Merge 40dd25d6b2 into 94fb55bb6f

crates/index-scheduler/src/batch.rs

@@ -39,7 +39,7 @@ use meilisearch_types::milli::update::{IndexDocumentsMethod, Settings as MilliSe
 use meilisearch_types::milli::vector::parsed_vectors::{
     ExplicitVectors, VectorOrArrayOfVectors, RESERVED_VECTORS_FIELD_NAME,
 };
-use meilisearch_types::milli::{self, Filter, ThreadPoolNoAbort, ThreadPoolNoAbortBuilder};
+use meilisearch_types::milli::{self, Filter};
 use meilisearch_types::settings::{apply_settings_to_builder, Settings, Unchecked};
 use meilisearch_types::tasks::{Details, IndexSwap, Kind, KindWithContent, Status, Task};
 use meilisearch_types::{compression, Index, VERSION_FILE_NAME};
@@ -1277,6 +1277,7 @@ impl IndexScheduler {
                 operations,
                 mut tasks,
             } => {
+                let indexer_config = self.index_mapper.indexer_config();
                 // TODO: at some point, for better efficiency we might want to reuse the bumpalo for successive batches.
                 // this is made difficult by the fact we're doing private clones of the index scheduler and sending it
                 // to a fresh thread.
@@ -1385,16 +1386,10 @@ impl IndexScheduler {
                 }
 
                 if tasks.iter().any(|res| res.error.is_none()) {
-                    let local_pool;
+                    /// TODO create a pool if needed
-                    let pool = match &self.index_mapper.indexer_config().thread_pool {
+                    // let pool = indexer_config.thread_pool.unwrap();
-                        Some(pool) => pool,
+                    let pool = rayon::ThreadPoolBuilder::new().build().unwrap();
-                        None => {
-                            local_pool = ThreadPoolNoAbortBuilder::new().build().unwrap();
-                            &local_pool
-                        }
-                    };
 
-                    // TODO we want to multithread this
                     let document_changes = indexer.into_changes(
                         &indexer_alloc,
                         index,
@@ -1403,20 +1398,18 @@ impl IndexScheduler {
                         &mut new_fields_ids_map,
                     )?;
 
-                    pool.install(|| {
                     indexer::index(
                         index_wtxn,
                         index,
                         &db_fields_ids_map,
                         new_fields_ids_map,
                         primary_key_has_been_set.then_some(primary_key),
+                        &pool,
                         &document_changes,
                         embedders,
                         &|| must_stop_processing.get(),
                         &send_progress,
-                        )
+                    )?;
-                    })
-                    .unwrap()?;
 
                     // tracing::info!(indexing_result = ?addition, processed_in = ?started_processing_at.elapsed(), "document indexing done");
                 }
@@ -1496,18 +1489,11 @@ impl IndexScheduler {
                 let result_count = Ok((candidates.len(), candidates.len())) as Result<_>;
 
                 if task.error.is_none() {
-                    let local_pool;
+                    /// TODO create a pool if needed
-                    let pool = match &self.index_mapper.indexer_config().thread_pool {
+                    // let pool = indexer_config.thread_pool.unwrap();
-                        Some(pool) => pool,
+                    let pool = rayon::ThreadPoolBuilder::new().build().unwrap();
-                        None => {
-                            local_pool = ThreadPoolNoAbortBuilder::new().build().unwrap();
-                            &local_pool
-                        }
-                    };
 
-                    pool.install(|| {
+                    let indexer = UpdateByFunction::new(candidates, context.clone(), code.clone());
-                        let indexer =
-                            UpdateByFunction::new(candidates, context.clone(), code.clone());
                     let document_changes = indexer.into_changes(&primary_key)?;
                     let embedders = index.embedding_configs(index_wtxn)?;
                     let embedders = self.embedders(embedders)?;
@@ -1518,16 +1504,13 @@ impl IndexScheduler {
                         &db_fields_ids_map,
                         new_fields_ids_map,
                         None, // cannot change primary key in DocumentEdition
+                        &pool,
                         &document_changes,
                         embedders,
                         &|| must_stop_processing.get(),
                         &send_progress,
                     )?;
 
-                        Result::Ok(())
-                    })
-                    .unwrap()?;
-
                     // tracing::info!(indexing_result = ?addition, processed_in = ?started_processing_at.elapsed(), "document indexing done");
                 }
 
@@ -1646,14 +1629,9 @@ impl IndexScheduler {
                     .map_err(milli::Error::from)?;
 
                 if !tasks.iter().all(|res| res.error.is_some()) {
-                    let local_pool;
+                    /// TODO create a pool if needed
-                    let pool = match &self.index_mapper.indexer_config().thread_pool {
+                    // let pool = indexer_config.thread_pool.unwrap();
-                        Some(pool) => pool,
+                    let pool = rayon::ThreadPoolBuilder::new().build().unwrap();
-                        None => {
-                            local_pool = ThreadPoolNoAbortBuilder::new().build().unwrap();
-                            &local_pool
-                        }
-                    };
 
                     let mut indexer = indexer::DocumentDeletion::new();
                     indexer.delete_documents_by_docids(to_delete);
@@ -1661,20 +1639,18 @@ impl IndexScheduler {
                     let embedders = index.embedding_configs(index_wtxn)?;
                     let embedders = self.embedders(embedders)?;
 
-                    pool.install(|| {
                     indexer::index(
                         index_wtxn,
                         index,
                         &db_fields_ids_map,
                         new_fields_ids_map,
                         None, // document deletion never changes primary key
+                        &pool,
                         &document_changes,
                         embedders,
                         &|| must_stop_processing.get(),
                         &send_progress,
-                        )
+                    )?;
-                    })
-                    .unwrap()?;
 
                     // tracing::info!(indexing_result = ?addition, processed_in = ?started_processing_at.elapsed(), "document indexing done");
                 }

crates/milli/src/search/new/matches/mod.rs

@@ -3,9 +3,6 @@ mod r#match;
 mod matching_words;
 mod simple_token_kind;
 
-use std::borrow::Cow;
-use std::cmp::{max, min};
-
 use charabia::{Language, SeparatorKind, Token, Tokenizer};
 use either::Either;
 pub use matching_words::MatchingWords;
@@ -13,6 +10,10 @@ use matching_words::{MatchType, PartialMatch};
 use r#match::{Match, MatchPosition};
 use serde::Serialize;
 use simple_token_kind::SimpleTokenKind;
+use std::{
+    borrow::Cow,
+    cmp::{max, min},
+};
 
 const DEFAULT_CROP_MARKER: &str = "…";
 const DEFAULT_HIGHLIGHT_PREFIX: &str = "<em>";

crates/milli/src/update/new/channel.rs

@@ -366,14 +366,14 @@ pub struct FieldIdDocidFacetSender<'a>(&'a ExtractorSender);
 impl FieldIdDocidFacetSender<'_> {
     pub fn write_facet_string(&self, key: &[u8], value: &[u8]) -> StdResult<(), SendError<()>> {
         debug_assert!(FieldDocIdFacetStringCodec::bytes_decode(key).is_ok());
-        let entry = EntryOperation::Write(KeyValueEntry::from_small_key_value(key, value));
+        let entry = EntryOperation::Write(KeyValueEntry::from_small_key_value(&key, &value));
         self.0
             .send_db_operation(DbOperation { database: Database::FieldIdDocidFacetStrings, entry })
     }
 
     pub fn write_facet_f64(&self, key: &[u8]) -> StdResult<(), SendError<()>> {
         debug_assert!(FieldDocIdFacetF64Codec::bytes_decode(key).is_ok());
-        let entry = EntryOperation::Write(KeyValueEntry::from_small_key_value(key, &[]));
+        let entry = EntryOperation::Write(KeyValueEntry::from_small_key_value(&key, &[]));
         self.0.send_db_operation(DbOperation { database: Database::FieldIdDocidFacetF64s, entry })
     }
 

crates/milli/src/update/new/extract/documents.rs

@@ -58,8 +58,7 @@ impl<'a, 'extractor> Extractor<'extractor> for DocumentsExtractor<'a> {
                         context.index,
                         &context.db_fields_ids_map,
                     )?;
-                    let geo_iter =
+                    let geo_iter = content.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
-                        content.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
                     for res in content.iter_top_level_fields().chain(geo_iter) {
                         let (f, _) = res?;
                         let entry = document_extractor_data
@@ -75,8 +74,7 @@ impl<'a, 'extractor> Extractor<'extractor> for DocumentsExtractor<'a> {
                     let docid = update.docid();
                     let content =
                         update.current(&context.rtxn, context.index, &context.db_fields_ids_map)?;
-                    let geo_iter =
+                        let geo_iter = content.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
-                        content.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
                     for res in content.iter_top_level_fields().chain(geo_iter) {
                         let (f, _) = res?;
                         let entry = document_extractor_data
@@ -86,8 +84,7 @@ impl<'a, 'extractor> Extractor<'extractor> for DocumentsExtractor<'a> {
                         *entry -= 1;
                     }
                     let content = update.updated();
-                    let geo_iter =
+                    let geo_iter = content.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
-                        content.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
                     for res in content.iter_top_level_fields().chain(geo_iter) {
                         let (f, _) = res?;
                         let entry = document_extractor_data
@@ -117,8 +114,7 @@ impl<'a, 'extractor> Extractor<'extractor> for DocumentsExtractor<'a> {
                 DocumentChange::Insertion(insertion) => {
                     let docid = insertion.docid();
                     let content = insertion.inserted();
-                    let geo_iter =
+                    let geo_iter = content.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
-                        content.geo_field().transpose().map(|res| res.map(|rv| ("_geo", rv)));
                     for res in content.iter_top_level_fields().chain(geo_iter) {
                         let (f, _) = res?;
                         let entry = document_extractor_data

crates/milli/src/update/new/extract/searchable/tokenize_document.rs

@@ -3,6 +3,7 @@ use std::collections::HashMap;
 use charabia::{SeparatorKind, Token, TokenKind, Tokenizer, TokenizerBuilder};
 use serde_json::Value;
 
+use crate::proximity::MAX_DISTANCE;
 use crate::update::new::document::Document;
 use crate::update::new::extract::perm_json_p::{
     seek_leaf_values_in_array, seek_leaf_values_in_object, select_field,
@@ -12,9 +13,6 @@ use crate::{
     MAX_WORD_LENGTH,
 };
 
-// todo: should be crate::proximity::MAX_DISTANCE but it has been forgotten
-const MAX_DISTANCE: u32 = 8;
-
 pub struct DocumentTokenizer<'a> {
     pub tokenizer: &'a Tokenizer<'a>,
     pub attribute_to_extract: Option<&'a [&'a str]>,
@@ -253,22 +251,22 @@ mod test {
             ]: "doggo",
             [
                 2,
-                8,
+                MAX_DISTANCE,
             ]: "doggo",
             [
                 2,
                 16,
             ]: "catto",
             [
-                5,
+                3,
                 0,
             ]: "10",
             [
-                6,
+                4,
                 0,
             ]: "pesti",
             [
-                7,
+                5,
                 0,
             ]: "23",
         }

crates/milli/src/update/new/facet_search_builder.rs

@@ -6,9 +6,11 @@ use grenad::Sorter;
 use heed::types::{Bytes, SerdeJson};
 use heed::{BytesDecode, BytesEncode, RoTxn, RwTxn};
 
+use super::extract::FacetKind;
 use super::fst_merger_builder::FstMergerBuilder;
 use super::KvReaderDelAdd;
-use crate::heed_codec::facet::FacetGroupKey;
+use crate::heed_codec::facet::{FacetGroupKey, FacetGroupKeyCodec};
+use crate::heed_codec::StrRefCodec;
 use crate::update::del_add::{DelAdd, KvWriterDelAdd};
 use crate::update::{create_sorter, MergeDeladdBtreesetString};
 use crate::{

crates/milli/src/update/new/fst_merger_builder.rs

@@ -1,12 +1,10 @@
-use std::fs::File;
+use std::{fs::File, io::BufWriter};
-use std::io::BufWriter;
 
 use fst::{Set, SetBuilder, Streamer};
 use memmap2::Mmap;
 use tempfile::tempfile;
 
-use crate::update::del_add::DelAdd;
+use crate::{update::del_add::DelAdd, InternalError, Result};
-use crate::{InternalError, Result};
 
 pub struct FstMergerBuilder<'a> {
     stream: Option<fst::set::Stream<'a>>,

crates/milli/src/update/new/indexer/document_deletion.rs

@@ -132,7 +132,7 @@ mod test {
         }
 
         let mut deletions = DocumentDeletion::new();
-        deletions.delete_documents_by_docids(Vec::<u32>::new().into_iter().collect());
+        deletions.delete_documents_by_docids(vec![0, 2, 42].into_iter().collect());
         let indexer = Bump::new();
 
         let index = TempIndex::new();

crates/milli/src/update/new/indexer/document_operation.rs

@@ -356,11 +356,11 @@ impl MergeChanges for MergeDocumentForUpdates {
         let has_deletion = last_deletion.is_some();
 
         if operations.is_empty() {
-            return if is_new {
+            return if !is_new {
-                Ok(None)
-            } else {
                 let deletion = Deletion::create(docid, external_docid);
                 Ok(Some(DocumentChange::Deletion(deletion)))
+            } else {
+                Ok(None)
             };
         }
 

crates/milli/src/update/new/indexer/mod.rs

@@ -13,6 +13,7 @@ use itertools::{merge_join_by, EitherOrBoth};
 pub use partial_dump::PartialDump;
 use rand::SeedableRng as _;
 use raw_collections::RawMap;
+use rayon::ThreadPool;
 use time::OffsetDateTime;
 pub use update_by_function::UpdateByFunction;
 
@@ -135,6 +136,7 @@ pub fn index<'pl, 'indexer, 'index, DC, MSP, SP>(
     db_fields_ids_map: &'indexer FieldsIdsMap,
     new_fields_ids_map: FieldsIdsMap,
     new_primary_key: Option<PrimaryKey<'pl>>,
+    pool: &ThreadPool,
     document_changes: &DC,
     embedders: EmbeddingConfigs,
     must_stop_processing: &'indexer MSP,
@@ -150,9 +152,9 @@ where
     let metadata_builder = MetadataBuilder::from_index(index, wtxn)?;
     let new_fields_ids_map = FieldIdMapWithMetadata::new(new_fields_ids_map, metadata_builder);
     let new_fields_ids_map = RwLock::new(new_fields_ids_map);
-    let fields_ids_map_store = ThreadLocal::with_capacity(rayon::current_num_threads());
+    let fields_ids_map_store = ThreadLocal::with_capacity(pool.current_num_threads());
-    let mut extractor_allocs = ThreadLocal::with_capacity(rayon::current_num_threads());
+    let mut extractor_allocs = ThreadLocal::with_capacity(pool.current_num_threads());
-    let doc_allocs = ThreadLocal::with_capacity(rayon::current_num_threads());
+    let doc_allocs = ThreadLocal::with_capacity(pool.current_num_threads());
 
     let indexing_context = IndexingContext {
         index,
@@ -177,6 +179,7 @@ where
         let document_ids = &mut document_ids;
         // TODO manage the errors correctly
         let extractor_handle = Builder::new().name(S("indexer-extractors")).spawn_scoped(s, move || {
+            let result = pool.in_place_scope(|_s| {
                 let span = tracing::trace_span!(target: "indexing::documents", parent: &indexer_span, "extract");
                 let _entered = span.enter();
 
@@ -185,17 +188,9 @@ where
                 // document but we need to create a function that collects and compresses documents.
                 let document_sender = extractor_sender.documents();
                 let document_extractor = DocumentsExtractor::new(&document_sender, embedders);
-            let datastore = ThreadLocal::with_capacity(rayon::current_num_threads());
+                let datastore = ThreadLocal::with_capacity(pool.current_num_threads());
                 let (finished_steps, step_name) = steps::extract_documents();
-            extract(document_changes,
+                extract(document_changes, &document_extractor, indexing_context, &mut extractor_allocs, &datastore, finished_steps, total_steps, step_name)?;
-                &document_extractor,
-                indexing_context,
-                &mut extractor_allocs,
-                &datastore,
-                finished_steps,
-                total_steps,
-                step_name,
-            )?;
 
                 for document_extractor_data in datastore {
                     let document_extractor_data = document_extractor_data.0.into_inner();
@@ -228,15 +223,7 @@ where
                     let (finished_steps, step_name) = steps::extract_facets();
 
                     facet_field_ids_delta = merge_and_send_facet_docids(
-                    FacetedDocidsExtractor::run_extraction(grenad_parameters,
+                        FacetedDocidsExtractor::run_extraction(grenad_parameters, document_changes, indexing_context, &mut extractor_allocs, &extractor_sender.field_id_docid_facet_sender(), finished_steps, total_steps, step_name)?,
-                        document_changes,
-                        indexing_context,
-                        &mut extractor_allocs,
-                        &extractor_sender.field_id_docid_facet_sender(),
-                        finished_steps,
-                        total_steps,
-                        step_name,
-                    )?,
                         FacetDatabases::new(index),
                         index,
                         extractor_sender.facet_docids(),
@@ -373,7 +360,7 @@ where
 
                     let embedding_sender = extractor_sender.embeddings();
                     let extractor = EmbeddingExtractor::new(embedders, &embedding_sender, field_distribution, request_threads());
-                let mut datastore = ThreadLocal::with_capacity(rayon::current_num_threads());
+                    let mut datastore = ThreadLocal::with_capacity(pool.current_num_threads());
                     let (finished_steps, step_name) = steps::extract_embeddings();
                     extract(document_changes, &extractor, indexing_context, &mut extractor_allocs, &datastore, finished_steps, total_steps, step_name)?;
 
@@ -396,7 +383,7 @@ where
                     let Some(extractor) = GeoExtractor::new(&rtxn, index, grenad_parameters)? else {
                         break 'geo;
                     };
-                let datastore = ThreadLocal::with_capacity(rayon::current_num_threads());
+                    let datastore = ThreadLocal::with_capacity(pool.current_num_threads());
                     let (finished_steps, step_name) = steps::extract_geo_points();
                     extract(document_changes,
                         &extractor,
@@ -432,6 +419,9 @@ where
                 // - [x] Extract fieldid facet number docids
                 // - [x] Extract fieldid facet string docids
 
+                Result::Ok(facet_field_ids_delta)
+            });
+
             {
                 let span = tracing::trace_span!(target: "indexing::documents::extract", "FINISH");
                 let _entered = span.enter();
@@ -439,7 +429,7 @@ where
                 (indexing_context.send_progress)(Progress { finished_steps, total_steps, step_name, finished_total_documents: None });
             }
 
-            Result::Ok(facet_field_ids_delta)
+            result
         })?;
 
         let global_fields_ids_map = GlobalFieldsIdsMap::new(&new_fields_ids_map);

crates/milli/src/update/new/merger.rs

@@ -1,4 +1,5 @@
 use std::cell::RefCell;
+use std::io;
 
 use hashbrown::HashSet;
 use heed::types::Bytes;

crates/milli/src/update/new/vector_document.rs

@@ -286,7 +286,7 @@ impl<'doc> MergedVectorDocument<'doc> {
     ) -> Result<Option<Self>> {
         let db = VectorDocumentFromDb::new(docid, index, rtxn, db_fields_ids_map, doc_alloc)?;
         let new_doc =
-            VectorDocumentFromVersions::new(external_document_id, versions, doc_alloc, embedders)?;
+            VectorDocumentFromVersions::new(&external_document_id, versions, doc_alloc, embedders)?;
         Ok(if db.is_none() && new_doc.is_none() { None } else { Some(Self { new_doc, db }) })
     }
 

crates/milli/src/update/new/word_fst_builder.rs

@@ -1,14 +1,13 @@
-use std::collections::HashSet;
 use std::io::BufWriter;
 
 use fst::{Set, SetBuilder, Streamer};
 use memmap2::Mmap;
+use std::collections::HashSet;
 use tempfile::tempfile;
 
+use crate::{index::PrefixSettings, update::del_add::DelAdd, InternalError, Prefix, Result};
+
 use super::fst_merger_builder::FstMergerBuilder;
-use crate::index::PrefixSettings;
-use crate::update::del_add::DelAdd;
-use crate::{InternalError, Prefix, Result};
 
 pub struct WordFstBuilder<'a> {
     word_fst_builder: FstMergerBuilder<'a>,