You are viewing a plain text version of this content. The canonical link for it is here.
Posted to github@datafusion.apache.org by "alamb (via GitHub)" <gi...@apache.org> on 2026/08/15 13:50:28 UTC
[PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
alamb opened a new pull request, #24395:
URL: https://github.com/apache/datafusion/pull/24395
## Which issue does this PR close?
- Part of https://github.com/apache/datafusion/issues/24393
- Related to https://github.com/apache/datafusion/issues/20324 and https://github.com/apache/datafusion/issues/3463
**Note: this is a prototype and is NOT READY FOR REVIEW.** It exists to evaluate whether decoupling the I/O pattern from filter pushdown avoids the I/O-related regressions that have blocked enabling `pushdown_filters` by default.
## Rationale for this change
Enabling `pushdown_filters` today also changes the I/O pattern: instead of one request per row group, the reader sequentially fetches the columns for each filter predicate, then (after evaluating the filters) the remaining projected columns. When a file has no offset index (e.g. the ClickBench dataset), those extra round trips cannot prune any bytes, so they are pure overhead.
This PR separates "filter evaluation mechanics" from the I/O pattern via a new config option, `datafusion.execution.parquet.progressive_io`:
| `pushdown_filters` | `progressive_io` | behavior |
|---|---|---|
| false | (any) | unchanged: one request per row group |
| true | **false (default)** | filters evaluated during decode, but all column chunks needed by the filters or the projection are fetched with **one request per row group** |
| true | true | progressive fetch, but only when the file has an offset index (where it can actually prune bytes); otherwise one request per row group |
## What changes are included in this PR?
- New `progressive_io` config option (config, proto, docs, `ParquetSource::with_progressive_io`), resolved table-or-session like `pushdown_filters`
- `RowFilterGenerator` now exposes the union of the filter predicates' `ProjectionMask`s
- In the opener, when one-shot I/O applies, the byte ranges of all column chunks in (filter mask βͺ projection mask) are precomputed per row group; `PushDecoderStreamState::transition` expands the first `NeedsData` request of each row group to those ranges, so the whole row group is fetched in one `get_byte_ranges` call. The decoder's buffered-range containment check then satisfies all subsequent phases with zero additional I/O. No arrow-rs changes were needed.
Known prototype limitations:
- When page-index pruning produces a row selection that actually prunes rows, the scan falls back to progressive I/O (one-shot whole-chunk ranges would defeat the page pruning). A future refinement could fetch selection-pruned page ranges in a single request instead.
- One-shot ranges are whole column chunks, so peak buffered bytes per row group equal the compressed size of all needed chunks (same as the non-pushdown reader today).
## Are these changes tested?
Yes:
- New object-store request-count tests in `object_store_access.rs` show that with `pushdown_filters=true` and the default `progressive_io=false`, the scan issues exactly one data request per row group with byte ranges identical to the non-pushdown scan, and that `progressive_io=true` restores the current progressive pattern
- Existing filter pushdown tests and sqllogictests pass unchanged
## Are there any user-facing changes?
Yes: a new configuration option, and (intentionally) a changed default I/O pattern for users who enable `pushdown_filters`. Benchmarks (`clickbench_pushdown`) to follow.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "adriangbot (via GitHub)" <gi...@apache.org>.
adriangbot commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5307179867
π€ Benchmark running (GKE) | [trigger](https://github.com/apache/datafusion/pull/24395#issuecomment-5307170440)
**Instance:** `c4a-highmem-16` (12 vCPU / 65 GiB) | `Linux bench-c5307170440-1617-45h8f 6.12.85+ #1 SMP Wed Jun 17 20:31:55 UTC 2026 aarch64 GNU/Linux`
<details><summary>CPU Details (lscpu)</summary>
```
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 alamb/progressive_io (4342770ee1ff6e31f9d7fc5ab201df3d027e3c31) to 00eba79 (merge-base) [diff](https://github.com/apache/datafusion/compare/00eba790b17f7f0a62e5b81627981197f9bfadb0..4342770ee1ff6e31f9d7fc5ab201df3d027e3c31)
<details><summary>Run configuration</summary>
```yaml
run benchmark clickbench_pushdown
```
</details>
Results will be posted here when complete
---
[File an issue](https://github.com/adriangb/datafusion-benchmarking/issues) against this benchmark runner
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "alamb (via GitHub)" <gi...@apache.org>.
alamb commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5439849424
So TLDR is that the benchmark results in https://github.com/apache/datafusion/pull/24395#issuecomment-5439527687 show that this setting avoids performance regressions under simulated latency, which is good (a pre-requsite for enabling filter pushdown by default).
Now I just need to bash out any other regressions and we'll ready to turn this into something real
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "adriangbot (via GitHub)" <gi...@apache.org>.
adriangbot commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5439153419
π€ Benchmark running (GKE) | [trigger](https://github.com/apache/datafusion/pull/24395#issuecomment-5439121941)
**Instance:** `c4a-highmem-16` (12 vCPU / 65 GiB) | `Linux bench-c5439121941-2018-wq2xg 6.12.85+ #1 SMP Sat Jun 27 09:31:30 UTC 2026 aarch64 GNU/Linux`
<details><summary>CPU Details (lscpu)</summary>
```
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 alamb/progressive_io (4342770ee1ff6e31f9d7fc5ab201df3d027e3c31) to 00eba79 (merge-base) [diff](https://github.com/apache/datafusion/compare/00eba790b17f7f0a62e5b81627981197f9bfadb0..4342770ee1ff6e31f9d7fc5ab201df3d027e3c31)
<details><summary>Run configuration</summary>
```yaml
run benchmark clickbench_pushdown
env:
SIMULATE_LATENCY: "true"
```
</details>
Results will be posted here when complete
---
[File an issue](https://github.com/adriangb/datafusion-benchmarking/issues) against this benchmark runner
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "alamb (via GitHub)" <gi...@apache.org>.
alamb commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5307170440
run benchmark clickbench_pushdown
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "alamb (via GitHub)" <gi...@apache.org>.
alamb commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5438730058
However, I don't understand the [`clickbench_pushdown` results ](https://github.com/apache/datafusion/pull/24395#issuecomment-5307247081) as I expect to see a change (ideally an improvement) in clickbench_pushdown compared to main, as main would be doing a different I/O pattern, but it seems as if the performance is basically the same.
I am having claude code look into that now
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "adriangbot (via GitHub)" <gi...@apache.org>.
adriangbot commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5307247081
π€ Benchmark completed (GKE) | [trigger](https://github.com/apache/datafusion/pull/24395#issuecomment-5307170440)
**Instance:** `c4a-highmem-16` (12 vCPU / 65 GiB)
Comparing alamb/progressive_io (4342770ee1ff6e31f9d7fc5ab201df3d027e3c31) to 00eba79 (merge-base) [diff](https://github.com/apache/datafusion/compare/00eba790b17f7f0a62e5b81627981197f9bfadb0..4342770ee1ff6e31f9d7fc5ab201df3d027e3c31)
<details><summary>Run configuration</summary>
```yaml
run benchmark clickbench_pushdown
```
</details>
<details><summary>CPU Details (lscpu)</summary>
```
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>
<details><summary>Details</summary>
<p>
```
Comparing HEAD and alamb_progressive_io
--------------------
Benchmark clickbench_pushdown.json
--------------------
βββββββββββββ³βββββββββββββ³βββββββββββββββββββββββ³ββββββββββββββββ
β Query β HEAD β alamb_progressive_io β Change β
β‘ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ©
β QQuery 0 β 1.23 ms β 1.24 ms β no change β
β QQuery 1 β 12.05 ms β 12.02 ms β no change β
β QQuery 2 β 35.87 ms β 36.41 ms β no change β
β QQuery 3 β 31.04 ms β 31.25 ms β no change β
β QQuery 4 β 222.36 ms β 223.81 ms β no change β
β QQuery 5 β 273.29 ms β 271.64 ms β no change β
β QQuery 6 β 1.30 ms β 1.27 ms β no change β
β QQuery 7 β 15.55 ms β 15.79 ms β no change β
β QQuery 8 β 325.73 ms β 320.85 ms β no change β
β QQuery 9 β 447.39 ms β 445.43 ms β no change β
β QQuery 10 β 92.47 ms β 92.81 ms β no change β
β QQuery 11 β 103.16 ms β 103.58 ms β no change β
β QQuery 12 β 291.74 ms β 296.16 ms β no change β
β QQuery 13 β 409.75 ms β 392.67 ms β no change β
β QQuery 14 β 311.60 ms β 310.33 ms β no change β
β QQuery 15 β 271.61 ms β 267.29 ms β no change β
β QQuery 16 β 607.02 ms β 612.19 ms β no change β
β QQuery 17 β 618.56 ms β 614.36 ms β no change β
β QQuery 18 β 1240.28 ms β 1261.16 ms β no change β
β QQuery 19 β 28.51 ms β 28.60 ms β no change β
β QQuery 20 β 514.57 ms β 516.16 ms β no change β
β QQuery 21 β 554.74 ms β 559.65 ms β no change β
β QQuery 22 β 904.31 ms β 915.27 ms β no change β
β QQuery 23 β 114.11 ms β 112.69 ms β no change β
β QQuery 24 β 36.72 ms β 34.12 ms β +1.08x faster β
β QQuery 25 β 138.42 ms β 141.03 ms β no change β
β QQuery 26 β 48.61 ms β 48.46 ms β no change β
β QQuery 27 β 568.29 ms β 552.57 ms β no change β
β QQuery 28 β 2905.12 ms β 2923.80 ms β no change β
β QQuery 29 β 40.66 ms β 41.17 ms β no change β
β QQuery 30 β 301.78 ms β 312.71 ms β no change β
β QQuery 31 β 278.73 ms β 284.82 ms β no change β
β QQuery 32 β 929.57 ms β 940.37 ms β no change β
β QQuery 33 β 1428.21 ms β 1433.23 ms β no change β
β QQuery 34 β 1464.64 ms β 1463.34 ms β no change β
β QQuery 35 β 279.84 ms β 280.79 ms β no change β
β QQuery 36 β 68.08 ms β 68.32 ms β no change β
β QQuery 37 β 38.05 ms β 35.62 ms β +1.07x faster β
β QQuery 38 β 34.77 ms β 35.80 ms β no change β
β QQuery 39 β 135.44 ms β 126.98 ms β +1.07x faster β
β QQuery 40 β 17.02 ms β 17.00 ms β no change β
β QQuery 41 β 16.15 ms β 15.89 ms β no change β
β QQuery 42 β 13.96 ms β 13.94 ms β no change β
βββββββββββββ΄βββββββββββββ΄βββββββββββββββββββββββ΄ββββββββββββββββ
βββββββββββββββββββββββββββββββββββββββ³βββββββββββββ
β Benchmark Summary β β
β‘βββββββββββββββββββββββββββββββββββββββββββββββββββ©
β Total Time (HEAD) β 16172.32ms β
β Total Time (alamb_progressive_io) β 16212.58ms β
β Average Time (HEAD) β 376.10ms β
β Average Time (alamb_progressive_io) β 377.04ms β
β Queries Faster β 3 β
β Queries Slower β 0 β
β Queries with No Change β 40 β
β Queries with Failure β 0 β
βββββββββββββββββββββββββββββββββββββββ΄βββββββββββββ
Distribution per query (min / mean Β±stddev / max):
Comparing HEAD and alamb_progressive_io
--------------------
Benchmark clickbench_pushdown.json
--------------------
βββββββββββββ³ββββββββββββββββββββββββββββββββββββββββ³ββββββββββββββββββββββββββββββββββββββββ³ββββββββββββββββ
β Query β HEAD β alamb_progressive_io β Change β
β‘ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ©
β QQuery 0 β 1.23 / 3.88 Β±5.25 / 14.37 ms β 1.24 / 3.90 Β±5.26 / 14.42 ms β no change β
β QQuery 1 β 12.05 / 12.26 Β±0.12 / 12.41 ms β 12.02 / 12.29 Β±0.16 / 12.44 ms β no change β
β QQuery 2 β 35.87 / 36.25 Β±0.24 / 36.49 ms β 36.41 / 36.95 Β±0.38 / 37.48 ms β no change β
β QQuery 3 β 31.04 / 31.65 Β±0.69 / 32.95 ms β 31.25 / 31.36 Β±0.14 / 31.62 ms β no change β
β QQuery 4 β 222.36 / 225.27 Β±2.42 / 228.70 ms β 223.81 / 224.93 Β±1.04 / 226.42 ms β no change β
β QQuery 5 β 273.29 / 274.61 Β±0.96 / 276.09 ms β 271.64 / 273.27 Β±1.56 / 275.50 ms β no change β
β QQuery 6 β 1.30 / 1.44 Β±0.22 / 1.86 ms β 1.27 / 1.41 Β±0.20 / 1.81 ms β no change β
β QQuery 7 β 15.55 / 15.76 Β±0.15 / 15.99 ms β 15.79 / 15.97 Β±0.12 / 16.14 ms β no change β
β QQuery 8 β 325.73 / 329.97 Β±2.62 / 332.82 ms β 320.85 / 326.09 Β±2.80 / 329.26 ms β no change β
β QQuery 9 β 447.39 / 454.93 Β±4.53 / 460.15 ms β 445.43 / 456.64 Β±8.04 / 469.64 ms β no change β
β QQuery 10 β 92.47 / 93.06 Β±0.80 / 94.64 ms β 92.81 / 93.46 Β±0.70 / 94.64 ms β no change β
β QQuery 11 β 103.16 / 104.20 Β±0.67 / 105.03 ms β 103.58 / 105.01 Β±1.71 / 108.01 ms β no change β
β QQuery 12 β 291.74 / 300.13 Β±6.32 / 305.52 ms β 296.16 / 299.40 Β±3.94 / 307.05 ms β no change β
β QQuery 13 β 409.75 / 424.08 Β±7.21 / 429.13 ms β 392.67 / 404.47 Β±8.50 / 416.90 ms β no change β
β QQuery 14 β 311.60 / 314.76 Β±3.43 / 319.14 ms β 310.33 / 317.32 Β±6.94 / 329.18 ms β no change β
β QQuery 15 β 271.61 / 274.36 Β±2.58 / 278.97 ms β 267.29 / 280.77 Β±15.42 / 310.36 ms β no change β
β QQuery 16 β 607.02 / 616.79 Β±6.03 / 624.37 ms β 612.19 / 623.43 Β±10.92 / 644.20 ms β no change β
β QQuery 17 β 618.56 / 626.61 Β±5.86 / 632.90 ms β 614.36 / 624.54 Β±6.41 / 631.56 ms β no change β
β QQuery 18 β 1240.28 / 1257.90 Β±14.01 / 1278.61 ms β 1261.16 / 1277.54 Β±16.95 / 1302.90 ms β no change β
β QQuery 19 β 28.51 / 35.21 Β±8.17 / 50.69 ms β 28.60 / 36.39 Β±10.31 / 54.44 ms β no change β
β QQuery 20 β 514.57 / 522.95 Β±11.45 / 545.58 ms β 516.16 / 521.96 Β±6.02 / 532.60 ms β no change β
β QQuery 21 β 554.74 / 574.41 Β±11.29 / 587.32 ms β 559.65 / 573.51 Β±9.11 / 582.61 ms β no change β
β QQuery 22 β 904.31 / 912.65 Β±7.48 / 925.47 ms β 915.27 / 925.54 Β±6.39 / 934.13 ms β no change β
β QQuery 23 β 114.11 / 116.02 Β±1.22 / 117.98 ms β 112.69 / 128.05 Β±12.91 / 146.25 ms β 1.10x slower β
β QQuery 24 β 36.72 / 43.00 Β±6.33 / 51.18 ms β 34.12 / 38.96 Β±7.71 / 54.35 ms β +1.10x faster β
β QQuery 25 β 138.42 / 145.60 Β±11.43 / 168.35 ms β 141.03 / 143.09 Β±1.76 / 145.77 ms β no change β
β QQuery 26 β 48.61 / 51.22 Β±3.02 / 56.65 ms β 48.46 / 49.93 Β±2.03 / 53.68 ms β no change β
β QQuery 27 β 568.29 / 575.36 Β±6.49 / 583.91 ms β 552.57 / 568.44 Β±10.98 / 582.70 ms β no change β
β QQuery 28 β 2905.12 / 2916.99 Β±10.62 / 2935.64 ms β 2923.80 / 2940.99 Β±16.93 / 2971.92 ms β no change β
β QQuery 29 β 40.66 / 44.31 Β±6.88 / 58.07 ms β 41.17 / 44.87 Β±6.45 / 57.72 ms β no change β
β QQuery 30 β 301.78 / 311.24 Β±7.64 / 321.89 ms β 312.71 / 323.19 Β±12.56 / 343.76 ms β no change β
β QQuery 31 β 278.73 / 293.33 Β±8.90 / 306.27 ms β 284.82 / 291.09 Β±3.79 / 295.07 ms β no change β
β QQuery 32 β 929.57 / 957.16 Β±24.10 / 986.73 ms β 940.37 / 970.79 Β±27.50 / 1021.68 ms β no change β
β QQuery 33 β 1428.21 / 1459.93 Β±25.27 / 1490.99 ms β 1433.23 / 1475.20 Β±22.60 / 1500.43 ms β no change β
β QQuery 34 β 1464.64 / 1510.69 Β±29.67 / 1545.88 ms β 1463.34 / 1488.49 Β±17.60 / 1511.26 ms β no change β
β QQuery 35 β 279.84 / 309.37 Β±34.26 / 365.66 ms β 280.79 / 298.94 Β±19.39 / 333.63 ms β no change β
β QQuery 36 β 68.08 / 72.07 Β±2.45 / 75.40 ms β 68.32 / 81.98 Β±10.58 / 92.10 ms β 1.14x slower β
β QQuery 37 β 38.05 / 43.51 Β±5.78 / 54.25 ms β 35.62 / 38.24 Β±3.00 / 43.90 ms β +1.14x faster β
β QQuery 38 β 34.77 / 37.53 Β±2.58 / 41.48 ms β 35.80 / 39.51 Β±4.45 / 47.47 ms β 1.05x slower β
β QQuery 39 β 135.44 / 149.79 Β±10.54 / 167.11 ms β 126.98 / 143.59 Β±8.40 / 149.32 ms β no change β
β QQuery 40 β 17.02 / 17.54 Β±0.46 / 18.23 ms β 17.00 / 17.84 Β±1.39 / 20.59 ms β no change β
β QQuery 41 β 16.15 / 16.29 Β±0.15 / 16.59 ms β 15.89 / 16.24 Β±0.33 / 16.77 ms β no change β
β QQuery 42 β 13.96 / 15.76 Β±3.13 / 22.00 ms β 13.94 / 15.70 Β±3.14 / 21.97 ms β no change β
βββββββββββββ΄ββββββββββββββββββββββββββββββββββββββββ΄ββββββββββββββββββββββββββββββββββββββββ΄ββββββββββββββββ
βββββββββββββββββββββββββββββββββββββββ³βββββββββββββ
β Benchmark Summary β β
β‘βββββββββββββββββββββββββββββββββββββββββββββββββββ©
β Total Time (HEAD) β 16529.85ms β
β Total Time (alamb_progressive_io) β 16581.30ms β
β Average Time (HEAD) β 384.42ms β
β Average Time (alamb_progressive_io) β 385.61ms β
β Queries Faster β 2 β
β Queries Slower β 3 β
β Queries with No Change β 38 β
β Queries with Failure β 0 β
βββββββββββββββββββββββββββββββββββββββ΄βββββββββββββ
```
</p>
</details>
<details><summary>Resource Usage</summary>
**clickbench_pushdown β base (merge-base)**
| Metric | Value |
|--------|-------|
| Wall time | 85.0s |
| Peak memory | 11.9 GiB |
| Avg memory | 4.8 GiB |
| CPU user | 839.8s |
| CPU sys | 58.8s |
| Peak spill | 0 B |
**clickbench_pushdown β branch**
| Metric | Value |
|--------|-------|
| Wall time | 85.0s |
| Peak memory | 12.4 GiB |
| Avg memory | 5.2 GiB |
| CPU user | 841.9s |
| CPU sys | 61.8s |
| Peak spill | 0 B |
</details>
---
[File an issue](https://github.com/adriangb/datafusion-benchmarking/issues) against this benchmark runner
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "alamb (via GitHub)" <gi...@apache.org>.
alamb commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5439121941
run benchmark clickbench_pushdown
env:
SIMULATE_LATENCY: "true"
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "alamb (via GitHub)" <gi...@apache.org>.
alamb closed pull request #24395: WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters
URL: https://github.com/apache/datafusion/pull/24395
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "adriangbot (via GitHub)" <gi...@apache.org>.
adriangbot commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5307250020
π€ Benchmark completed (GKE) | [trigger](https://github.com/apache/datafusion/pull/24395#issuecomment-5307167906)
**Instance:** `c4a-highmem-16` (12 vCPU / 65 GiB)
Comparing alamb/progressive_io (4342770ee1ff6e31f9d7fc5ab201df3d027e3c31) to 00eba79 (merge-base) [diff](https://github.com/apache/datafusion/compare/00eba790b17f7f0a62e5b81627981197f9bfadb0..4342770ee1ff6e31f9d7fc5ab201df3d027e3c31)
<details><summary>Run configuration</summary>
```yaml
run benchmark clickbench_partitioned
```
</details>
<details><summary>CPU Details (lscpu)</summary>
```
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>
<details><summary>Details</summary>
<p>
```
Comparing HEAD and alamb_progressive_io
--------------------
Benchmark clickbench_partitioned.json
--------------------
βββββββββββββ³βββββββββββββ³βββββββββββββββββββββββ³ββββββββββββββββ
β Query β HEAD β alamb_progressive_io β Change β
β‘ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ©
β QQuery 0 β 1.30 ms β 1.28 ms β no change β
β QQuery 1 β 12.11 ms β 12.63 ms β no change β
β QQuery 2 β 36.64 ms β 37.93 ms β no change β
β QQuery 3 β 32.12 ms β 32.58 ms β no change β
β QQuery 4 β 245.81 ms β 249.16 ms β no change β
β QQuery 5 β 292.41 ms β 294.88 ms β no change β
β QQuery 6 β 1.33 ms β 1.37 ms β no change β
β QQuery 7 β 13.72 ms β 14.00 ms β no change β
β QQuery 8 β 358.00 ms β 360.07 ms β no change β
β QQuery 9 β 490.53 ms β 518.93 ms β 1.06x slower β
β QQuery 10 β 72.28 ms β 74.79 ms β no change β
β QQuery 11 β 83.53 ms β 85.47 ms β no change β
β QQuery 12 β 287.55 ms β 296.48 ms β no change β
β QQuery 13 β 389.51 ms β 405.15 ms β no change β
β QQuery 14 β 304.94 ms β 311.65 ms β no change β
β QQuery 15 β 305.15 ms β 310.03 ms β no change β
β QQuery 16 β 667.64 ms β 662.01 ms β no change β
β QQuery 17 β 666.61 ms β 680.08 ms β no change β
β QQuery 18 β 1355.59 ms β 1390.22 ms β no change β
β QQuery 19 β 29.08 ms β 29.36 ms β no change β
β QQuery 20 β 529.59 ms β 526.98 ms β no change β
β QQuery 21 β 527.02 ms β 528.32 ms β no change β
β QQuery 22 β 1011.03 ms β 1025.95 ms β no change β
β QQuery 23 β 3190.66 ms β 3252.03 ms β no change β
β QQuery 24 β 42.27 ms β 42.17 ms β no change β
β QQuery 25 β 116.03 ms β 115.76 ms β no change β
β QQuery 26 β 42.31 ms β 43.30 ms β no change β
β QQuery 27 β 533.13 ms β 530.13 ms β no change β
β QQuery 28 β 2985.77 ms β 3021.03 ms β no change β
β QQuery 29 β 42.10 ms β 42.54 ms β no change β
β QQuery 30 β 329.55 ms β 328.23 ms β no change β
β QQuery 31 β 302.64 ms β 304.70 ms β no change β
β QQuery 32 β 1030.28 ms β 993.71 ms β no change β
β QQuery 33 β 1578.18 ms β 1598.79 ms β no change β
β QQuery 34 β 1608.59 ms β 1557.41 ms β no change β
β QQuery 35 β 320.77 ms β 330.93 ms β no change β
β QQuery 36 β 75.88 ms β 76.91 ms β no change β
β QQuery 37 β 38.12 ms β 37.16 ms β no change β
β QQuery 38 β 42.43 ms β 44.28 ms β no change β
β QQuery 39 β 160.27 ms β 150.02 ms β +1.07x faster β
β QQuery 40 β 15.42 ms β 15.26 ms β no change β
β QQuery 41 β 15.00 ms β 15.30 ms β no change β
β QQuery 42 β 14.19 ms β 14.43 ms β no change β
βββββββββββββ΄βββββββββββββ΄βββββββββββββββββββββββ΄ββββββββββββββββ
βββββββββββββββββββββββββββββββββββββββ³βββββββββββββ
β Benchmark Summary β β
β‘βββββββββββββββββββββββββββββββββββββββββββββββββββ©
β Total Time (HEAD) β 20197.03ms β
β Total Time (alamb_progressive_io) β 20363.45ms β
β Average Time (HEAD) β 469.70ms β
β Average Time (alamb_progressive_io) β 473.57ms β
β Queries Faster β 1 β
β Queries Slower β 1 β
β Queries with No Change β 41 β
β Queries with Failure β 0 β
βββββββββββββββββββββββββββββββββββββββ΄βββββββββββββ
Distribution per query (min / mean Β±stddev / max):
Comparing HEAD and alamb_progressive_io
--------------------
Benchmark clickbench_partitioned.json
--------------------
βββββββββββββ³ββββββββββββββββββββββββββββββββββββββββ³ββββββββββββββββββββββββββββββββββββββββ³ββββββββββββββββ
β Query β HEAD β alamb_progressive_io β Change β
β‘ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ©
β QQuery 0 β 1.30 / 4.20 Β±5.66 / 15.51 ms β 1.28 / 4.21 Β±5.70 / 15.61 ms β no change β
β QQuery 1 β 12.11 / 12.59 Β±0.27 / 12.87 ms β 12.63 / 12.93 Β±0.23 / 13.26 ms β no change β
β QQuery 2 β 36.64 / 37.17 Β±0.33 / 37.59 ms β 37.93 / 38.67 Β±0.51 / 39.27 ms β no change β
β QQuery 3 β 32.12 / 32.68 Β±0.72 / 34.09 ms β 32.58 / 32.86 Β±0.23 / 33.14 ms β no change β
β QQuery 4 β 245.81 / 250.19 Β±2.32 / 252.22 ms β 249.16 / 253.41 Β±3.57 / 258.91 ms β no change β
β QQuery 5 β 292.41 / 296.00 Β±3.67 / 301.12 ms β 294.88 / 300.41 Β±4.11 / 305.04 ms β no change β
β QQuery 6 β 1.33 / 1.49 Β±0.23 / 1.94 ms β 1.37 / 1.52 Β±0.23 / 1.98 ms β no change β
β QQuery 7 β 13.72 / 14.02 Β±0.23 / 14.30 ms β 14.00 / 14.34 Β±0.20 / 14.56 ms β no change β
β QQuery 8 β 358.00 / 360.50 Β±1.38 / 361.82 ms β 360.07 / 364.86 Β±4.40 / 372.19 ms β no change β
β QQuery 9 β 490.53 / 500.49 Β±8.41 / 515.39 ms β 518.93 / 537.20 Β±18.31 / 560.44 ms β 1.07x slower β
β QQuery 10 β 72.28 / 77.03 Β±6.57 / 89.99 ms β 74.79 / 75.53 Β±0.63 / 76.52 ms β no change β
β QQuery 11 β 83.53 / 84.72 Β±0.93 / 86.32 ms β 85.47 / 87.72 Β±1.91 / 91.17 ms β no change β
β QQuery 12 β 287.55 / 295.74 Β±7.46 / 308.49 ms β 296.48 / 301.32 Β±6.17 / 313.02 ms β no change β
β QQuery 13 β 389.51 / 400.26 Β±14.33 / 428.24 ms β 405.15 / 417.65 Β±11.94 / 432.61 ms β no change β
β QQuery 14 β 304.94 / 310.17 Β±4.78 / 317.62 ms β 311.65 / 318.73 Β±4.44 / 325.10 ms β no change β
β QQuery 15 β 305.15 / 311.63 Β±8.38 / 327.40 ms β 310.03 / 318.71 Β±12.22 / 342.51 ms β no change β
β QQuery 16 β 667.64 / 676.86 Β±7.07 / 689.19 ms β 662.01 / 671.67 Β±7.58 / 685.17 ms β no change β
β QQuery 17 β 666.61 / 671.40 Β±3.61 / 675.90 ms β 680.08 / 701.55 Β±19.08 / 735.42 ms β no change β
β QQuery 18 β 1355.59 / 1386.15 Β±24.30 / 1414.98 ms β 1390.22 / 1428.55 Β±23.15 / 1463.01 ms β no change β
β QQuery 19 β 29.08 / 45.56 Β±31.87 / 109.28 ms β 29.36 / 31.52 Β±2.61 / 36.24 ms β +1.45x faster β
β QQuery 20 β 529.59 / 539.84 Β±9.00 / 551.96 ms β 526.98 / 543.91 Β±12.03 / 557.80 ms β no change β
β QQuery 21 β 527.02 / 533.91 Β±4.52 / 540.19 ms β 528.32 / 537.44 Β±6.53 / 548.17 ms β no change β
β QQuery 22 β 1011.03 / 1026.48 Β±9.54 / 1037.42 ms β 1025.95 / 1032.31 Β±5.83 / 1043.33 ms β no change β
β QQuery 23 β 3190.66 / 3233.71 Β±25.48 / 3264.66 ms β 3252.03 / 3286.93 Β±25.95 / 3319.69 ms β no change β
β QQuery 24 β 42.27 / 42.94 Β±0.53 / 43.82 ms β 42.17 / 44.42 Β±3.86 / 52.11 ms β no change β
β QQuery 25 β 116.03 / 124.01 Β±14.98 / 153.95 ms β 115.76 / 125.81 Β±8.80 / 137.69 ms β no change β
β QQuery 26 β 42.31 / 43.33 Β±0.70 / 44.32 ms β 43.30 / 43.87 Β±0.88 / 45.62 ms β no change β
β QQuery 27 β 533.13 / 537.49 Β±5.76 / 548.28 ms β 530.13 / 542.60 Β±17.19 / 575.50 ms β no change β
β QQuery 28 β 2985.77 / 2998.65 Β±13.99 / 3023.66 ms β 3021.03 / 3041.54 Β±19.31 / 3074.08 ms β no change β
β QQuery 29 β 42.10 / 46.03 Β±5.69 / 57.06 ms β 42.54 / 52.18 Β±7.77 / 59.74 ms β 1.13x slower β
β QQuery 30 β 329.55 / 332.31 Β±2.43 / 336.87 ms β 328.23 / 343.00 Β±13.50 / 359.69 ms β no change β
β QQuery 31 β 302.64 / 311.36 Β±8.73 / 327.07 ms β 304.70 / 321.35 Β±12.10 / 337.39 ms β no change β
β QQuery 32 β 1030.28 / 1060.85 Β±26.92 / 1095.97 ms β 993.71 / 1020.24 Β±17.46 / 1046.30 ms β no change β
β QQuery 33 β 1578.18 / 1602.91 Β±24.34 / 1649.47 ms β 1598.79 / 1619.40 Β±17.50 / 1641.05 ms β no change β
β QQuery 34 β 1608.59 / 1623.04 Β±13.75 / 1645.60 ms β 1557.41 / 1610.57 Β±29.08 / 1644.31 ms β no change β
β QQuery 35 β 320.77 / 362.65 Β±34.84 / 410.24 ms β 330.93 / 346.63 Β±19.32 / 383.70 ms β no change β
β QQuery 36 β 75.88 / 81.76 Β±6.21 / 89.44 ms β 76.91 / 82.24 Β±5.31 / 91.81 ms β no change β
β QQuery 37 β 38.12 / 41.26 Β±4.06 / 48.73 ms β 37.16 / 37.92 Β±0.60 / 39.01 ms β +1.09x faster β
β QQuery 38 β 42.43 / 44.19 Β±1.34 / 46.10 ms β 44.28 / 47.50 Β±4.87 / 56.95 ms β 1.07x slower β
β QQuery 39 β 160.27 / 167.95 Β±5.47 / 175.32 ms β 150.02 / 167.72 Β±12.29 / 181.15 ms β no change β
β QQuery 40 β 15.42 / 15.76 Β±0.62 / 16.99 ms β 15.26 / 20.73 Β±7.32 / 34.07 ms β 1.32x slower β
β QQuery 41 β 15.00 / 15.18 Β±0.16 / 15.40 ms β 15.30 / 18.79 Β±2.86 / 21.61 ms β 1.24x slower β
β QQuery 42 β 14.19 / 14.70 Β±0.28 / 15.06 ms β 14.43 / 18.07 Β±5.88 / 29.79 ms β 1.23x slower β
βββββββββββββ΄ββββββββββββββββββββββββββββββββββββββββ΄ββββββββββββββββββββββββββββββββββββββββ΄ββββββββββββββββ
βββββββββββββββββββββββββββββββββββββββ³βββββββββββββ
β Benchmark Summary β β
β‘βββββββββββββββββββββββββββββββββββββββββββββββββββ©
β Total Time (HEAD) β 20569.17ms β
β Total Time (alamb_progressive_io) β 20818.51ms β
β Average Time (HEAD) β 478.35ms β
β Average Time (alamb_progressive_io) β 484.15ms β
β Queries Faster β 2 β
β Queries Slower β 6 β
β Queries with No Change β 35 β
β Queries with Failure β 0 β
βββββββββββββββββββββββββββββββββββββββ΄βββββββββββββ
```
</p>
</details>
<details><summary>Resource Usage</summary>
**clickbench_partitioned β base (merge-base)**
| Metric | Value |
|--------|-------|
| Wall time | 105.0s |
| Peak memory | 12.4 GiB |
| Avg memory | 4.7 GiB |
| CPU user | 1053.2s |
| CPU sys | 78.5s |
| Peak spill | 0 B |
**clickbench_partitioned β branch**
| Metric | Value |
|--------|-------|
| Wall time | 105.0s |
| Peak memory | 11.3 GiB |
| Avg memory | 4.2 GiB |
| CPU user | 1057.2s |
| CPU sys | 81.2s |
| Peak spill | 0 B |
</details>
---
[File an issue](https://github.com/adriangb/datafusion-benchmarking/issues) against this benchmark runner
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "adriangbot (via GitHub)" <gi...@apache.org>.
adriangbot commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5307177594
π€ Benchmark running (GKE) | [trigger](https://github.com/apache/datafusion/pull/24395#issuecomment-5307167906)
**Instance:** `c4a-highmem-16` (12 vCPU / 65 GiB) | `Linux bench-c5307167906-1616-gtt5l 6.12.85+ #1 SMP Wed Jun 17 20:31:55 UTC 2026 aarch64 GNU/Linux`
<details><summary>CPU Details (lscpu)</summary>
```
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 alamb/progressive_io (4342770ee1ff6e31f9d7fc5ab201df3d027e3c31) to 00eba79 (merge-base) [diff](https://github.com/apache/datafusion/compare/00eba790b17f7f0a62e5b81627981197f9bfadb0..4342770ee1ff6e31f9d7fc5ab201df3d027e3c31)
<details><summary>Run configuration</summary>
```yaml
run benchmark clickbench_partitioned
```
</details>
Results will be posted here when complete
---
[File an issue](https://github.com/adriangb/datafusion-benchmarking/issues) against this benchmark runner
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "alamb (via GitHub)" <gi...@apache.org>.
alamb commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5307167906
run benchmark clickbench_partitioned
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org
Re: [PR] WIP: NOT READY FOR REVIEW: Add parquet `progressive_io` option to keep one-shot row group I/O with pushdown filters [datafusion]
Posted by "adriangbot (via GitHub)" <gi...@apache.org>.
adriangbot commented on PR #24395:
URL: https://github.com/apache/datafusion/pull/24395#issuecomment-5439527687
π€ Benchmark completed (GKE) | [trigger](https://github.com/apache/datafusion/pull/24395#issuecomment-5439121941)
**Instance:** `c4a-highmem-16` (12 vCPU / 65 GiB)
Comparing alamb/progressive_io (4342770ee1ff6e31f9d7fc5ab201df3d027e3c31) to 00eba79 (merge-base) [diff](https://github.com/apache/datafusion/compare/00eba790b17f7f0a62e5b81627981197f9bfadb0..4342770ee1ff6e31f9d7fc5ab201df3d027e3c31)
<details><summary>Run configuration</summary>
```yaml
run benchmark clickbench_pushdown
env:
SIMULATE_LATENCY: "true"
```
</details>
<details><summary>CPU Details (lscpu)</summary>
```
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>
<details><summary>Details</summary>
<p>
```
Comparing HEAD and alamb_progressive_io
--------------------
Benchmark clickbench_pushdown.json
--------------------
βββββββββββββ³βββββββββββββ³βββββββββββββββββββββββ³ββββββββββββββββ
β Query β HEAD β alamb_progressive_io β Change β
β‘ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ©
β QQuery 0 β 1.26 ms β 1.23 ms β no change β
β QQuery 1 β 1854.01 ms β 1889.36 ms β no change β
β QQuery 2 β 2165.26 ms β 2166.09 ms β no change β
β QQuery 3 β 2164.80 ms β 2164.84 ms β no change β
β QQuery 4 β 2276.66 ms β 2282.44 ms β no change β
β QQuery 5 β 2358.42 ms β 2412.38 ms β no change β
β QQuery 6 β 1.26 ms β 1.29 ms β no change β
β QQuery 7 β 2000.04 ms β 1935.03 ms β no change β
β QQuery 8 β 2432.24 ms β 2413.12 ms β no change β
β QQuery 9 β 2605.68 ms β 2629.12 ms β no change β
β QQuery 10 β 4319.55 ms β 2213.87 ms β +1.95x faster β
β QQuery 11 β 4282.78 ms β 2231.74 ms β +1.92x faster β
β QQuery 12 β 2376.84 ms β 2410.46 ms β no change β
β QQuery 13 β 4596.14 ms β 2519.17 ms β +1.82x faster β
β QQuery 14 β 4509.86 ms β 2430.76 ms β +1.86x faster β
β QQuery 15 β 2354.43 ms β 2373.79 ms β no change β
β QQuery 16 β 2632.76 ms β 2652.07 ms β no change β
β QQuery 17 β 2637.38 ms β 2633.65 ms β no change β
β QQuery 18 β 3179.42 ms β 3156.56 ms β no change β
β QQuery 19 β 1792.72 ms β 1777.21 ms β no change β
β QQuery 20 β 2667.99 ms β 2630.83 ms β no change β
β QQuery 21 β 4946.26 ms β 3561.42 ms β +1.39x faster β
β QQuery 22 β 8629.78 ms β 3866.85 ms β +2.23x faster β
β QQuery 23 β 1087.27 ms β 747.43 ms β +1.45x faster β
β QQuery 24 β 810.50 ms β 456.12 ms β +1.78x faster β
β QQuery 25 β 2279.25 ms β 2280.79 ms β no change β
β QQuery 26 β 735.60 ms β 498.49 ms β +1.48x faster β
β QQuery 27 β 4589.76 ms β 2690.06 ms β +1.71x faster β
β QQuery 28 β 5051.92 ms β 4932.75 ms β no change β
β QQuery 29 β 2228.94 ms β 2175.79 ms β no change β
β QQuery 30 β 4516.30 ms β 2369.97 ms β +1.91x faster β
β QQuery 31 β 4497.03 ms β 2390.78 ms β +1.88x faster β
β QQuery 32 β 2599.50 ms β 2497.55 ms β no change β
β QQuery 33 β 3455.57 ms β 3388.76 ms β no change β
β QQuery 34 β 3470.31 ms β 3461.50 ms β no change β
β QQuery 35 β 2490.45 ms β 2372.85 ms β no change β
β QQuery 36 β 623.41 ms β 170.84 ms β +3.65x faster β
β QQuery 37 β 594.09 ms β 144.78 ms β +4.10x faster β
β QQuery 38 β 699.50 ms β 141.70 ms β +4.94x faster β
β QQuery 39 β 596.35 ms β 207.78 ms β +2.87x faster β
β QQuery 40 β 879.75 ms β 122.49 ms β +7.18x faster β
β QQuery 41 β 726.76 ms β 126.96 ms β +5.72x faster β
β QQuery 42 β 707.40 ms β 115.53 ms β +6.12x faster β
βββββββββββββ΄βββββββββββββ΄βββββββββββββββββββββββ΄ββββββββββββββββ
βββββββββββββββββββββββββββββββββββββββ³ββββββββββββββ
β Benchmark Summary β β
β‘ββββββββββββββββββββββββββββββββββββββββββββββββββββ©
β Total Time (HEAD) β 109425.21ms β
β Total Time (alamb_progressive_io) β 83646.21ms β
β Average Time (HEAD) β 2544.77ms β
β Average Time (alamb_progressive_io) β 1945.26ms β
β Queries Faster β 19 β
β Queries Slower β 0 β
β Queries with No Change β 24 β
β Queries with Failure β 0 β
βββββββββββββββββββββββββββββββββββββββ΄ββββββββββββββ
Distribution per query (min / mean Β±stddev / max):
Comparing HEAD and alamb_progressive_io
--------------------
Benchmark clickbench_pushdown.json
--------------------
βββββββββββββ³βββββββββββββββββββββββββββββββββββββββββ³βββββββββββββββββββββββββββββββββββββββββ³ββββββββββββββββ
β Query β HEAD β alamb_progressive_io β Change β
β‘ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ©
β QQuery 0 β 1.26 / 4.03 Β±5.46 / 14.95 ms β 1.23 / 4.02 Β±5.52 / 15.06 ms β no change β
β QQuery 1 β 1854.01 / 1906.41 Β±44.12 / 1965.86 ms β 1889.36 / 1919.38 Β±25.27 / 1955.72 ms β no change β
β QQuery 2 β 2165.26 / 2252.52 Β±50.34 / 2304.47 ms β 2166.09 / 2252.40 Β±47.51 / 2299.28 ms β no change β
β QQuery 3 β 2164.80 / 2212.54 Β±39.23 / 2276.02 ms β 2164.84 / 2205.59 Β±21.55 / 2225.67 ms β no change β
β QQuery 4 β 2276.66 / 2357.52 Β±53.77 / 2439.42 ms β 2282.44 / 2344.67 Β±52.53 / 2426.39 ms β no change β
β QQuery 5 β 2358.42 / 2426.36 Β±56.30 / 2518.78 ms β 2412.38 / 2431.06 Β±17.17 / 2456.13 ms β no change β
β QQuery 6 β 1.26 / 1.42 Β±0.23 / 1.87 ms β 1.29 / 1.44 Β±0.21 / 1.86 ms β no change β
β QQuery 7 β 2000.04 / 2013.26 Β±11.41 / 2029.70 ms β 1935.03 / 1999.52 Β±64.35 / 2105.23 ms β no change β
β QQuery 8 β 2432.24 / 2475.79 Β±32.41 / 2523.79 ms β 2413.12 / 2471.78 Β±66.87 / 2590.92 ms β no change β
β QQuery 9 β 2605.68 / 2645.05 Β±36.38 / 2690.50 ms β 2629.12 / 2676.42 Β±42.47 / 2740.68 ms β no change β
β QQuery 10 β 4319.55 / 4396.07 Β±47.29 / 4461.33 ms β 2213.87 / 2273.78 Β±45.31 / 2354.32 ms β +1.93x faster β
β QQuery 11 β 4282.78 / 4408.11 Β±83.30 / 4544.35 ms β 2231.74 / 2293.09 Β±44.40 / 2359.38 ms β +1.92x faster β
β QQuery 12 β 2376.84 / 2434.65 Β±47.71 / 2509.19 ms β 2410.46 / 2436.56 Β±23.61 / 2478.50 ms β no change β
β QQuery 13 β 4596.14 / 4656.25 Β±45.77 / 4716.40 ms β 2519.17 / 2568.57 Β±43.83 / 2645.15 ms β +1.81x faster β
β QQuery 14 β 4509.86 / 4561.56 Β±35.23 / 4605.15 ms β 2430.76 / 2494.11 Β±48.64 / 2557.32 ms β +1.83x faster β
β QQuery 15 β 2354.43 / 2434.87 Β±53.05 / 2508.93 ms β 2373.79 / 2434.78 Β±55.77 / 2501.27 ms β no change β
β QQuery 16 β 2632.76 / 2668.02 Β±29.82 / 2717.43 ms β 2652.07 / 2673.07 Β±22.84 / 2715.79 ms β no change β
β QQuery 17 β 2637.38 / 2713.13 Β±61.42 / 2786.57 ms β 2633.65 / 2700.18 Β±35.59 / 2738.28 ms β no change β
β QQuery 18 β 3179.42 / 3196.09 Β±21.29 / 3236.83 ms β 3156.56 / 3200.40 Β±30.84 / 3246.91 ms β no change β
β QQuery 19 β 1792.72 / 1854.14 Β±61.77 / 1962.22 ms β 1777.21 / 1819.76 Β±32.69 / 1870.07 ms β no change β
β QQuery 20 β 2667.99 / 2705.06 Β±39.22 / 2778.58 ms β 2630.83 / 2715.73 Β±58.10 / 2796.24 ms β no change β
β QQuery 21 β 4946.26 / 5005.06 Β±60.07 / 5110.72 ms β 3561.42 / 3622.29 Β±69.79 / 3756.06 ms β +1.38x faster β
β QQuery 22 β 8629.78 / 8734.13 Β±109.57 / 8914.96 ms β 3866.85 / 3987.18 Β±109.34 / 4173.45 ms β +2.19x faster β
β QQuery 23 β 1087.27 / 1126.74 Β±40.07 / 1201.68 ms β 747.43 / 804.66 Β±51.09 / 897.67 ms β +1.40x faster β
β QQuery 24 β 810.50 / 865.34 Β±48.93 / 944.78 ms β 456.12 / 551.39 Β±67.28 / 631.72 ms β +1.57x faster β
β QQuery 25 β 2279.25 / 2336.28 Β±53.45 / 2431.98 ms β 2280.79 / 2355.03 Β±51.37 / 2413.79 ms β no change β
β QQuery 26 β 735.60 / 854.85 Β±81.55 / 978.13 ms β 498.49 / 610.52 Β±121.97 / 838.98 ms β +1.40x faster β
β QQuery 27 β 4589.76 / 4712.75 Β±112.68 / 4874.99 ms β 2690.06 / 2833.82 Β±97.28 / 2936.73 ms β +1.66x faster β
β QQuery 28 β 5051.92 / 5154.51 Β±92.58 / 5324.01 ms β 4932.75 / 5018.88 Β±87.69 / 5188.17 ms β no change β
β QQuery 29 β 2228.94 / 2296.63 Β±45.20 / 2364.50 ms β 2175.79 / 2238.36 Β±55.54 / 2313.34 ms β no change β
β QQuery 30 β 4516.30 / 4690.30 Β±163.68 / 4888.41 ms β 2369.97 / 2518.39 Β±90.02 / 2651.59 ms β +1.86x faster β
β QQuery 31 β 4497.03 / 4603.64 Β±83.40 / 4730.67 ms β 2390.78 / 2437.45 Β±36.47 / 2498.01 ms β +1.89x faster β
β QQuery 32 β 2599.50 / 2654.76 Β±44.27 / 2703.24 ms β 2497.55 / 2586.99 Β±105.28 / 2787.03 ms β no change β
β QQuery 33 β 3455.57 / 3469.74 Β±10.58 / 3487.65 ms β 3388.76 / 3429.65 Β±37.08 / 3494.91 ms β no change β
β QQuery 34 β 3470.31 / 3509.77 Β±30.75 / 3547.76 ms β 3461.50 / 3508.21 Β±35.47 / 3558.45 ms β no change β
β QQuery 35 β 2490.45 / 2522.24 Β±29.17 / 2562.76 ms β 2372.85 / 2443.15 Β±50.42 / 2517.58 ms β no change β
β QQuery 36 β 623.41 / 828.74 Β±140.83 / 969.29 ms β 170.84 / 205.10 Β±43.64 / 284.70 ms β +4.04x faster β
β QQuery 37 β 594.09 / 735.44 Β±114.35 / 898.45 ms β 144.78 / 183.78 Β±42.56 / 258.47 ms β +4.00x faster β
β QQuery 38 β 699.50 / 990.60 Β±161.67 / 1162.10 ms β 141.70 / 183.00 Β±34.21 / 240.11 ms β +5.41x faster β
β QQuery 39 β 596.35 / 721.21 Β±111.12 / 875.78 ms β 207.78 / 244.46 Β±40.93 / 319.83 ms β +2.95x faster β
β QQuery 40 β 879.75 / 937.32 Β±55.44 / 1018.91 ms β 122.49 / 178.18 Β±44.62 / 252.35 ms β +5.26x faster β
β QQuery 41 β 726.76 / 875.34 Β±112.77 / 1002.63 ms β 126.96 / 176.97 Β±42.59 / 250.05 ms β +4.95x faster β
β QQuery 42 β 707.40 / 808.95 Β±91.35 / 917.27 ms β 115.53 / 175.92 Β±44.70 / 250.11 ms β +4.60x faster β
βββββββββββββ΄βββββββββββββββββββββββββββββββββββββββββ΄βββββββββββββββββββββββββββββββββββββββββ΄ββββββββββββββββ
βββββββββββββββββββββββββββββββββββββββ³ββββββββββββββ
β Benchmark Summary β β
β‘ββββββββββββββββββββββββββββββββββββββββββββββββββββ©
β Total Time (HEAD) β 112757.20ms β
β Total Time (alamb_progressive_io) β 86209.67ms β
β Average Time (HEAD) β 2622.26ms β
β Average Time (alamb_progressive_io) β 2004.88ms β
β Queries Faster β 19 β
β Queries Slower β 0 β
β Queries with No Change β 24 β
β Queries with Failure β 0 β
βββββββββββββββββββββββββββββββββββββββ΄ββββββββββββββ
```
</p>
</details>
<details><summary>Resource Usage</summary>
**clickbench_pushdown β base (merge-base)**
| Metric | Value |
|--------|-------|
| Wall time | 565.1s |
| Peak memory | 9.1 GiB |
| Avg memory | 1.7 GiB |
| CPU user | 847.3s |
| CPU sys | 116.1s |
| Peak spill | 0 B |
**clickbench_pushdown β branch**
| Metric | Value |
|--------|-------|
| Wall time | 435.1s |
| Peak memory | 8.9 GiB |
| Avg memory | 1.8 GiB |
| CPU user | 839.2s |
| CPU sys | 108.2s |
| Peak spill | 0 B |
</details>
---
[File an issue](https://github.com/adriangb/datafusion-benchmarking/issues) against this benchmark runner
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: github-unsubscribe@datafusion.apache.org
For additional commands, e-mail: github-help@datafusion.apache.org