Evaluate ClickHouse/ClickHouse#104431 (single-file Parquet parallelism) on ClickBench - #1436
Draft
alexey-milovidov wants to merge 8 commits into
Draft
Evaluate ClickHouse/ClickHouse#104431 (single-file Parquet parallelism) on ClickBench#1436alexey-milovidov wants to merge 8 commits into
alexey-milovidov wants to merge 8 commits into
Conversation
`clickhouse-parquet-104431` and `clickhouse-parquet-baseline` are copies of `clickhouse-parquet` whose `install` pins a specific RelWithDebInfo build from ClickHouse CI instead of downloading the latest release: - `clickhouse-parquet-104431` - head of ClickHouse/ClickHouse#104431 ("Parallelize reads from a single Parquet file in `StorageFile`, again"), `07303b95365362daace76759e4c51e39efefd25a`; - `clickhouse-parquet-baseline` - the master commit CI merged it into, `aec8b1d56234cdf035fcc0fc65afe31a9a0f224d`. Everything else - `queries.sql`, `create.sql`, `query`, the dataset - is byte identical to `clickhouse-parquet`, so the two directories differ only by the pull request. Not for merging: this is a measurement harness for one pull request. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…et-baseline (c6a.2xlarge, c6a.4xlarge, c6a.metal, c7a.metal-48xl, c8g.4xlarge, c8g.metal-48xl)
Contributor
|
Results for Logs:
|
…et-baseline (c6a.large, c6a.xlarge)
Contributor
|
Results for Logs:
|
…et-baseline (t3a.small)
Contributor
|
Results for Logs:
|
alexey-milovidov
marked this pull request as draft
August 17, 2026 05:52
… off
`clickhouse-parquet-104431-nosplit` runs the same pinned #104431 binary as
`clickhouse-parquet-104431`, with `input_format_parquet_min_bytes_to_split`
raised to 1 PB so no read clears the floor.
The decision still runs - the pull request computes the projected read size,
and so parses the footer, before comparing it against the floor - so the three
directories decompose the difference:
baseline -> nosplit cost of making the split decision
nosplit -> 104431 effect of actually splitting
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…into eval-ch-pr-104431
alexey-milovidov
had a problem deploying
to
benchmark-approval
August 17, 2026 06:00 — with 
GitHub Actions
Error
GitHub Actions
Error
…rge, c6a.4xlarge, c6a.metal, c6a.xlarge, c7a.metal-48xl, c8g.4xlarge, c8g.metal-48xl)
Contributor
|
Results for Logs:
|
Contributor
|
Results for Logs:
|
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
1 participant
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.
Measurement harness, not for merging.
Evaluates ClickHouse/ClickHouse#104431
("Parallelize reads from a single Parquet file in
StorageFile, again") on ClickBench,on every machine type of the daily runs.
clickhouse-parquet-104431andclickhouse-parquet-baselineare copies ofclickhouse-parquetwhoseinstallpins a specific RelWithDebInfo build fromClickHouse CI instead of downloading the latest release:
clickhouse-parquet-104431PRs/104431/.../build_{amd,arm}_release07303b95365362daace76759e4c51e39efefd25a(pull request head)clickhouse-parquet-baselineREFs/master/.../build_{amd,arm}_releaseaec8b1d56234cdf035fcc0fc65afe31a9a0f224d(master)aec8b1d5is the honest A/B parent: ClickHouse CI built the pull request as the merge commite350489c26= "Merge07303b95into2a539919", and2a539919's only difference from itsparent
aec8b1d5is test files, so the binaries are source identical.2a539919itself has nobuild_amd_release- master CI skips that job - whileaec8b1d5has both architectures.Everything else (
queries.sql,create.sql,query,benchmark.sh, the dataset) is byteidentical to
clickhouse-parquet, so the two directories differ only by the pull request.Neither directory carries
results/, so nothing is added to the website.The pull request only changes the single local Parquet file path, which is what
clickhouse-parquetexercises. With the pull request's defaultinput_format_parquet_min_bytes_to_split = 2 GiB, 12 of the 43 queries splithits.parquetinto several sources atmax_threads = 16(Q21-Q24, Q28, Q29, Q34, Q35,Q37-Q40 - the ones reading
URL,TitleorReferer); the rest keep a single source, so theymeasure the cost of making the decision. The number of queries that split is expected to vary
with the machine's core count, which is why this runs on all nine machine types.