snowykami/meilisearch
1
0
Fork 0
mirror of https://github.com/meilisearch/meilisearch.git synced 2025-09-04 03:36:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

Optimize the filter

Browse Source
This commit is contained in:
Mubelotix
2025-07-28 16:24:04 +02:00
parent 478f374b9d
commit e8a818f53d
1 changed files with 37 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
crates/milli/src/search/facet/filter.rs
View File

@ -18,6 +18,7 @@ use crate::heed_codec::facet::{
FacetGroupKey, FacetGroupKeyCodec, FacetGroupValue, FacetGroupValueCodec,
};
use crate::index::db_name::FACET_ID_STRING_DOCIDS;
use crate::search::facet::facet_range_search::find_docids_of_facet_within_bounds;
use crate::{
distance_between_two_points, lat_lng_to_xyz, FieldId, FieldsIdsMap,
FilterableAttributesFeatures, FilterableAttributesRule, Index, InternalError, Result,
@ -416,7 +417,43 @@ impl<'a> Filter<'a> {
return Ok(docids);
}
Condition::StartsWith { keyword: _, word } => {
// There are two algorithms:
//
// - The first one looks directly at level 0 of the facet group database.
// This pessimistic approach is more efficient when the value is unique.
//
// - The second one is recursive over levels.
// This is more efficient when the prefix is common among many values.
let value = crate::normalize_facet(word.value());
if value.len() <= 6 {
// 6 is abitrary, but it works well in practice
let mut value2 = value.as_bytes().to_owned();
if let Some(last) = value2.last_mut() {
if *last != 255 {
*last += 1;
if let Ok(value2) = String::from_utf8(value2) {
// The idea here is that "STARTS WITH baba" is the same as "baba <= value < babb".
// We just increase the last letter to find the upper bound.
// The result could be invalid utf8, so it can fallback.
let mut docids = RoaringBitmap::new();
find_docids_of_facet_within_bounds(
rtxn,
strings_db,
field_id,
&Included(&value),
&Excluded(&value2),
universe,
&mut docids,
)?;
return Ok(docids);
}
}
}
}
let base = FacetGroupKey { field_id, level: 0, left_bound: value.as_str() };
let docids = strings_db
.prefix_iter(rtxn, &base)?