quickwit-oss/tantivy

Make the Document binary serialization format more compact.

Open

#903 建立於 2020年10月5日

在 GitHub 查看
 (2 留言) (0 反應) (0 負責人)Rust (8,354 star) (499 fork)batch import
good first issuehelp wanted

描述

The DocStore builds blocks of contiguous document serialized in an adhoc simple binary format. These blocks are then compressed.

The format goes:

  • field: u32
  • type tag: u8
  • value: specific to the type but for instance u64 simply take 8 byts.

@ppodolsky noticed that despite the compression, it might be possible to shave off a few percent of storage by changing the encoding. This might be especially useful when running

Field, unsigned int, and date could be using variable byte encoding. Signed int could use zigzag encoding.

This issue is open to new contributor. Variable int encoding is available (search VInt). Zigzag is not implemented yet. Please add a bench to ensure there is not a catastrophic change in CPU usage when using LZ4, and please try to quantify the size impact on a pathological schema (100 uint for instance).

貢獻者指南

技術棧
rust
領域
database
議題類型
performance
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。
3
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。
3-5 days
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。
fresh
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
Rust programming basicsUnderstanding of variable integer encoding
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。
60
研究方向
Start by examining the current binary serialization format in the DocStore code (likely in src/docstore/...). Search for existing VInt encoding using `grep` in the codebase. Implement zigzag encoding for signed ints. Add a benchmark to measure CPU impact with LZ4 and quantify size reduction on a pathological schema with 100 uint fields.