meilisearch/milli/src/search/mod.rs

264 lines
8.9 KiB
Rust
Raw Normal View History

use std::borrow::Cow;
2021-04-14 01:10:58 +08:00
use std::collections::hash_map::{Entry, HashMap};
use std::fmt;
2021-04-14 18:18:13 +08:00
use std::mem::take;
use std::str::Utf8Error;
use std::time::Instant;
use fst::{IntoStreamer, Streamer};
2021-04-14 01:10:58 +08:00
use levenshtein_automata::{LevenshteinAutomatonBuilder as LevBuilder, DFA};
use log::debug;
2021-04-14 01:10:58 +08:00
use meilisearch_tokenizer::{Analyzer, AnalyzerConfig};
use once_cell::sync::Lazy;
use roaring::bitmap::RoaringBitmap;
2021-04-14 01:10:58 +08:00
use distinct::{Distinct, DocIter, FacetDistinct, MapDistinct, NoopDistinct};
2021-03-23 22:25:46 +08:00
use crate::search::criteria::r#final::{Final, FinalResult};
use crate::{Index, DocumentId};
pub use self::facet::{FacetCondition, FacetDistribution, FacetIter, Operator};
pub use self::query_tree::MatchingWords;
2021-04-14 01:10:58 +08:00
use self::query_tree::QueryTreeBuilder;
// Building these factories is not free.
static LEVDIST0: Lazy<LevBuilder> = Lazy::new(|| LevBuilder::new(0, true));
static LEVDIST1: Lazy<LevBuilder> = Lazy::new(|| LevBuilder::new(1, true));
static LEVDIST2: Lazy<LevBuilder> = Lazy::new(|| LevBuilder::new(2, true));
mod criteria;
mod distinct;
mod facet;
mod query_tree;
pub struct Search<'a> {
query: Option<String>,
facet_condition: Option<FacetCondition>,
offset: usize,
limit: usize,
optional_words: bool,
authorize_typos: bool,
2021-04-14 01:10:58 +08:00
words_limit: usize,
rtxn: &'a heed::RoTxn<'a>,
index: &'a Index,
}
impl<'a> Search<'a> {
pub fn new(rtxn: &'a heed::RoTxn, index: &'a Index) -> Search<'a> {
Search {
query: None,
facet_condition: None,
offset: 0,
limit: 20,
optional_words: true,
authorize_typos: true,
2021-04-14 01:10:58 +08:00
words_limit: 10,
rtxn,
index,
}
}
pub fn query(&mut self, query: impl Into<String>) -> &mut Search<'a> {
self.query = Some(query.into());
self
}
pub fn offset(&mut self, offset: usize) -> &mut Search<'a> {
self.offset = offset;
self
}
pub fn limit(&mut self, limit: usize) -> &mut Search<'a> {
self.limit = limit;
self
}
pub fn optional_words(&mut self, value: bool) -> &mut Search<'a> {
self.optional_words = value;
self
}
pub fn authorize_typos(&mut self, value: bool) -> &mut Search<'a> {
self.authorize_typos = value;
self
}
2021-04-14 01:10:58 +08:00
pub fn words_limit(&mut self, value: usize) -> &mut Search<'a> {
self.words_limit = value;
self
}
pub fn facet_condition(&mut self, condition: FacetCondition) -> &mut Search<'a> {
self.facet_condition = Some(condition);
self
}
pub fn execute(&self) -> anyhow::Result<SearchResult> {
// We create the query tree by spliting the query into tokens.
let before = Instant::now();
2021-05-04 19:44:55 +08:00
let (query_tree, primitive_query) = match self.query.as_ref() {
Some(query) => {
let mut builder = QueryTreeBuilder::new(self.rtxn, self.index);
builder.optional_words(self.optional_words);
builder.authorize_typos(self.authorize_typos);
2021-04-14 01:10:58 +08:00
builder.words_limit(self.words_limit);
// We make sure that the analyzer is aware of the stop words
// this ensures that the query builder is able to properly remove them.
let mut config = AnalyzerConfig::default();
let stop_words = self.index.stop_words(self.rtxn)?;
if let Some(ref stop_words) = stop_words {
config.stop_words(stop_words);
}
let analyzer = Analyzer::new(config);
let result = analyzer.analyze(query);
let tokens = result.tokens();
2021-05-04 19:44:55 +08:00
builder.build(tokens)?.map_or((None, None), |(qt, pq)| (Some(qt), Some(pq)))
},
2021-05-04 19:44:55 +08:00
None => (None, None),
};
debug!("query tree: {:?} took {:.02?}", query_tree, before.elapsed());
// We create the original candidates with the facet conditions results.
let before = Instant::now();
let facet_candidates = match &self.facet_condition {
Some(condition) => Some(condition.evaluate(self.rtxn, self.index)?),
None => None,
};
debug!("facet candidates: {:?} took {:.02?}", facet_candidates, before.elapsed());
let matching_words = match query_tree.as_ref() {
Some(query_tree) => MatchingWords::from_query_tree(&query_tree),
None => MatchingWords::default(),
};
2021-03-02 18:58:32 +08:00
let criteria_builder = criteria::CriteriaBuilder::new(self.rtxn, self.index)?;
2021-05-04 19:44:55 +08:00
let criteria = criteria_builder.build(query_tree, primitive_query, facet_candidates)?;
match self.index.distinct_attribute(self.rtxn)? {
None => self.perform_sort(NoopDistinct, matching_words, criteria),
Some(name) => {
let field_ids_map = self.index.fields_ids_map(self.rtxn)?;
let id = field_ids_map.id(name).expect("distinct not present in field map");
let faceted_fields = self.index.faceted_fields(self.rtxn)?;
if faceted_fields.contains(name) {
let distinct = FacetDistinct::new(id, self.index, self.rtxn);
self.perform_sort(distinct, matching_words, criteria)
} else {
let distinct = MapDistinct::new(id, self.index, self.rtxn);
self.perform_sort(distinct, matching_words, criteria)
}
}
}
}
fn perform_sort(
&self,
mut distinct: impl for<'c> Distinct<'c>,
matching_words: MatchingWords,
2021-03-23 22:25:46 +08:00
mut criteria: Final,
) -> anyhow::Result<SearchResult> {
let mut offset = self.offset;
let mut initial_candidates = RoaringBitmap::new();
let mut excluded_candidates = RoaringBitmap::new();
let mut documents_ids = Vec::with_capacity(self.limit);
while let Some(FinalResult { candidates, bucket_candidates, .. }) = criteria.next(&excluded_candidates)? {
2021-02-24 22:37:37 +08:00
debug!("Number of candidates found {}", candidates.len());
let excluded = take(&mut excluded_candidates);
let mut candidates = distinct.distinct(candidates, excluded);
2021-02-25 23:14:38 +08:00
initial_candidates.union_with(&bucket_candidates);
if offset != 0 {
let discarded = candidates.by_ref().take(offset).count();
offset = offset.saturating_sub(discarded);
}
for candidate in candidates.by_ref().take(self.limit - documents_ids.len()) {
documents_ids.push(candidate?);
}
if documents_ids.len() == self.limit { break }
excluded_candidates = candidates.into_excluded();
}
Ok(SearchResult { matching_words, candidates: initial_candidates, documents_ids })
}
}
impl fmt::Debug for Search<'_> {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
let Search {
query,
facet_condition,
offset,
limit,
optional_words,
authorize_typos,
2021-04-14 01:10:58 +08:00
words_limit,
rtxn: _,
index: _,
} = self;
f.debug_struct("Search")
.field("query", query)
.field("facet_condition", facet_condition)
.field("offset", offset)
.field("limit", limit)
.field("optional_words", optional_words)
.field("authorize_typos", authorize_typos)
2021-04-14 01:10:58 +08:00
.field("words_limit", words_limit)
.finish()
}
}
#[derive(Default)]
pub struct SearchResult {
pub matching_words: MatchingWords,
pub candidates: RoaringBitmap,
// TODO those documents ids should be associated with their criteria scores.
pub documents_ids: Vec<DocumentId>,
}
pub type WordDerivationsCache = HashMap<(String, bool, u8), Vec<(String, u8)>>;
pub fn word_derivations<'c>(
word: &str,
is_prefix: bool,
max_typo: u8,
fst: &fst::Set<Cow<[u8]>>,
cache: &'c mut WordDerivationsCache,
2021-04-14 01:10:58 +08:00
) -> Result<&'c [(String, u8)], Utf8Error> {
match cache.entry((word.to_string(), is_prefix, max_typo)) {
Entry::Occupied(entry) => Ok(entry.into_mut()),
Entry::Vacant(entry) => {
let mut derived_words = Vec::new();
let dfa = build_dfa(word, max_typo, is_prefix);
let mut stream = fst.search_with_state(&dfa).into_stream();
while let Some((word, state)) = stream.next() {
let word = std::str::from_utf8(word)?;
let distance = dfa.distance(state);
derived_words.push((word.to_string(), distance.to_u8()));
}
Ok(entry.insert(derived_words))
},
}
}
pub fn build_dfa(word: &str, typos: u8, is_prefix: bool) -> DFA {
let lev = match typos {
0 => &LEVDIST0,
1 => &LEVDIST1,
_ => &LEVDIST2,
};
if is_prefix {
lev.build_prefix_dfa(word)
} else {
lev.build_dfa(word)
}
}