 queries unchanged.
  • TPC-DS SF1 retry: 1.4-1.9% slower overall. The repeatable outliers are high-cardinality Q72 and sub-millisecond Q16. Balanced local runs measured Q72 at +2.5% and Q16 at -1.2%.
  • Wide schema: no material change; all cases are within 3%.

The small hosted Q41/Q42 ClickBench slowdowns did not reproduce over 200 local samples per side (+0.5% and +0.3%).

Testing

  • cargo test -p datafusion-datasource-parquet --lib (296 tests)
  • cargo clippy -p datafusion-datasource-parquet --all-targets --all-features -- -D warnings
  • repository pre-push checks, including full-workspace clippy and typos
  • cargo fmt --all -- --check

@github-actions github-actions Bot added documentation Improvements or additions to documentation sqllogictest SQL Logic Tests (.slt) common Related to common crate datasource Changes to the datasource crate labels Oct 31, 2026
@Rich-T-kid

Rich-T-kid commented Oct 31, 2026

Copy link
Copy Markdown
Contributor Author

Hosted ClickBench result: #26964 (comment)

With dictionary mode enabled on both sides, this change reduced total time from 22.22s to 18.87s (-15.1%) with 3 faster, 0 slower, and 40 unchanged queries in the primary comparison. Q10 improved 198.86ms -> 65.91ms (3.02x), Q11 211.87ms -> 76.34ms (2.78x), and Q23 6146.32ms -> 2992.38ms (2.05x). Peak RSS fell from 12.6 GiB to 11.3 GiB.

@Rich-T-kid Rich-T-kid closed this Nov 3, 2026
@Rich-T-kid Rich-T-kid reopened this Nov 9, 2026
@Rich-T-kid
Rich-T-kid force-pushed the codex/dictionary-wide-offset-reader branch from 42b286f to 75e94ab Compare November 9, 2026 04:25
@github-actions github-actions Bot removed documentation Improvements or additions to documentation sqllogictest SQL Logic Tests (.slt) common Related to common crate labels Nov 9, 2026
@Rich-T-kid

Rich-T-kid commented Nov 9, 2026

Copy link
Copy Markdown
Contributor Author

Re-ran the focused ClickBench comparison on the rebased branch using released Arrow 60.1.0.

Setup:

  • base: b8534c7fd836f23e99f3f95fccea53cd7e167bff
  • candidate: 75e94abfe45e7266ae1cd234154127189163c031
  • partitioned ClickBench dataset, 12 partitions, 32 GiB memory limit
  • datafusion.execution.parquet.enable_rle_to_dictionary=true on both sides
  • four balanced base/candidate/candidate/base runs, 50 iterations per query per run
  • separate Cargo target directories; base binary SHA-256 e906be784c24dff91285eb4a03b089b03769756d426c5a9e03af347158c97ac9, candidate 70c7aea70cffdc965626ded3aaeafc65ff7065bc305b780a4cca80658e8f2343

Pooled medians:

Query Main PR Change
Q10 176.26 ms 59.41 ms -66.3% / 2.97x faster
Q11 188.13 ms 70.61 ms -62.5% / 2.66x faster

The profiles now enter the Parquet dictionary decoder directly instead of spending roughly half of CPU time in DecoderProjection / pack_byte_to_dictionary rebuilding keys row by row.

@Rich-T-kid
Rich-T-kid force-pushed the codex/dictionary-wide-offset-reader branch from 75e94ab to 0b6467f Compare November 9, 2026 14:13
@Rich-T-kid

Rich-T-kid commented Nov 9, 2026

Copy link
Copy Markdown
Contributor Author

run benchmark tpch tpcds clickbench_partitioned wide_schema

env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

Benchmark for this request failed before finishing (Kubernetes reason: BackoffLimitExceeded).

Benchmarks requested: clickbench_partitioned

Runner log (last 40 lines) 
2026-11-09T14:16:43.885397Z  INFO runner starting benchmark runner bench_type=Datafusion, pr_url=https://github.com/apache/datafusion/pull/26977, benchmarks=clickbench_partitioned
2026-11-09T14:16:43.943253Z  INFO benchmark_controller::runner::bench_datafusion === Cloning PR branch ===
2026-11-09T14:16:43.943273Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["clone", "--depth=200", "https://github.com/apache/datafusion.git", "/workspace/datafusion-branch"], cwd="/"
2026-11-09T14:16:48.952070Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["fetch", "origin", "refs/pull/26977/head:codex/dictionary-wide-offset-reader", "main"], cwd="/workspace/datafusion-branch"
2026-11-09T14:16:53.953930Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["checkout", "codex/dictionary-wide-offset-reader"], cwd="/workspace/datafusion-branch"
2026-11-09T14:16:58.956793Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["merge-base", "HEAD", "origin/main"], cwd="/workspace/datafusion-branch"
2026-11-09T14:17:03.959255Z  INFO benchmark_controller::runner::bench_datafusion === Checking out custom changed ref === changed_ref=0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd
2026-11-09T14:17:03.959271Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["fetch", "origin", "refs/pull/26977/head"], cwd="/workspace/datafusion-branch"
2026-11-09T14:17:08.961412Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["fetch", "origin"], cwd="/workspace/datafusion-branch"
2026-11-09T14:17:13.963526Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["-c", "advice.detachedHead=false", "checkout", "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"], cwd="/workspace/datafusion-branch"
2026-11-09T14:17:18.965298Z  INFO benchmark_controller::runner::bench_datafusion === Cloning merge-base ===
2026-11-09T14:17:18.965315Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["clone", "--depth=200", "https://github.com/apache/datafusion.git", "/workspace/datafusion-base"], cwd="/"
2026-11-09T14:17:23.966859Z  INFO benchmark_controller::runner::bench_datafusion === Checking out custom baseline ref === baseline_ref=e66caad9042cf570a605da822e480a02845360b9
2026-11-09T14:17:23.966875Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["fetch", "origin", "refs/pull/26977/head"], cwd="/workspace/datafusion-base"
2026-11-09T14:17:28.968985Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["fetch", "origin"], cwd="/workspace/datafusion-base"
2026-11-09T14:17:33.971613Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["-c", "advice.detachedHead=false", "checkout", "e66caad9042cf570a605da822e480a02845360b9"], cwd="/workspace/datafusion-base"
2026-11-09T14:17:38.973157Z  INFO benchmark_controller::runner::shell running command cmd=rustc, args=["--version"], cwd="/"
2026-11-09T14:17:43.975586Z  INFO benchmark_controller::runner::shell running command cmd=cargo, args=["metadata", "--no-deps", "--format-version", "1"], cwd="/workspace/datafusion-branch/benchmarks"
2026-11-09T14:17:48.980198Z  INFO benchmark_controller::runner::bench_datafusion === Compiling dfbench for PR branch and merge-base in parallel ===
2026-11-09T14:17:49.004530Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["rev-parse", "HEAD"], cwd="/workspace/datafusion-branch"
2026-11-09T14:17:54.006315Z  INFO benchmark_controller::runner::shell running command cmd=git, args=["rev-parse", "HEAD"], cwd="/workspace/datafusion-base"
2026-11-09T14:17:59.007696Z  INFO benchmark_controller::runner::controller_client post_comment _repo=apache/datafusion, _pr_number=26977, job_id=4682
2026-11-09T14:19:14.302949Z ERROR runner benchmark failed error=send request: error sending request for url (http://benchmark-controller.benchmarking.svc.cluster.local:8080/jobs/4682/comment): client error (Connect): tcp connect error: Connection refused (os error 111)
Kubernetes message 
Job has reached the specified backoff limit

File an issue against this benchmark runner

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c6742235853-4683-pgpzd 6.12.105+ #1 SMP Sat Sep 19 07:55:25 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 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark wide_schema
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"

Results will be posted here when complete

File an issue against this benchmark runner

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c6742235853-4681-lmphh 6.12.105+ #1 SMP Sat Sep 19 07:55:25 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 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark tpcds
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"

Results will be posted here when complete

File an issue against this benchmark runner

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c6742235853-4680-nwq76 6.12.105+ #1 SMP Sat Sep 19 07:55:25 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 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark tpch
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"

Results will be posted here when complete

File an issue against this benchmark runner

@Rich-T-kid

Rich-T-kid commented Nov 9, 2026

Copy link
Copy Markdown
Contributor Author

run benchmark clickbench_partitioned

env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c6742354334-4684-x6lmd 6.12.105+ #1 SMP Sat Sep 19 07:55:25 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 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark clickbench_partitioned
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"

Results will be posted here when complete

File an issue against this benchmark runner

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark completed (GKE) | trigger

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

Comparing 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark tpch
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"
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_dictionary-wide-offset-reader
--------------------
Benchmark tpch_sf1.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Query     ┃     HEAD ┃ codex_dictionary-wide-offset-reader ┃    Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ QQuery 1  │ 40.56 ms │                            40.26 ms │ no change │
│ QQuery 2  │ 19.64 ms │                            19.43 ms │ no change │
│ QQuery 3  │ 29.71 ms │                            29.24 ms │ no change │
│ QQuery 4  │ 17.96 ms │                            17.84 ms │ no change │
│ QQuery 5  │ 37.83 ms │                            37.75 ms │ no change │
│ QQuery 6  │ 14.25 ms │                            14.12 ms │ no change │
│ QQuery 7  │ 44.98 ms │                            44.09 ms │ no change │
│ QQuery 8  │ 42.49 ms │                            41.73 ms │ no change │
│ QQuery 9  │ 52.73 ms │                            52.10 ms │ no change │
│ QQuery 10 │ 50.82 ms │                            50.78 ms │ no change │
│ QQuery 11 │ 14.49 ms │                            14.67 ms │ no change │
│ QQuery 12 │ 21.32 ms │                            21.50 ms │ no change │
│ QQuery 13 │ 44.26 ms │                            45.45 ms │ no change │
│ QQuery 14 │ 25.11 ms │                            25.58 ms │ no change │
│ QQuery 15 │ 32.62 ms │                            32.71 ms │ no change │
│ QQuery 16 │ 13.81 ms │                            13.80 ms │ no change │
│ QQuery 17 │ 75.14 ms │                            76.18 ms │ no change │
│ QQuery 18 │ 67.64 ms │                            70.11 ms │ no change │
│ QQuery 19 │ 30.77 ms │                            30.14 ms │ no change │
│ QQuery 20 │ 33.88 ms │                            33.47 ms │ no change │
│ QQuery 21 │ 56.72 ms │                            57.31 ms │ no change │
│ QQuery 22 │ 16.30 ms │                            16.41 ms │ no change │
└───────────┴──────────┴─────────────────────────────────────┴───────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃          ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 783.03ms │
│ Total Time (codex_dictionary-wide-offset-reader)   │ 784.67ms │
│ Average Time (HEAD)                                │  35.59ms │
│ Average Time (codex_dictionary-wide-offset-reader) │  35.67ms │
│ Queries Faster                                     │        0 │
│ Queries Slower                                     │        0 │
│ Queries with No Change                             │       22 │
│ Queries with Failure                               │        0 │
└────────────────────────────────────────────────────┴──────────┘

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

Comparing HEAD and codex_dictionary-wide-offset-reader
--------------------
Benchmark tpch_sf1.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Query     ┃                           HEAD ┃ codex_dictionary-wide-offset-reader ┃    Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ QQuery 1  │ 40.56 / 41.53 ±1.39 / 44.26 ms │      40.26 / 41.97 ±1.43 / 43.64 ms │ no change │
│ QQuery 2  │ 19.64 / 19.83 ±0.21 / 20.21 ms │      19.43 / 19.78 ±0.45 / 20.64 ms │ no change │
│ QQuery 3  │ 29.71 / 30.27 ±0.70 / 31.58 ms │      29.24 / 29.55 ±0.25 / 29.95 ms │ no change │
│ QQuery 4  │ 17.96 / 18.49 ±0.79 / 20.03 ms │      17.84 / 18.41 ±0.62 / 19.51 ms │ no change │
│ QQuery 5  │ 37.83 / 38.60 ±0.91 / 40.23 ms │      37.75 / 38.02 ±0.25 / 38.43 ms │ no change │
│ QQuery 6  │ 14.25 / 14.63 ±0.47 / 15.55 ms │      14.12 / 14.43 ±0.30 / 14.98 ms │ no change │
│ QQuery 7  │ 44.98 / 47.20 ±1.96 / 49.72 ms │      44.09 / 46.11 ±2.38 / 50.68 ms │ no change │
│ QQuery 8  │ 42.49 / 43.32 ±0.56 / 44.11 ms │      41.73 / 42.07 ±0.27 / 42.38 ms │ no change │
│ QQuery 9  │ 52.73 / 53.26 ±0.49 / 54.05 ms │      52.10 / 52.82 ±0.77 / 54.04 ms │ no change │
│ QQuery 10 │ 50.82 / 51.77 ±0.89 / 53.12 ms │      50.78 / 51.10 ±0.25 / 51.45 ms │ no change │
│ QQuery 11 │ 14.49 / 14.70 ±0.18 / 15.03 ms │      14.67 / 14.80 ±0.17 / 15.12 ms │ no change │
│ QQuery 12 │ 21.32 / 21.65 ±0.20 / 21.86 ms │      21.50 / 21.82 ±0.19 / 22.03 ms │ no change │
│ QQuery 13 │ 44.26 / 45.35 ±0.87 / 46.46 ms │      45.45 / 46.59 ±1.45 / 49.39 ms │ no change │
│ QQuery 14 │ 25.11 / 25.49 ±0.24 / 25.82 ms │      25.58 / 25.74 ±0.13 / 25.89 ms │ no change │
│ QQuery 15 │ 32.62 / 33.60 ±1.00 / 35.24 ms │      32.71 / 33.78 ±0.97 / 35.17 ms │ no change │
│ QQuery 16 │ 13.81 / 13.92 ±0.14 / 14.19 ms │      13.80 / 14.46 ±0.34 / 14.72 ms │ no change │
│ QQuery 17 │ 75.14 / 77.50 ±1.93 / 80.09 ms │      76.18 / 77.33 ±0.94 / 78.87 ms │ no change │
│ QQuery 18 │ 67.64 / 71.45 ±2.25 / 73.76 ms │      70.11 / 73.16 ±2.77 / 78.00 ms │ no change │
│ QQuery 19 │ 30.77 / 31.18 ±0.28 / 31.59 ms │      30.14 / 30.76 ±0.66 / 31.94 ms │ no change │
│ QQuery 20 │ 33.88 / 34.16 ±0.30 / 34.60 ms │      33.47 / 33.83 ±0.40 / 34.42 ms │ no change │
│ QQuery 21 │ 56.72 / 57.92 ±1.14 / 59.78 ms │      57.31 / 58.11 ±0.75 / 59.33 ms │ no change │
│ QQuery 22 │ 16.30 / 18.11 ±3.33 / 24.76 ms │      16.41 / 17.63 ±2.02 / 21.67 ms │ no change │
└───────────┴────────────────────────────────┴─────────────────────────────────────┴───────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃          ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 803.96ms │
│ Total Time (codex_dictionary-wide-offset-reader)   │ 802.29ms │
│ Average Time (HEAD)                                │  36.54ms │
│ Average Time (codex_dictionary-wide-offset-reader) │  36.47ms │
│ Queries Faster                                     │        0 │
│ Queries Slower                                     │        0 │
│ Queries with No Change                             │       22 │
│ Queries with Failure                               │        0 │
└────────────────────────────────────────────────────┴──────────┘

Memory Pool Peaks

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

Base: e66caad9042cf570a605da822e480a02845360b9 | Changed: 0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd

tpch — tpch_sf1

Query Base Changed Change
Query 1 30.1 MiB 30.1 MiB +0.0%
Query 2 23.1 MiB 23.1 MiB +0.0%
Query 3 11.5 MiB 11.5 MiB +0.0%
Query 4 32.2 MiB 22.2 MiB -30.9%
Query 5 42.8 MiB 32.9 MiB -23.1%
Query 6 832 B 832 B +0.0%
Query 7 107.6 MiB 107.6 MiB +0.0%
Query 8 39.8 MiB 39.8 MiB -0.0%
Query 9 155.6 MiB 155.6 MiB +0.0%
Query 10 49.1 MiB 47.4 MiB -3.5%
Query 11 61.7 MiB 61.7 MiB -0.0%
Query 12 22.5 MiB 22.5 MiB -0.0%
Query 13 52.6 MiB 72.5 MiB +37.8%
Query 14 7.7 MiB 7.7 MiB +0.0%
Query 15 12.7 MiB 12.7 MiB +0.0%
Query 16 42.5 MiB 43.4 MiB +2.1%
Query 17 150.3 MiB 150.3 MiB +0.0%
Query 18 127.3 MiB 127.3 MiB +0.0%
Query 19 228.6 KiB 228.6 KiB +0.0%
Query 20 60.0 MiB 59.4 MiB -1.1%
Query 21 37.7 MiB 37.7 MiB +0.0%
Query 22 32.0 MiB 41.6 MiB +30.1%

Pool accounting vs. process RSS

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

Benchmark Side Max pool peak Peak RSS Gap RSS / pool
tpch base (e66caad9042cf570a605da822e480a02845360b9) 155.6 MiB 1.1 GiB 1013.0 MiB 7.5×
tpch changed (0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd) 155.6 MiB 1.1 GiB 1011.6 MiB 7.5×
Resource Usage

tpch — base (merge-base)

Metric Value
Wall time 5.0s
Peak memory 1.1 GiB
Avg memory 649.1 MiB
CPU user 21.2s
CPU sys 1.9s
Peak spill 0 B

tpch — branch

Metric Value
Wall time 5.0s
Peak memory 1.1 GiB
Avg memory 639.0 MiB
CPU user 21.1s
CPU sys 1.9s
Peak spill 0 B

File an issue against this benchmark runner

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark completed (GKE) | trigger

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

Comparing 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark tpcds
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"
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_dictionary-wide-offset-reader
--------------------
Benchmark tpcds_sf1.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃       HEAD ┃ codex_dictionary-wide-offset-reader ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1  │    5.72 ms │                             5.41 ms │ +1.06x faster │
│ QQuery 2  │   77.33 ms │                            76.53 ms │     no change │
│ QQuery 3  │   28.77 ms │                            28.95 ms │     no change │
│ QQuery 4  │  458.81 ms │                           457.38 ms │     no change │
│ QQuery 5  │   51.17 ms │                            49.89 ms │     no change │
│ QQuery 6  │   34.33 ms │                            34.16 ms │     no change │
│ QQuery 7  │   71.64 ms │                            70.87 ms │     no change │
│ QQuery 8  │   38.38 ms │                            38.09 ms │     no change │
│ QQuery 9  │   51.23 ms │                            52.12 ms │     no change │
│ QQuery 10 │   58.67 ms │                            59.13 ms │     no change │
│ QQuery 11 │  278.82 ms │                           276.37 ms │     no change │
│ QQuery 12 │   28.94 ms │                            28.33 ms │     no change │
│ QQuery 13 │  111.44 ms │                           111.34 ms │     no change │
│ QQuery 14 │  382.62 ms │                           380.42 ms │     no change │
│ QQuery 15 │   52.02 ms │                            53.33 ms │     no change │
│ QQuery 16 │    6.28 ms │                             6.76 ms │  1.08x slower │
│ QQuery 17 │   76.19 ms │                            77.15 ms │     no change │
│ QQuery 18 │  106.59 ms │                           105.76 ms │     no change │
│ QQuery 19 │   41.47 ms │                            41.29 ms │     no change │
│ QQuery 20 │   35.34 ms │                            35.17 ms │     no change │
│ QQuery 21 │   17.52 ms │                            17.26 ms │     no change │
│ QQuery 22 │   69.90 ms │                            69.33 ms │     no change │
│ QQuery 23 │  300.30 ms │                           296.23 ms │     no change │
│ QQuery 24 │  193.09 ms │                           196.08 ms │     no change │
│ QQuery 25 │  105.37 ms │                           105.90 ms │     no change │
│ QQuery 26 │   45.80 ms │                            45.66 ms │     no change │
│ QQuery 27 │    6.84 ms │                             6.66 ms │     no change │
│ QQuery 28 │   57.60 ms │                            60.16 ms │     no change │
│ QQuery 29 │   93.04 ms │                            92.13 ms │     no change │
│ QQuery 30 │   33.58 ms │                            33.82 ms │     no change │
│ QQuery 31 │  110.48 ms │                           109.62 ms │     no change │
│ QQuery 32 │   20.74 ms │                            20.00 ms │     no change │
│ QQuery 33 │   36.52 ms │                            36.50 ms │     no change │
│ QQuery 34 │    9.40 ms │                             9.23 ms │     no change │
│ QQuery 35 │   70.46 ms │                            69.46 ms │     no change │
│ QQuery 36 │    6.43 ms │                             6.04 ms │ +1.06x faster │
│ QQuery 37 │    7.30 ms │                             7.03 ms │     no change │
│ QQuery 38 │   65.39 ms │                            65.14 ms │     no change │
│ QQuery 39 │   74.73 ms │                            73.13 ms │     no change │
│ QQuery 40 │   23.29 ms │                            23.56 ms │     no change │
│ QQuery 41 │   14.10 ms │                            14.16 ms │     no change │
│ QQuery 42 │   24.06 ms │                            23.97 ms │     no change │
│ QQuery 43 │    5.40 ms │                             5.07 ms │ +1.07x faster │
│ QQuery 44 │    9.35 ms │                             8.98 ms │     no change │
│ QQuery 45 │   36.59 ms │                            36.84 ms │     no change │
│ QQuery 46 │   11.75 ms │                            11.80 ms │     no change │
│ QQuery 47 │  210.72 ms │                           212.02 ms │     no change │
│ QQuery 48 │   91.97 ms │                            92.14 ms │     no change │
│ QQuery 49 │   69.52 ms │                            69.90 ms │     no change │
│ QQuery 50 │   57.98 ms │                            58.25 ms │     no change │
│ QQuery 51 │   96.15 ms │                            96.30 ms │     no change │
│ QQuery 52 │   24.47 ms │                            24.24 ms │     no change │
│ QQuery 53 │   29.61 ms │                            29.86 ms │     no change │
│ QQuery 54 │   54.05 ms │                            54.22 ms │     no change │
│ QQuery 55 │   24.08 ms │                            25.51 ms │  1.06x slower │
│ QQuery 56 │   38.36 ms │                            40.35 ms │  1.05x slower │
│ QQuery 57 │  156.60 ms │                           155.75 ms │     no change │
│ QQuery 58 │  105.33 ms │                           105.30 ms │     no change │
│ QQuery 59 │  108.38 ms │                           108.40 ms │     no change │
│ QQuery 60 │   38.84 ms │                            39.02 ms │     no change │
│ QQuery 61 │   10.78 ms │                            10.88 ms │     no change │
│ QQuery 62 │   38.62 ms │                            38.26 ms │     no change │
│ QQuery 63 │   29.81 ms │                            29.50 ms │     no change │
│ QQuery 64 │  359.94 ms │                           362.09 ms │     no change │
│ QQuery 65 │   75.92 ms │                            75.15 ms │     no change │
│ QQuery 66 │   74.60 ms │                            74.78 ms │     no change │
│ QQuery 67 │  241.44 ms │                           240.71 ms │     no change │
│ QQuery 68 │   11.97 ms │                            11.48 ms │     no change │
│ QQuery 69 │   55.42 ms │                            53.88 ms │     no change │
│ QQuery 70 │   97.82 ms │                            97.72 ms │     no change │
│ QQuery 71 │   35.34 ms │                            35.28 ms │     no change │
│ QQuery 72 │ 1562.49 ms │                          1719.23 ms │  1.10x slower │
│ QQuery 73 │    9.07 ms │                             9.34 ms │     no change │
│ QQuery 74 │  196.91 ms │                           204.78 ms │     no change │
│ QQuery 75 │  139.14 ms │                           137.72 ms │     no change │
│ QQuery 76 │   34.55 ms │                            34.96 ms │     no change │
│ QQuery 77 │   60.56 ms │                            60.39 ms │     no change │
│ QQuery 78 │  271.53 ms │                           278.82 ms │     no change │
│ QQuery 79 │   66.22 ms │                            68.21 ms │     no change │
│ QQuery 80 │   94.60 ms │                            95.76 ms │     no change │
│ QQuery 81 │   27.43 ms │                            28.17 ms │     no change │
│ QQuery 82 │   16.84 ms │                            17.07 ms │     no change │
│ QQuery 83 │   32.33 ms │                            33.26 ms │     no change │
│ QQuery 84 │   30.16 ms │                            30.38 ms │     no change │
│ QQuery 85 │   95.65 ms │                            95.67 ms │     no change │
│ QQuery 86 │   24.49 ms │                            24.62 ms │     no change │
│ QQuery 87 │   65.78 ms │                            66.73 ms │     no change │
│ QQuery 88 │   59.98 ms │                            59.61 ms │     no change │
│ QQuery 89 │   35.85 ms │                            35.90 ms │     no change │
│ QQuery 90 │   16.72 ms │                            16.67 ms │     no change │
│ QQuery 91 │   40.69 ms │                            40.26 ms │     no change │
│ QQuery 92 │   29.74 ms │                            29.49 ms │     no change │
│ QQuery 93 │   49.14 ms │                            49.10 ms │     no change │
│ QQuery 94 │   38.57 ms │                            38.67 ms │     no change │
│ QQuery 95 │   80.82 ms │                            81.72 ms │     no change │
│ QQuery 96 │   24.05 ms │                            24.07 ms │     no change │
│ QQuery 97 │   51.48 ms │                            50.72 ms │     no change │
│ QQuery 98 │   41.67 ms │                            41.80 ms │     no change │
│ QQuery 99 │   56.97 ms │                            56.99 ms │     no change │
└───────────┴────────────┴─────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 8835.97ms │
│ Total Time (codex_dictionary-wide-offset-reader)   │ 9003.30ms │
│ Average Time (HEAD)                                │   89.25ms │
│ Average Time (codex_dictionary-wide-offset-reader) │   90.94ms │
│ Queries Faster                                     │         3 │
│ Queries Slower                                     │         4 │
│ Queries with No Change                             │        92 │
│ Queries with Failure                               │         0 │
└────────────────────────────────────────────────────┴───────────┘

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

Comparing HEAD and codex_dictionary-wide-offset-reader
--------------------
Benchmark tpcds_sf1.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃                                  HEAD ┃   codex_dictionary-wide-offset-reader ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1  │           5.72 / 6.36 ±1.11 / 8.57 ms │           5.41 / 6.03 ±1.02 / 8.05 ms │ +1.05x faster │
│ QQuery 2  │        77.33 / 77.66 ±0.26 / 78.10 ms │        76.53 / 77.17 ±0.43 / 77.57 ms │     no change │
│ QQuery 3  │        28.77 / 29.04 ±0.20 / 29.38 ms │        28.95 / 29.07 ±0.10 / 29.26 ms │     no change │
│ QQuery 4  │     458.81 / 471.47 ±7.34 / 479.00 ms │     457.38 / 464.34 ±5.56 / 472.88 ms │     no change │
│ QQuery 5  │        51.17 / 51.67 ±0.42 / 52.27 ms │        49.89 / 52.47 ±3.02 / 58.36 ms │     no change │
│ QQuery 6  │        34.33 / 34.55 ±0.35 / 35.26 ms │        34.16 / 34.53 ±0.39 / 35.16 ms │     no change │
│ QQuery 7  │        71.64 / 72.44 ±0.62 / 73.51 ms │        70.87 / 71.29 ±0.34 / 71.73 ms │     no change │
│ QQuery 8  │        38.38 / 38.90 ±0.39 / 39.55 ms │        38.09 / 38.31 ±0.20 / 38.64 ms │     no change │
│ QQuery 9  │        51.23 / 54.82 ±3.94 / 62.10 ms │        52.12 / 53.54 ±1.09 / 55.45 ms │     no change │
│ QQuery 10 │        58.67 / 59.53 ±0.51 / 60.16 ms │        59.13 / 59.33 ±0.24 / 59.79 ms │     no change │
│ QQuery 11 │     278.82 / 287.80 ±8.19 / 301.27 ms │     276.37 / 289.02 ±7.91 / 301.01 ms │     no change │
│ QQuery 12 │        28.94 / 30.66 ±2.41 / 35.35 ms │        28.33 / 29.71 ±1.98 / 33.63 ms │     no change │
│ QQuery 13 │     111.44 / 112.66 ±1.32 / 115.12 ms │     111.34 / 112.60 ±1.07 / 113.89 ms │     no change │
│ QQuery 14 │     382.62 / 386.96 ±3.73 / 392.93 ms │     380.42 / 384.52 ±3.15 / 388.67 ms │     no change │
│ QQuery 15 │        52.02 / 52.85 ±0.67 / 53.81 ms │        53.33 / 54.76 ±1.38 / 57.37 ms │     no change │
│ QQuery 16 │           6.28 / 6.41 ±0.14 / 6.68 ms │           6.76 / 6.84 ±0.10 / 7.03 ms │  1.07x slower │
│ QQuery 17 │        76.19 / 79.53 ±4.44 / 88.19 ms │        77.15 / 78.33 ±1.12 / 79.91 ms │     no change │
│ QQuery 18 │     106.59 / 106.96 ±0.45 / 107.83 ms │     105.76 / 107.00 ±0.77 / 107.82 ms │     no change │
│ QQuery 19 │        41.47 / 41.76 ±0.25 / 42.17 ms │        41.29 / 41.58 ±0.19 / 41.79 ms │     no change │
│ QQuery 20 │        35.34 / 36.18 ±1.34 / 38.83 ms │        35.17 / 36.20 ±1.34 / 38.85 ms │     no change │
│ QQuery 21 │        17.52 / 17.84 ±0.20 / 18.03 ms │        17.26 / 17.63 ±0.33 / 18.22 ms │     no change │
│ QQuery 22 │        69.90 / 71.11 ±0.86 / 72.36 ms │        69.33 / 70.58 ±1.10 / 71.85 ms │     no change │
│ QQuery 23 │     300.30 / 306.85 ±6.42 / 318.18 ms │     296.23 / 301.26 ±3.63 / 305.32 ms │     no change │
│ QQuery 24 │     193.09 / 199.26 ±5.58 / 209.25 ms │     196.08 / 203.42 ±5.91 / 211.87 ms │     no change │
│ QQuery 25 │     105.37 / 107.62 ±1.71 / 109.37 ms │     105.90 / 107.88 ±1.99 / 110.29 ms │     no change │
│ QQuery 26 │        45.80 / 46.12 ±0.17 / 46.29 ms │        45.66 / 46.07 ±0.46 / 46.97 ms │     no change │
│ QQuery 27 │           6.84 / 6.89 ±0.04 / 6.95 ms │           6.66 / 6.75 ±0.06 / 6.83 ms │     no change │
│ QQuery 28 │        57.60 / 60.60 ±2.92 / 65.44 ms │        60.16 / 61.70 ±0.84 / 62.56 ms │     no change │
│ QQuery 29 │        93.04 / 94.51 ±1.92 / 97.97 ms │       92.13 / 94.69 ±2.96 / 100.25 ms │     no change │
│ QQuery 30 │        33.58 / 34.12 ±0.33 / 34.59 ms │        33.82 / 34.02 ±0.18 / 34.33 ms │     no change │
│ QQuery 31 │     110.48 / 112.23 ±1.50 / 114.39 ms │     109.62 / 110.76 ±0.88 / 112.06 ms │     no change │
│ QQuery 32 │        20.74 / 20.99 ±0.31 / 21.59 ms │        20.00 / 20.20 ±0.16 / 20.43 ms │     no change │
│ QQuery 33 │        36.52 / 38.69 ±2.65 / 43.80 ms │        36.50 / 36.86 ±0.35 / 37.50 ms │     no change │
│ QQuery 34 │          9.40 / 9.70 ±0.21 / 10.02 ms │           9.23 / 9.43 ±0.21 / 9.81 ms │     no change │
│ QQuery 35 │        70.46 / 73.42 ±2.93 / 78.16 ms │        69.46 / 71.48 ±1.80 / 74.37 ms │     no change │
│ QQuery 36 │           6.43 / 6.61 ±0.15 / 6.81 ms │           6.04 / 6.24 ±0.27 / 6.77 ms │ +1.06x faster │
│ QQuery 37 │           7.30 / 7.39 ±0.07 / 7.50 ms │           7.03 / 7.17 ±0.14 / 7.42 ms │     no change │
│ QQuery 38 │        65.39 / 66.12 ±0.65 / 67.18 ms │        65.14 / 65.93 ±0.79 / 66.91 ms │     no change │
│ QQuery 39 │        74.73 / 75.30 ±0.83 / 76.95 ms │        73.13 / 75.12 ±2.36 / 79.64 ms │     no change │
│ QQuery 40 │        23.29 / 24.57 ±1.91 / 28.35 ms │        23.56 / 24.28 ±0.51 / 24.93 ms │     no change │
│ QQuery 41 │        14.10 / 14.32 ±0.21 / 14.72 ms │        14.16 / 14.68 ±0.61 / 15.87 ms │     no change │
│ QQuery 42 │        24.06 / 24.67 ±0.40 / 25.15 ms │        23.97 / 24.47 ±0.46 / 25.05 ms │     no change │
│ QQuery 43 │           5.40 / 5.51 ±0.11 / 5.70 ms │           5.07 / 5.18 ±0.15 / 5.48 ms │ +1.06x faster │
│ QQuery 44 │           9.35 / 9.40 ±0.06 / 9.50 ms │           8.98 / 9.07 ±0.05 / 9.14 ms │     no change │
│ QQuery 45 │        36.59 / 37.70 ±0.86 / 39.21 ms │        36.84 / 37.75 ±0.54 / 38.49 ms │     no change │
│ QQuery 46 │        11.75 / 12.74 ±1.74 / 16.22 ms │        11.80 / 11.98 ±0.17 / 12.28 ms │ +1.06x faster │
│ QQuery 47 │     210.72 / 216.80 ±3.98 / 221.79 ms │     212.02 / 216.87 ±4.37 / 224.43 ms │     no change │
│ QQuery 48 │        91.97 / 92.72 ±0.74 / 93.65 ms │        92.14 / 92.80 ±0.79 / 94.32 ms │     no change │
│ QQuery 49 │        69.52 / 70.54 ±0.84 / 71.97 ms │        69.90 / 70.38 ±0.64 / 71.61 ms │     no change │
│ QQuery 50 │        57.98 / 60.59 ±3.00 / 66.18 ms │        58.25 / 60.23 ±3.69 / 67.60 ms │     no change │
│ QQuery 51 │       96.15 / 98.52 ±1.83 / 100.21 ms │       96.30 / 98.70 ±2.10 / 101.94 ms │     no change │
│ QQuery 52 │        24.47 / 24.74 ±0.26 / 25.21 ms │        24.24 / 24.76 ±0.28 / 25.09 ms │     no change │
│ QQuery 53 │        29.61 / 29.86 ±0.30 / 30.45 ms │        29.86 / 30.02 ±0.23 / 30.48 ms │     no change │
│ QQuery 54 │        54.05 / 56.29 ±3.39 / 63.01 ms │        54.22 / 57.81 ±4.65 / 66.79 ms │     no change │
│ QQuery 55 │        24.08 / 24.54 ±0.40 / 25.06 ms │        25.51 / 25.84 ±0.35 / 26.50 ms │  1.05x slower │
│ QQuery 56 │        38.36 / 39.05 ±0.56 / 40.08 ms │        40.35 / 41.84 ±0.87 / 42.86 ms │  1.07x slower │
│ QQuery 57 │     156.60 / 158.41 ±1.60 / 160.93 ms │     155.75 / 157.32 ±1.04 / 158.27 ms │     no change │
│ QQuery 58 │     105.33 / 107.63 ±2.15 / 111.67 ms │     105.30 / 107.32 ±2.04 / 109.82 ms │     no change │
│ QQuery 59 │     108.38 / 109.32 ±0.79 / 110.75 ms │     108.40 / 109.94 ±0.82 / 110.68 ms │     no change │
│ QQuery 60 │        38.84 / 39.83 ±0.76 / 40.87 ms │        39.02 / 39.81 ±0.83 / 41.34 ms │     no change │
│ QQuery 61 │        10.78 / 10.96 ±0.18 / 11.27 ms │        10.88 / 11.03 ±0.19 / 11.39 ms │     no change │
│ QQuery 62 │        38.62 / 38.95 ±0.32 / 39.55 ms │        38.26 / 38.45 ±0.18 / 38.72 ms │     no change │
│ QQuery 63 │        29.81 / 31.40 ±2.50 / 36.39 ms │        29.50 / 29.77 ±0.19 / 30.03 ms │ +1.05x faster │
│ QQuery 64 │     359.94 / 366.48 ±3.93 / 371.67 ms │     362.09 / 370.21 ±8.37 / 386.35 ms │     no change │
│ QQuery 65 │        75.92 / 80.00 ±3.26 / 85.16 ms │        75.15 / 76.60 ±1.32 / 78.67 ms │     no change │
│ QQuery 66 │        74.60 / 76.77 ±2.05 / 80.34 ms │        74.78 / 76.02 ±1.17 / 78.04 ms │     no change │
│ QQuery 67 │     241.44 / 245.42 ±2.81 / 248.92 ms │     240.71 / 243.05 ±1.79 / 245.53 ms │     no change │
│ QQuery 68 │        11.97 / 12.19 ±0.12 / 12.33 ms │        11.48 / 11.69 ±0.17 / 11.97 ms │     no change │
│ QQuery 69 │        55.42 / 57.27 ±3.12 / 63.48 ms │        53.88 / 57.16 ±5.10 / 67.30 ms │     no change │
│ QQuery 70 │       97.82 / 99.57 ±1.11 / 101.24 ms │       97.72 / 98.97 ±1.17 / 100.89 ms │     no change │
│ QQuery 71 │        35.34 / 35.54 ±0.24 / 36.00 ms │        35.28 / 35.42 ±0.13 / 35.66 ms │     no change │
│ QQuery 72 │ 1562.49 / 1684.95 ±83.59 / 1814.43 ms │ 1719.23 / 1756.50 ±22.94 / 1780.73 ms │     no change │
│ QQuery 73 │           9.07 / 9.28 ±0.23 / 9.72 ms │           9.34 / 9.51 ±0.19 / 9.87 ms │     no change │
│ QQuery 74 │     196.91 / 204.39 ±4.53 / 209.86 ms │     204.78 / 211.57 ±6.34 / 223.27 ms │     no change │
│ QQuery 75 │     139.14 / 139.88 ±1.17 / 142.21 ms │     137.72 / 140.19 ±1.75 / 142.25 ms │     no change │
│ QQuery 76 │        34.55 / 34.96 ±0.25 / 35.29 ms │        34.96 / 35.38 ±0.28 / 35.79 ms │     no change │
│ QQuery 77 │        60.56 / 60.98 ±0.41 / 61.70 ms │        60.39 / 61.16 ±0.53 / 61.79 ms │     no change │
│ QQuery 78 │     271.53 / 279.72 ±7.16 / 290.17 ms │     278.82 / 290.36 ±7.50 / 299.09 ms │     no change │
│ QQuery 79 │        66.22 / 70.57 ±3.58 / 77.10 ms │        68.21 / 69.76 ±0.98 / 70.78 ms │     no change │
│ QQuery 80 │        94.60 / 96.21 ±1.57 / 99.18 ms │        95.76 / 96.59 ±0.79 / 97.67 ms │     no change │
│ QQuery 81 │        27.43 / 27.70 ±0.17 / 27.96 ms │        28.17 / 31.30 ±6.08 / 43.45 ms │  1.13x slower │
│ QQuery 82 │        16.84 / 17.17 ±0.46 / 18.07 ms │        17.07 / 17.58 ±0.91 / 19.40 ms │     no change │
│ QQuery 83 │        32.33 / 34.66 ±3.91 / 42.45 ms │        33.26 / 33.57 ±0.23 / 33.92 ms │     no change │
│ QQuery 84 │        30.16 / 30.65 ±0.94 / 32.53 ms │        30.38 / 30.98 ±0.48 / 31.67 ms │     no change │
│ QQuery 85 │       95.65 / 97.43 ±2.35 / 102.07 ms │       95.67 / 99.24 ±4.15 / 107.24 ms │     no change │
│ QQuery 86 │        24.49 / 24.93 ±0.39 / 25.66 ms │        24.62 / 24.91 ±0.25 / 25.26 ms │     no change │
│ QQuery 87 │        65.78 / 66.88 ±0.79 / 68.18 ms │        66.73 / 67.46 ±0.60 / 68.45 ms │     no change │
│ QQuery 88 │        59.98 / 60.85 ±1.09 / 62.89 ms │        59.61 / 60.02 ±0.22 / 60.22 ms │     no change │
│ QQuery 89 │        35.85 / 36.35 ±0.50 / 37.23 ms │        35.90 / 36.06 ±0.10 / 36.17 ms │     no change │
│ QQuery 90 │        16.72 / 17.01 ±0.15 / 17.12 ms │        16.67 / 17.89 ±2.17 / 22.24 ms │  1.05x slower │
│ QQuery 91 │        40.69 / 41.20 ±0.49 / 42.04 ms │        40.26 / 40.99 ±0.61 / 41.88 ms │     no change │
│ QQuery 92 │        29.74 / 30.69 ±0.73 / 31.97 ms │        29.49 / 30.19 ±0.41 / 30.61 ms │     no change │
│ QQuery 93 │        49.14 / 50.55 ±0.81 / 51.38 ms │        49.10 / 50.40 ±0.94 / 52.03 ms │     no change │
│ QQuery 94 │        38.57 / 39.66 ±1.14 / 41.77 ms │        38.67 / 39.01 ±0.20 / 39.26 ms │     no change │
│ QQuery 95 │        80.82 / 83.25 ±2.91 / 88.90 ms │        81.72 / 82.29 ±0.41 / 82.93 ms │     no change │
│ QQuery 96 │        24.05 / 24.28 ±0.21 / 24.67 ms │        24.07 / 24.29 ±0.18 / 24.60 ms │     no change │
│ QQuery 97 │        51.48 / 52.51 ±1.49 / 55.47 ms │        50.72 / 52.26 ±1.69 / 54.95 ms │     no change │
│ QQuery 98 │        41.67 / 42.83 ±0.72 / 43.87 ms │        41.80 / 42.43 ±0.38 / 42.95 ms │     no change │
│ QQuery 99 │        56.97 / 57.57 ±0.51 / 58.52 ms │        56.99 / 58.14 ±1.62 / 61.30 ms │     no change │
└───────────┴───────────────────────────────────────┴───────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 9122.95ms │
│ Total Time (codex_dictionary-wide-offset-reader)   │ 9193.31ms │
│ Average Time (HEAD)                                │   92.15ms │
│ Average Time (codex_dictionary-wide-offset-reader) │   92.86ms │
│ Queries Faster                                     │         5 │
│ Queries Slower                                     │         5 │
│ Queries with No Change                             │        89 │
│ Queries with Failure                               │         0 │
└────────────────────────────────────────────────────┴───────────┘

Memory Pool Peaks

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

Base: e66caad9042cf570a605da822e480a02845360b9 | Changed: 0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd

tpcds — tpcds_sf1

Query Base Changed Change
Query 1 135.4 KiB 197.5 KiB +45.8%
Query 2 70.9 MiB 70.9 MiB -0.0%
Query 3 1.9 MiB 1.8 MiB -2.3%
Query 4 81.9 MiB 81.8 MiB -0.0%
Query 5 13.0 MiB 13.0 MiB -0.1%
Query 6 4.4 MiB 5.0 MiB +14.1%
Query 7 3.7 MiB 3.6 MiB -2.2%
Query 8 1.7 MiB 1.7 MiB +0.0%
Query 9 32.1 KiB 32.1 KiB +0.0%
Query 10 2.0 MiB 2.0 MiB +0.0%
Query 11 56.6 MiB 56.7 MiB +0.1%
Query 12 116.3 MiB 116.3 MiB -0.0%
Query 13 13.7 MiB 13.7 MiB +0.0%
Query 14 170.9 MiB 170.9 MiB +0.0%
Query 15 4.2 MiB 4.2 MiB -0.2%
Query 16 118.6 KiB 118.6 KiB +0.0%
Query 17 41.1 MiB 41.6 MiB +1.1%
Query 18 21.2 MiB 21.2 MiB +0.0%
Query 19 2.1 MiB 2.1 MiB -1.0%
Query 20 116.6 MiB 116.6 MiB -0.0%
Query 21 4.0 MiB 3.3 MiB -16.3%
Query 22 61.8 MiB 61.8 MiB +0.0%
Query 23 212.7 MiB 211.9 MiB -0.4%
Query 24 38.7 MiB 38.7 MiB +0.0%
Query 25 99.7 MiB 99.7 MiB +0.0%
Query 26 3.6 MiB 3.6 MiB -0.6%
Query 27 1.6 MiB 1.6 MiB +0.0%
Query 28 1.3 MiB 1.5 MiB +9.9%
Query 29 104.8 MiB 104.8 MiB +0.0%
Query 30 4.1 MiB 4.0 MiB -2.9%
Query 31 13.4 MiB 13.4 MiB +0.0%
Query 32 2.9 MiB 2.7 MiB -7.5%
Query 33 4.2 MiB 4.2 MiB +0.0%
Query 34 6.0 MiB 6.0 MiB +0.0%
Query 35 16.6 MiB 17.5 MiB +5.3%
Query 36 10.3 MiB 10.3 MiB +0.0%
Query 37 3.9 MiB 3.9 MiB +0.0%
Query 38 10.4 MiB 10.8 MiB +3.4%
Query 39 43.3 MiB 52.0 MiB +20.1%
Query 40 12.2 MiB 12.2 MiB +0.0%
Query 41 2.1 MiB 2.1 MiB +0.0%
Query 42 1.5 MiB 1.3 MiB -14.1%
Query 43 275.0 KiB 275.0 KiB +0.0%
Query 44 233.2 KiB 233.2 KiB +0.0%
Query 45 4.7 MiB 4.5 MiB -4.0%
Query 46 6.2 MiB 6.2 MiB +0.0%
Query 47 128.5 MiB 128.5 MiB +0.0%
Query 48 11.8 MiB 11.8 MiB +0.0%
Query 49 43.2 MiB 42.5 MiB -1.5%
Query 50 17.3 MiB 17.3 MiB +0.1%
Query 51 177.7 MiB 177.7 MiB -0.0%
Query 52 1.5 MiB 1.5 MiB -0.7%
Query 53 30.4 MiB 30.6 MiB +0.5%
Query 54 2.2 MiB 2.1 MiB -4.2%
Query 55 1.3 MiB 1.7 MiB +34.4%
Query 56 2.7 MiB 3.0 MiB +8.4%
Query 57 125.6 MiB 120.8 MiB -3.8%
Query 58 4.4 MiB 4.2 MiB -4.8%
Query 59 11.7 MiB 11.7 MiB +0.0%
Query 60 5.5 MiB 5.4 MiB -0.3%
Query 61 1.2 MiB 1001.1 KiB -15.0%
Query 62 3.7 MiB 3.3 MiB -9.1%
Query 63 30.6 MiB 30.6 MiB +0.0%
Query 64 116.2 MiB 95.3 MiB -18.0%
Query 65 27.0 MiB 27.0 MiB +0.1%
Query 66 14.9 MiB 14.9 MiB +0.0%
Query 67 141.7 MiB 136.9 MiB -3.4%
Query 68 6.2 MiB 6.2 MiB +0.0%
Query 69 3.6 MiB 4.1 MiB +14.6%
Query 70 44.0 MiB 44.0 MiB -0.0%
Query 71 40.3 MiB 40.6 MiB +0.7%
Query 72 97.3 MiB 97.3 MiB +0.0%
Query 73 6.0 MiB 6.0 MiB +0.0%
Query 74 36.0 MiB 36.0 MiB +0.0%
Query 75 23.1 MiB 23.1 MiB -0.0%
Query 76 4.3 MiB 4.3 MiB +0.0%
Query 77 4.1 MiB 3.9 MiB -3.6%
Query 78 335.7 MiB 335.7 MiB +0.0%
Query 79 8.7 MiB 8.7 MiB +0.0%
Query 80 42.4 MiB 41.7 MiB -1.7%
Query 81 4.2 MiB 4.2 MiB -0.1%
Query 82 3.9 MiB 3.9 MiB +0.0%
Query 83 9.0 MiB 9.7 MiB +7.7%
Query 84 11.9 MiB 11.9 MiB +0.0%
Query 85 17.0 MiB 17.0 MiB +0.0%
Query 86 82.1 MiB 82.1 MiB +0.0%
Query 87 12.9 MiB 12.9 MiB +0.0%
Query 88 1.8 MiB 1.7 MiB -8.0%
Query 89 112.2 MiB 112.2 MiB +0.0%
Query 90 351.7 KiB 357.5 KiB +1.6%
Query 91 31.4 MiB 41.3 MiB +31.3%
Query 92 2.7 MiB 2.7 MiB +0.0%
Query 93 15.4 MiB 15.1 MiB -1.6%
Query 94 3.4 MiB 3.7 MiB +8.2%
Query 95 24.4 MiB 24.4 MiB +0.0%
Query 96 224.7 KiB 224.7 KiB +0.0%
Query 97 33.3 MiB 33.1 MiB -0.5%
Query 98 116.7 MiB 116.7 MiB +0.0%
Query 99 1.9 MiB 1.6 MiB -15.1%

Pool accounting vs. process RSS

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

Benchmark Side Max pool peak Peak RSS Gap RSS / pool
tpcds base (e66caad9042cf570a605da822e480a02845360b9) 335.7 MiB 2.0 GiB 1.7 GiB 6.2×
tpcds changed (0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd) 335.7 MiB 1.9 GiB 1.6 GiB 5.8×
Resource Usage

tpcds — base (merge-base)

Metric Value
Wall time 50.0s
Peak memory 2.0 GiB
Avg memory 1.4 GiB
CPU user 189.1s
CPU sys 5.3s
Peak spill 0 B

tpcds — branch

Metric Value
Wall time 50.0s
Peak memory 1.9 GiB
Avg memory 1.3 GiB
CPU user 195.8s
CPU sys 5.4s
Peak spill 0 B

File an issue against this benchmark runner

@Rich-T-kid

Rich-T-kid commented Nov 9, 2026

Copy link
Copy Markdown
Contributor Author

run benchmark tpcds

env:
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark completed (GKE) | trigger

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

Comparing 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark clickbench_partitioned
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"
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_dictionary-wide-offset-reader
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃       HEAD ┃ codex_dictionary-wide-offset-reader ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0  │    1.20 ms │                             1.17 ms │     no change │
│ QQuery 1  │   11.76 ms │                            12.30 ms │     no change │
│ QQuery 2  │   36.34 ms │                            35.60 ms │     no change │
│ QQuery 3  │   31.47 ms │                            31.97 ms │     no change │
│ QQuery 4  │  249.70 ms │                           258.00 ms │     no change │
│ QQuery 5  │  280.00 ms │                           287.46 ms │     no change │
│ QQuery 6  │    1.21 ms │                             1.22 ms │     no change │
│ QQuery 7  │   13.40 ms │                            13.41 ms │     no change │
│ QQuery 8  │  348.53 ms │                           351.79 ms │     no change │
│ QQuery 9  │  474.38 ms │                           489.34 ms │     no change │
│ QQuery 10 │  200.11 ms │                            66.05 ms │ +3.03x faster │
│ QQuery 11 │  213.04 ms │                            76.67 ms │ +2.78x faster │
│ QQuery 12 │  279.88 ms │                           279.37 ms │     no change │
│ QQuery 13 │  381.91 ms │                           378.39 ms │     no change │
│ QQuery 14 │  293.07 ms │                           293.03 ms │     no change │
│ QQuery 15 │  297.59 ms │                           297.95 ms │     no change │
│ QQuery 16 │  663.46 ms │                           632.43 ms │     no change │
│ QQuery 17 │  659.37 ms │                           636.72 ms │     no change │
│ QQuery 18 │ 1286.44 ms │                          1324.36 ms │     no change │
│ QQuery 19 │   28.57 ms │                            27.99 ms │     no change │
│ QQuery 20 │  523.57 ms │                           525.74 ms │     no change │
│ QQuery 21 │  506.16 ms │                           511.00 ms │     no change │
│ QQuery 22 │  996.41 ms │                           988.57 ms │     no change │
│ QQuery 23 │ 6228.05 ms │                          3095.13 ms │ +2.01x faster │
│ QQuery 24 │   41.70 ms │                            41.40 ms │     no change │
│ QQuery 25 │  108.13 ms │                           109.75 ms │     no change │
│ QQuery 26 │   41.73 ms │                            42.66 ms │     no change │
│ QQuery 27 │  537.62 ms │                           513.57 ms │     no change │
│ QQuery 28 │ 2923.90 ms │                          2927.75 ms │     no change │
│ QQuery 29 │   38.63 ms │                            39.67 ms │     no change │
│ QQuery 30 │  318.42 ms │                           323.67 ms │     no change │
│ QQuery 31 │  293.10 ms │                           292.35 ms │     no change │
│ QQuery 32 │ 1020.03 ms │                          1009.68 ms │     no change │
│ QQuery 33 │ 1480.25 ms │                          1535.66 ms │     no change │
│ QQuery 34 │ 1558.99 ms │                          1525.17 ms │     no change │
│ QQuery 35 │  322.89 ms │                           296.93 ms │ +1.09x faster │
│ QQuery 36 │   67.91 ms │                            67.62 ms │     no change │
│ QQuery 37 │   36.56 ms │                            35.92 ms │     no change │
│ QQuery 38 │   41.52 ms │                            40.35 ms │     no change │
│ QQuery 39 │  148.73 ms │                           148.44 ms │     no change │
│ QQuery 40 │   14.33 ms │                            14.80 ms │     no change │
│ QQuery 41 │   13.37 ms │                            14.04 ms │  1.05x slower │
│ QQuery 42 │   12.89 ms │                            13.74 ms │  1.07x slower │
└───────────┴────────────┴─────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 23026.35ms │
│ Total Time (codex_dictionary-wide-offset-reader)   │ 19608.82ms │
│ Average Time (HEAD)                                │   535.50ms │
│ Average Time (codex_dictionary-wide-offset-reader) │   456.02ms │
│ Queries Faster                                     │          4 │
│ Queries Slower                                     │          2 │
│ Queries with No Change                             │         37 │
│ Queries with Failure                               │          0 │
└────────────────────────────────────────────────────┴────────────┘

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

Comparing HEAD and codex_dictionary-wide-offset-reader
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃                                  HEAD ┃   codex_dictionary-wide-offset-reader ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0  │           1.20 / 1.87 ±1.19 / 4.25 ms │           1.17 / 1.81 ±1.11 / 4.03 ms │     no change │
│ QQuery 1  │        11.76 / 12.01 ±0.18 / 12.23 ms │        12.30 / 12.79 ±0.28 / 13.14 ms │  1.06x slower │
│ QQuery 2  │        36.34 / 37.09 ±0.79 / 38.45 ms │        35.60 / 36.33 ±0.55 / 37.21 ms │     no change │
│ QQuery 3  │        31.47 / 32.28 ±0.70 / 33.40 ms │        31.97 / 32.60 ±0.73 / 33.89 ms │     no change │
│ QQuery 4  │     249.70 / 253.04 ±4.87 / 262.64 ms │     258.00 / 261.97 ±2.21 / 264.70 ms │     no change │
│ QQuery 5  │     280.00 / 286.49 ±5.08 / 294.52 ms │     287.46 / 291.60 ±2.83 / 295.24 ms │     no change │
│ QQuery 6  │           1.21 / 1.37 ±0.26 / 1.88 ms │           1.22 / 1.38 ±0.25 / 1.87 ms │     no change │
│ QQuery 7  │        13.40 / 15.25 ±3.53 / 22.31 ms │        13.41 / 13.83 ±0.23 / 14.04 ms │ +1.10x faster │
│ QQuery 8  │     348.53 / 353.19 ±3.71 / 357.50 ms │     351.79 / 355.30 ±3.08 / 360.01 ms │     no change │
│ QQuery 9  │     474.38 / 486.12 ±9.35 / 500.18 ms │     489.34 / 498.25 ±9.64 / 515.20 ms │     no change │
│ QQuery 10 │     200.11 / 203.69 ±4.11 / 211.74 ms │        66.05 / 67.93 ±1.68 / 70.69 ms │ +3.00x faster │
│ QQuery 11 │     213.04 / 214.20 ±0.66 / 215.04 ms │        76.67 / 78.89 ±1.45 / 81.16 ms │ +2.72x faster │
│ QQuery 12 │     279.88 / 286.36 ±5.23 / 294.95 ms │     279.37 / 290.46 ±9.91 / 308.86 ms │     no change │
│ QQuery 13 │    381.91 / 392.57 ±12.14 / 415.31 ms │    378.39 / 398.05 ±20.49 / 431.85 ms │     no change │
│ QQuery 14 │     293.07 / 299.82 ±4.34 / 305.10 ms │     293.03 / 301.65 ±9.83 / 314.52 ms │     no change │
│ QQuery 15 │    297.59 / 316.35 ±19.80 / 344.65 ms │    297.95 / 311.73 ±13.81 / 334.51 ms │     no change │
│ QQuery 16 │    663.46 / 674.74 ±11.08 / 694.95 ms │     632.43 / 647.61 ±8.89 / 656.76 ms │     no change │
│ QQuery 17 │    659.37 / 672.98 ±18.32 / 709.22 ms │    636.72 / 662.11 ±18.99 / 686.22 ms │     no change │
│ QQuery 18 │ 1286.44 / 1315.49 ±28.72 / 1364.27 ms │ 1324.36 / 1352.26 ±24.33 / 1380.91 ms │     no change │
│ QQuery 19 │        28.57 / 29.36 ±0.79 / 30.68 ms │        27.99 / 30.09 ±3.70 / 37.48 ms │     no change │
│ QQuery 20 │    523.57 / 534.30 ±10.06 / 548.13 ms │    525.74 / 540.72 ±19.58 / 578.71 ms │     no change │
│ QQuery 21 │    506.16 / 526.95 ±14.90 / 547.40 ms │     511.00 / 520.53 ±5.25 / 526.85 ms │     no change │
│ QQuery 22 │  996.41 / 1011.75 ±12.92 / 1028.53 ms │   988.57 / 1004.02 ±8.49 / 1011.86 ms │     no change │
│ QQuery 23 │ 6228.05 / 6276.12 ±30.89 / 6317.43 ms │ 3095.13 / 3172.77 ±47.29 / 3223.53 ms │ +1.98x faster │
│ QQuery 24 │       41.70 / 50.08 ±13.95 / 77.90 ms │        41.40 / 42.20 ±1.16 / 44.45 ms │ +1.19x faster │
│ QQuery 25 │     108.13 / 109.77 ±1.14 / 111.02 ms │     109.75 / 111.43 ±1.05 / 112.46 ms │     no change │
│ QQuery 26 │        41.73 / 43.72 ±1.66 / 46.00 ms │        42.66 / 43.95 ±1.32 / 46.33 ms │     no change │
│ QQuery 27 │    537.62 / 559.46 ±11.75 / 569.11 ms │     513.57 / 526.67 ±9.22 / 540.45 ms │ +1.06x faster │
│ QQuery 28 │ 2923.90 / 2955.47 ±20.64 / 2986.98 ms │ 2927.75 / 2954.75 ±19.67 / 2985.59 ms │     no change │
│ QQuery 29 │        38.63 / 39.15 ±0.50 / 40.01 ms │       39.67 / 51.82 ±11.58 / 71.07 ms │  1.32x slower │
│ QQuery 30 │    318.42 / 332.57 ±13.59 / 352.89 ms │     323.67 / 332.86 ±6.78 / 341.28 ms │     no change │
│ QQuery 31 │    293.10 / 306.37 ±18.06 / 341.01 ms │    292.35 / 303.67 ±10.17 / 321.18 ms │     no change │
│ QQuery 32 │ 1020.03 / 1054.36 ±28.14 / 1103.48 ms │ 1009.68 / 1030.66 ±25.26 / 1078.57 ms │     no change │
│ QQuery 33 │ 1480.25 / 1544.41 ±34.04 / 1580.24 ms │ 1535.66 / 1580.93 ±37.66 / 1628.93 ms │     no change │
│ QQuery 34 │ 1558.99 / 1592.48 ±27.21 / 1634.06 ms │ 1525.17 / 1588.17 ±49.48 / 1668.63 ms │     no change │
│ QQuery 35 │    322.89 / 360.51 ±48.14 / 454.14 ms │    296.93 / 317.58 ±16.95 / 348.64 ms │ +1.14x faster │
│ QQuery 36 │        67.91 / 73.64 ±5.78 / 83.57 ms │        67.62 / 72.51 ±5.19 / 81.76 ms │     no change │
│ QQuery 37 │        36.56 / 42.74 ±4.16 / 47.20 ms │        35.92 / 40.49 ±7.63 / 55.70 ms │ +1.06x faster │
│ QQuery 38 │        41.52 / 47.26 ±4.65 / 55.44 ms │        40.35 / 42.63 ±1.66 / 44.54 ms │ +1.11x faster │
│ QQuery 39 │     148.73 / 153.07 ±2.81 / 156.92 ms │     148.44 / 161.87 ±7.22 / 169.80 ms │  1.06x slower │
│ QQuery 40 │        14.33 / 16.68 ±3.59 / 23.79 ms │        14.80 / 15.53 ±0.55 / 16.45 ms │ +1.07x faster │
│ QQuery 41 │        13.37 / 15.25 ±3.28 / 21.81 ms │        14.04 / 14.28 ±0.17 / 14.51 ms │ +1.07x faster │
│ QQuery 42 │        12.89 / 13.19 ±0.24 / 13.51 ms │        13.74 / 14.12 ±0.27 / 14.34 ms │  1.07x slower │
└───────────┴───────────────────────────────────────┴───────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 23543.56ms │
│ Total Time (codex_dictionary-wide-offset-reader)   │ 20130.79ms │
│ Average Time (HEAD)                                │   547.52ms │
│ Average Time (codex_dictionary-wide-offset-reader) │   468.16ms │
│ Queries Faster                                     │         11 │
│ Queries Slower                                     │          4 │
│ Queries with No Change                             │         28 │
│ Queries with Failure                               │          0 │
└────────────────────────────────────────────────────┴────────────┘

Memory Pool Peaks

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

Base: e66caad9042cf570a605da822e480a02845360b9 | Changed: 0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd

clickbench_partitioned

Query Base Changed Change
Query 0 0 B 0 B 0.0%
Query 1 104 B 104 B +0.0%
Query 2 936 B 936 B +0.0%
Query 3 312 B 312 B +0.0%
Query 4 829.0 MiB 810.9 MiB -2.2%
Query 5 1.2 GiB 1.2 GiB +5.2%
Query 6 0 B 0 B 0.0%
Query 7 50.3 MiB 60.2 MiB +19.5%
Query 8 954.1 MiB 965.3 MiB +1.2%
Query 9 552.0 MiB 636.0 MiB +15.2%
Query 10 73.3 MiB 63.8 MiB -13.1%
Query 11 77.4 MiB 70.3 MiB -9.1%
Query 12 1.3 GiB 1.4 GiB +3.1%
Query 13 1.6 GiB 1.7 GiB +5.6%
Query 14 1.3 GiB 1.3 GiB -2.1%
Query 15 1.2 GiB 1.2 GiB -1.5%
Query 16 3.4 GiB 3.4 GiB +1.6%
Query 17 3.4 GiB 3.5 GiB +1.6%
Query 18 8.3 GiB 8.4 GiB +1.4%
Query 19 0 B 0 B 0.0%
Query 20 104 B 104 B +0.0%
Query 21 3.7 MiB 4.0 MiB +8.9%
Query 22 3.6 MiB 3.4 MiB -4.7%
Query 23 11.0 MiB 11.9 MiB +7.8%
Query 24 62.1 MiB 58.3 MiB -6.1%
Query 25 188.5 MiB 170.2 MiB -9.7%
Query 26 61.5 MiB 60.6 MiB -1.4%
Query 27 2.2 MiB 2.5 MiB +9.7%
Query 28 1.8 GiB 1.8 GiB +3.4%
Query 29 624 B 624 B +0.0%
Query 30 704.1 MiB 722.4 MiB +2.6%
Query 31 1.5 GiB 1.5 GiB +3.1%
Query 32 7.1 GiB 7.1 GiB -0.0%
Query 33 7.7 GiB 8.2 GiB +5.9%
Query 34 8.1 GiB 7.8 GiB -3.5%
Query 35 664.8 MiB 678.4 MiB +2.0%
Query 36 117.5 MiB 118.5 MiB +0.8%
Query 37 6.6 MiB 7.3 MiB +9.4%
Query 38 5.3 MiB 5.8 MiB +9.9%
Query 39 308.8 MiB 308.8 MiB -0.0%
Query 40 2.6 MiB 2.6 MiB -0.3%
Query 41 3.3 MiB 3.3 MiB +0.0%
Query 42 1.7 MiB 1.9 MiB +10.7%

Pool accounting vs. process RSS

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

Benchmark Side Max pool peak Peak RSS Gap RSS / pool
clickbench_partitioned base (e66caad9042cf570a605da822e480a02845360b9) 8.3 GiB 12.5 GiB 4.2 GiB 1.5×
clickbench_partitioned changed (0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd) 8.4 GiB 13.0 GiB 4.6 GiB 1.5×
Resource Usage

clickbench_partitioned — base (merge-base)

Metric Value
Wall time 120.0s
Peak memory 12.5 GiB
Avg memory 4.2 GiB
CPU user 1221.1s
CPU sys 77.0s
Peak spill 0 B

clickbench_partitioned — branch

Metric Value
Wall time 105.0s
Peak memory 13.0 GiB
Avg memory 4.5 GiB
CPU user 1026.3s
CPU sys 75.6s
Peak spill 0 B

File an issue against this benchmark runner

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark running (GKE) | trigger
Instance: c4a-highmem-16 (12 vCPU / 65 GiB) | Linux bench-c6742572322-4685-548rk 6.12.105+ #1 SMP Sat Sep 19 07:55:25 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 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark tpcds
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"

Results will be posted here when complete

File an issue against this benchmark runner

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark completed (GKE) | trigger

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

Comparing 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark tpcds
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"
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_dictionary-wide-offset-reader
--------------------
Benchmark tpcds_sf1.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┓
┃ Query     ┃       HEAD ┃ codex_dictionary-wide-offset-reader ┃       Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━┩
│ QQuery 1  │    5.48 ms │                             5.26 ms │    no change │
│ QQuery 2  │   76.36 ms │                            76.86 ms │    no change │
│ QQuery 3  │   29.07 ms │                            28.89 ms │    no change │
│ QQuery 4  │  436.04 ms │                           440.93 ms │    no change │
│ QQuery 5  │   49.85 ms │                            51.69 ms │    no change │
│ QQuery 6  │   34.26 ms │                            33.81 ms │    no change │
│ QQuery 7  │   71.58 ms │                            71.95 ms │    no change │
│ QQuery 8  │   38.39 ms │                            38.33 ms │    no change │
│ QQuery 9  │   53.61 ms │                            51.52 ms │    no change │
│ QQuery 10 │   59.07 ms │                            59.00 ms │    no change │
│ QQuery 11 │  268.56 ms │                           277.74 ms │    no change │
│ QQuery 12 │   27.68 ms │                            28.17 ms │    no change │
│ QQuery 13 │  111.36 ms │                           112.44 ms │    no change │
│ QQuery 14 │  377.95 ms │                           380.15 ms │    no change │
│ QQuery 15 │   52.04 ms │                            52.07 ms │    no change │
│ QQuery 16 │    5.93 ms │                             6.54 ms │ 1.10x slower │
│ QQuery 17 │   75.81 ms │                            75.97 ms │    no change │
│ QQuery 18 │  105.31 ms │                           104.88 ms │    no change │
│ QQuery 19 │   41.43 ms │                            41.39 ms │    no change │
│ QQuery 20 │   34.35 ms │                            34.64 ms │    no change │
│ QQuery 21 │   17.26 ms │                            17.37 ms │    no change │
│ QQuery 22 │   69.82 ms │                            70.04 ms │    no change │
│ QQuery 23 │  291.57 ms │                           301.13 ms │    no change │
│ QQuery 24 │  193.27 ms │                           199.34 ms │    no change │
│ QQuery 25 │  104.27 ms │                           105.06 ms │    no change │
│ QQuery 26 │   45.41 ms │                            44.89 ms │    no change │
│ QQuery 27 │    6.31 ms │                             6.38 ms │    no change │
│ QQuery 28 │   57.24 ms │                            56.98 ms │    no change │
│ QQuery 29 │   92.07 ms │                            91.64 ms │    no change │
│ QQuery 30 │   33.10 ms │                            32.96 ms │    no change │
│ QQuery 31 │  109.88 ms │                           109.38 ms │    no change │
│ QQuery 32 │   20.27 ms │                            20.26 ms │    no change │
│ QQuery 33 │   36.71 ms │                            36.54 ms │    no change │
│ QQuery 34 │    9.62 ms │                             9.25 ms │    no change │
│ QQuery 35 │   71.31 ms │                            70.25 ms │    no change │
│ QQuery 36 │    5.80 ms │                             5.88 ms │    no change │
│ QQuery 37 │    7.07 ms │                             6.99 ms │    no change │
│ QQuery 38 │   63.92 ms │                            64.88 ms │    no change │
│ QQuery 39 │   71.89 ms │                            72.29 ms │    no change │
│ QQuery 40 │   23.02 ms │                            23.09 ms │    no change │
│ QQuery 41 │   13.91 ms │                            13.79 ms │    no change │
│ QQuery 42 │   23.78 ms │                            24.06 ms │    no change │
│ QQuery 43 │    4.89 ms │                             4.86 ms │    no change │
│ QQuery 44 │    8.76 ms │                             8.73 ms │    no change │
│ QQuery 45 │   34.67 ms │                            35.29 ms │    no change │
│ QQuery 46 │   11.14 ms │                            11.09 ms │    no change │
│ QQuery 47 │  204.27 ms │                           205.57 ms │    no change │
│ QQuery 48 │   89.45 ms │                            90.37 ms │    no change │
│ QQuery 49 │   68.76 ms │                            69.39 ms │    no change │
│ QQuery 50 │   56.73 ms │                            57.21 ms │    no change │
│ QQuery 51 │   94.03 ms │                            93.51 ms │    no change │
│ QQuery 52 │   24.36 ms │                            24.67 ms │    no change │
│ QQuery 53 │   29.64 ms │                            29.49 ms │    no change │
│ QQuery 54 │   53.61 ms │                            53.96 ms │    no change │
│ QQuery 55 │   23.86 ms │                            24.36 ms │    no change │
│ QQuery 56 │   37.73 ms │                            38.17 ms │    no change │
│ QQuery 57 │  152.90 ms │                           153.57 ms │    no change │
│ QQuery 58 │  104.20 ms │                           104.07 ms │    no change │
│ QQuery 59 │  106.65 ms │                           107.66 ms │    no change │
│ QQuery 60 │   38.45 ms │                            38.54 ms │    no change │
│ QQuery 61 │   10.64 ms │                            10.31 ms │    no change │
│ QQuery 62 │   37.91 ms │                            37.99 ms │    no change │
│ QQuery 63 │   29.32 ms │                            29.57 ms │    no change │
│ QQuery 64 │  354.93 ms │                           355.83 ms │    no change │
│ QQuery 65 │   74.26 ms │                            74.22 ms │    no change │
│ QQuery 66 │   71.15 ms │                            72.86 ms │    no change │
│ QQuery 67 │  228.80 ms │                           229.33 ms │    no change │
│ QQuery 68 │   11.26 ms │                            11.37 ms │    no change │
│ QQuery 69 │   53.49 ms │                            53.76 ms │    no change │
│ QQuery 70 │   97.03 ms │                            97.70 ms │    no change │
│ QQuery 71 │   34.91 ms │                            35.06 ms │    no change │
│ QQuery 72 │ 1553.86 ms │                          1658.75 ms │ 1.07x slower │
│ QQuery 73 │    9.38 ms │                             9.29 ms │    no change │
│ QQuery 74 │  199.36 ms │                           200.50 ms │    no change │
│ QQuery 75 │  139.00 ms │                           137.46 ms │    no change │
│ QQuery 76 │   34.73 ms │                            34.17 ms │    no change │
│ QQuery 77 │   59.82 ms │                            59.67 ms │    no change │
│ QQuery 78 │  267.12 ms │                           262.70 ms │    no change │
│ QQuery 79 │   67.91 ms │                            67.05 ms │    no change │
│ QQuery 80 │   93.61 ms │                            94.74 ms │    no change │
│ QQuery 81 │   26.59 ms │                            27.29 ms │    no change │
│ QQuery 82 │   16.34 ms │                            16.66 ms │    no change │
│ QQuery 83 │   32.63 ms │                            32.49 ms │    no change │
│ QQuery 84 │   29.97 ms │                            29.81 ms │    no change │
│ QQuery 85 │   94.40 ms │                            95.80 ms │    no change │
│ QQuery 86 │   24.18 ms │                            24.41 ms │    no change │
│ QQuery 87 │   64.48 ms │                            64.45 ms │    no change │
│ QQuery 88 │   59.27 ms │                            59.06 ms │    no change │
│ QQuery 89 │   35.73 ms │                            36.11 ms │    no change │
│ QQuery 90 │   16.08 ms │                            16.07 ms │    no change │
│ QQuery 91 │   39.34 ms │                            40.27 ms │    no change │
│ QQuery 92 │   28.54 ms │                            29.19 ms │    no change │
│ QQuery 93 │   49.05 ms │                            49.22 ms │    no change │
│ QQuery 94 │   37.99 ms │                            38.67 ms │    no change │
│ QQuery 95 │   77.80 ms │                            80.74 ms │    no change │
│ QQuery 96 │   23.68 ms │                            23.74 ms │    no change │
│ QQuery 97 │   50.33 ms │                            50.28 ms │    no change │
│ QQuery 98 │   40.87 ms │                            41.41 ms │    no change │
│ QQuery 99 │   56.59 ms │                            56.32 ms │    no change │
└───────────┴────────────┴─────────────────────────────────────┴──────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 8698.10ms │
│ Total Time (codex_dictionary-wide-offset-reader)   │ 8849.50ms │
│ Average Time (HEAD)                                │   87.86ms │
│ Average Time (codex_dictionary-wide-offset-reader) │   89.39ms │
│ Queries Faster                                     │         0 │
│ Queries Slower                                     │         2 │
│ Queries with No Change                             │        97 │
│ Queries with Failure                               │         0 │
└────────────────────────────────────────────────────┴───────────┘

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

Comparing HEAD and codex_dictionary-wide-offset-reader
--------------------
Benchmark tpcds_sf1.json
--------------------
┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query     ┃                                  HEAD ┃   codex_dictionary-wide-offset-reader ┃        Change ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1  │           5.48 / 6.07 ±1.02 / 8.10 ms │           5.26 / 5.94 ±1.06 / 8.04 ms │     no change │
│ QQuery 2  │        76.36 / 76.67 ±0.35 / 77.32 ms │        76.86 / 77.23 ±0.36 / 77.77 ms │     no change │
│ QQuery 3  │        29.07 / 29.27 ±0.14 / 29.44 ms │        28.89 / 29.23 ±0.21 / 29.48 ms │     no change │
│ QQuery 4  │     436.04 / 446.25 ±8.77 / 457.46 ms │     440.93 / 447.20 ±3.36 / 450.41 ms │     no change │
│ QQuery 5  │        49.85 / 52.89 ±4.70 / 62.26 ms │        51.69 / 53.45 ±1.76 / 56.37 ms │     no change │
│ QQuery 6  │        34.26 / 34.41 ±0.12 / 34.61 ms │        33.81 / 34.48 ±0.46 / 34.91 ms │     no change │
│ QQuery 7  │        71.58 / 71.85 ±0.18 / 72.14 ms │        71.95 / 72.12 ±0.25 / 72.60 ms │     no change │
│ QQuery 8  │        38.39 / 38.79 ±0.32 / 39.30 ms │        38.33 / 38.63 ±0.28 / 39.10 ms │     no change │
│ QQuery 9  │        53.61 / 58.12 ±5.67 / 68.99 ms │        51.52 / 54.89 ±3.73 / 61.87 ms │ +1.06x faster │
│ QQuery 10 │        59.07 / 59.40 ±0.27 / 59.72 ms │        59.00 / 59.27 ±0.25 / 59.71 ms │     no change │
│ QQuery 11 │     268.56 / 270.77 ±3.14 / 276.99 ms │     277.74 / 280.22 ±2.32 / 284.18 ms │     no change │
│ QQuery 12 │        27.68 / 28.31 ±0.34 / 28.71 ms │        28.17 / 28.46 ±0.25 / 28.75 ms │     no change │
│ QQuery 13 │     111.36 / 112.30 ±0.62 / 113.14 ms │     112.44 / 113.74 ±1.23 / 115.33 ms │     no change │
│ QQuery 14 │     377.95 / 381.16 ±2.97 / 384.84 ms │     380.15 / 384.86 ±4.54 / 392.88 ms │     no change │
│ QQuery 15 │        52.04 / 53.53 ±1.71 / 56.86 ms │        52.07 / 53.09 ±0.68 / 54.09 ms │     no change │
│ QQuery 16 │           5.93 / 6.24 ±0.29 / 6.79 ms │           6.54 / 6.64 ±0.07 / 6.73 ms │  1.06x slower │
│ QQuery 17 │        75.81 / 76.54 ±0.53 / 77.31 ms │        75.97 / 77.46 ±1.80 / 80.84 ms │     no change │
│ QQuery 18 │     105.31 / 107.15 ±2.13 / 111.29 ms │     104.88 / 106.12 ±1.26 / 108.38 ms │     no change │
│ QQuery 19 │        41.43 / 42.06 ±0.48 / 42.81 ms │        41.39 / 41.64 ±0.24 / 42.02 ms │     no change │
│ QQuery 20 │        34.35 / 35.05 ±0.51 / 35.78 ms │        34.64 / 34.99 ±0.41 / 35.70 ms │     no change │
│ QQuery 21 │        17.26 / 17.42 ±0.18 / 17.75 ms │        17.37 / 17.54 ±0.27 / 18.08 ms │     no change │
│ QQuery 22 │        69.82 / 70.77 ±0.98 / 72.18 ms │        70.04 / 70.64 ±0.52 / 71.41 ms │     no change │
│ QQuery 23 │     291.57 / 294.36 ±2.23 / 297.41 ms │     301.13 / 306.07 ±4.28 / 313.83 ms │     no change │
│ QQuery 24 │     193.27 / 196.85 ±3.11 / 201.26 ms │     199.34 / 203.35 ±5.56 / 214.38 ms │     no change │
│ QQuery 25 │     104.27 / 105.86 ±1.22 / 107.57 ms │     105.06 / 108.39 ±3.64 / 115.02 ms │     no change │
│ QQuery 26 │        45.41 / 46.86 ±2.03 / 50.81 ms │        44.89 / 45.55 ±0.39 / 46.11 ms │     no change │
│ QQuery 27 │           6.31 / 6.42 ±0.11 / 6.62 ms │           6.38 / 6.52 ±0.09 / 6.64 ms │     no change │
│ QQuery 28 │        57.24 / 60.60 ±1.72 / 61.89 ms │        56.98 / 60.80 ±2.11 / 62.78 ms │     no change │
│ QQuery 29 │        92.07 / 93.93 ±2.10 / 97.77 ms │        91.64 / 92.91 ±1.14 / 94.63 ms │     no change │
│ QQuery 30 │        33.10 / 34.87 ±2.94 / 40.71 ms │        32.96 / 33.45 ±0.41 / 34.16 ms │     no change │
│ QQuery 31 │     109.88 / 110.66 ±0.62 / 111.78 ms │     109.38 / 110.29 ±0.91 / 111.61 ms │     no change │
│ QQuery 32 │        20.27 / 20.81 ±0.50 / 21.62 ms │        20.26 / 20.55 ±0.19 / 20.80 ms │     no change │
│ QQuery 33 │        36.71 / 36.84 ±0.08 / 36.95 ms │        36.54 / 38.99 ±3.76 / 46.48 ms │  1.06x slower │
│ QQuery 34 │         9.62 / 10.81 ±2.16 / 15.13 ms │           9.25 / 9.41 ±0.20 / 9.79 ms │ +1.15x faster │
│ QQuery 35 │        71.31 / 71.75 ±0.32 / 72.29 ms │        70.25 / 70.58 ±0.36 / 71.25 ms │     no change │
│ QQuery 36 │           5.80 / 5.96 ±0.24 / 6.44 ms │           5.88 / 6.09 ±0.29 / 6.65 ms │     no change │
│ QQuery 37 │           7.07 / 7.19 ±0.08 / 7.31 ms │           6.99 / 7.32 ±0.58 / 8.46 ms │     no change │
│ QQuery 38 │        63.92 / 64.65 ±0.82 / 66.24 ms │        64.88 / 65.20 ±0.30 / 65.65 ms │     no change │
│ QQuery 39 │        71.89 / 73.26 ±0.87 / 74.61 ms │        72.29 / 75.01 ±2.37 / 79.18 ms │     no change │
│ QQuery 40 │        23.02 / 23.64 ±0.51 / 24.32 ms │        23.09 / 23.53 ±0.28 / 23.91 ms │     no change │
│ QQuery 41 │        13.91 / 14.10 ±0.13 / 14.23 ms │        13.79 / 13.96 ±0.14 / 14.20 ms │     no change │
│ QQuery 42 │        23.78 / 24.08 ±0.22 / 24.35 ms │        24.06 / 24.94 ±0.70 / 25.95 ms │     no change │
│ QQuery 43 │           4.89 / 5.08 ±0.23 / 5.52 ms │           4.86 / 5.01 ±0.19 / 5.38 ms │     no change │
│ QQuery 44 │           8.76 / 8.96 ±0.12 / 9.12 ms │           8.73 / 8.80 ±0.05 / 8.87 ms │     no change │
│ QQuery 45 │        34.67 / 36.29 ±1.72 / 39.49 ms │        35.29 / 35.75 ±0.47 / 36.52 ms │     no change │
│ QQuery 46 │        11.14 / 11.40 ±0.19 / 11.71 ms │        11.09 / 11.33 ±0.23 / 11.78 ms │     no change │
│ QQuery 47 │     204.27 / 206.79 ±2.81 / 212.21 ms │     205.57 / 207.53 ±2.35 / 212.11 ms │     no change │
│ QQuery 48 │        89.45 / 91.22 ±1.43 / 93.43 ms │       90.37 / 93.24 ±3.94 / 100.97 ms │     no change │
│ QQuery 49 │        68.76 / 69.29 ±0.65 / 70.55 ms │        69.39 / 70.07 ±0.93 / 71.89 ms │     no change │
│ QQuery 50 │        56.73 / 57.85 ±0.92 / 59.41 ms │        57.21 / 59.02 ±2.94 / 64.88 ms │     no change │
│ QQuery 51 │       94.03 / 96.57 ±3.64 / 103.76 ms │        93.51 / 95.82 ±1.86 / 98.62 ms │     no change │
│ QQuery 52 │        24.36 / 24.66 ±0.31 / 25.06 ms │        24.67 / 24.96 ±0.32 / 25.43 ms │     no change │
│ QQuery 53 │        29.64 / 30.13 ±0.84 / 31.81 ms │        29.49 / 29.65 ±0.11 / 29.81 ms │     no change │
│ QQuery 54 │        53.61 / 53.85 ±0.18 / 54.03 ms │        53.96 / 56.13 ±3.13 / 62.27 ms │     no change │
│ QQuery 55 │        23.86 / 26.28 ±3.69 / 33.59 ms │        24.36 / 24.68 ±0.27 / 24.96 ms │ +1.07x faster │
│ QQuery 56 │        37.73 / 38.42 ±0.71 / 39.69 ms │        38.17 / 38.50 ±0.39 / 39.26 ms │     no change │
│ QQuery 57 │     152.90 / 153.87 ±1.09 / 155.98 ms │     153.57 / 155.23 ±1.97 / 158.73 ms │     no change │
│ QQuery 58 │     104.20 / 106.15 ±3.01 / 112.12 ms │     104.07 / 106.14 ±1.25 / 107.84 ms │     no change │
│ QQuery 59 │     106.65 / 108.83 ±2.51 / 113.10 ms │     107.66 / 108.29 ±0.56 / 109.12 ms │     no change │
│ QQuery 60 │        38.45 / 39.01 ±0.53 / 40.00 ms │        38.54 / 39.39 ±0.51 / 40.09 ms │     no change │
│ QQuery 61 │        10.64 / 10.70 ±0.04 / 10.74 ms │        10.31 / 10.49 ±0.23 / 10.93 ms │     no change │
│ QQuery 62 │        37.91 / 38.34 ±0.23 / 38.61 ms │        37.99 / 38.66 ±0.57 / 39.69 ms │     no change │
│ QQuery 63 │        29.32 / 30.10 ±0.87 / 31.77 ms │        29.57 / 29.84 ±0.24 / 30.21 ms │     no change │
│ QQuery 64 │     354.93 / 360.43 ±4.73 / 368.88 ms │     355.83 / 360.94 ±4.02 / 368.18 ms │     no change │
│ QQuery 65 │        74.26 / 75.27 ±1.12 / 77.43 ms │        74.22 / 74.75 ±0.51 / 75.50 ms │     no change │
│ QQuery 66 │        71.15 / 72.55 ±1.49 / 75.45 ms │        72.86 / 75.38 ±4.22 / 83.81 ms │     no change │
│ QQuery 67 │     228.80 / 231.84 ±3.43 / 236.33 ms │     229.33 / 230.75 ±1.31 / 232.52 ms │     no change │
│ QQuery 68 │        11.26 / 11.48 ±0.14 / 11.68 ms │        11.37 / 11.54 ±0.11 / 11.70 ms │     no change │
│ QQuery 69 │        53.49 / 54.25 ±1.04 / 56.27 ms │        53.76 / 54.13 ±0.26 / 54.49 ms │     no change │
│ QQuery 70 │       97.03 / 99.02 ±2.10 / 102.84 ms │      97.70 / 100.86 ±3.13 / 106.55 ms │     no change │
│ QQuery 71 │        34.91 / 35.77 ±0.95 / 37.61 ms │        35.06 / 35.28 ±0.19 / 35.64 ms │     no change │
│ QQuery 72 │ 1553.86 / 1612.95 ±36.28 / 1656.94 ms │ 1658.75 / 1695.85 ±41.67 / 1772.52 ms │  1.05x slower │
│ QQuery 73 │           9.38 / 9.55 ±0.20 / 9.94 ms │         9.29 / 11.11 ±3.24 / 17.59 ms │  1.16x slower │
│ QQuery 74 │     199.36 / 204.41 ±4.77 / 212.47 ms │     200.50 / 205.00 ±2.54 / 207.53 ms │     no change │
│ QQuery 75 │     139.00 / 140.00 ±0.79 / 141.31 ms │     137.46 / 139.47 ±1.85 / 142.77 ms │     no change │
│ QQuery 76 │        34.73 / 35.22 ±0.53 / 36.18 ms │        34.17 / 34.40 ±0.17 / 34.63 ms │     no change │
│ QQuery 77 │        59.82 / 60.77 ±1.02 / 62.73 ms │        59.67 / 61.36 ±1.22 / 63.07 ms │     no change │
│ QQuery 78 │     267.12 / 280.91 ±9.09 / 289.61 ms │     262.70 / 270.26 ±6.69 / 281.51 ms │     no change │
│ QQuery 79 │        67.91 / 70.84 ±2.95 / 75.65 ms │        67.05 / 67.64 ±0.58 / 68.75 ms │     no change │
│ QQuery 80 │        93.61 / 95.64 ±1.13 / 97.01 ms │       94.74 / 97.19 ±3.99 / 105.12 ms │     no change │
│ QQuery 81 │        26.59 / 26.81 ±0.21 / 27.19 ms │        27.29 / 27.54 ±0.13 / 27.67 ms │     no change │
│ QQuery 82 │        16.34 / 16.54 ±0.20 / 16.89 ms │        16.66 / 17.51 ±1.28 / 20.06 ms │  1.06x slower │
│ QQuery 83 │        32.63 / 33.71 ±1.72 / 37.14 ms │        32.49 / 32.84 ±0.23 / 33.09 ms │     no change │
│ QQuery 84 │        29.97 / 30.66 ±0.68 / 31.85 ms │        29.81 / 29.97 ±0.12 / 30.10 ms │     no change │
│ QQuery 85 │        94.40 / 95.29 ±0.66 / 96.28 ms │       95.80 / 98.73 ±4.81 / 108.31 ms │     no change │
│ QQuery 86 │        24.18 / 24.50 ±0.28 / 24.92 ms │        24.41 / 24.59 ±0.16 / 24.86 ms │     no change │
│ QQuery 87 │        64.48 / 64.69 ±0.17 / 64.97 ms │        64.45 / 65.22 ±0.63 / 66.28 ms │     no change │
│ QQuery 88 │        59.27 / 59.58 ±0.19 / 59.85 ms │        59.06 / 61.52 ±3.23 / 67.93 ms │     no change │
│ QQuery 89 │        35.73 / 36.08 ±0.31 / 36.64 ms │        36.11 / 36.87 ±0.57 / 37.89 ms │     no change │
│ QQuery 90 │        16.08 / 16.18 ±0.16 / 16.51 ms │        16.07 / 16.26 ±0.17 / 16.51 ms │     no change │
│ QQuery 91 │        39.34 / 39.55 ±0.12 / 39.67 ms │        40.27 / 40.87 ±0.60 / 42.00 ms │     no change │
│ QQuery 92 │        28.54 / 30.20 ±1.69 / 33.14 ms │        29.19 / 29.32 ±0.14 / 29.57 ms │     no change │
│ QQuery 93 │        49.05 / 49.51 ±0.28 / 49.92 ms │        49.22 / 50.86 ±1.68 / 53.57 ms │     no change │
│ QQuery 94 │        37.99 / 39.02 ±1.25 / 41.45 ms │        38.67 / 39.32 ±0.71 / 40.32 ms │     no change │
│ QQuery 95 │        77.80 / 79.25 ±0.90 / 80.63 ms │        80.74 / 81.24 ±0.56 / 82.30 ms │     no change │
│ QQuery 96 │        23.68 / 25.14 ±2.69 / 30.51 ms │        23.74 / 23.88 ±0.09 / 23.98 ms │ +1.05x faster │
│ QQuery 97 │        50.33 / 51.45 ±1.60 / 54.61 ms │        50.28 / 51.09 ±0.83 / 52.68 ms │     no change │
│ QQuery 98 │        40.87 / 41.80 ±0.63 / 42.81 ms │        41.41 / 42.58 ±1.16 / 44.69 ms │     no change │
│ QQuery 99 │        56.59 / 56.96 ±0.30 / 57.30 ms │        56.32 / 56.86 ±0.53 / 57.84 ms │     no change │
└───────────┴───────────────────────────────────────┴───────────────────────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                                  ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                                  │ 8890.15ms │
│ Total Time (codex_dictionary-wide-offset-reader)   │ 9012.22ms │
│ Average Time (HEAD)                                │   89.80ms │
│ Average Time (codex_dictionary-wide-offset-reader) │   91.03ms │
│ Queries Faster                                     │         4 │
│ Queries Slower                                     │         5 │
│ Queries with No Change                             │        90 │
│ Queries with Failure                               │         0 │
└────────────────────────────────────────────────────┴───────────┘

Memory Pool Peaks

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

Base: e66caad9042cf570a605da822e480a02845360b9 | Changed: 0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd

tpcds — tpcds_sf1

Query Base Changed Change
Query 1 135.4 KiB 138.5 KiB +2.2%
Query 2 82.8 MiB 68.3 MiB -17.5%
Query 3 1.9 MiB 1.9 MiB -0.6%
Query 4 81.9 MiB 81.8 MiB -0.1%
Query 5 13.0 MiB 13.5 MiB +4.0%
Query 6 4.2 MiB 5.0 MiB +19.3%
Query 7 3.6 MiB 3.9 MiB +7.2%
Query 8 2.1 MiB 1.8 MiB -12.9%
Query 9 32.1 KiB 32.1 KiB +0.0%
Query 10 2.0 MiB 2.0 MiB +0.0%
Query 11 56.7 MiB 56.7 MiB -0.0%
Query 12 116.2 MiB 125.9 MiB +8.3%
Query 13 13.0 MiB 13.7 MiB +5.1%
Query 14 170.9 MiB 171.4 MiB +0.3%
Query 15 4.2 MiB 4.2 MiB +0.0%
Query 16 118.6 KiB 118.6 KiB +0.0%
Query 17 42.1 MiB 41.1 MiB -2.2%
Query 18 21.2 MiB 21.3 MiB +0.1%
Query 19 2.1 MiB 2.1 MiB +1.8%
Query 20 116.6 MiB 116.6 MiB -0.0%
Query 21 4.0 MiB 3.2 MiB -18.4%
Query 22 61.8 MiB 61.8 MiB +0.0%
Query 23 213.2 MiB 212.6 MiB -0.3%
Query 24 38.7 MiB 38.7 MiB +0.0%
Query 25 99.7 MiB 99.7 MiB +0.0%
Query 26 3.8 MiB 3.7 MiB -2.0%
Query 27 1.6 MiB 1.6 MiB +0.0%
Query 28 1.2 MiB 1.5 MiB +22.1%
Query 29 104.8 MiB 104.8 MiB +0.0%
Query 30 4.0 MiB 4.0 MiB +2.1%
Query 31 13.4 MiB 13.4 MiB +0.0%
Query 32 2.7 MiB 2.7 MiB +0.0%
Query 33 4.0 MiB 5.8 MiB +42.4%
Query 34 6.0 MiB 6.0 MiB +0.0%
Query 35 16.6 MiB 16.6 MiB +0.0%
Query 36 10.3 MiB 10.3 MiB +0.0%
Query 37 3.9 MiB 3.9 MiB +0.0%
Query 38 10.4 MiB 10.8 MiB +3.1%
Query 39 42.4 MiB 52.4 MiB +23.6%
Query 40 12.2 MiB 12.2 MiB +0.0%
Query 41 2.1 MiB 2.1 MiB +0.0%
Query 42 1.1 MiB 1.3 MiB +20.0%
Query 43 292.2 KiB 275.0 KiB -5.9%
Query 44 233.2 KiB 233.2 KiB +0.0%
Query 45 4.5 MiB 4.5 MiB +0.0%
Query 46 6.2 MiB 6.2 MiB +0.0%
Query 47 128.5 MiB 128.5 MiB -0.0%
Query 48 11.8 MiB 11.8 MiB +0.0%
Query 49 52.8 MiB 52.8 MiB -0.0%
Query 50 17.3 MiB 17.3 MiB +0.0%
Query 51 177.7 MiB 177.7 MiB +0.0%
Query 52 1.8 MiB 1.8 MiB -0.1%
Query 53 20.6 MiB 20.6 MiB +0.0%
Query 54 2.1 MiB 2.2 MiB +4.4%
Query 55 1.5 MiB 1.7 MiB +14.7%
Query 56 3.2 MiB 2.7 MiB -13.4%
Query 57 120.8 MiB 120.8 MiB +0.0%
Query 58 4.4 MiB 4.4 MiB +0.0%
Query 59 11.7 MiB 11.7 MiB +0.0%
Query 60 5.2 MiB 5.2 MiB -0.0%
Query 61 1.1 MiB 961.7 KiB -11.1%
Query 62 3.7 MiB 4.0 MiB +9.0%
Query 63 30.6 MiB 30.6 MiB +0.0%
Query 64 114.3 MiB 106.2 MiB -7.1%
Query 65 26.4 MiB 26.1 MiB -1.3%
Query 66 11.9 MiB 14.9 MiB +26.1%
Query 67 139.8 MiB 137.9 MiB -1.4%
Query 68 6.2 MiB 6.2 MiB +0.0%
Query 69 4.3 MiB 3.8 MiB -12.8%
Query 70 44.0 MiB 44.0 MiB -0.0%
Query 71 50.4 MiB 40.4 MiB -19.8%
Query 72 97.3 MiB 97.3 MiB +0.0%
Query 73 6.0 MiB 6.0 MiB +0.0%
Query 74 36.0 MiB 36.0 MiB -0.0%
Query 75 24.1 MiB 23.1 MiB -4.1%
Query 76 4.3 MiB 4.3 MiB -0.2%
Query 77 4.1 MiB 4.1 MiB +0.0%
Query 78 335.7 MiB 335.7 MiB -0.0%
Query 79 9.3 MiB 9.3 MiB +0.0%
Query 80 41.6 MiB 41.7 MiB +0.3%
Query 81 4.2 MiB 4.2 MiB +1.8%
Query 82 3.9 MiB 3.9 MiB +0.0%
Query 83 9.0 MiB 9.0 MiB +0.5%
Query 84 11.9 MiB 11.9 MiB +0.0%
Query 85 17.0 MiB 17.0 MiB +0.0%
Query 86 82.1 MiB 82.1 MiB +0.0%
Query 87 12.9 MiB 12.9 MiB +0.0%
Query 88 1.6 MiB 1.8 MiB +10.4%
Query 89 112.2 MiB 112.2 MiB +0.0%
Query 90 348.6 KiB 351.7 KiB +0.9%
Query 91 31.5 MiB 31.3 MiB -0.6%
Query 92 2.7 MiB 2.7 MiB +0.0%
Query 93 15.4 MiB 14.9 MiB -3.3%
Query 94 3.7 MiB 3.7 MiB +0.0%
Query 95 23.9 MiB 23.0 MiB -3.7%
Query 96 224.7 KiB 224.7 KiB +0.0%
Query 97 33.1 MiB 33.1 MiB -0.2%
Query 98 116.7 MiB 116.2 MiB -0.5%
Query 99 1.9 MiB 1.6 MiB -15.0%

Pool accounting vs. process RSS

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

Benchmark Side Max pool peak Peak RSS Gap RSS / pool
tpcds base (e66caad9042cf570a605da822e480a02845360b9) 335.7 MiB 2.0 GiB 1.6 GiB 6.0×
tpcds changed (0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd) 335.7 MiB 2.1 GiB 1.8 GiB 6.4×
Resource Usage

tpcds — base (merge-base)

Metric Value
Wall time 45.0s
Peak memory 2.0 GiB
Avg memory 1.5 GiB
CPU user 184.9s
CPU sys 5.5s
Peak spill 0 B

tpcds — branch

Metric Value
Wall time 50.0s
Peak memory 2.1 GiB
Avg memory 1.4 GiB
CPU user 188.3s
CPU sys 5.5s
Peak spill 0 B

File an issue against this benchmark runner

@adriangbot

adriangbot commented Nov 9, 2026

Copy link
Copy Markdown

🤖 Benchmark completed (GKE) | trigger

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

Comparing 0b6467f (0b6467f) to e66caad diff

Run configuration 
run benchmark wide_schema
env:
  DATAFUSION_EXECUTION_PARQUET_ENABLE_RLE_TO_DICTIONARY: "true"
  DATAFUSION_RUNTIME_MEMORY_LIMIT: "32G"
baseline:
  ref: "e66caad9042cf570a605da822e480a02845360b9"
changed:
  ref: "0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd"
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                                   codex_dictionary-wide-offset-reader
-----                     ----                                   -----------------------------------
wide_schema/Q01_narrow    1.01     83.5±0.81ms        ? ?/sec    1.00     82.4±0.89ms        ? ?/sec
wide_schema/Q01_wide      1.02    411.9±5.62ms        ? ?/sec    1.00    405.0±5.55ms        ? ?/sec
wide_schema/Q02_narrow    1.00      6.9±0.25ms        ? ?/sec    1.03      7.1±0.31ms        ? ?/sec
wide_schema/Q02_wide      1.00    216.6±3.46ms        ? ?/sec    1.02    221.2±2.92ms        ? ?/sec
wide_schema/Q03_narrow    1.02     11.2±0.19ms        ? ?/sec    1.00     11.0±0.15ms        ? ?/sec
wide_schema/Q03_wide      1.02    230.7±2.94ms        ? ?/sec    1.00    225.7±4.49ms        ? ?/sec
wide_schema/Q04_narrow    1.00     37.2±0.40ms        ? ?/sec    1.01     37.8±0.66ms        ? ?/sec
wide_schema/Q04_wide      1.01    369.8±5.77ms        ? ?/sec    1.00    367.7±6.10ms        ? ?/sec

Memory Pool Peaks

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

Base: e66caad9042cf570a605da822e480a02845360b9 | Changed: 0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd

wide_schema — wide_schema_narrow

Query Base Changed Change
wide_schema/Q01_narrow 37.0 MiB 37.0 MiB +0.0%
wide_schema/Q02_narrow 0 B 0 B 0.0%
wide_schema/Q03_narrow 0 B 0 B 0.0%
wide_schema/Q04_narrow 0 B 0 B 0.0%

wide_schema — wide_schema_wide

Query Base Changed Change
wide_schema/Q01_wide 36.1 MiB 36.1 MiB +0.0%
wide_schema/Q02_wide 0 B 0 B 0.0%
wide_schema/Q03_wide 0 B 0 B 0.0%
wide_schema/Q04_wide 0 B 0 B 0.0%

Pool accounting vs. process RSS

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

Benchmark Side Max pool peak Peak RSS Gap RSS / pool
wide_schema base (e66caad9042cf570a605da822e480a02845360b9) 37.0 MiB 969.9 MiB 932.9 MiB 26.2×
wide_schema changed (0b6467ffa88219a4045fb8b8c98a23fc0cb0f8fd) 37.0 MiB 1002.3 MiB 965.3 MiB 27.1×
Resource Usage

wide_schema — base (merge-base)

Metric Value
Wall time 940.3s
Peak memory 969.9 MiB
Avg memory 82.8 MiB
CPU user 575.9s
CPU sys 65.5s
Peak spill 0 B

wide_schema — branch

Metric Value
Wall time 645.3s
Peak memory 1002.3 MiB
Avg memory 113.4 MiB
CPU user 568.6s
CPU sys 65.3s
Peak spill 0 B

File an issue against this benchmark runner

@Rich-T-kid

Rich-T-kid commented Nov 9, 2026

Copy link
Copy Markdown
Contributor Author

Final exact-SHA benchmark summary (e66caad9 -> 0b6467ff, Arrow 60.1.0, dictionary reads enabled on both sides):

  • ClickBench: 23.03 s -> 19.61 s (-14.8%). Q10 3.03x faster, Q11 2.78x, Q23 2.01x.
  • TPC-H SF1: all 22 queries unchanged.
  • TPC-DS SF1: two hosted runs were 1.4-1.9% slower overall. Q72 is the repeatable large outlier and groups by high-cardinality i_item_desc; Q16 moved by about 0.6 ms. Balanced local runs measured Q72 at +2.5% and Q16 at -1.2%.
  • Wide schema: all cases within 3%.

I also reran the small ClickBench Q41/Q42 outliers over 200 samples per side. They were neutral locally at +0.5% and +0.3%.

@Rich-T-kid Rich-T-kid closed this Nov 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

datasource Changes to the datasource crate

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Avoid per-row dictionary casts for binary Parquet columns

2 participants

@Rich-T-kid @adriangbot