Slow GeoShapeQuery Performance in Bleve (v2.4.3)

[MeonAigeanta]

I'm experiencing slow performance with Bleve's GeoShapeQuery in my Go (v1.22) application. The query takes around 355 ms to return 51,092 hits, and I'm hoping to optimize it for faster results and understand why my code is running slow.

I'm using the following code to create the GeoShapeQuery:

geojson := `{"type":"Polygon","coordinates":[[[-152,62],[-152,59],[-147,59],[-147,62],[-152,62]]]}`
shape, _ := geo.ParseGeoJSONShape([]byte(geojson))
gsq := &query.GeoShapeQuery{
    FieldVal: "geog",
    Geometry: query.Geometry{
      Shape:    shape,
      Relation: "within",
    },
  }
sreq := bleve.NewSearchRequest(gsq)

I varied the batch size used to build the index then searched with one of the polygons:

Batch	> 1000	> 500	> 100	> 50	No batching
ms	683	492	355	351	364

I non-randomly varied the number of entries (batch > 100):

Entries	100,000	200,000	300,000	400,000	526,286
ms	91	139	197	261	373

I varied the size of the polygon (batch > 100):

polygon	1	2	3	4	5
ms	354	366	390	372	327
hits	51,092	35,812	12,474	363	6

Setting the relation to “within” is faster than using “intersects”.

I'd appreciate any guidance on optimizing my GeoShapeQuery performance. Are there any potential issues with my code or configuration that could be causing the slow results?

I've created a complete minimal example on github: 
MeonAigeanta/bleve-search-test@a7d782c

[abhinavdangeti]

Hey @Likith101 would you look into this one when you can please.

[Likith101]

Hey @MeonAigeanta , apologies for the late response.

I went through your code and I only find one thing that could help improve your performance and that is multithreading the queries. For a use case with one query at a time however, this is the maximum as of now.

There are a few work in progress commits to improve performance in this area as of now. You could perhaps try cherry picking this (blevesearch/zapx#258) commit on a local bleve/zapx build and run the performance numbers again.

[MeonAigeanta]

Thank you for looking at my code and for your suggestions.