Repository navigation
Add the nested TPC-H benchmark (tpch-nested) to vortex-bench - #11191
Merged
Merged
Conversation
Contributor
|
can you rebase on develop now? I have fixed the export bug |
mprammer
force-pushed
the
mp/tpch-nested
branch
from
November 6, 2026 00:41
f60fde0 to
e28333f
Compare
Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: Martin Prammer <martin@spiraldb.com>
mprammer
force-pushed
the
mp/tpch-nested
branch
from
November 6, 2026 00:44
e28333f to
c8ca0ed
Compare
robert3005
approved these changes
Nov 6, 2026
robert3005
left a comment
Contributor
There was a problem hiding this comment.
We need to wire that into the runner but this gives us the scaffolding for the benchmark itself
Contributor
|
What I meant is we need to add it to github actions orchestration |
lwwmanning
added a commit
that referenced
this pull request
Nov 21, 2026
…ed TPC-H S3 data path (#11726) This changes the nightly SQL benchmark matrix in `bench-orchestrator/bench_orchestrator/ci_matrix/catalog.py`. The develop and PR presets are unchanged. Before, nightly ran DataFusion's datasource integration on the v1 scan for every benchmark, which #11685 added. On its first night, six of those jobs failed. TPC-DS SF=10 hung on q62 until the 120-minute job timeout, and the runner shut down during TPC-DS SF=100 q62 and TPC-H SF=100 q18. After, nightly skips datasource v1. The file v1 scans on DataFusion and DuckDB stay, because they pass. Develop and `pr-scan` still run datasource v1 on ClickBench and TPC-H SF=1. Before, Nested TPC-H on S3 failed every night since #11191. prepare-data wrote the data under `tpch-nested/100/list`, but the upload read `tpch-nested/100.0/list`. After, the upload reads the directory that prepare-data writes. A new test checks that each S3 data path names the scale factor its job passes. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Signed-off-by: Will Manning <will@willmanning.io>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds
tpch-nested, a port of Apache DataFusion Comet's nested TPC-H benchmark, to vx-bench. Each table keeps its join keys and one partition column (p_brand,c_mktsegment,o_orderdate,l_shipdate) at the top level and folds every other column into a one-element<table>_data: list<struct<...>>; each query unnests the lists it reads. Because every list holds exactly one element, the queries return the flat TPC-H answers and the suite reuses the flat row-count checks. It gives us a known external nested-data workload to measure the list layout against Parquet on both engines.Changes
The data comes from the flat suite's generator with a nesting step (
tpch::nest), so it keeps vx-bench's types, 1 Mi-row row groups, and file splits rather than Comet's Spark output;nested_schemareproduces Comet's per-table column placement and is unit-tested against it. The 22 queries insql/tpch-nested/list/are the flat vx-bench queries with Comet's derived-table rewrite (unnest(lineitem_data) as lineitem_data, thenlineitem_data.l_quantity), one SQL text per query shared by DataFusion and DuckDB. The suite takes ashapeoption (onlylisttoday) so a differently nested variant of the same data can be added without renaming anything. The list layout needs the unstable bench edition and the V1 scan, so the orchestrator pins this suite to--scan-v1 truelike the other list-layout suites. CI runs SF1 on NVMe inpr-fullanddevelopand SF100 on NVMe and S3 innightly, mirroring flat TPC-H; every lane in those presets was run at SF1 locally with row counts validated.The first commit is the DuckDB exporter fix from #11190, which this suite needs for the DuckDB lanes; it drops out on rebase once that merges.
🤖 Generated with Claude Code