Call for real world benchmarks / use cases

未关闭
  • 哪里不对？
#27,330 1 条评论 4 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
35/100
Issue 类型
文档
描述清晰度
基本清楚
活跃度
活跃
技术栈
rust
领域
performance

调研方向

Start with the benchmark examples linked in the issue and the contributor guidance. The requested work is to submit a real-world workload with a runnable reproducer, performance details, and one query pattern; contributors should reproduce it and post an analysis before proposing optimizations. The issue provides an example query but no specific workload to implement, so a code change cannot be scoped from this issue alone.

由索引模型根据 Issue 内容生成。

描述

Background

A core value proposition for DataFusion is its performance.

We have lots of benchmarks, but there are many real world use cases that are not covered. For example, queries on STRUCTs / LISTSs and other nested types, but there are many more.

There are many people looking for ways to help, and I want optimization effort to go to workloads people actually run rather than over optimize benchmarks like ClickBench or random things.

Open Call / Request

To that end, I this is an open call for users of DataFusion to file issues describing queries they would like to go faster. Obviously we will not be able to help everyone, however, supplying the information below I think will maximize the chance of making it happen.

Why: Ensure DataFusion stays focused on real users and actual use cases.

Who: Real users of DataFusion -- you are using DataFusion and would like some particular workload to go faster.

How: Open a new issue with the following information

Format

Please provide the following information, concisely (please don't dump multiple pages of AI generated spew with every imaginable detail 🙏 ).

Use case: 1-2 sentences (max) describing how you are using DataFusion. Examples: "Log data analysis from Parquet files on S3", "Full text search engine on local Vortex files", etc.

Local reproducer:

  1. Data files. In order of preference:

    • a script that generates the data
    • a synthetic or anonymized sample with the same shape as your real data
    • a public URL to the files

    Ideally the data is no more than 10s of GB and the query runs in under a minute on a laptop.

  2. A simple benchmark command (ideally a .sql file we can run in datafusion-cli) that runs against your data files.

  3. 1-2 sentences explaining what the data and query mean in English (for example the query descriptions in our current benchmarks)

If you can not share data, the schema and EXPLAIN ANALYZE output may still be useful, but issues with a runnable reproducer are far more likely to get attention and be acted on.

Current performance:

  1. DataFusion version, and any non-default configuration settings
  2. How you run DataFusion (e.g. SQL or dataframe, do you have custom optimzer rules, etc)
  3. Hardware
  4. Why you think it could be faster (for example, the time another engine takes, or a profile / explain plan showing where the time goes)

Restrictions:

  1. Only include 1 query or query pattern in each issue (not your whole workload, please) to keep the conversation focused.

Then please post a link to the new ticket below.

For contributors

If you are looking for a way to help, check out any usses

The first thing is not to code (or have your agent code) Instead, you should reproduce reported performance and analyze where the time is going and post a writeup to the ticket. Please include flamegraphs, diagrams, etc , not walls of AI generated analysis.

Then, explore possible optimizations (AI tools are great at this). Review each idea carefully, and if you find one that looks general and improves performance write it up (1 paragraph) explaining the high level optimization, why it works, and what classes of queries are affected.

Example

Here is an example using ClickBench Q22.

Use case: Web analytics over a single Parquet table (ClickBench hits).

Data: 14 GB, 100M rows, 105 columns

wget https://datasets.clickhouse.com/hits_compatible/hits.parquet

Query (q22.sql, run with datafusion-cli -f q22.sql):

CREATE EXTERNAL TABLE hits STORED AS PARQUET LOCATION 'hits.parquet';

SELECT "SearchPhrase", MIN("URL"), MIN("Title"), COUNT(*) AS c, COUNT(DISTINCT "UserID")
FROM hits
WHERE "Title" LIKE '%Google%' AND "URL" NOT LIKE '%.google.%' AND "SearchPhrase" <> ''
GROUP BY "SearchPhrase" ORDER BY c DESC LIMIT 10;

How this matches my real use case: TODO 1-2 sentences

DataFusion Version: 56.0.0

主要语言
Rust
星标
9.6k
派生
2.6k
平均合并
3 天 4 小时
30 天内合并 PR
345

环境准备

在 Codespaces 中打开

在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

apache/datafusion 的其他 Issue

查看 apache/datafusion 的全部 Issue

相似的 Issue

更多 Rust Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。