pingcap/tidb

Improve the performance of `Sort` by parallel

Open

#14,417 建立於 2020年1月9日

在 GitHub 查看
 (1 留言) (0 反應) (0 負責人)Go (40,090 star) (6,186 fork)batch import
help wantedsig/executiontype/enhancement

描述

Feature Request

Is your feature request related to a problem? Please describe:

Now the Sort executor is executed serially. We can run it parallelly to speed up.

Describe the feature you'd like:

Considering using a parallel framework, to run each sort partition in parallel. And then using a merge sort to generate the final result. We can refer to this proposal https://github.com/pingcap/tidb/pull/14238#issuecomment-569880893 and reuse it (after this pr merge).

Describe alternatives you've considered:

Task:

  1. Implement parallel sort.
  • Implement merge sort algorithm above Shuffle operator.
  • Implement plan builder for parallel sort.
  1. Cost Model, explain analyze information.
  • Change the cost model of sort if it will run parallelly.
  • Change the explain analyze result for parallel sort.

Teachability, Documentation, Adoption, Migration Strategy:

貢獻者指南

技術棧
go
領域
backenddatabaseperformance
議題類型
feature
難度面向新貢獻者的預計實作難度，1 表示很小改動，5 表示專家級工作。
5
預計時間有經驗貢獻者完成調查、實作、測試並準備 pull request 的粗略時間範圍。
over 1 week
活動狀態議題目前的可參與程度：新鮮、活躍、陳舊、阻塞或等待維護者輸入。
stale
清晰度議題是否清楚說明預期改動、驗收標準和下一步。
clear
前置要求
GoTiDB internalsMerge sort algorithmParallel execution
新手友善度1-100 的估計分數，表示該議題對首次貢獻者的友善程度。
10
研究方向
Review the referenced proposal in PR #14238 and the existing sort executor code in TiDB. Focus on integrating a merge sort above the Shuffle operator and modifying the plan builder. Understand the current cost model and explain analyze output to plan changes for parallel sort.