meilisearch/MeiliSearch

Improve the geosearch

Open

#3,377 opened on Sep 9, 2021

View on GitHub
 (3 comments) (2 reactions) (0 assignees)Rust (20,887 stars) (733 forks)batch import
help wantedmilliperformance

Description

Multiple things can / should be done to improve the geosearch:

  • Currently, we need to Deserialize the whole RTree in memory before being able to do anything. This consumes a lot of RAM and thus is quite slow. => We should patch rstar to add the possibility to parse the RTree part by part
  • When creating an RTree for the first time we should use the function bulk_load
  • Add the possibility to sort the position by descending order. It will also require to modify rstar

Contributor guide

Tech stack
rust
Domain
backendsearchperformance
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
3-5 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
Rust programmingspatial indexing conceptsrstar library internals
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
20
Research direction
This issue involves three tasks: (1) patching the rstar crate to allow parsing RTree incrementally, (2) using bulk load during initial RTree creation in milli/src/update/index documents/typed chunk.rs, and (3) adding descending sort order. The primary challenge is modifying rstar to support streaming deserialization, which may require forking or contributing upstream. Research should begin by examining rstar's current deserialization implementation and understanding the bulk load API. The maintainers have not provided additional guidance, so the contributor may need to propose a design before coding.