Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
28 commits
Select commit Hold shift + click to select a range
4e4ed9d
Improve Scan V2 TopK feedback and sparse payload retrieval
joseph-isaacs Oct 29, 2026
d0be576
Choose Scan V2 access from input row selections
joseph-isaacs Oct 29, 2026
3f5325a
Choose sparse dictionary access from predicate and row demand
joseph-isaacs Oct 30, 2026
0679729
Simplify Scan V2 dictionary retrieval to one execution path
joseph-isaacs Oct 30, 2026
f01efc1
Default DuckDB to Scan V2 and size work using file parallelism
joseph-isaacs Oct 30, 2026
f94b70a
Bind DuckDB file scans to their stored column order
joseph-isaacs Oct 30, 2026
61b075d
Size wide filtered Scan V2 work by bytes
joseph-isaacs Oct 30, 2026
2c6d5e6
Integrate sparse scans with layout-aware DuckDB work scheduling
joseph-isaacs Oct 30, 2026
6c489ce
Overlap reads across two DuckDB scan work items per thread
joseph-isaacs Nov 3, 2026
963fdfa
Amortize narrow fixed-width scans over larger row chunks
joseph-isaacs Nov 3, 2026
a991c95
Compare scan versions on TPC-H, TPC-DS and ClickBench in Actions
joseph-isaacs Nov 3, 2026
9c678bb
Identify paired benchmark runs in CI
joseph-isaacs Nov 3, 2026
f3bf313
Verify benchmark artifacts against build manifests
joseph-isaacs Nov 3, 2026
8a6097d
Respect DataFusion scan partition parallelism
joseph-isaacs Nov 3, 2026
05b8b05
Continue scan comparisons after recording query failures
joseph-isaacs Nov 3, 2026
4efceab
Keep inline DataFusion scans on one task across IO waits
joseph-isaacs Nov 3, 2026
a8499d1
Keep benchmark cache paths consistent on NVMe
joseph-isaacs Nov 3, 2026
9d0e271
Stream stored projection pieces through filtered scans
joseph-isaacs Nov 3, 2026
44e1952
Preserve DataFusion read overlap by default
joseph-isaacs Nov 3, 2026
bd26538
Keep incomplete comparison warnings outside result tables
joseph-isaacs Nov 3, 2026
a114a2a
Reuse ordered segment locations for V2 range lookup
joseph-isaacs Nov 3, 2026
7245606
Bound DataFusion Arrow batches within large V2 work chunks
joseph-isaacs Nov 3, 2026
6c3f592
Allow longer TPC-DS SF50 benchmark sample processes
joseph-isaacs Nov 3, 2026
0ac01b8
Capture dedicated Scan V2 diagnostic profiles in Actions
joseph-isaacs Nov 3, 2026
f62d4e3
Bound Arrow materialization for narrow fixed-width scans
joseph-isaacs Nov 3, 2026
ddf747c
Stream completed static filter pieces into projection
joseph-isaacs Nov 3, 2026
9ad658b
Drop filter piece streaming after the TPC-H comparison
joseph-isaacs Nov 3, 2026
ac26753
Coarsen narrow unfiltered DataFusion scan partitions
joseph-isaacs Nov 3, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
41 changes: 41 additions & 0 deletions .github/workflows/pr-bench-sql.yml
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,14 @@ on:
required: false
type: string
default: "pr"
profile_run_id:
required: false
type: string
default: ""
profile_source_sha:
required: false
type: string
default: ""
workflow_dispatch:
inputs:
matrix_preset:
Expand All @@ -29,6 +37,18 @@ on:
- "pr-compact"
- "pr-all"
- "pr-full"
- "scan-v2"
- "scan-v2-profile"
profile_run_id:
description: "Existing scan comparison run supplying profile binaries"
required: false
type: string
default: ""
profile_source_sha:
description: "Exact source SHA in the profile binary build manifest"
required: false
type: string
default: ""

permissions:
contents: read
Expand All @@ -37,8 +57,29 @@ permissions:

jobs:
sql:
if: inputs.matrix_preset != 'scan-v2' && inputs.matrix_preset != 'scan-v2-profile'
uses: ./.github/workflows/sql-bench-matrix.yml
secrets: inherit
with:
mode: "pr"
matrix_preset: ${{ inputs.matrix_preset }}

scan-comparison:
if: inputs.matrix_preset == 'scan-v2'
uses: ./.github/workflows/scan-v2-comparison.yml
secrets: inherit
with:
head_sha: ${{ github.event.pull_request.head.sha || github.sha }}
base_ref: ${{ github.event.pull_request.base.sha || 'origin/develop' }}

scan-profile:
if: inputs.matrix_preset == 'scan-v2-profile'
permissions:
actions: read
contents: read
id-token: write
uses: ./.github/workflows/scan-v2-profile.yml
secrets: inherit
with:
source_run_id: ${{ inputs.profile_run_id }}
source_sha: ${{ inputs.profile_source_sha }}
213 changes: 213 additions & 0 deletions .github/workflows/scan-v2-comparison.yml
Original file line number Diff line number Diff line change
@@ -0,0 +1,213 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright the Vortex contributors

name: Paired Scan V2 Comparisons

on:
workflow_call:
inputs:
head_sha:
required: true
type: string
base_ref:
required: true
type: string

permissions:
contents: read
id-token: write

concurrency:
group: scan-v2-comparison-${{ inputs.head_sha }}
cancel-in-progress: false

jobs:
resolve:
runs-on: ubuntu-latest
timeout-minutes: 10
outputs:
base_sha: ${{ steps.base.outputs.sha }}
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
ref: ${{ inputs.head_sha }}
fetch-depth: 0
persist-credentials: false
- id: base
name: Pin the comparison base
env:
BASE_REF: ${{ inputs.base_ref }}
run: |
sha=$(git rev-parse --verify "${BASE_REF}^{commit}")
echo "sha=$sha" >> "$GITHUB_OUTPUT"

build:
needs: resolve
timeout-minutes: 60
strategy:
fail-fast: false
matrix:
revision: [base, head]
runs-on: >-
${{ github.repository == 'vortex-data/vortex'
&& format('runs-on={0}/region=us-east-1/runner=bench-dedicated/family=c8gd.8xlarge+c8g.8xlarge+m8gd.8xlarge+m8g.8xlarge+r8gd.8xlarge+r8g.8xlarge/image=ubuntu26-full-arm64-pre-v2/tag=scan-{1}/extras=s3-cache', github.run_id, matrix.revision)
|| 'ubuntu-latest' }}
steps:
- uses: runs-on/action@v2
if: github.repository == 'vortex-data/vortex'
with:
sccache: s3
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
ref: ${{ matrix.revision == 'base' && needs.resolve.outputs.base_sha || inputs.head_sha }}
persist-credentials: false
- uses: ./.github/actions/assert-cpu
if: github.repository == 'vortex-data/vortex'
- uses: ./.github/actions/setup-prebuild
with:
repo-token: ${{ secrets.GITHUB_TOKEN }}
enable-sccache: ${{ github.repository == 'vortex-data/vortex' && 'true' || 'false' }}
- name: Build identical native benchmark targets
env:
RUSTFLAGS: "-C target-cpu=native -C force-frame-pointers=yes"
run: |
cargo build --locked --bin data-gen --bin datafusion-bench --bin duckdb-bench --profile release_debug
python3 - <<'PY'
import hashlib
import json
import subprocess
from pathlib import Path
directory = Path("target/release_debug")
binaries = {}
for name in ("data-gen", "datafusion-bench", "duckdb-bench"):
with (directory / name).open("rb") as executable:
binaries[name] = hashlib.file_digest(executable, "sha256").hexdigest()
manifest = {"commit": subprocess.check_output(["git", "rev-parse", "HEAD"], text=True).strip(),
"binaries": binaries}
(directory / "build-manifest.json").write_text(json.dumps(manifest, indent=2))
PY
- uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7
with:
name: scan-comparison-${{ matrix.revision }}-binaries
path: |
target/release_debug/data-gen
target/release_debug/datafusion-bench
target/release_debug/duckdb-bench
target/release_debug/build-manifest.json
target/release_debug/build/vortex-duckdb-cache/duckdb-lib-*-prebuilt/libduckdb.so

bench:
needs: [resolve, build]
timeout-minutes: 180
strategy:
fail-fast: false
matrix:
include:
- id: tpch-10
suite: tpch
scale: "10.0"
- id: tpch-50
suite: tpch
scale: "50.0"
- id: tpcds-10
suite: tpcds
scale: "10.0"
- id: tpcds-50
suite: tpcds
scale: "50.0"
- id: clickbench
suite: clickbench
scale: ""
runs-on: >-
${{ github.repository == 'vortex-data/vortex'
&& format('runs-on={0}/region=us-east-1/runner=bench-dedicated/family=c8gd.metal-24xl/image=ubuntu26-full-arm64-pre-v2/tag=scan-{1}/extras=s3-cache', github.run_id, matrix.id)
|| 'ubuntu-latest' }}
env:
FLAT_LAYOUT_INLINE_ARRAY_NODE: "1"
steps:
- uses: runs-on/action@v2
if: github.repository == 'vortex-data/vortex'
with:
sccache: s3
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
ref: ${{ inputs.head_sha }}
persist-credentials: false
- uses: ./.github/actions/assert-cpu
if: github.repository == 'vortex-data/vortex'
- uses: ./.github/actions/setup-prebuild
with:
repo-token: ${{ secrets.GITHUB_TOKEN }}
- uses: ./.github/actions/setup-duckdb
- uses: ./.github/actions/system-info
- uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8
with:
name: scan-comparison-head-binaries
path: target/scan-comparison/head
- uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8
with:
name: scan-comparison-base-binaries
path: target/scan-comparison/base
- name: Place inputs and DuckDB spill files on local NVMe
env:
CASE_ID: ${{ matrix.id }}
run: |
data_root="$RUNNER_TEMP/scan-comparison-data"
if [[ -d /mnt/ephemeral ]]; then
data_root=/mnt/ephemeral/scan-comparison-data
fi
data_dir="$data_root/$GITHUB_RUN_ID/$CASE_ID"
sudo install -d -o "$(id -u)" -g "$(id -g)" "$data_dir" "$data_dir/data" "$data_dir/tmp"
mv target "$data_dir/target"
ln -s "$data_dir/target" target
ln -s "$data_dir/data" vortex-bench/data
mkdir -p scan-comparison-results
ln -s "$data_dir/tmp" scan-comparison-results/tmp
- name: Prepare shared benchmark files
env:
SUITE: ${{ matrix.suite }}
SCALE: ${{ matrix.scale }}
run: |
chmod +x target/scan-comparison/{head,base}/{data-gen,datafusion-bench,duckdb-bench}
options=()
if [[ -n "$SCALE" ]]; then
options+=(--opt "scale-factor=$SCALE")
fi
target/scan-comparison/head/data-gen "$SUITE" --formats vortex "${options[@]}"
- name: Prepare benchmark CPUs
run: sudo bash scripts/setup-benchmark.sh
- name: Compare base V2, head V2 and head V1 sequentially
env:
SUITE: ${{ matrix.suite }}
SCALE: ${{ matrix.scale }}
BASE_SHA: ${{ needs.resolve.outputs.base_sha }}
HEAD_SHA: ${{ inputs.head_sha }}
QUERY_TIMEOUT_SECONDS: ${{ matrix.id == 'tpcds-50' && '900' || '300' }}
BENCH_ITERATIONS: ${{ matrix.id == 'tpcds-50' && '7' || '11' }}
BENCH_ROUNDS: ${{ matrix.id == 'tpcds-50' && '2' || '4' }}
run: |
options=()
if [[ -n "$SCALE" ]]; then
options+=(--scale-factor "$SCALE")
fi
bash scripts/bench-taskset.sh python3 scripts/compare-scan-versions.py "$SUITE" \
--base-dir target/scan-comparison/base --head-dir target/scan-comparison/head \
--base-sha "$BASE_SHA" --head-sha "$HEAD_SHA" \
--iterations "$BENCH_ITERATIONS" --warmup 2 --rounds "$BENCH_ROUNDS" \
--query-timeout "$QUERY_TIMEOUT_SECONDS" \
--output-dir scan-comparison-results "${options[@]}"
- name: Publish comparison summary
if: always()
run: |
if [[ -f scan-comparison-results/summary.md ]]; then
cat scan-comparison-results/summary.md >> "$GITHUB_STEP_SUMMARY"
else
echo "Comparison incomplete; see job logs and partial measurement artifacts." >> "$GITHUB_STEP_SUMMARY"
fi
- uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7
if: always()
with:
name: scan-comparison-${{ matrix.id }}
path: |
scan-comparison-results/
!scan-comparison-results/tmp/**
Loading
Loading