Hacktoberfest 2026：维护者为十月标记出来的 issue，仍然开放、适合新手。 浏览 Hacktoberfest issue

perf(results): build typed columns directly from the driver stream instead of collecting List<List<Object?>> first

未关闭
  • 哪里不对？
#1,552 2 条评论 0 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
4/5
预计耗时
3-5 天
新手友好度
42/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
活跃
技术栈
dart, mysql, postgresql, sqlite
领域
database, performance

调研方向

Read postgres_sql_workspace.dart, mysql_sql_workspace.dart, and sqlite_sql_workspace.dart to trace how each driver currently collects rows and builds CompactResultDataset; then inspect CompactResultDataset.fromRawRows and the related work in #1548. Define a shared builder that preserves current mixed-type and NULL behavior, and use existing dataset tests to check it across all three drivers. Done means those tests pass and the PR reports peak RSS and timings for 5k and 50k rows.

由索引模型根据 Issue 内容生成。

描述

backend core enhancement P3 performance
Problem

Each SQL driver workspace builds an intermediate List<List<Object?>> of the whole result and only then packs it into typed columns (CompactResultDataset.fromRawRows):

  • PostgreSQL (postgres_sql_workspace.dart): rawRows.add(row.toList()) for every row.
  • MySQL (mysql_sql_workspace.dart): taken.items (all ResultSetRows kept) and then [for (i...) row.colAt(i)] per row.
  • SQLite: same pattern (sqlite_sql_workspace.dart).

So every cell is copied into a per-row list once, kept alive until the dataset is built, and then read again column by column.

Measurements

Local PostgreSQL 16 / MySQL 8.4 containers, 100k rows x 20 columns, dart run, median of 3:

Driver Rows Network + decode Extra row copy Longest event-loop stall during it
PostgreSQL 5,000 21-22 ms 1-3 ms (row.toList) 0 ms
PostgreSQL 50,000 257-271 ms 10-19 ms 0-2 ms
MySQL 5,000 31-38 ms 2-5 ms (colAt) 4-7 ms
MySQL 50,000 296-320 ms 33-42 ms 7-15 ms

(The decode itself yields between socket chunks: the PostgreSQL driver blocks the event loop for 0-2 ms even at 50k rows, so moving the driver into a worker isolate would not remove a visible hitch; it only adds connect cost and a result hand-off. Not recommended.)

Solution

Let the driver loop append into column builders (int / double / string / generic with a null bitmap) as rows arrive, deciding the kind from the first rows and falling back to generic when a value does not fit, instead of collecting List<List<Object?>>. A small ResultColumnsBuilder would be shared by the three drivers and by #1548's sampled detection.

Expected: removes the 1-3 ms (5k) / 10-40 ms (50k) copy and the peak memory of the intermediate list, and makes the "decode / pack" step incremental.

Acceptance
  • A shared builder used by PostgreSQL, MySQL and SQLite workspaces; existing dataset tests unchanged.
  • Mixed-type and NULL-heavy columns still end up as generic / string exactly as today (tests).
  • Peak RSS and time for 5k and 50k rows reported in the PR.

Related: #1548, #1549.

主要语言
Dart
星标
23
派生
1
平均合并
31 分钟
30 天内合并 PR
570

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。

QueryaHub/Querya-Desktop 的其他 Issue

查看 QueryaHub/Querya-Desktop 的全部 Issue

相似的 Issue

更多 Dart Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。