Only run word pair proximity docids extraction if proximity_precision enables it

index-scheduler/src/batch.rs

@@ -33,13 +33,11 @@ use meilisearch_types::milli::heed::CompactionOption;
 use meilisearch_types::milli::update::new::indexer::{
     self, retrieve_or_guess_primary_key, UpdateByFunction,
 };
-use meilisearch_types::milli::update::{
+use meilisearch_types::milli::update::{IndexDocumentsMethod, Settings as MilliSettings};
-    IndexDocumentsConfig, IndexDocumentsMethod, IndexerConfig, Settings as MilliSettings,
-};
 use meilisearch_types::milli::vector::parsed_vectors::{
     ExplicitVectors, VectorOrArrayOfVectors, RESERVED_VECTORS_FIELD_NAME,
 };
-use meilisearch_types::milli::{self, Filter, Object};
+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};
@@ -50,7 +48,7 @@ use uuid::Uuid;
 
 use crate::autobatcher::{self, BatchKind};
 use crate::utils::{self, swap_index_uid_in_task};
-use crate::{Error, IndexScheduler, MustStopProcessing, ProcessingTasks, Result, TaskId};
+use crate::{Error, IndexScheduler, ProcessingTasks, Result, TaskId};
 
 /// Represents a combination of tasks that can all be processed at the same time.
 ///

meili-snap/Cargo.toml

@@ -11,6 +11,6 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-insta = { version = "^1.39.0", features = ["json", "redactions"] }
+insta = { version = "=1.39.0", features = ["json", "redactions"] }
 md5 = "0.7.0"
 once_cell = "1.19"

milli/Cargo.toml

@@ -94,10 +94,12 @@ tracing = "0.1.40"
 ureq = { version = "2.10.0", features = ["json"] }
 url = "2.5.2"
 rayon-par-bridge = "0.1.0"
-hashbrown = "0.14.5"
+hashbrown = "0.15.0"
 raw-collections = { git = "https://github.com/dureuill/raw-collections.git", version = "0.1.0" }
 bumpalo = "3.16.0"
 thread_local = "1.1.8"
+allocator-api2 = "0.2.18"
+rustc-hash = "2.0.0"
 
 [dev-dependencies]
 mimalloc = { version = "0.1.43", default-features = false }

milli/src/update/index_documents/helpers/grenad_helpers.rs

@@ -83,6 +83,8 @@ pub fn writer_into_reader(
     grenad::Reader::new(BufReader::new(file)).map_err(Into::into)
 }
 
+/// # Safety
+/// We use memory mapping inside. So, according to the Rust community, it's unsafe.
 pub unsafe fn as_cloneable_grenad(
     reader: &grenad::Reader<BufReader<File>>,
 ) -> Result<grenad::Reader<CursorClonableMmap>> {

milli/src/update/new/channel.rs

@@ -1,42 +1,33 @@
-use std::fs::File;
 use std::marker::PhantomData;
+use std::sync::atomic::{AtomicUsize, Ordering};
 
 use crossbeam_channel::{IntoIter, Receiver, SendError, Sender};
-use grenad::Merger;
 use heed::types::Bytes;
-use memmap2::Mmap;
 use roaring::RoaringBitmap;
 
 use super::extract::FacetKind;
 use super::StdResult;
-use crate::index::main_key::{DOCUMENTS_IDS_KEY, WORDS_FST_KEY, WORDS_PREFIXES_FST_KEY};
+use crate::index::main_key::DOCUMENTS_IDS_KEY;
 use crate::update::new::KvReaderFieldId;
-use crate::update::MergeDeladdCboRoaringBitmaps;
 use crate::{DocumentId, Index};
 
 /// The capacity of the channel is currently in number of messages.
-pub fn merger_writer_channel(cap: usize) -> (MergerSender, WriterReceiver) {
+pub fn extractor_writer_channel(cap: usize) -> (ExtractorSender, WriterReceiver) {
     let (sender, receiver) = crossbeam_channel::bounded(cap);
     (
-        MergerSender {
+        ExtractorSender {
             sender,
             send_count: Default::default(),
             writer_contentious_count: Default::default(),
-            merger_contentious_count: Default::default(),
+            extractor_contentious_count: Default::default(),
         },
         WriterReceiver(receiver),
     )
 }
 
-/// The capacity of the channel is currently in number of messages.
+pub struct KeyValueEntry {
-pub fn extractors_merger_channels(cap: usize) -> (ExtractorSender, MergerReceiver) {
+    pub key_length: usize,
-    let (sender, receiver) = crossbeam_channel::bounded(cap);
+    pub data: Box<[u8]>,
-    (ExtractorSender(sender), MergerReceiver(receiver))
-}
-
-pub enum KeyValueEntry {
-    SmallInMemory { key_length: usize, data: Box<[u8]> },
-    LargeOnDisk { key: Box<[u8]>, value: Mmap },
 }
 
 impl KeyValueEntry {
@@ -44,32 +35,22 @@ impl KeyValueEntry {
         let mut data = Vec::with_capacity(key.len() + value.len());
         data.extend_from_slice(key);
         data.extend_from_slice(value);
-        KeyValueEntry::SmallInMemory { key_length: key.len(), data: data.into_boxed_slice() }
+        KeyValueEntry { key_length: key.len(), data: data.into_boxed_slice() }
     }
 
     pub fn from_small_key_bitmap(key: &[u8], bitmap: RoaringBitmap) -> Self {
         let mut data = Vec::with_capacity(key.len() + bitmap.serialized_size());
         data.extend_from_slice(key);
         bitmap.serialize_into(&mut data).unwrap();
-        KeyValueEntry::SmallInMemory { key_length: key.len(), data: data.into_boxed_slice() }
+        KeyValueEntry { key_length: key.len(), data: data.into_boxed_slice() }
-    }
-
-    pub fn from_large_key_value(key: &[u8], value: Mmap) -> Self {
-        KeyValueEntry::LargeOnDisk { key: key.to_vec().into_boxed_slice(), value }
     }
 
     pub fn key(&self) -> &[u8] {
-        match self {
+        &self.data[..self.key_length]
-            KeyValueEntry::SmallInMemory { key_length, data } => &data.as_ref()[..*key_length],
-            KeyValueEntry::LargeOnDisk { key, value: _ } => key.as_ref(),
-        }
     }
 
     pub fn value(&self) -> &[u8] {
-        match self {
+        &self.data[self.key_length..]
-            KeyValueEntry::SmallInMemory { key_length, data } => &data.as_ref()[*key_length..],
-            KeyValueEntry::LargeOnDisk { key: _, value } => value.as_ref(),
-        }
     }
 }
 
@@ -92,37 +73,6 @@ pub enum EntryOperation {
     Write(KeyValueEntry),
 }
 
-pub struct DocumentEntry {
-    docid: DocumentId,
-    content: Box<[u8]>,
-}
-
-impl DocumentEntry {
-    pub fn new_uncompressed(docid: DocumentId, content: Box<KvReaderFieldId>) -> Self {
-        DocumentEntry { docid, content: content.into() }
-    }
-
-    pub fn new_compressed(docid: DocumentId, content: Box<[u8]>) -> Self {
-        DocumentEntry { docid, content }
-    }
-
-    pub fn key(&self) -> [u8; 4] {
-        self.docid.to_be_bytes()
-    }
-
-    pub fn content(&self) -> &[u8] {
-        &self.content
-    }
-}
-
-pub struct DocumentDeletionEntry(DocumentId);
-
-impl DocumentDeletionEntry {
-    pub fn key(&self) -> [u8; 4] {
-        self.0.to_be_bytes()
-    }
-}
-
 pub struct WriterOperation {
     database: Database,
     entry: EntryOperation,
@@ -206,34 +156,32 @@ impl IntoIterator for WriterReceiver {
     }
 }
 
-pub struct MergerSender {
+pub struct ExtractorSender {
     sender: Sender<WriterOperation>,
-    /// The number of message we send in total in the channel.
+    /// The number of message we sent in total in the channel.
-    send_count: std::cell::Cell<usize>,
+    send_count: AtomicUsize,
     /// The number of times we sent something in a channel that was full.
-    writer_contentious_count: std::cell::Cell<usize>,
+    writer_contentious_count: AtomicUsize,
     /// The number of times we sent something in a channel that was empty.
-    merger_contentious_count: std::cell::Cell<usize>,
+    extractor_contentious_count: AtomicUsize,
 }
 
-impl Drop for MergerSender {
+impl Drop for ExtractorSender {
     fn drop(&mut self) {
+        let send_count = *self.send_count.get_mut();
+        let writer_contentious_count = *self.writer_contentious_count.get_mut();
+        let extractor_contentious_count = *self.extractor_contentious_count.get_mut();
         eprintln!(
-            "Merger channel stats: {} sends, {} writer contentions ({}%), {} merger contentions ({}%)",
+            "Extractor channel stats: {send_count} sends, \
-            self.send_count.get(),
+            {writer_contentious_count} writer contentions ({}%), \
-            self.writer_contentious_count.get(),
+            {extractor_contentious_count} extractor contentions ({}%)",
-            (self.writer_contentious_count.get() as f32 / self.send_count.get() as f32) * 100.0,
+            (writer_contentious_count as f32 / send_count as f32) * 100.0,
-            self.merger_contentious_count.get(),
+            (extractor_contentious_count as f32 / send_count as f32) * 100.0
-            (self.merger_contentious_count.get() as f32 / self.send_count.get() as f32) * 100.0
         )
     }
 }
 
-impl MergerSender {
+impl ExtractorSender {
-    pub fn main(&self) -> MainSender<'_> {
-        MainSender(self)
-    }
-
     pub fn docids<D: DatabaseType>(&self) -> WordDocidsSender<'_, D> {
         WordDocidsSender { sender: self, _marker: PhantomData }
     }
@@ -263,12 +211,12 @@ impl MergerSender {
 
     fn send(&self, op: WriterOperation) -> StdResult<(), SendError<()>> {
         if self.sender.is_full() {
-            self.writer_contentious_count.set(self.writer_contentious_count.get() + 1);
+            self.writer_contentious_count.fetch_add(1, Ordering::SeqCst);
         }
         if self.sender.is_empty() {
-            self.merger_contentious_count.set(self.merger_contentious_count.get() + 1);
+            self.extractor_contentious_count.fetch_add(1, Ordering::SeqCst);
         }
-        self.send_count.set(self.send_count.get() + 1);
+        self.send_count.fetch_add(1, Ordering::SeqCst);
         match self.sender.send(op) {
             Ok(()) => Ok(()),
             Err(SendError(_)) => Err(SendError(())),
@@ -276,129 +224,48 @@ impl MergerSender {
     }
 }
 
-pub struct MainSender<'a>(&'a MergerSender);
-
-impl MainSender<'_> {
-    pub fn write_words_fst(&self, value: Mmap) -> StdResult<(), SendError<()>> {
-        let entry = EntryOperation::Write(KeyValueEntry::from_large_key_value(
-            WORDS_FST_KEY.as_bytes(),
-            value,
-        ));
-        match self.0.send(WriterOperation { database: Database::Main, entry }) {
-            Ok(()) => Ok(()),
-            Err(SendError(_)) => Err(SendError(())),
-        }
-    }
-
-    pub fn write_words_prefixes_fst(&self, value: Mmap) -> StdResult<(), SendError<()>> {
-        let entry = EntryOperation::Write(KeyValueEntry::from_large_key_value(
-            WORDS_PREFIXES_FST_KEY.as_bytes(),
-            value,
-        ));
-        match self.0.send(WriterOperation { database: Database::Main, entry }) {
-            Ok(()) => Ok(()),
-            Err(SendError(_)) => Err(SendError(())),
-        }
-    }
-
-    pub fn delete(&self, key: &[u8]) -> StdResult<(), SendError<()>> {
-        let entry = EntryOperation::Delete(KeyEntry::from_key(key));
-        match self.0.send(WriterOperation { database: Database::Main, entry }) {
-            Ok(()) => Ok(()),
-            Err(SendError(_)) => Err(SendError(())),
-        }
-    }
-}
-
 pub enum ExactWordDocids {}
 pub enum FidWordCountDocids {}
 pub enum WordDocids {}
 pub enum WordFidDocids {}
 pub enum WordPairProximityDocids {}
 pub enum WordPositionDocids {}
-pub enum FacetDocids {}
 
 pub trait DatabaseType {
     const DATABASE: Database;
 }
 
-pub trait MergerOperationType {
-    fn new_merger_operation(merger: Merger<File, MergeDeladdCboRoaringBitmaps>) -> MergerOperation;
-}
-
 impl DatabaseType for ExactWordDocids {
     const DATABASE: Database = Database::ExactWordDocids;
 }
 
-impl MergerOperationType for ExactWordDocids {
-    fn new_merger_operation(merger: Merger<File, MergeDeladdCboRoaringBitmaps>) -> MergerOperation {
-        MergerOperation::ExactWordDocidsMerger(merger)
-    }
-}
-
 impl DatabaseType for FidWordCountDocids {
     const DATABASE: Database = Database::FidWordCountDocids;
 }
 
-impl MergerOperationType for FidWordCountDocids {
-    fn new_merger_operation(merger: Merger<File, MergeDeladdCboRoaringBitmaps>) -> MergerOperation {
-        MergerOperation::FidWordCountDocidsMerger(merger)
-    }
-}
-
 impl DatabaseType for WordDocids {
     const DATABASE: Database = Database::WordDocids;
 }
 
-impl MergerOperationType for WordDocids {
-    fn new_merger_operation(merger: Merger<File, MergeDeladdCboRoaringBitmaps>) -> MergerOperation {
-        MergerOperation::WordDocidsMerger(merger)
-    }
-}
-
 impl DatabaseType for WordFidDocids {
     const DATABASE: Database = Database::WordFidDocids;
 }
 
-impl MergerOperationType for WordFidDocids {
-    fn new_merger_operation(merger: Merger<File, MergeDeladdCboRoaringBitmaps>) -> MergerOperation {
-        MergerOperation::WordFidDocidsMerger(merger)
-    }
-}
-
 impl DatabaseType for WordPairProximityDocids {
     const DATABASE: Database = Database::WordPairProximityDocids;
 }
 
-impl MergerOperationType for WordPairProximityDocids {
-    fn new_merger_operation(merger: Merger<File, MergeDeladdCboRoaringBitmaps>) -> MergerOperation {
-        MergerOperation::WordPairProximityDocidsMerger(merger)
-    }
-}
-
 impl DatabaseType for WordPositionDocids {
     const DATABASE: Database = Database::WordPositionDocids;
 }
 
-impl MergerOperationType for WordPositionDocids {
-    fn new_merger_operation(merger: Merger<File, MergeDeladdCboRoaringBitmaps>) -> MergerOperation {
-        MergerOperation::WordPositionDocidsMerger(merger)
-    }
-}
-
-impl MergerOperationType for FacetDocids {
-    fn new_merger_operation(merger: Merger<File, MergeDeladdCboRoaringBitmaps>) -> MergerOperation {
-        MergerOperation::FacetDocidsMerger(merger)
-    }
-}
-
 pub trait DocidsSender {
     fn write(&self, key: &[u8], value: &[u8]) -> StdResult<(), SendError<()>>;
     fn delete(&self, key: &[u8]) -> StdResult<(), SendError<()>>;
 }
 
 pub struct WordDocidsSender<'a, D> {
-    sender: &'a MergerSender,
+    sender: &'a ExtractorSender,
     _marker: PhantomData<D>,
 }
 
@@ -421,7 +288,7 @@ impl<D: DatabaseType> DocidsSender for WordDocidsSender<'_, D> {
 }
 
 pub struct FacetDocidsSender<'a> {
-    sender: &'a MergerSender,
+    sender: &'a ExtractorSender,
 }
 
 impl DocidsSender for FacetDocidsSender<'_> {
@@ -456,7 +323,7 @@ impl DocidsSender for FacetDocidsSender<'_> {
 }
 
 pub struct FacetSearchableSender<'a> {
-    sender: &'a MergerSender,
+    sender: &'a ExtractorSender,
 }
 
 impl FacetSearchableSender<'_> {
@@ -481,25 +348,9 @@ impl FacetSearchableSender<'_> {
             Err(SendError(_)) => Err(SendError(())),
         }
     }
-
-    pub fn write_fst(&self, key: &[u8], value: Mmap) -> StdResult<(), SendError<()>> {
-        let entry = EntryOperation::Write(KeyValueEntry::from_large_key_value(key, value));
-        match self.sender.send(WriterOperation { database: Database::FacetIdStringFst, entry }) {
-            Ok(()) => Ok(()),
-            Err(SendError(_)) => Err(SendError(())),
-        }
-    }
-
-    pub fn delete_fst(&self, key: &[u8]) -> StdResult<(), SendError<()>> {
-        let entry = EntryOperation::Delete(KeyEntry::from_key(key));
-        match self.sender.send(WriterOperation { database: Database::FacetIdStringFst, entry }) {
-            Ok(()) => Ok(()),
-            Err(SendError(_)) => Err(SendError(())),
-        }
-    }
 }
 
-pub struct DocumentsSender<'a>(&'a MergerSender);
+pub struct DocumentsSender<'a>(&'a ExtractorSender);
 
 impl DocumentsSender<'_> {
     /// TODO do that efficiently
@@ -542,86 +393,3 @@ impl DocumentsSender<'_> {
         }
     }
 }
-
-pub enum MergerOperation {
-    ExactWordDocidsMerger(Merger<File, MergeDeladdCboRoaringBitmaps>),
-    FidWordCountDocidsMerger(Merger<File, MergeDeladdCboRoaringBitmaps>),
-    WordDocidsMerger(Merger<File, MergeDeladdCboRoaringBitmaps>),
-    WordFidDocidsMerger(Merger<File, MergeDeladdCboRoaringBitmaps>),
-    WordPairProximityDocidsMerger(Merger<File, MergeDeladdCboRoaringBitmaps>),
-    WordPositionDocidsMerger(Merger<File, MergeDeladdCboRoaringBitmaps>),
-    FacetDocidsMerger(Merger<File, MergeDeladdCboRoaringBitmaps>),
-    DeleteDocument { docid: DocumentId, external_id: String },
-    InsertDocument { docid: DocumentId, external_id: String, document: Box<KvReaderFieldId> },
-    FinishedDocument,
-}
-
-pub struct MergerReceiver(Receiver<MergerOperation>);
-
-impl IntoIterator for MergerReceiver {
-    type Item = MergerOperation;
-    type IntoIter = IntoIter<Self::Item>;
-
-    fn into_iter(self) -> Self::IntoIter {
-        self.0.into_iter()
-    }
-}
-
-pub struct ExtractorSender(Sender<MergerOperation>);
-
-impl ExtractorSender {
-    pub fn document_sender(&self) -> DocumentSender<'_> {
-        DocumentSender(Some(&self.0))
-    }
-
-    pub fn send_searchable<D: MergerOperationType>(
-        &self,
-        merger: Merger<File, MergeDeladdCboRoaringBitmaps>,
-    ) -> StdResult<(), SendError<()>> {
-        match self.0.send(D::new_merger_operation(merger)) {
-            Ok(()) => Ok(()),
-            Err(SendError(_)) => Err(SendError(())),
-        }
-    }
-}
-
-pub struct DocumentSender<'a>(Option<&'a Sender<MergerOperation>>);
-
-impl DocumentSender<'_> {
-    pub fn insert(
-        &self,
-        docid: DocumentId,
-        external_id: String,
-        document: Box<KvReaderFieldId>,
-    ) -> StdResult<(), SendError<()>> {
-        let sender = self.0.unwrap();
-        match sender.send(MergerOperation::InsertDocument { docid, external_id, document }) {
-            Ok(()) => Ok(()),
-            Err(SendError(_)) => Err(SendError(())),
-        }
-    }
-
-    pub fn delete(&self, docid: DocumentId, external_id: String) -> StdResult<(), SendError<()>> {
-        let sender = self.0.unwrap();
-        match sender.send(MergerOperation::DeleteDocument { docid, external_id }) {
-            Ok(()) => Ok(()),
-            Err(SendError(_)) => Err(SendError(())),
-        }
-    }
-
-    pub fn finish(mut self) -> StdResult<(), SendError<()>> {
-        let sender = self.0.take().unwrap();
-        match sender.send(MergerOperation::FinishedDocument) {
-            Ok(()) => Ok(()),
-            Err(SendError(_)) => Err(SendError(())),
-        }
-    }
-}
-
-impl Drop for DocumentSender<'_> {
-    fn drop(&mut self) {
-        if let Some(sender) = self.0.take() {
-            let _ = sender.send(MergerOperation::FinishedDocument);
-        }
-    }
-}

milli/src/update/new/extract/cache.rs

@@ -1,299 +1,611 @@
-use std::fmt::Write as _;
+//! # How the Merge Algorithm works
-use std::mem;
+//!
-use std::num::NonZeroUsize;
+//! Each extractor create #Threads caches and balances the entries
+//! based on the hash of the keys. To do that we can use the
+//! hashbrown::hash_map::RawEntryBuilderMut::from_key_hashed_nocheck.
+//! This way we can compute the hash on our own, decide on the cache to
+//! target, and insert it into the right HashMap.
+//!
+//! #Thread  -> caches
+//! t1       -> [t1c1, t1c2, t1c3]
+//! t2       -> [t2c1, t2c2, t2c3]
+//! t3       -> [t3c1, t3c2, t3c3]
+//!
+//! When the extractors are done filling the caches, we want to merge
+//! the content of all the caches. We do a transpose and each thread is
+//! assigned the associated cache. By doing that we know that every key
+//! is put in a known cache and will collide with keys in the other
+//! caches of the other threads.
+//!
+//! #Thread  -> caches
+//! t1       -> [t1c1, t2c1, t3c1]
+//! t2       -> [t1c2, t2c2, t3c2]
+//! t3       -> [t1c3, t2c3, t3c3]
+//!
+//! When we encountered a miss in the other caches we must still try
+//! to find it in the spilled entries. This is the reason why we use
+//! a grenad sorter/reader so that we can seek "efficiently" for a key.
+//!
+//! ## More Detailled Algorithm
+//!
+//! Each sub-cache has an in-memory HashMap and some spilled
+//! lexicographically ordered entries on disk (grenad). We first iterate
+//! over the spilled entries of all the caches at once by using a merge
+//! join algorithm. This algorithm will merge the entries by using its
+//! merge function.
+//!
+//! Everytime a merged entry is emited by the merge join algorithm we also
+//! fetch the value from the other in-memory caches (HashMaps) to finish
+//! the merge. Everytime we retrieve an entry from the in-memory caches
+//! we mark them with a tombstone for later.
+//!
+//! Once we are done with the spilled entries we iterate over the in-memory
+//! HashMaps. We iterate over the first one, retrieve the content from the
+//! other onces and mark them with a tombstone again. We also make sure
+//! to ignore the dead (tombstoned) ones.
+//!
+//! ## Memory Control
+//!
+//! We can detect that there are no more memory available when the
+//! bump allocator reaches a threshold. When this is the case we
+//! freeze the cache. There is one bump allocator by thread and the
+//! memory must be well balanced as we manage one type of extraction
+//! at a time with well-balanced documents.
+//!
+//! It means that the unknown new keys added to the
+//! cache are directly spilled to disk: basically a key followed by a
+//! del/add bitmap. For the known keys we can keep modifying them in
+//! the materialized version in the cache: update the del/add bitmaps.
+//!
+//! For now we can use a grenad sorter for spilling even thought I think
+//! it's not the most efficient way (too many files open, sorting entries).
 
-use grenad::{MergeFunction, Sorter};
+use std::cmp::Ordering;
-use roaring::bitmap::Statistics;
+use std::collections::binary_heap::PeekMut;
+use std::collections::BinaryHeap;
+use std::fs::File;
+use std::hash::BuildHasher;
+use std::io::BufReader;
+use std::{io, iter, mem};
+
+use bumpalo::Bump;
+use grenad::ReaderCursor;
+use hashbrown::hash_map::RawEntryMut;
+use hashbrown::HashMap;
+use raw_collections::map::FrozenMap;
 use roaring::RoaringBitmap;
-use smallvec::SmallVec;
+use rustc_hash::FxBuildHasher;
 
-use super::lru::Lru;
 use crate::update::del_add::{DelAdd, KvWriterDelAdd};
-use crate::CboRoaringBitmapCodec;
+use crate::update::new::indexer::document_changes::MostlySend;
+use crate::update::new::KvReaderDelAdd;
+use crate::update::MergeDeladdCboRoaringBitmaps;
+use crate::{CboRoaringBitmapCodec, Result};
 
-const KEY_SIZE: usize = 12;
+/// A cache that stores bytes keys associated to CboDelAddRoaringBitmaps.
+///
+/// Internally balances the content over `N` buckets for future merging.
+pub struct BalancedCaches<'extractor> {
+    hasher: FxBuildHasher,
+    alloc: &'extractor Bump,
+    max_memory: Option<usize>,
+    caches: InnerCaches<'extractor>,
+}
 
-#[derive(Debug)]
+enum InnerCaches<'extractor> {
-pub struct CboCachedSorter<MF> {
+    Normal(NormalCaches<'extractor>),
-    cache: Lru<SmallVec<[u8; KEY_SIZE]>, DelAddRoaringBitmap>,
+    Spilling(SpillingCaches<'extractor>),
-    sorter: Sorter<MF>,
+}
+
+impl<'extractor> BalancedCaches<'extractor> {
+    pub fn new_in(buckets: usize, max_memory: Option<usize>, alloc: &'extractor Bump) -> Self {
+        Self {
+            hasher: FxBuildHasher,
+            max_memory,
+            caches: InnerCaches::Normal(NormalCaches {
+                caches: iter::repeat_with(|| HashMap::with_hasher_in(FxBuildHasher, alloc))
+                    .take(buckets)
+                    .collect(),
+            }),
+            alloc,
+        }
+    }
+
+    fn buckets(&self) -> usize {
+        match &self.caches {
+            InnerCaches::Normal(caches) => caches.caches.len(),
+            InnerCaches::Spilling(caches) => caches.caches.len(),
+        }
+    }
+
+    pub fn insert_del_u32(&mut self, key: &[u8], n: u32) -> Result<()> {
+        if self.max_memory.map_or(false, |mm| self.alloc.allocated_bytes() >= mm) {
+            self.start_spilling()?;
+        }
+
+        let buckets = self.buckets();
+        match &mut self.caches {
+            InnerCaches::Normal(normal) => {
+                normal.insert_del_u32(&self.hasher, self.alloc, buckets, key, n);
+                Ok(())
+            }
+            InnerCaches::Spilling(spilling) => {
+                spilling.insert_del_u32(&self.hasher, buckets, key, n)
+            }
+        }
+    }
+
+    pub fn insert_add_u32(&mut self, key: &[u8], n: u32) -> Result<()> {
+        if self.max_memory.map_or(false, |mm| self.alloc.allocated_bytes() >= mm) {
+            self.start_spilling()?;
+        }
+
+        let buckets = self.buckets();
+        match &mut self.caches {
+            InnerCaches::Normal(normal) => {
+                normal.insert_add_u32(&self.hasher, self.alloc, buckets, key, n);
+                Ok(())
+            }
+            InnerCaches::Spilling(spilling) => {
+                spilling.insert_add_u32(&self.hasher, buckets, key, n)
+            }
+        }
+    }
+
+    /// Make sure the cache is no longer allocating data
+    /// and writes every new and unknow entry to disk.
+    fn start_spilling(&mut self) -> Result<()> {
+        let BalancedCaches { hasher: _, alloc, max_memory: _, caches } = self;
+
+        if let InnerCaches::Normal(normal_caches) = caches {
+            eprintln!(
+                "We are spilling after we allocated {} bytes on thread #{}",
+                alloc.allocated_bytes(),
+                rayon::current_thread_index().unwrap_or(0)
+            );
+
+            let allocated: usize = normal_caches.caches.iter().map(|m| m.allocation_size()).sum();
+            eprintln!("The last allocated HasMap took {allocated} bytes");
+
+            let dummy = NormalCaches { caches: Vec::new() };
+            let NormalCaches { caches: cache_maps } = mem::replace(normal_caches, dummy);
+            *caches = InnerCaches::Spilling(SpillingCaches::from_cache_maps(cache_maps));
+        }
+
+        Ok(())
+    }
+
+    pub fn freeze(&mut self) -> Result<Vec<FrozenCache<'_, 'extractor>>> {
+        match &mut self.caches {
+            InnerCaches::Normal(NormalCaches { caches }) => caches
+                .iter_mut()
+                .enumerate()
+                .map(|(bucket, map)| {
+                    Ok(FrozenCache { bucket, cache: FrozenMap::new(map), spilled: Vec::new() })
+                })
+                .collect(),
+            InnerCaches::Spilling(SpillingCaches { caches, spilled_entries, .. }) => caches
+                .iter_mut()
+                .zip(mem::take(spilled_entries))
+                .enumerate()
+                .map(|(bucket, (map, sorter))| {
+                    let spilled = sorter
+                        .into_reader_cursors()?
+                        .into_iter()
+                        .map(ReaderCursor::into_inner)
+                        .map(BufReader::new)
+                        .map(|bufreader| grenad::Reader::new(bufreader).map_err(Into::into))
+                        .collect::<Result<_>>()?;
+                    Ok(FrozenCache { bucket, cache: FrozenMap::new(map), spilled })
+                })
+                .collect(),
+        }
+    }
+}
+
+unsafe impl MostlySend for BalancedCaches<'_> {}
+
+struct NormalCaches<'extractor> {
+    caches: Vec<HashMap<&'extractor [u8], DelAddRoaringBitmap, FxBuildHasher, &'extractor Bump>>,
+}
+
+impl<'extractor> NormalCaches<'extractor> {
+    pub fn insert_del_u32(
+        &mut self,
+        hasher: &FxBuildHasher,
+        alloc: &'extractor Bump,
+        buckets: usize,
+        key: &[u8],
+        n: u32,
+    ) {
+        let hash = hasher.hash_one(key);
+        let bucket = compute_bucket_from_hash(buckets, hash);
+
+        match self.caches[bucket].raw_entry_mut().from_hash(hash, |&k| k == key) {
+            RawEntryMut::Occupied(mut entry) => {
+                entry.get_mut().del.get_or_insert_with(RoaringBitmap::default).insert(n);
+            }
+            RawEntryMut::Vacant(entry) => {
+                entry.insert_hashed_nocheck(
+                    hash,
+                    alloc.alloc_slice_copy(key),
+                    DelAddRoaringBitmap::new_del_u32(n),
+                );
+            }
+        }
+    }
+
+    pub fn insert_add_u32(
+        &mut self,
+        hasher: &FxBuildHasher,
+        alloc: &'extractor Bump,
+        buckets: usize,
+        key: &[u8],
+        n: u32,
+    ) {
+        let hash = hasher.hash_one(key);
+        let bucket = compute_bucket_from_hash(buckets, hash);
+        match self.caches[bucket].raw_entry_mut().from_hash(hash, |&k| k == key) {
+            RawEntryMut::Occupied(mut entry) => {
+                entry.get_mut().add.get_or_insert_with(RoaringBitmap::default).insert(n);
+            }
+            RawEntryMut::Vacant(entry) => {
+                entry.insert_hashed_nocheck(
+                    hash,
+                    alloc.alloc_slice_copy(key),
+                    DelAddRoaringBitmap::new_add_u32(n),
+                );
+            }
+        }
+    }
+}
+
+struct SpillingCaches<'extractor> {
+    caches: Vec<HashMap<&'extractor [u8], DelAddRoaringBitmap, FxBuildHasher, &'extractor Bump>>,
+    spilled_entries: Vec<grenad::Sorter<MergeDeladdCboRoaringBitmaps>>,
     deladd_buffer: Vec<u8>,
     cbo_buffer: Vec<u8>,
-    total_insertions: usize,
-    fitted_in_key: usize,
 }
 
-impl<MF> CboCachedSorter<MF> {
+impl<'extractor> SpillingCaches<'extractor> {
-    pub fn new(cap: NonZeroUsize, sorter: Sorter<MF>) -> Self {
+    fn from_cache_maps(
-        CboCachedSorter {
+        caches: Vec<
-            cache: Lru::new(cap),
+            HashMap<&'extractor [u8], DelAddRoaringBitmap, FxBuildHasher, &'extractor Bump>,
-            sorter,
+        >,
+    ) -> SpillingCaches<'extractor> {
+        SpillingCaches {
+            spilled_entries: iter::repeat_with(|| {
+                let mut builder = grenad::SorterBuilder::new(MergeDeladdCboRoaringBitmaps);
+                builder.dump_threshold(0);
+                builder.allow_realloc(false);
+                builder.build()
+            })
+            .take(caches.len())
+            .collect(),
+            caches,
             deladd_buffer: Vec::new(),
             cbo_buffer: Vec::new(),
-            total_insertions: 0,
-            fitted_in_key: 0,
         }
     }
-}
 
-impl<MF: MergeFunction> CboCachedSorter<MF> {
+    pub fn insert_del_u32(
-    pub fn insert_del_u32(&mut self, key: &[u8], n: u32) -> grenad::Result<(), MF::Error> {
-        match self.cache.get_mut(key) {
-            Some(DelAddRoaringBitmap { del, add: _ }) => {
-                del.get_or_insert_with(RoaringBitmap::default).insert(n);
-            }
-            None => {
-                self.total_insertions += 1;
-                self.fitted_in_key += (key.len() <= KEY_SIZE) as usize;
-                let value = DelAddRoaringBitmap::new_del_u32(n);
-                if let Some((key, deladd)) = self.cache.push(key.into(), value) {
-                    self.write_entry(key, deladd)?;
-                }
-            }
-        }
-
-        Ok(())
-    }
-
-    pub fn insert_del(
         &mut self,
+        hasher: &FxBuildHasher,
+        buckets: usize,
         key: &[u8],
-        bitmap: RoaringBitmap,
+        n: u32,
-    ) -> grenad::Result<(), MF::Error> {
+    ) -> Result<()> {
-        match self.cache.get_mut(key) {
+        let hash = hasher.hash_one(key);
-            Some(DelAddRoaringBitmap { del, add: _ }) => {
+        let bucket = compute_bucket_from_hash(buckets, hash);
-                *del.get_or_insert_with(RoaringBitmap::default) |= bitmap;
+        match self.caches[bucket].raw_entry_mut().from_hash(hash, |&k| k == key) {
+            RawEntryMut::Occupied(mut entry) => {
+                entry.get_mut().del.get_or_insert_with(RoaringBitmap::default).insert(n);
+                Ok(())
             }
-            None => {
+            RawEntryMut::Vacant(_entry) => {
-                self.total_insertions += 1;
+                let deladd = DelAddRoaringBitmap::new_del_u32(n);
-                self.fitted_in_key += (key.len() <= KEY_SIZE) as usize;
+                spill_entry_to_sorter(
-                let value = DelAddRoaringBitmap::new_del(bitmap);
+                    &mut self.spilled_entries[bucket],
-                if let Some((key, deladd)) = self.cache.push(key.into(), value) {
+                    &mut self.deladd_buffer,
-                    self.write_entry(key, deladd)?;
+                    &mut self.cbo_buffer,
-                }
+                    key,
+                    deladd,
+                )
             }
         }
-
-        Ok(())
     }
 
-    pub fn insert_add_u32(&mut self, key: &[u8], n: u32) -> grenad::Result<(), MF::Error> {
+    pub fn insert_add_u32(
-        match self.cache.get_mut(key) {
-            Some(DelAddRoaringBitmap { del: _, add }) => {
-                add.get_or_insert_with(RoaringBitmap::default).insert(n);
-            }
-            None => {
-                self.total_insertions += 1;
-                self.fitted_in_key += (key.len() <= KEY_SIZE) as usize;
-                let value = DelAddRoaringBitmap::new_add_u32(n);
-                if let Some((key, deladd)) = self.cache.push(key.into(), value) {
-                    self.write_entry(key, deladd)?;
-                }
-            }
-        }
-
-        Ok(())
-    }
-
-    pub fn insert_add(
         &mut self,
+        hasher: &FxBuildHasher,
+        buckets: usize,
         key: &[u8],
-        bitmap: RoaringBitmap,
+        n: u32,
-    ) -> grenad::Result<(), MF::Error> {
+    ) -> Result<()> {
-        match self.cache.get_mut(key) {
+        let hash = hasher.hash_one(key);
-            Some(DelAddRoaringBitmap { del: _, add }) => {
+        let bucket = compute_bucket_from_hash(buckets, hash);
-                *add.get_or_insert_with(RoaringBitmap::default) |= bitmap;
+        match self.caches[bucket].raw_entry_mut().from_hash(hash, |&k| k == key) {
+            RawEntryMut::Occupied(mut entry) => {
+                entry.get_mut().add.get_or_insert_with(RoaringBitmap::default).insert(n);
+                Ok(())
             }
-            None => {
+            RawEntryMut::Vacant(_entry) => {
-                self.total_insertions += 1;
+                let deladd = DelAddRoaringBitmap::new_add_u32(n);
-                self.fitted_in_key += (key.len() <= KEY_SIZE) as usize;
+                spill_entry_to_sorter(
-                let value = DelAddRoaringBitmap::new_add(bitmap);
+                    &mut self.spilled_entries[bucket],
-                if let Some((key, deladd)) = self.cache.push(key.into(), value) {
+                    &mut self.deladd_buffer,
-                    self.write_entry(key, deladd)?;
+                    &mut self.cbo_buffer,
+                    key,
+                    deladd,
+                )
+            }
+        }
+    }
+}
+
+#[inline]
+fn compute_bucket_from_hash(buckets: usize, hash: u64) -> usize {
+    hash as usize % buckets
+}
+
+fn spill_entry_to_sorter(
+    spilled_entries: &mut grenad::Sorter<MergeDeladdCboRoaringBitmaps>,
+    deladd_buffer: &mut Vec<u8>,
+    cbo_buffer: &mut Vec<u8>,
+    key: &[u8],
+    deladd: DelAddRoaringBitmap,
+) -> Result<()> {
+    deladd_buffer.clear();
+    let mut value_writer = KvWriterDelAdd::new(deladd_buffer);
+
+    match deladd {
+        DelAddRoaringBitmap { del: Some(del), add: None } => {
+            cbo_buffer.clear();
+            CboRoaringBitmapCodec::serialize_into(&del, cbo_buffer);
+            value_writer.insert(DelAdd::Deletion, &cbo_buffer)?;
+        }
+        DelAddRoaringBitmap { del: None, add: Some(add) } => {
+            cbo_buffer.clear();
+            CboRoaringBitmapCodec::serialize_into(&add, cbo_buffer);
+            value_writer.insert(DelAdd::Addition, &cbo_buffer)?;
+        }
+        DelAddRoaringBitmap { del: Some(del), add: Some(add) } => {
+            cbo_buffer.clear();
+            CboRoaringBitmapCodec::serialize_into(&del, cbo_buffer);
+            value_writer.insert(DelAdd::Deletion, &cbo_buffer)?;
+
+            cbo_buffer.clear();
+            CboRoaringBitmapCodec::serialize_into(&add, cbo_buffer);
+            value_writer.insert(DelAdd::Addition, &cbo_buffer)?;
+        }
+        DelAddRoaringBitmap { del: None, add: None } => return Ok(()),
+    }
+
+    let bytes = value_writer.into_inner().unwrap();
+    spilled_entries.insert(key, bytes).map_err(Into::into)
+}
+
+pub struct FrozenCache<'a, 'extractor> {
+    bucket: usize,
+    cache: FrozenMap<'a, 'extractor, &'extractor [u8], DelAddRoaringBitmap, FxBuildHasher>,
+    spilled: Vec<grenad::Reader<BufReader<File>>>,
+}
+
+pub fn transpose_and_freeze_caches<'a, 'extractor>(
+    caches: &'a mut [BalancedCaches<'extractor>],
+) -> Result<Vec<Vec<FrozenCache<'a, 'extractor>>>> {
+    let width = caches.first().map(BalancedCaches::buckets).unwrap_or(0);
+    let mut bucket_caches: Vec<_> = iter::repeat_with(Vec::new).take(width).collect();
+
+    for thread_cache in caches {
+        for frozen in thread_cache.freeze()? {
+            bucket_caches[frozen.bucket].push(frozen);
+        }
+    }
+
+    Ok(bucket_caches)
+}
+
+/// Merges the caches that must be all associated to the same bucket.
+///
+/// # Panics
+///
+/// - If the bucket IDs in these frozen caches are not exactly the same.
+pub fn merge_caches<F>(frozen: Vec<FrozenCache>, mut f: F) -> Result<()>
+where
+    F: for<'a> FnMut(&'a [u8], DelAddRoaringBitmap) -> Result<()>,
+{
+    let mut maps = Vec::new();
+    let mut readers = Vec::new();
+    let mut current_bucket = None;
+    for FrozenCache { bucket, cache, ref mut spilled } in frozen {
+        assert_eq!(*current_bucket.get_or_insert(bucket), bucket);
+        maps.push(cache);
+        readers.append(spilled);
+    }
+
+    // First manage the spilled entries by looking into the HashMaps,
+    // merge them and mark them as dummy.
+    let mut heap = BinaryHeap::new();
+    for (source_index, source) in readers.into_iter().enumerate() {
+        let mut cursor = source.into_cursor()?;
+        if cursor.move_on_next()?.is_some() {
+            heap.push(Entry { cursor, source_index });
+        }
+    }
+
+    loop {
+        let mut first_entry = match heap.pop() {
+            Some(entry) => entry,
+            None => break,
+        };
+
+        let (first_key, first_value) = match first_entry.cursor.current() {
+            Some((key, value)) => (key, value),
+            None => break,
+        };
+
+        let mut output = DelAddRoaringBitmap::from_bytes(first_value)?;
+        while let Some(mut entry) = heap.peek_mut() {
+            if let Some((key, _value)) = entry.cursor.current() {
+                if first_key == key {
+                    let new = DelAddRoaringBitmap::from_bytes(first_value)?;
+                    output = output.merge(new);
+                    // When we are done we the current value of this entry move make
+                    // it move forward and let the heap reorganize itself (on drop)
+                    if entry.cursor.move_on_next()?.is_none() {
+                        PeekMut::pop(entry);
+                    }
+                } else {
+                    break;
                 }
             }
         }
 
-        Ok(())
+        // Once we merged all of the spilled bitmaps we must also
-    }
+        // fetch the entries from the non-spilled entries (the HashMaps).
-
+        for (map_index, map) in maps.iter_mut().enumerate() {
-    pub fn insert_del_add_u32(&mut self, key: &[u8], n: u32) -> grenad::Result<(), MF::Error> {
+            if first_entry.source_index != map_index {
-        match self.cache.get_mut(key) {
+                if let Some(new) = map.get_mut(first_key) {
-            Some(DelAddRoaringBitmap { del, add }) => {
+                    output = output.merge(mem::take(new));
-                del.get_or_insert_with(RoaringBitmap::default).insert(n);
-                add.get_or_insert_with(RoaringBitmap::default).insert(n);
-            }
-            None => {
-                self.total_insertions += 1;
-                self.fitted_in_key += (key.len() <= KEY_SIZE) as usize;
-                let value = DelAddRoaringBitmap::new_del_add_u32(n);
-                if let Some((key, deladd)) = self.cache.push(key.into(), value) {
-                    self.write_entry(key, deladd)?;
                 }
             }
         }
 
-        Ok(())
+        // We send the merged entry outside.
+        (f)(first_key, output)?;
+
+        // Don't forget to put the first entry back into the heap.
+        if first_entry.cursor.move_on_next()?.is_some() {
+            heap.push(first_entry)
+        }
     }
 
-    fn write_entry<A: AsRef<[u8]>>(
+    // Then manage the content on the HashMap entries that weren't taken (mem::take).
-        &mut self,
+    while let Some(mut map) = maps.pop() {
-        key: A,
+        for (key, output) in map.iter_mut() {
-        deladd: DelAddRoaringBitmap,
+            let mut output = mem::take(output);
-    ) -> grenad::Result<(), MF::Error> {
-        /// TODO we must create a serialization trait to correctly serialize bitmaps
-        self.deladd_buffer.clear();
-        let mut value_writer = KvWriterDelAdd::new(&mut self.deladd_buffer);
-        match deladd {
-            DelAddRoaringBitmap { del: Some(del), add: None } => {
-                self.cbo_buffer.clear();
-                CboRoaringBitmapCodec::serialize_into(&del, &mut self.cbo_buffer);
-                value_writer.insert(DelAdd::Deletion, &self.cbo_buffer)?;
-            }
-            DelAddRoaringBitmap { del: None, add: Some(add) } => {
-                self.cbo_buffer.clear();
-                CboRoaringBitmapCodec::serialize_into(&add, &mut self.cbo_buffer);
-                value_writer.insert(DelAdd::Addition, &self.cbo_buffer)?;
-            }
-            DelAddRoaringBitmap { del: Some(del), add: Some(add) } => {
-                self.cbo_buffer.clear();
-                CboRoaringBitmapCodec::serialize_into(&del, &mut self.cbo_buffer);
-                value_writer.insert(DelAdd::Deletion, &self.cbo_buffer)?;
 
-                self.cbo_buffer.clear();
+            // Make sure we don't try to work with entries already managed by the spilled
-                CboRoaringBitmapCodec::serialize_into(&add, &mut self.cbo_buffer);
+            if !output.is_empty() {
-                value_writer.insert(DelAdd::Addition, &self.cbo_buffer)?;
+                for rhs in maps.iter_mut() {
+                    if let Some(new) = rhs.get_mut(key) {
+                        output = output.merge(mem::take(new));
+                    }
+                }
+
+                // We send the merged entry outside.
+                (f)(key, output)?;
             }
-            DelAddRoaringBitmap { del: None, add: None } => return Ok(()),
         }
-        let bytes = value_writer.into_inner().unwrap();
-        self.sorter.insert(key, bytes)
     }
 
-    pub fn direct_insert(&mut self, key: &[u8], val: &[u8]) -> grenad::Result<(), MF::Error> {
+    Ok(())
-        self.sorter.insert(key, val)
+}
-    }
 
-    pub fn into_sorter(mut self) -> grenad::Result<Sorter<MF>, MF::Error> {
+struct Entry<R> {
-        let mut all_n_containers = Vec::new();
+    cursor: ReaderCursor<R>,
-        let mut all_n_array_containers = Vec::new();
+    source_index: usize,
-        let mut all_n_bitset_containers = Vec::new();
+}
-        let mut all_n_values_array_containers = Vec::new();
-        let mut all_n_values_bitset_containers = Vec::new();
-        let mut all_cardinality = Vec::new();
 
-        let default_arc = Lru::new(NonZeroUsize::MIN);
+impl<R> Ord for Entry<R> {
-        for (key, deladd) in mem::replace(&mut self.cache, default_arc) {
+    fn cmp(&self, other: &Entry<R>) -> Ordering {
-            for bitmap in [&deladd.del, &deladd.add].into_iter().flatten() {
+        let skey = self.cursor.current().map(|(k, _)| k);
-                let Statistics {
+        let okey = other.cursor.current().map(|(k, _)| k);
-                    n_containers,
+        skey.cmp(&okey).then(self.source_index.cmp(&other.source_index)).reverse()
-                    n_array_containers,
-                    n_bitset_containers,
-                    n_values_array_containers,
-                    n_values_bitset_containers,
-                    cardinality,
-                    ..
-                } = bitmap.statistics();
-                all_n_containers.push(n_containers);
-                all_n_array_containers.push(n_array_containers);
-                all_n_bitset_containers.push(n_bitset_containers);
-                all_n_values_array_containers.push(n_values_array_containers);
-                all_n_values_bitset_containers.push(n_values_bitset_containers);
-                all_cardinality.push(cardinality as u32);
-            }
-
-            self.write_entry(key, deladd)?;
-        }
-
-        let mut output = String::new();
-
-        for (name, mut slice) in [
-            ("n_containers", all_n_containers),
-            ("n_array_containers", all_n_array_containers),
-            ("n_bitset_containers", all_n_bitset_containers),
-            ("n_values_array_containers", all_n_values_array_containers),
-            ("n_values_bitset_containers", all_n_values_bitset_containers),
-            ("cardinality", all_cardinality),
-        ] {
-            let _ = writeln!(&mut output, "{name} (p100) {:?}", Stats::from_slice(&mut slice));
-            // let _ = writeln!(&mut output, "{name} (p99)  {:?}", Stats::from_slice_p99(&mut slice));
-        }
-
-        let _ = writeln!(
-            &mut output,
-            "LruCache stats: {} <= {KEY_SIZE} bytes ({}%) on a total of {} insertions",
-            self.fitted_in_key,
-            (self.fitted_in_key as f32 / self.total_insertions as f32) * 100.0,
-            self.total_insertions,
-        );
-
-        eprintln!("{output}");
-
-        Ok(self.sorter)
     }
 }
 
-#[derive(Default, Debug)]
+impl<R> Eq for Entry<R> {}
-struct Stats {
-    pub len: usize,
-    pub average: f32,
-    pub mean: u32,
-    pub min: u32,
-    pub max: u32,
-}
 
-impl Stats {
+impl<R> PartialEq for Entry<R> {
-    fn from_slice(slice: &mut [u32]) -> Stats {
+    fn eq(&self, other: &Entry<R>) -> bool {
-        slice.sort_unstable();
+        self.cmp(other) == Ordering::Equal
-        Self::from_sorted_slice(slice)
-    }
-
-    fn from_slice_p99(slice: &mut [u32]) -> Stats {
-        slice.sort_unstable();
-        let new_len = slice.len() - (slice.len() as f32 / 100.0) as usize;
-        match slice.get(..new_len) {
-            Some(slice) => Self::from_sorted_slice(slice),
-            None => Stats::default(),
-        }
-    }
-
-    fn from_sorted_slice(slice: &[u32]) -> Stats {
-        let sum: f64 = slice.iter().map(|i| *i as f64).sum();
-        let average = (sum / slice.len() as f64) as f32;
-        let mean = *slice.len().checked_div(2).and_then(|middle| slice.get(middle)).unwrap_or(&0);
-        let min = *slice.first().unwrap_or(&0);
-        let max = *slice.last().unwrap_or(&0);
-        Stats { len: slice.len(), average, mean, min, max }
     }
 }
 
-#[derive(Debug, Clone)]
+impl<R> PartialOrd for Entry<R> {
+    fn partial_cmp(&self, other: &Entry<R>) -> Option<Ordering> {
+        Some(self.cmp(other))
+    }
+}
+
+#[derive(Debug, Default, Clone)]
 pub struct DelAddRoaringBitmap {
-    pub(crate) del: Option<RoaringBitmap>,
+    pub del: Option<RoaringBitmap>,
-    pub(crate) add: Option<RoaringBitmap>,
+    pub add: Option<RoaringBitmap>,
 }
 
 impl DelAddRoaringBitmap {
-    fn new_del_add_u32(n: u32) -> Self {
+    fn from_bytes(bytes: &[u8]) -> io::Result<DelAddRoaringBitmap> {
-        DelAddRoaringBitmap {
+        let reader = KvReaderDelAdd::from_slice(bytes);
-            del: Some(RoaringBitmap::from([n])),
+
-            add: Some(RoaringBitmap::from([n])),
+        let del = match reader.get(DelAdd::Deletion) {
-        }
+            Some(bytes) => CboRoaringBitmapCodec::deserialize_from(bytes).map(Some)?,
+            None => None,
+        };
+
+        let add = match reader.get(DelAdd::Addition) {
+            Some(bytes) => CboRoaringBitmapCodec::deserialize_from(bytes).map(Some)?,
+            None => None,
+        };
+
+        Ok(DelAddRoaringBitmap { del, add })
     }
 
-    fn new_del(bitmap: RoaringBitmap) -> Self {
+    pub fn empty() -> DelAddRoaringBitmap {
-        DelAddRoaringBitmap { del: Some(bitmap), add: None }
+        DelAddRoaringBitmap { del: None, add: None }
     }
 
-    fn new_del_u32(n: u32) -> Self {
+    pub fn is_empty(&self) -> bool {
+        let DelAddRoaringBitmap { del, add } = self;
+        del.is_none() && add.is_none()
+    }
+
+    pub fn insert_del_u32(&mut self, n: u32) {
+        self.del.get_or_insert_with(RoaringBitmap::new).insert(n);
+    }
+
+    pub fn insert_add_u32(&mut self, n: u32) {
+        self.add.get_or_insert_with(RoaringBitmap::new).insert(n);
+    }
+
+    pub fn new_del_u32(n: u32) -> Self {
         DelAddRoaringBitmap { del: Some(RoaringBitmap::from([n])), add: None }
     }
 
-    fn new_add(bitmap: RoaringBitmap) -> Self {
+    pub fn new_add_u32(n: u32) -> Self {
-        DelAddRoaringBitmap { del: None, add: Some(bitmap) }
-    }
-
-    fn new_add_u32(n: u32) -> Self {
         DelAddRoaringBitmap { del: None, add: Some(RoaringBitmap::from([n])) }
     }
+
+    pub fn merge(self, rhs: DelAddRoaringBitmap) -> DelAddRoaringBitmap {
+        let DelAddRoaringBitmap { del, add } = self;
+        let DelAddRoaringBitmap { del: ndel, add: nadd } = rhs;
+
+        let del = match (del, ndel) {
+            (None, None) => None,
+            (None, Some(del)) | (Some(del), None) => Some(del),
+            (Some(del), Some(ndel)) => Some(del | ndel),
+        };
+
+        let add = match (add, nadd) {
+            (None, None) => None,
+            (None, Some(add)) | (Some(add), None) => Some(add),
+            (Some(add), Some(nadd)) => Some(add | nadd),
+        };
+
+        DelAddRoaringBitmap { del, add }
+    }
+
+    pub fn apply_to(&self, documents_ids: &mut RoaringBitmap) {
+        let DelAddRoaringBitmap { del, add } = self;
+
+        if let Some(del) = del {
+            *documents_ids -= del;
+        }
+
+        if let Some(add) = add {
+            *documents_ids |= add;
+        }
+    }
 }

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

@@ -0,0 +1,73 @@
+use std::cell::RefCell;
+
+use bumpalo::Bump;
+
+use super::DelAddRoaringBitmap;
+use crate::update::new::channel::DocumentsSender;
+use crate::update::new::document::write_to_obkv;
+use crate::update::new::indexer::document_changes::{
+    DocumentChangeContext, Extractor, FullySend, RefCellExt as _,
+};
+use crate::update::new::DocumentChange;
+use crate::Result;
+
+pub struct DocumentsExtractor<'a> {
+    documents_sender: &'a DocumentsSender<'a>,
+}
+
+impl<'a> DocumentsExtractor<'a> {
+    pub fn new(documents_sender: &'a DocumentsSender<'a>) -> Self {
+        Self { documents_sender }
+    }
+}
+
+impl<'a, 'extractor> Extractor<'extractor> for DocumentsExtractor<'a> {
+    type Data = FullySend<RefCell<DelAddRoaringBitmap>>;
+
+    fn init_data(&self, _extractor_alloc: &'extractor Bump) -> Result<Self::Data> {
+        Ok(FullySend(RefCell::new(DelAddRoaringBitmap::empty())))
+    }
+
+    fn process(
+        &self,
+        change: DocumentChange,
+        context: &DocumentChangeContext<Self::Data>,
+    ) -> Result<()> {
+        let mut document_buffer = Vec::new();
+        let mut delta_documents_ids = context.data.0.borrow_mut_or_yield();
+
+        let new_fields_ids_map = context.new_fields_ids_map.borrow_or_yield();
+        let new_fields_ids_map = &*new_fields_ids_map;
+        let new_fields_ids_map = new_fields_ids_map.local_map();
+
+        let external_docid = change.external_docid().to_owned();
+
+        // document but we need to create a function that collects and compresses documents.
+        match change {
+            DocumentChange::Deletion(deletion) => {
+                let docid = deletion.docid();
+                self.documents_sender.delete(docid, external_docid).unwrap();
+                delta_documents_ids.insert_del_u32(docid);
+            }
+            /// TODO: change NONE by SOME(vector) when implemented
+            DocumentChange::Update(update) => {
+                let docid = update.docid();
+                let content =
+                    update.new(&context.txn, context.index, &context.db_fields_ids_map)?;
+                let content =
+                    write_to_obkv(&content, None, new_fields_ids_map, &mut document_buffer)?;
+                self.documents_sender.uncompressed(docid, external_docid, content).unwrap();
+            }
+            DocumentChange::Insertion(insertion) => {
+                let docid = insertion.docid();
+                let content = insertion.new();
+                let content =
+                    write_to_obkv(&content, None, new_fields_ids_map, &mut document_buffer)?;
+                self.documents_sender.uncompressed(docid, external_docid, content).unwrap();
+                delta_documents_ids.insert_add_u32(docid);
+                // extracted_dictionary_sender.send(self, dictionary: &[u8]);
+            }
+        }
+        Ok(())
+    }
+}

milli/src/update/new/extract/faceted/extract_facets.rs

@@ -1,16 +1,12 @@
 use std::cell::RefCell;
 use std::collections::HashSet;
-use std::fmt::Debug;
-use std::fs::File;
 use std::ops::DerefMut as _;
 
 use bumpalo::Bump;
-use grenad::{MergeFunction, Merger};
 use heed::RoTxn;
-use rayon::iter::{ParallelBridge as _, ParallelIterator as _};
 use serde_json::Value;
 
-use super::super::cache::CboCachedSorter;
+use super::super::cache::BalancedCaches;
 use super::facet_document::extract_document_facets;
 use super::FacetKind;
 use crate::facet::value_encoding::f64_into_bytes;
@@ -20,44 +16,30 @@ use crate::update::new::indexer::document_changes::{
     IndexingContext, RefCellExt, ThreadLocal,
 };
 use crate::update::new::DocumentChange;
-use crate::update::{create_sorter, GrenadParameters, MergeDeladdCboRoaringBitmaps};
+use crate::update::GrenadParameters;
 use crate::{DocumentId, FieldId, Index, Result, MAX_FACET_VALUE_LENGTH};
 
-pub struct FacetedExtractorData<'extractor> {
+pub struct FacetedExtractorData<'a> {
-    attributes_to_extract: &'extractor [&'extractor str],
+    attributes_to_extract: &'a [&'a str],
     grenad_parameters: GrenadParameters,
-    max_memory: Option<usize>,
+    buckets: usize,
 }
 
-impl<'extractor> Extractor<'extractor> for FacetedExtractorData<'extractor> {
+impl<'a, 'extractor> Extractor<'extractor> for FacetedExtractorData<'a> {
-    type Data = FullySend<RefCell<CboCachedSorter<MergeDeladdCboRoaringBitmaps>>>;
+    type Data = RefCell<BalancedCaches<'extractor>>;
 
-    fn init_data(
+    fn init_data(&self, extractor_alloc: &'extractor Bump) -> Result<Self::Data> {
-        &self,
+        Ok(RefCell::new(BalancedCaches::new_in(
-        _extractor_alloc: raw_collections::alloc::RefBump<'extractor>,
+            self.buckets,
-    ) -> Result<Self::Data> {
+            self.grenad_parameters.max_memory,
-        Ok(FullySend(RefCell::new(CboCachedSorter::new(
+            extractor_alloc,
-            // TODO use a better value
+        )))
-            1_000_000.try_into().unwrap(),
-            create_sorter(
-                grenad::SortAlgorithm::Stable,
-                MergeDeladdCboRoaringBitmaps,
-                self.grenad_parameters.chunk_compression_type,
-                self.grenad_parameters.chunk_compression_level,
-                self.grenad_parameters.max_nb_chunks,
-                self.max_memory,
-                // *NOTE*: this must not be set to true:
-                // 1. we're already using max parallelism in the pool, so it wouldn't help
-                // 2. it creates correctness issues if it causes to yield a borrow-mut wielding task
-                false,
-            ),
-        ))))
     }
 
     fn process(
         &self,
         change: DocumentChange,
-        context: &crate::update::new::indexer::document_changes::DocumentChangeContext<Self::Data>,
+        context: &DocumentChangeContext<Self::Data>,
     ) -> Result<()> {
         FacetedDocidsExtractor::extract_document_change(context, self.attributes_to_extract, change)
     }
@@ -67,16 +49,14 @@ pub struct FacetedDocidsExtractor;
 
 impl FacetedDocidsExtractor {
     fn extract_document_change(
-        context: &DocumentChangeContext<
+        context: &DocumentChangeContext<RefCell<BalancedCaches>>,
-            FullySend<RefCell<CboCachedSorter<MergeDeladdCboRoaringBitmaps>>>,
-        >,
         attributes_to_extract: &[&str],
         document_change: DocumentChange,
     ) -> Result<()> {
         let index = &context.index;
         let rtxn = &context.txn;
         let mut new_fields_ids_map = context.new_fields_ids_map.borrow_mut_or_yield();
-        let mut cached_sorter = context.data.0.borrow_mut_or_yield();
+        let mut cached_sorter = context.data.borrow_mut_or_yield();
         match document_change {
             DocumentChange::Deletion(inner) => extract_document_facets(
                 attributes_to_extract,
@@ -86,7 +66,7 @@ impl FacetedDocidsExtractor {
                     Self::facet_fn_with_options(
                         &context.doc_alloc,
                         cached_sorter.deref_mut(),
-                        CboCachedSorter::insert_del_u32,
+                        BalancedCaches::insert_del_u32,
                         inner.docid(),
                         fid,
                         value,
@@ -102,7 +82,7 @@ impl FacetedDocidsExtractor {
                         Self::facet_fn_with_options(
                             &context.doc_alloc,
                             cached_sorter.deref_mut(),
-                            CboCachedSorter::insert_del_u32,
+                            BalancedCaches::insert_del_u32,
                             inner.docid(),
                             fid,
                             value,
@@ -118,7 +98,7 @@ impl FacetedDocidsExtractor {
                         Self::facet_fn_with_options(
                             &context.doc_alloc,
                             cached_sorter.deref_mut(),
-                            CboCachedSorter::insert_add_u32,
+                            BalancedCaches::insert_add_u32,
                             inner.docid(),
                             fid,
                             value,
@@ -134,7 +114,7 @@ impl FacetedDocidsExtractor {
                     Self::facet_fn_with_options(
                         &context.doc_alloc,
                         cached_sorter.deref_mut(),
-                        CboCachedSorter::insert_add_u32,
+                        BalancedCaches::insert_add_u32,
                         inner.docid(),
                         fid,
                         value,
@@ -144,25 +124,20 @@ impl FacetedDocidsExtractor {
         }
     }
 
-    fn facet_fn_with_options<MF>(
+    fn facet_fn_with_options<'extractor>(
         doc_alloc: &Bump,
-        cached_sorter: &mut CboCachedSorter<MF>,
+        cached_sorter: &mut BalancedCaches<'extractor>,
-        cache_fn: impl Fn(&mut CboCachedSorter<MF>, &[u8], u32) -> grenad::Result<(), MF::Error>,
+        cache_fn: impl Fn(&mut BalancedCaches<'extractor>, &[u8], u32) -> Result<()>,
         docid: DocumentId,
         fid: FieldId,
         value: &Value,
-    ) -> Result<()>
+    ) -> Result<()> {
-    where
-        MF: MergeFunction,
-        MF::Error: Debug,
-        grenad::Error<MF::Error>: Into<crate::Error>,
-    {
         let mut buffer = bumpalo::collections::Vec::new_in(doc_alloc);
         // Exists
         // key: fid
         buffer.push(FacetKind::Exists as u8);
         buffer.extend_from_slice(&fid.to_be_bytes());
-        cache_fn(cached_sorter, &buffer, docid).map_err(Into::into)?;
+        cache_fn(cached_sorter, &buffer, docid)?;
 
         match value {
             // Number
@@ -177,8 +152,7 @@ impl FacetedDocidsExtractor {
                     buffer.push(0); // level 0
                     buffer.extend_from_slice(&ordered);
                     buffer.extend_from_slice(&n.to_be_bytes());
-
+                    cache_fn(cached_sorter, &buffer, docid)
-                    cache_fn(cached_sorter, &buffer, docid).map_err(Into::into)
                 } else {
                     Ok(())
                 }
@@ -193,7 +167,7 @@ impl FacetedDocidsExtractor {
                 buffer.extend_from_slice(&fid.to_be_bytes());
                 buffer.push(0); // level 0
                 buffer.extend_from_slice(truncated.as_bytes());
-                cache_fn(cached_sorter, &buffer, docid).map_err(Into::into)
+                cache_fn(cached_sorter, &buffer, docid)
             }
             // Null
             // key: fid
@@ -201,7 +175,7 @@ impl FacetedDocidsExtractor {
                 buffer.clear();
                 buffer.push(FacetKind::Null as u8);
                 buffer.extend_from_slice(&fid.to_be_bytes());
-                cache_fn(cached_sorter, &buffer, docid).map_err(Into::into)
+                cache_fn(cached_sorter, &buffer, docid)
             }
             // Empty
             // key: fid
@@ -209,13 +183,13 @@ impl FacetedDocidsExtractor {
                 buffer.clear();
                 buffer.push(FacetKind::Empty as u8);
                 buffer.extend_from_slice(&fid.to_be_bytes());
-                cache_fn(cached_sorter, &buffer, docid).map_err(Into::into)
+                cache_fn(cached_sorter, &buffer, docid)
             }
             Value::Object(o) if o.is_empty() => {
                 buffer.clear();
                 buffer.push(FacetKind::Empty as u8);
                 buffer.extend_from_slice(&fid.to_be_bytes());
-                cache_fn(cached_sorter, &buffer, docid).map_err(Into::into)
+                cache_fn(cached_sorter, &buffer, docid)
             }
             // Otherwise, do nothing
             /// TODO: What about Value::Bool?
@@ -242,16 +216,13 @@ fn truncate_str(s: &str) -> &str {
 
 impl DocidsExtractor for FacetedDocidsExtractor {
     #[tracing::instrument(level = "trace", skip_all, target = "indexing::extract::faceted")]
-    fn run_extraction<'pl, 'fid, 'indexer, 'index, DC: DocumentChanges<'pl>>(
+    fn run_extraction<'pl, 'fid, 'indexer, 'index, 'extractor, DC: DocumentChanges<'pl>>(
         grenad_parameters: GrenadParameters,
         document_changes: &DC,
         indexing_context: IndexingContext<'fid, 'indexer, 'index>,
-        extractor_allocs: &mut ThreadLocal<FullySend<RefCell<Bump>>>,
+        extractor_allocs: &'extractor mut ThreadLocal<FullySend<Bump>>,
-    ) -> Result<Merger<File, MergeDeladdCboRoaringBitmaps>> {
+    ) -> Result<Vec<BalancedCaches<'extractor>>> {
-        let max_memory = grenad_parameters.max_memory_by_thread();
-
         let index = indexing_context.index;
-
         let rtxn = index.read_txn()?;
         let attributes_to_extract = Self::attributes_to_extract(&rtxn, index)?;
         let attributes_to_extract: Vec<_> =
@@ -266,7 +237,7 @@ impl DocidsExtractor for FacetedDocidsExtractor {
             let extractor = FacetedExtractorData {
                 attributes_to_extract: &attributes_to_extract,
                 grenad_parameters,
-                max_memory,
+                buckets: rayon::current_num_threads(),
             };
             for_each_document_change(
                 document_changes,
@@ -276,26 +247,7 @@ impl DocidsExtractor for FacetedDocidsExtractor {
                 &datastore,
             )?;
         }
-        {
-            let mut builder = grenad::MergerBuilder::new(MergeDeladdCboRoaringBitmaps);
-            let span =
-                tracing::trace_span!(target: "indexing::documents::extract", "merger_building");
-            let _entered = span.enter();
 
-            let readers: Vec<_> = datastore
+        Ok(datastore.into_iter().map(RefCell::into_inner).collect())
-                .into_iter()
-                .par_bridge()
-                .map(|cached_sorter| {
-                    let cached_sorter = cached_sorter.0.into_inner();
-                    let sorter = cached_sorter.into_sorter()?;
-                    sorter.into_reader_cursors()
-                })
-                .collect();
-
-            for reader in readers {
-                builder.extend(reader?);
-            }
-            Ok(builder.build())
-        }
     }
 }

milli/src/update/new/extract/lru.rs

@@ -1,234 +0,0 @@
-use std::borrow::Borrow;
-use std::hash::{BuildHasher, Hash};
-use std::iter::repeat_with;
-use std::mem;
-use std::num::NonZeroUsize;
-
-use hashbrown::hash_map::{DefaultHashBuilder, Entry};
-use hashbrown::HashMap;
-
-#[derive(Debug)]
-pub struct Lru<K, V, S = DefaultHashBuilder> {
-    lookup: HashMap<K, usize, S>,
-    storage: FixedSizeList<LruNode<K, V>>,
-}
-
-impl<K: Eq + Hash, V> Lru<K, V> {
-    /// Creates a new LRU cache that holds at most `capacity` elements.
-    pub fn new(capacity: NonZeroUsize) -> Self {
-        Self { lookup: HashMap::new(), storage: FixedSizeList::new(capacity.get()) }
-    }
-}
-
-impl<K: Eq + Hash, V, S: BuildHasher> Lru<K, V, S> {
-    /// Creates a new LRU cache that holds at most `capacity` elements
-    /// and uses the provided hash builder to hash keys.
-    pub fn with_hasher(capacity: NonZeroUsize, hash_builder: S) -> Lru<K, V, S> {
-        Self {
-            lookup: HashMap::with_hasher(hash_builder),
-            storage: FixedSizeList::new(capacity.get()),
-        }
-    }
-}
-
-impl<K: Eq + Hash, V, S: BuildHasher> Lru<K, V, S> {
-    /// Returns a mutable reference to the value of the key in the cache or `None` if it is not present in the cache.
-    ///
-    /// Moves the key to the head of the LRU list if it exists.
-    pub fn get_mut<Q>(&mut self, key: &Q) -> Option<&mut V>
-    where
-        K: Borrow<Q>,
-        Q: Hash + Eq + ?Sized,
-    {
-        let idx = *self.lookup.get(key)?;
-        self.storage.move_front(idx).map(|node| &mut node.value)
-    }
-}
-
-impl<K: Clone + Eq + Hash, V, S: BuildHasher> Lru<K, V, S> {
-    pub fn push(&mut self, key: K, value: V) -> Option<(K, V)> {
-        match self.lookup.entry(key) {
-            Entry::Occupied(occ) => {
-                // It's fine to unwrap here because:
-                // * the entry already exists
-                let node = self.storage.move_front(*occ.get()).unwrap();
-                let old_value = mem::replace(&mut node.value, value);
-                let old_key = occ.replace_key();
-                Some((old_key, old_value))
-            }
-            Entry::Vacant(vac) => {
-                let key = vac.key().clone();
-                if self.storage.is_full() {
-                    // It's fine to unwrap here because:
-                    // * the cache capacity is non zero
-                    // * the cache is full
-                    let idx = self.storage.back_idx();
-                    let node = self.storage.move_front(idx).unwrap();
-                    let LruNode { key, value } = mem::replace(node, LruNode { key, value });
-                    vac.insert(idx);
-                    self.lookup.remove(&key);
-                    Some((key, value))
-                } else {
-                    // It's fine to unwrap here because:
-                    // * the cache capacity is non zero
-                    // * the cache is not full
-                    let (idx, _) = self.storage.push_front(LruNode { key, value }).unwrap();
-                    vac.insert(idx);
-                    None
-                }
-            }
-        }
-    }
-}
-
-impl<K, V, S> IntoIterator for Lru<K, V, S> {
-    type Item = (K, V);
-    type IntoIter = IntoIter<K, V>;
-
-    fn into_iter(self) -> Self::IntoIter {
-        IntoIter { lookup_iter: self.lookup.into_iter(), nodes: self.storage.nodes }
-    }
-}
-
-pub struct IntoIter<K, V> {
-    lookup_iter: hashbrown::hash_map::IntoIter<K, usize>,
-    nodes: Box<[Option<FixedSizeListNode<LruNode<K, V>>>]>,
-}
-
-impl<K, V> Iterator for IntoIter<K, V> {
-    type Item = (K, V);
-
-    fn next(&mut self) -> Option<Self::Item> {
-        let (_key, idx) = self.lookup_iter.next()?;
-        let LruNode { key, value } = self.nodes.get_mut(idx)?.take()?.data;
-        Some((key, value))
-    }
-}
-
-#[derive(Debug)]
-struct LruNode<K, V> {
-    key: K,
-    value: V,
-}
-
-#[derive(Debug)]
-struct FixedSizeListNode<T> {
-    prev: usize,
-    next: usize,
-    data: T,
-}
-
-#[derive(Debug)]
-struct FixedSizeList<T> {
-    nodes: Box<[Option<FixedSizeListNode<T>>]>,
-    /// Also corresponds to the first `None` in the nodes.
-    length: usize,
-    // TODO Also, we probably do not need one of the front and back cursors.
-    front: usize,
-    back: usize,
-}
-
-impl<T> FixedSizeList<T> {
-    fn new(capacity: usize) -> Self {
-        Self {
-            nodes: repeat_with(|| None).take(capacity).collect::<Vec<_>>().into_boxed_slice(),
-            length: 0,
-            front: usize::MAX,
-            back: usize::MAX,
-        }
-    }
-
-    #[inline]
-    fn capacity(&self) -> usize {
-        self.nodes.len()
-    }
-
-    #[inline]
-    fn len(&self) -> usize {
-        self.length
-    }
-
-    #[inline]
-    fn is_empty(&self) -> bool {
-        self.len() == 0
-    }
-
-    #[inline]
-    fn is_full(&self) -> bool {
-        self.len() == self.capacity()
-    }
-
-    #[inline]
-    fn back_idx(&self) -> usize {
-        self.back
-    }
-
-    #[inline]
-    fn next(&mut self) -> Option<usize> {
-        if self.is_full() {
-            None
-        } else {
-            let current_free = self.length;
-            self.length += 1;
-            Some(current_free)
-        }
-    }
-
-    #[inline]
-    fn node_mut(&mut self, idx: usize) -> Option<&mut FixedSizeListNode<T>> {
-        self.nodes.get_mut(idx).and_then(|node| node.as_mut())
-    }
-
-    #[inline]
-    fn node_ref(&self, idx: usize) -> Option<&FixedSizeListNode<T>> {
-        self.nodes.get(idx).and_then(|node| node.as_ref())
-    }
-
-    #[inline]
-    fn move_front(&mut self, idx: usize) -> Option<&mut T> {
-        let node = self.nodes.get_mut(idx)?.take()?;
-        if let Some(prev) = self.node_mut(node.prev) {
-            prev.next = node.next;
-        } else {
-            self.front = node.next;
-        }
-        if let Some(next) = self.node_mut(node.next) {
-            next.prev = node.prev;
-        } else {
-            self.back = node.prev;
-        }
-
-        if let Some(front) = self.node_mut(self.front) {
-            front.prev = idx;
-        }
-        if self.node_ref(self.back).is_none() {
-            self.back = idx;
-        }
-
-        let node = self.nodes.get_mut(idx).unwrap().insert(FixedSizeListNode {
-            prev: usize::MAX,
-            next: self.front,
-            data: node.data,
-        });
-        self.front = idx;
-        Some(&mut node.data)
-    }
-
-    #[inline]
-    fn push_front(&mut self, data: T) -> Option<(usize, &mut T)> {
-        let idx = self.next()?;
-        if let Some(front) = self.node_mut(self.front) {
-            front.prev = idx;
-        }
-        if self.node_ref(self.back).is_none() {
-            self.back = idx;
-        }
-        let node = self.nodes.get_mut(idx).unwrap().insert(FixedSizeListNode {
-            prev: usize::MAX,
-            next: self.front,
-            data,
-        });
-        self.front = idx;
-        Some((idx, &mut node.data))
-    }
-}

milli/src/update/new/extract/mod.rs

@@ -1,27 +1,25 @@
 mod cache;
+mod documents;
 mod faceted;
-mod lru;
 mod searchable;
 
-use std::cell::RefCell;
-use std::fs::File;
-
 use bumpalo::Bump;
+pub use cache::{merge_caches, transpose_and_freeze_caches, BalancedCaches, DelAddRoaringBitmap};
+pub use documents::*;
 pub use faceted::*;
-use grenad::Merger;
 pub use searchable::*;
 
 use super::indexer::document_changes::{DocumentChanges, FullySend, IndexingContext, ThreadLocal};
-use crate::update::{GrenadParameters, MergeDeladdCboRoaringBitmaps};
+use crate::update::GrenadParameters;
 use crate::Result;
 
 pub trait DocidsExtractor {
-    fn run_extraction<'pl, 'fid, 'indexer, 'index, DC: DocumentChanges<'pl>>(
+    fn run_extraction<'pl, 'fid, 'indexer, 'index, 'extractor, DC: DocumentChanges<'pl>>(
         grenad_parameters: GrenadParameters,
         document_changes: &DC,
         indexing_context: IndexingContext<'fid, 'indexer, 'index>,
-        extractor_allocs: &mut ThreadLocal<FullySend<RefCell<Bump>>>,
+        extractor_allocs: &'extractor mut ThreadLocal<FullySend<Bump>>,
-    ) -> Result<Merger<File, MergeDeladdCboRoaringBitmaps>>;
+    ) -> Result<Vec<BalancedCaches<'extractor>>>;
 }
 
 /// TODO move in permissive json pointer

milli/src/update/new/extract/searchable/extract_word_docids.rs

@@ -1,113 +1,46 @@
 use std::cell::RefCell;
 use std::collections::HashMap;
-use std::fs::File;
 use std::mem::size_of;
-use std::num::NonZero;
 use std::ops::DerefMut as _;
 
 use bumpalo::collections::vec::Vec as BumpVec;
 use bumpalo::Bump;
-use grenad::{Merger, MergerBuilder};
 use heed::RoTxn;
 
 use super::tokenize_document::{tokenizer_builder, DocumentTokenizer};
-use crate::update::new::extract::cache::CboCachedSorter;
+use crate::update::new::extract::cache::BalancedCaches;
 use crate::update::new::extract::perm_json_p::contained_in;
 use crate::update::new::indexer::document_changes::{
     for_each_document_change, DocumentChangeContext, DocumentChanges, Extractor, FullySend,
-    IndexingContext, RefCellExt, ThreadLocal,
+    IndexingContext, MostlySend, RefCellExt, ThreadLocal,
 };
 use crate::update::new::DocumentChange;
-use crate::update::{create_sorter, GrenadParameters, MergeDeladdCboRoaringBitmaps};
+use crate::update::GrenadParameters;
 use crate::{bucketed_position, DocumentId, FieldId, Index, Result, MAX_POSITION_PER_ATTRIBUTE};
 
 const MAX_COUNTED_WORDS: usize = 30;
 
-pub struct WordDocidsCachedSorters {
+pub struct WordDocidsBalancedCaches<'extractor> {
-    word_fid_docids: CboCachedSorter<MergeDeladdCboRoaringBitmaps>,
+    word_fid_docids: BalancedCaches<'extractor>,
-    word_docids: CboCachedSorter<MergeDeladdCboRoaringBitmaps>,
+    word_docids: BalancedCaches<'extractor>,
-    exact_word_docids: CboCachedSorter<MergeDeladdCboRoaringBitmaps>,
+    exact_word_docids: BalancedCaches<'extractor>,
-    word_position_docids: CboCachedSorter<MergeDeladdCboRoaringBitmaps>,
+    word_position_docids: BalancedCaches<'extractor>,
-    fid_word_count_docids: CboCachedSorter<MergeDeladdCboRoaringBitmaps>,
+    fid_word_count_docids: BalancedCaches<'extractor>,
     fid_word_count: HashMap<FieldId, (usize, usize)>,
     current_docid: Option<DocumentId>,
 }
 
-impl WordDocidsCachedSorters {
+unsafe impl<'extractor> MostlySend for WordDocidsBalancedCaches<'extractor> {}
-    pub fn new(
-        indexer: GrenadParameters,
-        max_memory: Option<usize>,
-        capacity: NonZero<usize>,
-    ) -> Self {
-        let max_memory = max_memory.map(|max_memory| max_memory / 4);
-
-        let word_fid_docids = CboCachedSorter::new(
-            capacity,
-            create_sorter(
-                grenad::SortAlgorithm::Stable,
-                MergeDeladdCboRoaringBitmaps,
-                indexer.chunk_compression_type,
-                indexer.chunk_compression_level,
-                indexer.max_nb_chunks,
-                max_memory,
-                false,
-            ),
-        );
-        let word_docids = CboCachedSorter::new(
-            capacity,
-            create_sorter(
-                grenad::SortAlgorithm::Stable,
-                MergeDeladdCboRoaringBitmaps,
-                indexer.chunk_compression_type,
-                indexer.chunk_compression_level,
-                indexer.max_nb_chunks,
-                max_memory,
-                false,
-            ),
-        );
-        let exact_word_docids = CboCachedSorter::new(
-            capacity,
-            create_sorter(
-                grenad::SortAlgorithm::Stable,
-                MergeDeladdCboRoaringBitmaps,
-                indexer.chunk_compression_type,
-                indexer.chunk_compression_level,
-                indexer.max_nb_chunks,
-                max_memory,
-                false,
-            ),
-        );
-        let word_position_docids = CboCachedSorter::new(
-            capacity,
-            create_sorter(
-                grenad::SortAlgorithm::Stable,
-                MergeDeladdCboRoaringBitmaps,
-                indexer.chunk_compression_type,
-                indexer.chunk_compression_level,
-                indexer.max_nb_chunks,
-                max_memory,
-                false,
-            ),
-        );
-        let fid_word_count_docids = CboCachedSorter::new(
-            capacity,
-            create_sorter(
-                grenad::SortAlgorithm::Stable,
-                MergeDeladdCboRoaringBitmaps,
-                indexer.chunk_compression_type,
-                indexer.chunk_compression_level,
-                indexer.max_nb_chunks,
-                max_memory,
-                false,
-            ),
-        );
 
+impl<'extractor> WordDocidsBalancedCaches<'extractor> {
+    /// TODO Make sure to give the same max_memory to all of them, without splitting it
+    pub fn new_in(buckets: usize, max_memory: Option<usize>, alloc: &'extractor Bump) -> Self {
         Self {
-            word_fid_docids,
+            word_fid_docids: BalancedCaches::new_in(buckets, max_memory, alloc),
-            word_docids,
+            word_docids: BalancedCaches::new_in(buckets, max_memory, alloc),
-            exact_word_docids,
+            exact_word_docids: BalancedCaches::new_in(buckets, max_memory, alloc),
-            word_position_docids,
+            word_position_docids: BalancedCaches::new_in(buckets, max_memory, alloc),
-            fid_word_count_docids,
+            fid_word_count_docids: BalancedCaches::new_in(buckets, max_memory, alloc),
             fid_word_count: HashMap::new(),
             current_docid: None,
         }
@@ -198,6 +131,7 @@ impl WordDocidsCachedSorters {
             .entry(field_id)
             .and_modify(|(current_count, _new_count)| *current_count += 1)
             .or_insert((1, 0));
+
         self.current_docid = Some(docid);
 
         Ok(())
@@ -227,37 +161,29 @@ impl WordDocidsCachedSorters {
     }
 }
 
-struct WordDocidsMergerBuilders {
+pub struct WordDocidsCaches<'extractor> {
-    word_fid_docids: MergerBuilder<File, MergeDeladdCboRoaringBitmaps>,
+    pub word_docids: Vec<BalancedCaches<'extractor>>,
-    word_docids: MergerBuilder<File, MergeDeladdCboRoaringBitmaps>,
+    pub word_fid_docids: Vec<BalancedCaches<'extractor>>,
-    exact_word_docids: MergerBuilder<File, MergeDeladdCboRoaringBitmaps>,
+    pub exact_word_docids: Vec<BalancedCaches<'extractor>>,
-    word_position_docids: MergerBuilder<File, MergeDeladdCboRoaringBitmaps>,
+    pub word_position_docids: Vec<BalancedCaches<'extractor>>,
-    fid_word_count_docids: MergerBuilder<File, MergeDeladdCboRoaringBitmaps>,
+    pub fid_word_count_docids: Vec<BalancedCaches<'extractor>>,
 }
 
-pub struct WordDocidsMergers {
+impl<'extractor> WordDocidsCaches<'extractor> {
-    pub word_fid_docids: Merger<File, MergeDeladdCboRoaringBitmaps>,
-    pub word_docids: Merger<File, MergeDeladdCboRoaringBitmaps>,
-    pub exact_word_docids: Merger<File, MergeDeladdCboRoaringBitmaps>,
-    pub word_position_docids: Merger<File, MergeDeladdCboRoaringBitmaps>,
-    pub fid_word_count_docids: Merger<File, MergeDeladdCboRoaringBitmaps>,
-}
-
-impl WordDocidsMergerBuilders {
     fn new() -> Self {
         Self {
-            word_fid_docids: MergerBuilder::new(MergeDeladdCboRoaringBitmaps),
+            word_docids: Vec::new(),
-            word_docids: MergerBuilder::new(MergeDeladdCboRoaringBitmaps),
+            word_fid_docids: Vec::new(),
-            exact_word_docids: MergerBuilder::new(MergeDeladdCboRoaringBitmaps),
+            exact_word_docids: Vec::new(),
-            word_position_docids: MergerBuilder::new(MergeDeladdCboRoaringBitmaps),
+            word_position_docids: Vec::new(),
-            fid_word_count_docids: MergerBuilder::new(MergeDeladdCboRoaringBitmaps),
+            fid_word_count_docids: Vec::new(),
         }
     }
 
-    fn add_sorters(&mut self, other: WordDocidsCachedSorters) -> Result<()> {
+    fn push(&mut self, other: WordDocidsBalancedCaches<'extractor>) -> Result<()> {
-        let WordDocidsCachedSorters {
+        let WordDocidsBalancedCaches {
-            word_fid_docids,
             word_docids,
+            word_fid_docids,
             exact_word_docids,
             word_position_docids,
             fid_word_count_docids,
@@ -265,78 +191,37 @@ impl WordDocidsMergerBuilders {
             current_docid: _,
         } = other;
 
-        let mut word_fid_docids_readers = Ok(vec![]);
+        self.word_docids.push(word_docids);
-        let mut word_docids_readers = Ok(vec![]);
+        self.word_fid_docids.push(word_fid_docids);
-        let mut exact_word_docids_readers = Ok(vec![]);
+        self.exact_word_docids.push(exact_word_docids);
-        let mut word_position_docids_readers = Ok(vec![]);
+        self.word_position_docids.push(word_position_docids);
-        let mut fid_word_count_docids_readers = Ok(vec![]);
+        self.fid_word_count_docids.push(fid_word_count_docids);
-        rayon::scope(|s| {
-            s.spawn(|_| {
-                word_fid_docids_readers =
-                    word_fid_docids.into_sorter().and_then(|s| s.into_reader_cursors());
-            });
-            s.spawn(|_| {
-                word_docids_readers =
-                    word_docids.into_sorter().and_then(|s| s.into_reader_cursors());
-            });
-            s.spawn(|_| {
-                exact_word_docids_readers =
-                    exact_word_docids.into_sorter().and_then(|s| s.into_reader_cursors());
-            });
-            s.spawn(|_| {
-                word_position_docids_readers =
-                    word_position_docids.into_sorter().and_then(|s| s.into_reader_cursors());
-            });
-            s.spawn(|_| {
-                fid_word_count_docids_readers =
-                    fid_word_count_docids.into_sorter().and_then(|s| s.into_reader_cursors());
-            });
-        });
-        self.word_fid_docids.extend(word_fid_docids_readers?);
-        self.word_docids.extend(word_docids_readers?);
-        self.exact_word_docids.extend(exact_word_docids_readers?);
-        self.word_position_docids.extend(word_position_docids_readers?);
-        self.fid_word_count_docids.extend(fid_word_count_docids_readers?);
 
         Ok(())
     }
-
-    fn build(self) -> WordDocidsMergers {
-        WordDocidsMergers {
-            word_fid_docids: self.word_fid_docids.build(),
-            word_docids: self.word_docids.build(),
-            exact_word_docids: self.exact_word_docids.build(),
-            word_position_docids: self.word_position_docids.build(),
-            fid_word_count_docids: self.fid_word_count_docids.build(),
-        }
-    }
 }
 
-pub struct WordDocidsExtractorData<'extractor> {
+pub struct WordDocidsExtractorData<'a> {
-    tokenizer: &'extractor DocumentTokenizer<'extractor>,
+    tokenizer: &'a DocumentTokenizer<'a>,
     grenad_parameters: GrenadParameters,
-    max_memory: Option<usize>,
+    buckets: usize,
 }
 
-impl<'extractor> Extractor<'extractor> for WordDocidsExtractorData<'extractor> {
+impl<'a, 'extractor> Extractor<'extractor> for WordDocidsExtractorData<'a> {
-    type Data = FullySend<RefCell<WordDocidsCachedSorters>>;
+    type Data = RefCell<Option<WordDocidsBalancedCaches<'extractor>>>;
 
-    fn init_data(
+    fn init_data(&self, extractor_alloc: &'extractor Bump) -> Result<Self::Data> {
-        &self,
+        Ok(RefCell::new(Some(WordDocidsBalancedCaches::new_in(
-        _extractor_alloc: raw_collections::alloc::RefBump<'extractor>,
+            self.buckets,
-    ) -> Result<Self::Data> {
+            self.grenad_parameters.max_memory,
-        Ok(FullySend(RefCell::new(WordDocidsCachedSorters::new(
+            extractor_alloc,
-            self.grenad_parameters,
-            self.max_memory,
-            // TODO use a better value
-            200_000.try_into().unwrap(),
         ))))
     }
 
     fn process(
         &self,
         change: DocumentChange,
-        context: &crate::update::new::indexer::document_changes::DocumentChangeContext<Self::Data>,
+        context: &DocumentChangeContext<Self::Data>,
     ) -> Result<()> {
         WordDocidsExtractors::extract_document_change(context, self.tokenizer, change)
     }
@@ -345,16 +230,15 @@ impl<'extractor> Extractor<'extractor> for WordDocidsExtractorData<'extractor> {
 pub struct WordDocidsExtractors;
 
 impl WordDocidsExtractors {
-    pub fn run_extraction<'pl, 'fid, 'indexer, 'index, DC: DocumentChanges<'pl>>(
+    pub fn run_extraction<'pl, 'fid, 'indexer, 'index, 'extractor, DC: DocumentChanges<'pl>>(
         grenad_parameters: GrenadParameters,
         document_changes: &DC,
         indexing_context: IndexingContext<'fid, 'indexer, 'index>,
-        extractor_allocs: &mut ThreadLocal<FullySend<RefCell<Bump>>>,
+        extractor_allocs: &'extractor mut ThreadLocal<FullySend<Bump>>,
-    ) -> Result<WordDocidsMergers> {
+    ) -> Result<WordDocidsCaches<'extractor>> {
-        let max_memory = grenad_parameters.max_memory_by_thread();
         let index = indexing_context.index;
-
         let rtxn = index.read_txn()?;
+
         let stop_words = index.stop_words(&rtxn)?;
         let allowed_separators = index.allowed_separators(&rtxn)?;
         let allowed_separators: Option<Vec<_>> =
@@ -392,7 +276,7 @@ impl WordDocidsExtractors {
             let extractor = WordDocidsExtractorData {
                 tokenizer: &document_tokenizer,
                 grenad_parameters,
-                max_memory,
+                buckets: rayon::current_num_threads(),
             };
 
             for_each_document_change(
@@ -404,28 +288,23 @@ impl WordDocidsExtractors {
             )?;
         }
 
-        {
+        let mut merger = WordDocidsCaches::new();
-            let span =
+        for cache in datastore.into_iter().flat_map(RefCell::into_inner) {
-                tracing::trace_span!(target: "indexing::documents::extract", "merger_building");
+            merger.push(cache)?;
-            let _entered = span.enter();
-            let mut builder = WordDocidsMergerBuilders::new();
-            for cache in datastore.into_iter().map(|cache| cache.0.into_inner()) {
-                builder.add_sorters(cache)?;
-            }
-
-            Ok(builder.build())
         }
+
+        Ok(merger)
     }
 
     fn extract_document_change(
-        context: &DocumentChangeContext<FullySend<RefCell<WordDocidsCachedSorters>>>,
+        context: &DocumentChangeContext<RefCell<Option<WordDocidsBalancedCaches>>>,
         document_tokenizer: &DocumentTokenizer,
         document_change: DocumentChange,
     ) -> Result<()> {
         let index = &context.index;
         let rtxn = &context.txn;
-        let mut cached_sorter = context.data.0.borrow_mut_or_yield();
+        let mut cached_sorter_ref = context.data.borrow_mut_or_yield();
-        let cached_sorter = cached_sorter.deref_mut();
+        let cached_sorter = cached_sorter_ref.as_mut().unwrap();
         let mut new_fields_ids_map = context.new_fields_ids_map.borrow_mut_or_yield();
         let new_fields_ids_map = new_fields_ids_map.deref_mut();
         let doc_alloc = &context.doc_alloc;
@@ -436,16 +315,14 @@ impl WordDocidsExtractors {
         match document_change {
             DocumentChange::Deletion(inner) => {
                 let mut token_fn = |fname: &str, fid, pos, word: &str| {
-                    cached_sorter
+                    cached_sorter.insert_del_u32(
-                        .insert_del_u32(
+                        fid,
-                            fid,
+                        pos,
-                            pos,
+                        word,
-                            word,
+                        is_exact_attribute(fname),
-                            is_exact_attribute(fname),
+                        inner.docid(),
-                            inner.docid(),
+                        doc_alloc,
-                            doc_alloc,
+                    )
-                        )
-                        .map_err(crate::Error::from)
                 };
                 document_tokenizer.tokenize_document(
                     inner.current(rtxn, index, context.db_fields_ids_map)?,
@@ -455,16 +332,14 @@ impl WordDocidsExtractors {
             }
             DocumentChange::Update(inner) => {
                 let mut token_fn = |fname: &str, fid, pos, word: &str| {
-                    cached_sorter
+                    cached_sorter.insert_del_u32(
-                        .insert_del_u32(
+                        fid,
-                            fid,
+                        pos,
-                            pos,
+                        word,
-                            word,
+                        is_exact_attribute(fname),
-                            is_exact_attribute(fname),
+                        inner.docid(),
-                            inner.docid(),
+                        doc_alloc,
-                            doc_alloc,
+                    )
-                        )
-                        .map_err(crate::Error::from)
                 };
                 document_tokenizer.tokenize_document(
                     inner.current(rtxn, index, context.db_fields_ids_map)?,
@@ -473,16 +348,14 @@ impl WordDocidsExtractors {
                 )?;
 
                 let mut token_fn = |fname: &str, fid, pos, word: &str| {
-                    cached_sorter
+                    cached_sorter.insert_add_u32(
-                        .insert_add_u32(
+                        fid,
-                            fid,
+                        pos,
-                            pos,
+                        word,
-                            word,
+                        is_exact_attribute(fname),
-                            is_exact_attribute(fname),
+                        inner.docid(),
-                            inner.docid(),
+                        doc_alloc,
-                            doc_alloc,
+                    )
-                        )
-                        .map_err(crate::Error::from)
                 };
                 document_tokenizer.tokenize_document(
                     inner.new(rtxn, index, context.db_fields_ids_map)?,
@@ -492,16 +365,14 @@ impl WordDocidsExtractors {
             }
             DocumentChange::Insertion(inner) => {
                 let mut token_fn = |fname: &str, fid, pos, word: &str| {
-                    cached_sorter
+                    cached_sorter.insert_add_u32(
-                        .insert_add_u32(
+                        fid,
-                            fid,
+                        pos,
-                            pos,
+                        word,
-                            word,
+                        is_exact_attribute(fname),
-                            is_exact_attribute(fname),
+                        inner.docid(),
-                            inner.docid(),
+                        doc_alloc,
-                            doc_alloc,
+                    )
-                        )
-                        .map_err(crate::Error::from)
                 };
                 document_tokenizer.tokenize_document(
                     inner.new(),

milli/src/update/new/extract/searchable/extract_word_pair_proximity_docids.rs

@@ -8,13 +8,13 @@ use super::tokenize_document::DocumentTokenizer;
 use super::SearchableExtractor;
 use crate::proximity::{index_proximity, MAX_DISTANCE};
 use crate::update::new::document::Document;
-use crate::update::new::extract::cache::CboCachedSorter;
+use crate::update::new::extract::cache::BalancedCaches;
-use crate::update::new::indexer::document_changes::{DocumentChangeContext, FullySend, RefCellExt};
+use crate::update::new::indexer::document_changes::{DocumentChangeContext, RefCellExt};
 use crate::update::new::DocumentChange;
-use crate::update::MergeDeladdCboRoaringBitmaps;
 use crate::{FieldId, GlobalFieldsIdsMap, Index, Result};
 
 pub struct WordPairProximityDocidsExtractor;
+
 impl SearchableExtractor for WordPairProximityDocidsExtractor {
     fn attributes_to_extract<'a>(
         rtxn: &'a RoTxn,
@@ -28,11 +28,10 @@ impl SearchableExtractor for WordPairProximityDocidsExtractor {
     }
 
     // This method is reimplemented to count the number of words in the document in each field
-    // and to store the docids of the documents that have a number of words in a given field equal to or under than MAX_COUNTED_WORDS.
+    // and to store the docids of the documents that have a number of words in a given field
+    // equal to or under than MAX_COUNTED_WORDS.
     fn extract_document_change(
-        context: &DocumentChangeContext<
+        context: &DocumentChangeContext<RefCell<BalancedCaches>>,
-            FullySend<RefCell<CboCachedSorter<MergeDeladdCboRoaringBitmaps>>>,
-        >,
         document_tokenizer: &DocumentTokenizer,
         document_change: DocumentChange,
     ) -> Result<()> {
@@ -48,7 +47,7 @@ impl SearchableExtractor for WordPairProximityDocidsExtractor {
         let mut new_fields_ids_map = context.new_fields_ids_map.borrow_mut_or_yield();
         let new_fields_ids_map = &mut *new_fields_ids_map;
 
-        let mut cached_sorter = context.data.0.borrow_mut_or_yield();
+        let mut cached_sorter = context.data.borrow_mut_or_yield();
         let cached_sorter = &mut *cached_sorter;
 
         // is a vecdequeue, and will be smol, so can stay on the heap for now
@@ -139,7 +138,7 @@ fn build_key<'a>(
 fn word_positions_into_word_pair_proximity(
     word_positions: &mut VecDeque<(Rc<str>, u16)>,
     word_pair_proximity: &mut impl FnMut((Rc<str>, Rc<str>), u8),
-) -> Result<()> {
+) {
     let (head_word, head_position) = word_positions.pop_front().unwrap();
     for (word, position) in word_positions.iter() {
         let prox = index_proximity(head_position as u32, *position as u32) as u8;
@@ -147,7 +146,6 @@ fn word_positions_into_word_pair_proximity(
             word_pair_proximity((head_word.clone(), word.clone()), prox);
         }
     }
-    Ok(())
 }
 
 fn process_document_tokens<'doc>(
@@ -163,7 +161,7 @@ fn process_document_tokens<'doc>(
             .front()
             .map_or(false, |(_w, p)| index_proximity(*p as u32, pos as u32) >= MAX_DISTANCE)
         {
-            word_positions_into_word_pair_proximity(word_positions, word_pair_proximity)?;
+            word_positions_into_word_pair_proximity(word_positions, word_pair_proximity);
         }
 
         // insert the new word.
@@ -173,7 +171,7 @@ fn process_document_tokens<'doc>(
     document_tokenizer.tokenize_document(document, fields_ids_map, &mut token_fn)?;
 
     while !word_positions.is_empty() {
-        word_positions_into_word_pair_proximity(word_positions, word_pair_proximity)?;
+        word_positions_into_word_pair_proximity(word_positions, word_pair_proximity);
     }
 
     Ok(())

milli/src/update/new/extract/searchable/mod.rs

@@ -3,76 +3,60 @@ mod extract_word_pair_proximity_docids;
 mod tokenize_document;
 
 use std::cell::RefCell;
-use std::fs::File;
 use std::marker::PhantomData;
 
 use bumpalo::Bump;
-pub use extract_word_docids::{WordDocidsExtractors, WordDocidsMergers};
+pub use extract_word_docids::{WordDocidsCaches, WordDocidsExtractors};
 pub use extract_word_pair_proximity_docids::WordPairProximityDocidsExtractor;
-use grenad::Merger;
 use heed::RoTxn;
-use rayon::iter::{ParallelBridge, ParallelIterator};
 use tokenize_document::{tokenizer_builder, DocumentTokenizer};
 
-use super::cache::CboCachedSorter;
+use super::cache::BalancedCaches;
 use super::DocidsExtractor;
 use crate::update::new::indexer::document_changes::{
     for_each_document_change, DocumentChangeContext, DocumentChanges, Extractor, FullySend,
     IndexingContext, ThreadLocal,
 };
 use crate::update::new::DocumentChange;
-use crate::update::{create_sorter, GrenadParameters, MergeDeladdCboRoaringBitmaps};
+use crate::update::GrenadParameters;
 use crate::{Index, Result, MAX_POSITION_PER_ATTRIBUTE};
 
-pub struct SearchableExtractorData<'extractor, EX: SearchableExtractor> {
+pub struct SearchableExtractorData<'a, EX: SearchableExtractor> {
-    tokenizer: &'extractor DocumentTokenizer<'extractor>,
+    tokenizer: &'a DocumentTokenizer<'a>,
     grenad_parameters: GrenadParameters,
-    max_memory: Option<usize>,
+    buckets: usize,
     _ex: PhantomData<EX>,
 }
 
-impl<'extractor, EX: SearchableExtractor + Sync> Extractor<'extractor>
+impl<'a, 'extractor, EX: SearchableExtractor + Sync> Extractor<'extractor>
-    for SearchableExtractorData<'extractor, EX>
+    for SearchableExtractorData<'a, EX>
 {
-    type Data = FullySend<RefCell<CboCachedSorter<MergeDeladdCboRoaringBitmaps>>>;
+    type Data = RefCell<BalancedCaches<'extractor>>;
 
-    fn init_data(
+    fn init_data(&self, extractor_alloc: &'extractor Bump) -> Result<Self::Data> {
-        &self,
+        Ok(RefCell::new(BalancedCaches::new_in(
-        _extractor_alloc: raw_collections::alloc::RefBump<'extractor>,
+            self.buckets,
-    ) -> Result<Self::Data> {
+            self.grenad_parameters.max_memory,
-        Ok(FullySend(RefCell::new(CboCachedSorter::new(
+            extractor_alloc,
-            // TODO use a better value
+        )))
-            1_000_000.try_into().unwrap(),
-            create_sorter(
-                grenad::SortAlgorithm::Stable,
-                MergeDeladdCboRoaringBitmaps,
-                self.grenad_parameters.chunk_compression_type,
-                self.grenad_parameters.chunk_compression_level,
-                self.grenad_parameters.max_nb_chunks,
-                self.max_memory,
-                false,
-            ),
-        ))))
     }
 
     fn process(
         &self,
         change: DocumentChange,
-        context: &crate::update::new::indexer::document_changes::DocumentChangeContext<Self::Data>,
+        context: &DocumentChangeContext<Self::Data>,
     ) -> Result<()> {
         EX::extract_document_change(context, self.tokenizer, change)
     }
 }
 
 pub trait SearchableExtractor: Sized + Sync {
-    fn run_extraction<'pl, 'fid, 'indexer, 'index, DC: DocumentChanges<'pl>>(
+    fn run_extraction<'pl, 'fid, 'indexer, 'index, 'extractor, DC: DocumentChanges<'pl>>(
         grenad_parameters: GrenadParameters,
         document_changes: &DC,
         indexing_context: IndexingContext<'fid, 'indexer, 'index>,
-        extractor_allocs: &mut ThreadLocal<FullySend<RefCell<Bump>>>,
+        extractor_allocs: &'extractor mut ThreadLocal<FullySend<Bump>>,
-    ) -> Result<Merger<File, MergeDeladdCboRoaringBitmaps>> {
+    ) -> Result<Vec<BalancedCaches<'extractor>>> {
-        let max_memory = grenad_parameters.max_memory_by_thread();
-
         let rtxn = indexing_context.index.read_txn()?;
         let stop_words = indexing_context.index.stop_words(&rtxn)?;
         let allowed_separators = indexing_context.index.allowed_separators(&rtxn)?;
@@ -104,7 +88,7 @@ pub trait SearchableExtractor: Sized + Sync {
         let extractor_data: SearchableExtractorData<Self> = SearchableExtractorData {
             tokenizer: &document_tokenizer,
             grenad_parameters,
-            max_memory,
+            buckets: rayon::current_num_threads(),
             _ex: PhantomData,
         };
 
@@ -122,37 +106,12 @@ pub trait SearchableExtractor: Sized + Sync {
                 &datastore,
             )?;
         }
-        {
-            let mut builder = grenad::MergerBuilder::new(MergeDeladdCboRoaringBitmaps);
-            let span =
-                tracing::trace_span!(target: "indexing::documents::extract", "merger_building");
-            let _entered = span.enter();
 
-            let readers: Vec<_> = datastore
+        Ok(datastore.into_iter().map(RefCell::into_inner).collect())
-                .into_iter()
-                .par_bridge()
-                .map(|cache_entry| {
-                    let cached_sorter: FullySend<
-                        RefCell<CboCachedSorter<MergeDeladdCboRoaringBitmaps>>,
-                    > = cache_entry;
-                    let cached_sorter = cached_sorter.0.into_inner();
-                    let sorter = cached_sorter.into_sorter()?;
-                    sorter.into_reader_cursors()
-                })
-                .collect();
-
-            for reader in readers {
-                builder.extend(reader?);
-            }
-
-            Ok(builder.build())
-        }
     }
 
     fn extract_document_change(
-        context: &DocumentChangeContext<
+        context: &DocumentChangeContext<RefCell<BalancedCaches>>,
-            FullySend<RefCell<CboCachedSorter<MergeDeladdCboRoaringBitmaps>>>,
-        >,
         document_tokenizer: &DocumentTokenizer,
         document_change: DocumentChange,
     ) -> Result<()>;
@@ -164,12 +123,12 @@ pub trait SearchableExtractor: Sized + Sync {
 }
 
 impl<T: SearchableExtractor> DocidsExtractor for T {
-    fn run_extraction<'pl, 'fid, 'indexer, 'index, DC: DocumentChanges<'pl>>(
+    fn run_extraction<'pl, 'fid, 'indexer, 'index, 'extractor, DC: DocumentChanges<'pl>>(
         grenad_parameters: GrenadParameters,
         document_changes: &DC,
         indexing_context: IndexingContext<'fid, 'indexer, 'index>,
-        extractor_allocs: &mut ThreadLocal<FullySend<RefCell<Bump>>>,
+        extractor_allocs: &'extractor mut ThreadLocal<FullySend<Bump>>,
-    ) -> Result<Merger<File, MergeDeladdCboRoaringBitmaps>> {
+    ) -> Result<Vec<BalancedCaches<'extractor>>> {
         Self::run_extraction(
             grenad_parameters,
             document_changes,

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

@@ -171,7 +171,6 @@ mod test {
     use bumpalo::Bump;
     use charabia::TokenizerBuilder;
     use meili_snap::snapshot;
-    
     use raw_collections::RawMap;
     use serde_json::json;
     use serde_json::value::RawValue;

milli/src/update/new/facet_search_builder.rs

@@ -1,31 +1,24 @@
 use std::collections::{BTreeSet, HashMap};
 
-use charabia::{normalizer::NormalizerOption, Language, Normalize, StrDetection, Token};
+use charabia::normalizer::NormalizerOption;
+use charabia::{Language, Normalize, StrDetection, Token};
 use grenad::Sorter;
-use heed::{
+use heed::types::{Bytes, SerdeJson};
-    types::{Bytes, SerdeJson},
+use heed::{BytesDecode, BytesEncode, RoTxn, RwTxn};
-    BytesDecode, BytesEncode, RoTxn,
-};
 
+use super::channel::FacetSearchableSender;
+use super::extract::FacetKind;
+use super::fst_merger_builder::FstMergerBuilder;
+use super::KvReaderDelAdd;
+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::{
-    heed_codec::{
-        facet::{FacetGroupKey, FacetGroupKeyCodec},
-        StrRefCodec,
-    },
-    update::{
-        create_sorter,
-        del_add::{DelAdd, KvWriterDelAdd},
-        MergeDeladdBtreesetString,
-    },
     BEU16StrCodec, FieldId, GlobalFieldsIdsMap, Index, LocalizedAttributesRule, Result,
     MAX_FACET_VALUE_LENGTH,
 };
 
-use super::{
-    channel::FacetSearchableSender, extract::FacetKind, fst_merger_builder::FstMergerBuilder,
-    KvReaderDelAdd,
-};
-
 pub struct FacetSearchBuilder<'indexer> {
     registered_facets: HashMap<FieldId, usize>,
     normalized_facet_string_docids_sorter: Sorter<MergeDeladdBtreesetString>,
@@ -49,6 +42,7 @@ impl<'indexer> FacetSearchBuilder<'indexer> {
             None,
             None,
             Some(0),
+            false,
         );
 
         Self {
@@ -84,7 +78,7 @@ impl<'indexer> FacetSearchBuilder<'indexer> {
         }
 
         let locales = self.locales(field_id);
-        let hyper_normalized_value = normalize_facet_string(left_bound, locales.as_deref());
+        let hyper_normalized_value = normalize_facet_string(left_bound, locales);
 
         let set = BTreeSet::from_iter(std::iter::once(left_bound));
 
@@ -103,7 +97,7 @@ impl<'indexer> FacetSearchBuilder<'indexer> {
     }
 
     fn locales(&mut self, field_id: FieldId) -> Option<&[Language]> {
-        if self.localized_field_ids.get(&field_id).is_none() {
+        if !self.localized_field_ids.contains_key(&field_id) {
             let Some(field_name) = self.global_fields_ids_map.name(field_id) else {
                 unreachable!("Field id {} not found in the global fields ids map", field_id);
             };
@@ -124,7 +118,8 @@ impl<'indexer> FacetSearchBuilder<'indexer> {
     pub fn merge_and_send(
         self,
         index: &Index,
-        rtxn: &RoTxn<'_>,
+        wtxn: &mut RwTxn,
+        rtxn: &RoTxn,
         sender: FacetSearchableSender,
     ) -> Result<()> {
         let reader = self.normalized_facet_string_docids_sorter.into_reader_cursors()?;
@@ -139,13 +134,14 @@ impl<'indexer> FacetSearchBuilder<'indexer> {
         let mut fst_merger_builder: Option<FstMergerBuilder> = None;
         while let Some((key, deladd)) = merger_iter.next()? {
             let (field_id, normalized_facet_string) =
-                BEU16StrCodec::bytes_decode(&key).map_err(heed::Error::Encoding)?;
+                BEU16StrCodec::bytes_decode(key).map_err(heed::Error::Encoding)?;
 
             if current_field_id != Some(field_id) {
                 if let Some(fst_merger_builder) = fst_merger_builder {
                     // send the previous fst to the channel
                     let mmap = fst_merger_builder.build(&mut callback)?;
-                    sender.write_fst(&field_id.to_be_bytes(), mmap).unwrap();
+                    // sender.write_fst(&field_id.to_be_bytes(), mmap).unwrap();
+                    todo!("What to do");
                 }
 
                 println!("getting fst for field_id: {}", field_id);
@@ -198,7 +194,8 @@ impl<'indexer> FacetSearchBuilder<'indexer> {
 
         if let (Some(field_id), Some(fst_merger_builder)) = (current_field_id, fst_merger_builder) {
             let mmap = fst_merger_builder.build(&mut callback)?;
-            sender.write_fst(&field_id.to_be_bytes(), mmap).unwrap();
+            // sender.write_fst(&field_id.to_be_bytes(), mmap).unwrap();
+            todo!("What to do");
         }
 
         Ok(())
@@ -209,7 +206,7 @@ fn callback(_bytes: &[u8], _deladd: DelAdd, _is_modified: bool) -> Result<()> {
     Ok(())
 }
 
-fn merge_btreesets<'a>(
+fn merge_btreesets(
     current: Option<&[u8]>,
     del: Option<&[u8]>,
     add: Option<&[u8]>,

milli/src/update/new/indexer/de.rs

@@ -49,7 +49,7 @@ impl<'de, 'p, 'indexer: 'de, Mapper: MutFieldIdMapper> Visitor<'de>
                 visitor: MutFieldIdMapVisitor(self.fields_ids_map),
             })?
         {
-            let Some(fid) = fid else {
+            let Some(_fid) = fid else {
                 return Ok(Err(crate::UserError::AttributeLimitReached));
             };
             self.fields_ids_map = fields_ids_map;

milli/src/update/new/indexer/document_changes.rs

@@ -3,7 +3,6 @@ use std::sync::{Arc, RwLock};
 
 use bumpalo::Bump;
 use heed::RoTxn;
-use raw_collections::alloc::RefBump;
 use rayon::iter::IndexedParallelIterator;
 
 use super::super::document_change::DocumentChange;
@@ -104,6 +103,10 @@ pub struct FullySend<T>(pub T);
 // SAFETY: a type **fully** send is always mostly send as well.
 unsafe impl<T> MostlySend for FullySend<T> where T: Send {}
 
+unsafe impl<T> MostlySend for RefCell<T> where T: MostlySend {}
+
+unsafe impl<T> MostlySend for Option<T> where T: MostlySend {}
+
 impl<T> FullySend<T> {
     pub fn into(self) -> T {
         self.0
@@ -256,7 +259,7 @@ pub struct DocumentChangeContext<
     pub doc_alloc: Bump,
 
     /// Data allocated in this allocator is not cleared between each call to `process`, unless the data spills.
-    pub extractor_alloc: RefBump<'extractor>,
+    pub extractor_alloc: &'extractor Bump,
 
     /// Pool of doc allocators, used to retrieve the doc allocator we provided for the documents
     doc_allocs: &'doc ThreadLocal<FullySend<Cell<Bump>>>,
@@ -279,14 +282,14 @@ impl<
         index: &'indexer Index,
         db_fields_ids_map: &'indexer FieldsIdsMap,
         new_fields_ids_map: &'fid RwLock<FieldsIdsMap>,
-        extractor_allocs: &'extractor ThreadLocal<FullySend<RefCell<Bump>>>,
+        extractor_allocs: &'extractor ThreadLocal<FullySend<Bump>>,
         doc_allocs: &'doc ThreadLocal<FullySend<Cell<Bump>>>,
         datastore: &'data ThreadLocal<T>,
         fields_ids_map_store: &'doc ThreadLocal<FullySend<RefCell<GlobalFieldsIdsMap<'fid>>>>,
         init_data: F,
     ) -> Result<Self>
     where
-        F: FnOnce(RefBump<'extractor>) -> Result<T>,
+        F: FnOnce(&'extractor Bump) -> Result<T>,
     {
         let doc_alloc =
             doc_allocs.get_or(|| FullySend(Cell::new(Bump::with_capacity(1024 * 1024 * 1024))));
@@ -297,9 +300,7 @@ impl<
         let fields_ids_map = &fields_ids_map.0;
         let extractor_alloc = extractor_allocs.get_or_default();
 
-        let extractor_alloc = RefBump::new(extractor_alloc.0.borrow_or_yield());
+        let data = datastore.get_or_try(move || init_data(&extractor_alloc.0))?;
-
-        let data = datastore.get_or_try(|| init_data(RefBump::clone(&extractor_alloc)))?;
 
         let txn = index.read_txn()?;
         Ok(DocumentChangeContext {
@@ -308,7 +309,7 @@ impl<
             db_fields_ids_map,
             new_fields_ids_map: fields_ids_map,
             doc_alloc,
-            extractor_alloc,
+            extractor_alloc: &extractor_alloc.0,
             data,
             doc_allocs,
         })
@@ -319,7 +320,7 @@ impl<
 pub trait Extractor<'extractor>: Sync {
     type Data: MostlySend;
 
-    fn init_data<'doc>(&'doc self, extractor_alloc: RefBump<'extractor>) -> Result<Self::Data>;
+    fn init_data<'doc>(&'doc self, extractor_alloc: &'extractor Bump) -> Result<Self::Data>;
 
     fn process<'doc>(
         &'doc self,
@@ -375,15 +376,17 @@ pub fn for_each_document_change<
         doc_allocs,
         fields_ids_map_store,
     }: IndexingContext<'fid, 'indexer, 'index>,
-    extractor_allocs: &'extractor mut ThreadLocal<FullySend<RefCell<Bump>>>,
+    extractor_allocs: &'extractor mut ThreadLocal<FullySend<Bump>>,
     datastore: &'data ThreadLocal<EX::Data>,
 ) -> Result<()>
 where
     EX: Extractor<'extractor>,
 {
+    eprintln!("We are resetting the extractor allocators");
     // Clean up and reuse the extractor allocs
     for extractor_alloc in extractor_allocs.iter_mut() {
-        extractor_alloc.0.get_mut().reset();
+        eprintln!("\tWith {} bytes resetted", extractor_alloc.0.allocated_bytes());
+        extractor_alloc.0.reset();
     }
 
     let pi = document_changes.iter();

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

@@ -80,7 +80,6 @@ mod test {
     use std::sync::RwLock;
 
     use bumpalo::Bump;
-    use raw_collections::alloc::RefBump;
 
     use crate::index::tests::TempIndex;
     use crate::update::new::indexer::document_changes::{
@@ -95,11 +94,7 @@ mod test {
     fn test_deletions() {
         struct DeletionWithData<'extractor> {
             deleted: RefCell<
-                hashbrown::HashSet<
+                hashbrown::HashSet<DocumentId, hashbrown::DefaultHashBuilder, &'extractor Bump>,
-                    DocumentId,
-                    hashbrown::hash_map::DefaultHashBuilder,
-                    RefBump<'extractor>,
-                >,
             >,
         }
 
@@ -110,10 +105,7 @@ mod test {
         impl<'extractor> Extractor<'extractor> for TrackDeletion<'extractor> {
             type Data = DeletionWithData<'extractor>;
 
-            fn init_data(
+            fn init_data(&self, extractor_alloc: &'extractor Bump) -> crate::Result<Self::Data> {
-                &self,
-                extractor_alloc: raw_collections::alloc::RefBump<'extractor>,
-            ) -> crate::Result<Self::Data> {
                 let deleted = RefCell::new(hashbrown::HashSet::new_in(extractor_alloc));
                 Ok(DeletionWithData { deleted })
             }
@@ -173,8 +165,7 @@ mod test {
                 println!("deleted by {index}: {:?}", data.deleted.borrow());
             }
             for alloc in extractor_allocs.iter_mut() {
-                let alloc = &mut alloc.0;
+                alloc.0.reset();
-                alloc.get_mut().reset();
             }
         }
     }

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

@@ -1,26 +1,24 @@
-use std::cell::RefCell;
+use std::cmp::Ordering;
 use std::sync::RwLock;
 use std::thread::{self, Builder};
 
 use big_s::S;
-use bumpalo::Bump;
 use document_changes::{
-    for_each_document_change, DocumentChanges, Extractor, FullySend, IndexingContext, RefCellExt,
+    for_each_document_change, DocumentChanges, FullySend, IndexingContext, ThreadLocal,
-    ThreadLocal,
 };
 pub use document_deletion::DocumentDeletion;
 pub use document_operation::DocumentOperation;
+use heed::types::{Bytes, DecodeIgnore, Str};
 use heed::{RoTxn, RwTxn};
+use itertools::{merge_join_by, EitherOrBoth};
 pub use partial_dump::PartialDump;
 use rayon::ThreadPool;
 use time::OffsetDateTime;
 pub use update_by_function::UpdateByFunction;
 
 use super::channel::*;
-use super::document::write_to_obkv;
-use super::document_change::DocumentChange;
 use super::extract::*;
-use super::merger::{merge_grenad_entries, FacetFieldIdsDelta};
+use super::merger::{FacetDatabases, FacetFieldIdsDelta};
 use super::word_fst_builder::PrefixDelta;
 use super::words_prefix_docids::{
     compute_word_prefix_docids, compute_word_prefix_fid_docids, compute_word_prefix_position_docids,
@@ -28,75 +26,23 @@ use super::words_prefix_docids::{
 use super::{StdResult, TopLevelMap};
 use crate::documents::{PrimaryKey, DEFAULT_PRIMARY_KEY};
 use crate::facet::FacetType;
+use crate::index::main_key::{WORDS_FST_KEY, WORDS_PREFIXES_FST_KEY};
 use crate::proximity::ProximityPrecision;
-use crate::update::new::channel::ExtractorSender;
+use crate::update::del_add::DelAdd;
+use crate::update::new::word_fst_builder::{PrefixData, WordFstBuilder};
 use crate::update::new::words_prefix_docids::compute_exact_word_prefix_docids;
+use crate::update::new::{merge_and_send_docids, merge_and_send_facet_docids};
 use crate::update::settings::InnerIndexSettings;
 use crate::update::{FacetsUpdateBulk, GrenadParameters};
-use crate::{Error, FieldsIdsMap, GlobalFieldsIdsMap, Index, Result, UserError};
+use crate::{FieldsIdsMap, GlobalFieldsIdsMap, Index, Result, UserError};
 
-pub(crate) mod de;
+pub mod de;
 pub mod document_changes;
 mod document_deletion;
 mod document_operation;
 mod partial_dump;
 mod update_by_function;
 
-struct DocumentExtractor<'a> {
-    document_sender: &'a DocumentSender<'a>,
-}
-
-impl<'a, 'extractor> Extractor<'extractor> for DocumentExtractor<'a> {
-    type Data = FullySend<()>;
-
-    fn init_data(
-        &self,
-        _extractor_alloc: raw_collections::alloc::RefBump<'extractor>,
-    ) -> Result<Self::Data> {
-        Ok(FullySend(()))
-    }
-
-    fn process(
-        &self,
-        change: DocumentChange,
-        context: &document_changes::DocumentChangeContext<Self::Data>,
-    ) -> Result<()> {
-        let mut document_buffer = Vec::new();
-
-        let new_fields_ids_map = context.new_fields_ids_map.borrow_or_yield();
-        let new_fields_ids_map = &*new_fields_ids_map;
-        let new_fields_ids_map = new_fields_ids_map.local_map();
-
-        let external_docid = change.external_docid().to_owned();
-
-        // document but we need to create a function that collects and compresses documents.
-        match change {
-            DocumentChange::Deletion(deletion) => {
-                let docid = deletion.docid();
-                self.document_sender.delete(docid, external_docid).unwrap();
-            }
-            /// TODO: change NONE by SOME(vector) when implemented
-            DocumentChange::Update(update) => {
-                let docid = update.docid();
-                let content =
-                    update.new(&context.txn, context.index, &context.db_fields_ids_map)?;
-                let content =
-                    write_to_obkv(&content, None, new_fields_ids_map, &mut document_buffer)?;
-                self.document_sender.insert(docid, external_docid, content.boxed()).unwrap();
-            }
-            DocumentChange::Insertion(insertion) => {
-                let docid = insertion.docid();
-                let content = insertion.new();
-                let content =
-                    write_to_obkv(&content, None, new_fields_ids_map, &mut document_buffer)?;
-                self.document_sender.insert(docid, external_docid, content.boxed()).unwrap();
-                // extracted_dictionary_sender.send(self, dictionary: &[u8]);
-            }
-        }
-        Ok(())
-    }
-}
-
 /// This is the main function of this crate.
 ///
 /// Give it the output of the [`Indexer::document_changes`] method and it will execute it in the [`rayon::ThreadPool`].
@@ -114,12 +60,11 @@ pub fn index<'pl, 'indexer, 'index, DC>(
 where
     DC: DocumentChanges<'pl>,
 {
-    let (merger_sender, writer_receiver) = merger_writer_channel(10_000);
+    // TODO find a better channel limit
-    // This channel acts as a rendezvous point to ensure that we are one task ahead
+    let (extractor_sender, writer_receiver) = extractor_writer_channel(10_000);
-    let (extractor_sender, merger_receiver) = extractors_merger_channels(4);
-
     let new_fields_ids_map = RwLock::new(new_fields_ids_map);
 
+    let global_fields_ids_map = GlobalFieldsIdsMap::new(&new_fields_ids_map);
     let fields_ids_map_store = ThreadLocal::with_capacity(pool.current_num_threads());
     let mut extractor_allocs = ThreadLocal::with_capacity(pool.current_num_threads());
     let doc_allocs = ThreadLocal::with_capacity(pool.current_num_threads());
@@ -132,124 +77,171 @@ where
         fields_ids_map_store: &fields_ids_map_store,
     };
 
-    thread::scope(|s| {
+    thread::scope(|s| -> crate::Result<_> {
         let indexer_span = tracing::Span::current();
         // TODO manage the errors correctly
-        let handle = Builder::new().name(S("indexer-extractors")).spawn_scoped(s, move || {
+        let extractor_handle = Builder::new().name(S("indexer-extractors")).spawn_scoped(s, move || {
             pool.in_place_scope(|_s| {
-                    let span = tracing::trace_span!(target: "indexing::documents", parent: &indexer_span, "extract");
+                let span = tracing::trace_span!(target: "indexing::documents", parent: &indexer_span, "extract");
+                let _entered = span.enter();
+
+                // document but we need to create a function that collects and compresses documents.
+                let rtxn = index.read_txn().unwrap();
+                let document_sender = extractor_sender.documents();
+                let document_extractor = DocumentsExtractor::new(&document_sender);
+                let datastore = ThreadLocal::with_capacity(pool.current_num_threads());
+                for_each_document_change(document_changes, &document_extractor, indexing_context, &mut extractor_allocs, &datastore)?;
+
+                let mut documents_ids = index.documents_ids(&rtxn)?;
+                let delta_documents_ids = datastore.into_iter().map(|FullySend(d)| d.into_inner()).reduce(DelAddRoaringBitmap::merge).unwrap_or_default();
+                delta_documents_ids.apply_to(&mut documents_ids);
+                extractor_sender.send_documents_ids(documents_ids).unwrap();
+
+                // document_sender.finish().unwrap();
+
+                const TEN_GIB: usize = 10 * 1024 * 1024 * 1024;
+                let current_num_threads = rayon::current_num_threads();
+                let max_memory = TEN_GIB / current_num_threads;
+                eprintln!("A maximum of {max_memory} bytes will be used for each of the {current_num_threads} threads");
+                let grenad_parameters = GrenadParameters {
+                    max_memory: Some(max_memory),
+                    ..GrenadParameters::default()
+                };
+
+                let facet_field_ids_delta;
+
+                {
+                    let span = tracing::trace_span!(target: "indexing::documents::extract", "faceted");
+                    let _entered = span.enter();
+                    facet_field_ids_delta = merge_and_send_facet_docids(
+                        global_fields_ids_map,
+                        FacetedDocidsExtractor::run_extraction(grenad_parameters, document_changes, indexing_context, &mut extractor_allocs)?,
+                        FacetDatabases::new(index),
+                        index,
+                        extractor_sender.facet_docids(),
+                    )?;
+                }
+
+                {
+                    let span = tracing::trace_span!(target: "indexing::documents::extract", "word_docids");
                     let _entered = span.enter();
 
-                    // document but we need to create a function that collects and compresses documents.
+                    let WordDocidsCaches {
-                    let document_sender = extractor_sender.document_sender();
+                        word_docids,
-                    let document_extractor = DocumentExtractor { document_sender: &document_sender};
+                        word_fid_docids,
-                    let datastore = ThreadLocal::with_capacity(pool.current_num_threads());
+                        exact_word_docids,
-                    for_each_document_change(document_changes, &document_extractor, indexing_context, &mut extractor_allocs, &datastore)?;
+                        word_position_docids,
-
+                        fid_word_count_docids,
-                    document_sender.finish().unwrap();
+                    } = WordDocidsExtractors::run_extraction(grenad_parameters, document_changes, indexing_context, &mut extractor_allocs)?;
-
-                    const TEN_GIB: usize = 10 * 1024 * 1024 * 1024;
-                    let max_memory = TEN_GIB / dbg!(rayon::current_num_threads());
-                    let grenad_parameters = GrenadParameters {
-                        max_memory: Some(max_memory),
-                        ..GrenadParameters::default()
-                    };
 
+                    // TODO Word Docids Merger
+                    // extractor_sender.send_searchable::<WordDocids>(word_docids).unwrap();
                     {
-                        let span = tracing::trace_span!(target: "indexing::documents::extract", "faceted");
+                        let span = tracing::trace_span!(target: "indexing::documents::merge", "word_docids");
                         let _entered = span.enter();
-                        extract_and_send_docids::<
+                        merge_and_send_docids(
-                            _,
-                            FacetedDocidsExtractor,
-                            FacetDocids,
-                        >(
-                            grenad_parameters,
-                            document_changes,
-                            indexing_context,
-                            &mut extractor_allocs,
-                            &extractor_sender,
-                        )?;
-                    }
-
-                    {
-                        let span = tracing::trace_span!(target: "indexing::documents::extract", "word_docids");
-                        let _entered = span.enter();
-
-                        let WordDocidsMergers {
-                            word_fid_docids,
                             word_docids,
-                            exact_word_docids,
+                            index.word_docids.remap_types(),
-                            word_position_docids,
+                            index,
-                            fid_word_count_docids,
+                            extractor_sender.docids::<WordDocids>(),
-                        } = WordDocidsExtractors::run_extraction(grenad_parameters, document_changes, indexing_context, &mut extractor_allocs)?;
-                        extractor_sender.send_searchable::<WordDocids>(word_docids).unwrap();
-                        extractor_sender.send_searchable::<WordFidDocids>(word_fid_docids).unwrap();
-                        extractor_sender.send_searchable::<ExactWordDocids>(exact_word_docids).unwrap();
-                        extractor_sender.send_searchable::<WordPositionDocids>(word_position_docids).unwrap();
-                        extractor_sender.send_searchable::<FidWordCountDocids>(fid_word_count_docids).unwrap();
-                    }
-
-                    // run the proximity extraction only if the precision is by word
-                    // this works only if the settings didn't change during this transaction.
-                    let rtxn = index.read_txn().unwrap();
-                    let proximity_precision = index.proximity_precision(&rtxn)?.unwrap_or_default();
-                    if proximity_precision == ProximityPrecision::ByWord {
-                        let span = tracing::trace_span!(target: "indexing::documents::extract", "word_pair_proximity_docids");
-                        let _entered = span.enter();
-                        extract_and_send_docids::<
-                            _,
-                            WordPairProximityDocidsExtractor,
-                            WordPairProximityDocids,
-                        >(
-                            grenad_parameters,
-                            document_changes,
-                            indexing_context,
-                      &mut extractor_allocs,
-                            &extractor_sender,
                         )?;
                     }
 
+                    // Word Fid Docids Merging
+                    // extractor_sender.send_searchable::<WordFidDocids>(word_fid_docids).unwrap();
                     {
-                        let span = tracing::trace_span!(target: "indexing::documents::extract", "FINISH");
+                        let span = tracing::trace_span!(target: "indexing::documents::merge", "word_fid_docids");
                         let _entered = span.enter();
+                        merge_and_send_docids(
+                            word_fid_docids,
+                            index.word_fid_docids.remap_types(),
+                            index,
+                            extractor_sender.docids::<WordFidDocids>()
+                        )?;
                     }
 
-                    // TODO THIS IS TOO MUCH
+                    // Exact Word Docids Merging
-                    // - [ ] Extract fieldid docid facet number
+                    // extractor_sender.send_searchable::<ExactWordDocids>(exact_word_docids).unwrap();
-                    // - [ ] Extract fieldid docid facet string
+                    {
-                    // - [ ] Extract facetid string fst
+                        let span = tracing::trace_span!(target: "indexing::documents::merge", "exact_word_docids");
-                    // - [ ] Extract facetid normalized string strings
+                        let _entered = span.enter();
+                        merge_and_send_docids(
+                            exact_word_docids,
+                            index.exact_word_docids.remap_types(),
+                            index,
+                            extractor_sender.docids::<ExactWordDocids>(),
+                        )?;
+                    }
 
-                    // TODO Inverted Indexes again
+                    // Word Position Docids Merging
-                    // - [x] Extract fieldid facet isempty docids
+                    // extractor_sender.send_searchable::<WordPositionDocids>(word_position_docids).unwrap();
-                    // - [x] Extract fieldid facet isnull docids
+                    {
-                    // - [x] Extract fieldid facet exists docids
+                        let span = tracing::trace_span!(target: "indexing::documents::merge", "word_position_docids");
+                        let _entered = span.enter();
+                        merge_and_send_docids(
+                            word_position_docids,
+                            index.word_position_docids.remap_types(),
+                            index,
+                            extractor_sender.docids::<WordPositionDocids>(),
+                        )?;
+                    }
 
-                    // TODO This is the normal system
+                    // Fid Word Count Docids Merging
-                    // - [x] Extract fieldid facet number docids
+                    // extractor_sender.send_searchable::<FidWordCountDocids>(fid_word_count_docids).unwrap();
-                    // - [x] Extract fieldid facet string docids
+                    {
+                        let span = tracing::trace_span!(target: "indexing::documents::merge", "fid_word_count_docids");
+                        let _entered = span.enter();
+                        merge_and_send_docids(
+                            fid_word_count_docids,
+                            index.field_id_word_count_docids.remap_types(),
+                            index,
+                            extractor_sender.docids::<FidWordCountDocids>(),
+                        )?;
+                    }
+                }
 
-                    Ok(()) as Result<_>
+                // run the proximity extraction only if the precision is by word
-                })
+                // this works only if the settings didn't change during this transaction.
+                let proximity_precision = index.proximity_precision(&rtxn)?.unwrap_or_default();
+                if proximity_precision == ProximityPrecision::ByWord {
+                    let span = tracing::trace_span!(target: "indexing::documents::extract", "word_pair_proximity_docids");
+                    let _entered = span.enter();
+                    let caches = <WordPairProximityDocidsExtractor as DocidsExtractor>::run_extraction(grenad_parameters, document_changes, indexing_context, &mut extractor_allocs)?;
+                    merge_and_send_docids(
+                        caches,
+                        index.word_pair_proximity_docids.remap_types(),
+                        index,
+                        extractor_sender.docids::<WordPairProximityDocids>(),
+                    )?;
+                }
+
+                {
+                    let span = tracing::trace_span!(target: "indexing::documents::extract", "FINISH");
+                    let _entered = span.enter();
+                }
+
+                // TODO THIS IS TOO MUCH
+                // - [ ] Extract fieldid docid facet number
+                // - [ ] Extract fieldid docid facet string
+                // - [ ] Extract facetid string fst
+                // - [ ] Extract facetid normalized string strings
+
+                // TODO Inverted Indexes again
+                // - [x] Extract fieldid facet isempty docids
+                // - [x] Extract fieldid facet isnull docids
+                // - [x] Extract fieldid facet exists docids
+
+                // TODO This is the normal system
+                // - [x] Extract fieldid facet number docids
+                // - [x] Extract fieldid facet string docids
+
+                // TODO use None when needed
+                Result::Ok(facet_field_ids_delta)
+            })
         })?;
 
         let global_fields_ids_map = GlobalFieldsIdsMap::new(&new_fields_ids_map);
-
         let indexer_span = tracing::Span::current();
-        // TODO manage the errors correctly
-        let merger_thread = Builder::new().name(S("indexer-merger")).spawn_scoped(s, move || {
-            let span =
-                tracing::trace_span!(target: "indexing::documents", parent: &indexer_span, "merge");
-            let _entered = span.enter();
-            let rtxn = index.read_txn().unwrap();
-            merge_grenad_entries(
-                merger_receiver,
-                merger_sender,
-                &rtxn,
-                index,
-                global_fields_ids_map,
-            )
-        })?;
 
         for operation in writer_receiver {
             let database = operation.database(index);
@@ -264,18 +256,66 @@ where
         }
 
         /// TODO handle the panicking threads
-        handle.join().unwrap()?;
+        let facet_field_ids_delta = extractor_handle.join().unwrap()?;
-        let merger_result = merger_thread.join().unwrap()?;
 
-        if let Some(facet_field_ids_delta) = merger_result.facet_field_ids_delta {
+        let prefix_delta = {
-            compute_facet_level_database(index, wtxn, facet_field_ids_delta)?;
+            let rtxn = index.read_txn()?;
-        }
+            let words_fst = index.words_fst(&rtxn)?;
+            let mut word_fst_builder = WordFstBuilder::new(&words_fst)?;
+            let prefix_settings = index.prefix_settings(&rtxn)?;
+            word_fst_builder.with_prefix_settings(prefix_settings);
 
-        if let Some(prefix_delta) = merger_result.prefix_delta {
+            let previous_words = index.word_docids.iter(&rtxn)?.remap_data_type::<DecodeIgnore>();
+            let current_words = index.word_docids.iter(wtxn)?.remap_data_type::<DecodeIgnore>();
+            for eob in merge_join_by(previous_words, current_words, |lhs, rhs| match (lhs, rhs) {
+                (Ok((l, _)), Ok((r, _))) => l.cmp(r),
+                (Err(_), _) | (_, Err(_)) => Ordering::Equal,
+            }) {
+                match eob {
+                    EitherOrBoth::Both(lhs, rhs) => {
+                        if let Some(e) = lhs.err().or(rhs.err()) {
+                            return Err(e.into());
+                        }
+                    }
+                    EitherOrBoth::Left(result) => {
+                        let (word, _) = result?;
+                        word_fst_builder.register_word(DelAdd::Deletion, word.as_ref())?;
+                    }
+                    EitherOrBoth::Right(result) => {
+                        let (word, _) = result?;
+                        word_fst_builder.register_word(DelAdd::Addition, word.as_ref())?;
+                    }
+                }
+            }
+
+            let span = tracing::trace_span!(target: "indexing::documents::merge", "words_fst");
+            let _entered = span.enter();
+
+            let (word_fst_mmap, prefix_data) = word_fst_builder.build(index, &rtxn)?;
+            // extractor_sender.main().write_words_fst(word_fst_mmap).unwrap();
+            index.main.remap_types::<Str, Bytes>().put(wtxn, WORDS_FST_KEY, &word_fst_mmap)?;
+            if let Some(PrefixData { prefixes_fst_mmap, prefix_delta }) = prefix_data {
+                // extractor_sender.main().write_words_prefixes_fst(prefixes_fst_mmap).unwrap();
+                index.main.remap_types::<Str, Bytes>().put(
+                    wtxn,
+                    WORDS_PREFIXES_FST_KEY,
+                    &prefixes_fst_mmap,
+                )?;
+                Some(prefix_delta)
+            } else {
+                None
+            }
+        };
+
+        // if let Some(facet_field_ids_delta) = merger_result.facet_field_ids_delta {
+        //     compute_facet_level_database(index, wtxn, facet_field_ids_delta)?;
+        // }
+
+        if let Some(prefix_delta) = prefix_delta {
             compute_prefix_database(index, wtxn, prefix_delta)?;
         }
 
-        Ok(()) as Result<_>
+        Result::Ok(())
     })?;
 
     // required to into_inner the new_fields_ids_map
@@ -347,30 +387,6 @@ fn compute_facet_level_database(
     Ok(())
 }
 
-/// TODO: GrenadParameters::default() should be removed in favor a passed parameter
-/// TODO: manage the errors correctly
-/// TODO: we must have a single trait that also gives the extractor type
-fn extract_and_send_docids<
-    'pl,
-    'fid,
-    'indexer,
-    'index,
-    DC: DocumentChanges<'pl>,
-    E: DocidsExtractor,
-    D: MergerOperationType,
->(
-    grenad_parameters: GrenadParameters,
-    document_changes: &DC,
-    indexing_context: IndexingContext<'fid, 'indexer, 'index>,
-    extractor_allocs: &mut ThreadLocal<FullySend<RefCell<Bump>>>,
-    sender: &ExtractorSender,
-) -> Result<()> {
-    let merger =
-        E::run_extraction(grenad_parameters, document_changes, indexing_context, extractor_allocs)?;
-    sender.send_searchable::<D>(merger).unwrap();
-    Ok(())
-}
-
 /// Returns the primary key that has already been set for this index or the
 /// one we will guess by searching for the first key that contains "id" as a substring,
 /// and whether the primary key changed

milli/src/update/new/indexer/partial_dump.rs

@@ -1,12 +1,10 @@
 use std::ops::DerefMut;
 
 use rayon::iter::IndexedParallelIterator;
-use serde::Deserializer;
 use serde_json::value::RawValue;
 
-use super::de::FieldAndDocidExtractor;
 use super::document_changes::{DocumentChangeContext, DocumentChanges, MostlySend, RefCellExt};
-use crate::documents::{DocumentIdExtractionError, PrimaryKey};
+use crate::documents::PrimaryKey;
 use crate::update::concurrent_available_ids::ConcurrentAvailableIds;
 use crate::update::new::document::DocumentFromVersions;
 use crate::update::new::document_change::Versions;

milli/src/update/new/indexer/update_by_function.rs

@@ -1,5 +1,3 @@
-use std::collections::BTreeMap;
-
 use raw_collections::RawMap;
 use rayon::iter::{IndexedParallelIterator, IntoParallelRefIterator, ParallelIterator};
 use rhai::{Dynamic, Engine, OptimizationLevel, Scope, AST};
@@ -12,8 +10,8 @@ use crate::documents::PrimaryKey;
 use crate::error::{FieldIdMapMissingEntry, InternalError};
 use crate::update::new::document::DocumentFromVersions;
 use crate::update::new::document_change::Versions;
-use crate::update::new::{Deletion, DocumentChange, KvReaderFieldId, KvWriterFieldId, Update};
+use crate::update::new::{Deletion, DocumentChange, KvReaderFieldId, Update};
-use crate::{all_obkv_to_json, Error, FieldsIdsMap, GlobalFieldsIdsMap, Object, Result, UserError};
+use crate::{all_obkv_to_json, Error, FieldsIdsMap, Object, Result, UserError};
 
 pub struct UpdateByFunction {
     documents: RoaringBitmap,

milli/src/update/new/merger.rs

@@ -1,222 +1,20 @@
-use std::fs::File;
 use std::io::{self};
 
 use bincode::ErrorKind;
-use grenad::Merger;
 use hashbrown::HashSet;
 use heed::types::Bytes;
 use heed::{Database, RoTxn};
+use rayon::iter::{IntoParallelIterator, ParallelIterator};
 use roaring::RoaringBitmap;
 
 use super::channel::*;
-use super::extract::FacetKind;
+use super::extract::{
-use super::facet_search_builder::FacetSearchBuilder;
+    merge_caches, transpose_and_freeze_caches, BalancedCaches, DelAddRoaringBitmap, FacetKind,
-use super::word_fst_builder::{PrefixData, PrefixDelta};
-use super::{Deletion, DocumentChange, KvReaderDelAdd, KvReaderFieldId};
-use crate::update::del_add::DelAdd;
-use crate::update::new::channel::MergerOperation;
-use crate::update::new::word_fst_builder::WordFstBuilder;
-use crate::update::MergeDeladdCboRoaringBitmaps;
-use crate::{
-    localized_attributes_rules, CboRoaringBitmapCodec, Error, FieldId, GeoPoint,
-    GlobalFieldsIdsMap, Index, Result,
 };
-
+use super::facet_search_builder::FacetSearchBuilder;
-/// TODO We must return some infos/stats
+use super::DocumentChange;
-#[tracing::instrument(level = "trace", skip_all, target = "indexing::documents", name = "merge")]
+use crate::update::del_add::DelAdd;
-pub fn merge_grenad_entries(
+use crate::{CboRoaringBitmapCodec, Error, FieldId, GeoPoint, GlobalFieldsIdsMap, Index, Result};
-    receiver: MergerReceiver,
-    sender: MergerSender,
-    rtxn: &RoTxn,
-    index: &Index,
-    global_fields_ids_map: GlobalFieldsIdsMap<'_>,
-) -> Result<MergerResult> {
-    let mut buffer: Vec<u8> = Vec::new();
-    let mut documents_ids = index.documents_ids(rtxn)?;
-    let mut geo_extractor = GeoExtractor::new(rtxn, index)?;
-    let mut merger_result = MergerResult::default();
-
-    for merger_operation in receiver {
-        match merger_operation {
-            MergerOperation::ExactWordDocidsMerger(merger) => {
-                let span =
-                    tracing::trace_span!(target: "indexing::documents::merge", "exact_word_docids");
-                let _entered = span.enter();
-                merge_and_send_docids(
-                    merger,
-                    /// TODO do a MergerOperation::database(&Index) -> Database<Bytes, Bytes>.
-                    index.exact_word_docids.remap_types(),
-                    rtxn,
-                    &mut buffer,
-                    sender.docids::<ExactWordDocids>(),
-                    |_, _key| Ok(()),
-                )?;
-            }
-            MergerOperation::FidWordCountDocidsMerger(merger) => {
-                let span = tracing::trace_span!(target: "indexing::documents::merge", "fid_word_count_docids");
-                let _entered = span.enter();
-                merge_and_send_docids(
-                    merger,
-                    index.field_id_word_count_docids.remap_types(),
-                    rtxn,
-                    &mut buffer,
-                    sender.docids::<FidWordCountDocids>(),
-                    |_, _key| Ok(()),
-                )?;
-            }
-            MergerOperation::WordDocidsMerger(merger) => {
-                let words_fst = index.words_fst(rtxn)?;
-                let mut word_fst_builder = WordFstBuilder::new(&words_fst)?;
-                let prefix_settings = index.prefix_settings(rtxn)?;
-                word_fst_builder.with_prefix_settings(prefix_settings);
-
-                {
-                    let span =
-                        tracing::trace_span!(target: "indexing::documents::merge", "word_docids");
-                    let _entered = span.enter();
-
-                    merge_and_send_docids(
-                        merger,
-                        index.word_docids.remap_types(),
-                        rtxn,
-                        &mut buffer,
-                        sender.docids::<WordDocids>(),
-                        |deladd, key| word_fst_builder.register_word(deladd, key),
-                    )?;
-                }
-
-                {
-                    let span =
-                        tracing::trace_span!(target: "indexing::documents::merge", "words_fst");
-                    let _entered = span.enter();
-
-                    let (word_fst_mmap, prefix_data) = word_fst_builder.build(index, rtxn)?;
-                    sender.main().write_words_fst(word_fst_mmap).unwrap();
-                    if let Some(PrefixData { prefixes_fst_mmap, prefix_delta }) = prefix_data {
-                        sender.main().write_words_prefixes_fst(prefixes_fst_mmap).unwrap();
-                        merger_result.prefix_delta = Some(prefix_delta);
-                    }
-                }
-            }
-            MergerOperation::WordFidDocidsMerger(merger) => {
-                let span =
-                    tracing::trace_span!(target: "indexing::documents::merge", "word_fid_docids");
-                let _entered = span.enter();
-                merge_and_send_docids(
-                    merger,
-                    index.word_fid_docids.remap_types(),
-                    rtxn,
-                    &mut buffer,
-                    sender.docids::<WordFidDocids>(),
-                    |_, _key| Ok(()),
-                )?;
-            }
-            MergerOperation::WordPairProximityDocidsMerger(merger) => {
-                let span = tracing::trace_span!(target: "indexing::documents::merge", "word_pair_proximity_docids");
-                let _entered = span.enter();
-                merge_and_send_docids(
-                    merger,
-                    index.word_pair_proximity_docids.remap_types(),
-                    rtxn,
-                    &mut buffer,
-                    sender.docids::<WordPairProximityDocids>(),
-                    |_, _key| Ok(()),
-                )?;
-            }
-            MergerOperation::WordPositionDocidsMerger(merger) => {
-                let span = tracing::trace_span!(target: "indexing::documents::merge", "word_position_docids");
-                let _entered = span.enter();
-                merge_and_send_docids(
-                    merger,
-                    index.word_position_docids.remap_types(),
-                    rtxn,
-                    &mut buffer,
-                    sender.docids::<WordPositionDocids>(),
-                    |_, _key| Ok(()),
-                )?;
-            }
-            MergerOperation::InsertDocument { docid, external_id, document } => {
-                let span =
-                    tracing::trace_span!(target: "indexing::documents::merge", "insert_document");
-                let _entered = span.enter();
-                documents_ids.insert(docid);
-                sender.documents().uncompressed(docid, external_id.clone(), &document).unwrap();
-
-                if let Some(geo_extractor) = geo_extractor.as_mut() {
-                    let current = index.documents.remap_data_type::<Bytes>().get(rtxn, &docid)?;
-                    let current: Option<&KvReaderFieldId> = current.map(Into::into);
-                    let change = match current {
-                        Some(current) => DocumentChange::Update(todo!()),
-                        None => DocumentChange::Insertion(todo!()),
-                    };
-                    geo_extractor.manage_change(&mut global_fields_ids_map, &change)?;
-                }
-            }
-            MergerOperation::DeleteDocument { docid, external_id } => {
-                let span =
-                    tracing::trace_span!(target: "indexing::documents::merge", "delete_document");
-                let _entered = span.enter();
-                if !documents_ids.remove(docid) {
-                    unreachable!("Tried deleting a document that we do not know about");
-                }
-                sender.documents().delete(docid, external_id.clone()).unwrap();
-
-                if let Some(geo_extractor) = geo_extractor.as_mut() {
-                    let change = DocumentChange::Deletion(Deletion::create(docid, todo!()));
-                    geo_extractor.manage_change(&mut global_fields_ids_map, &change)?;
-                }
-            }
-            MergerOperation::FinishedDocument => {
-                // send the rtree
-            }
-            MergerOperation::FacetDocidsMerger(merger) => {
-                let span =
-                    tracing::trace_span!(target: "indexing::documents::merge", "facet_docids");
-                let _entered = span.enter();
-                let mut facet_field_ids_delta = FacetFieldIdsDelta::new();
-                let localized_attributes_rules =
-                    index.localized_attributes_rules(rtxn)?.unwrap_or_default();
-                let mut facet_search_builder = FacetSearchBuilder::new(
-                    global_fields_ids_map.clone(),
-                    localized_attributes_rules,
-                );
-                merge_and_send_facet_docids(
-                    merger,
-                    FacetDatabases::new(index),
-                    rtxn,
-                    &mut buffer,
-                    sender.facet_docids(),
-                    &mut facet_field_ids_delta,
-                    &mut facet_search_builder,
-                )?;
-
-                merger_result.facet_field_ids_delta = Some(facet_field_ids_delta);
-                // merge and send the facet fst and the searchable facet values
-                facet_search_builder.merge_and_send(index, rtxn, sender.facet_searchable())?;
-            }
-        }
-    }
-
-    {
-        let span = tracing::trace_span!(target: "indexing::documents::merge", "documents_ids");
-        let _entered = span.enter();
-
-        // Send the documents ids unionized with the current one
-        sender.send_documents_ids(documents_ids).unwrap();
-    }
-
-    // ...
-
-    Ok(merger_result)
-}
-
-#[derive(Default, Debug)]
-pub struct MergerResult {
-    /// The delta of the prefixes
-    pub prefix_delta: Option<PrefixDelta>,
-    /// The field ids that have been modified
-    pub facet_field_ids_delta: Option<FacetFieldIdsDelta>,
-}
 
 pub struct GeoExtractor {
     rtree: Option<rstar::RTree<GeoPoint>>,
@@ -267,80 +65,92 @@ impl GeoExtractor {
 }
 
 #[tracing::instrument(level = "trace", skip_all, target = "indexing::merge")]
-fn merge_and_send_docids(
+pub fn merge_and_send_docids<'extractor>(
-    merger: Merger<File, MergeDeladdCboRoaringBitmaps>,
+    mut caches: Vec<BalancedCaches<'extractor>>,
     database: Database<Bytes, Bytes>,
-    rtxn: &RoTxn<'_>,
+    index: &Index,
-    buffer: &mut Vec<u8>,
+    docids_sender: impl DocidsSender + Sync,
-    docids_sender: impl DocidsSender,
-    mut register_key: impl FnMut(DelAdd, &[u8]) -> Result<()>,
 ) -> Result<()> {
-    let mut merger_iter = merger.into_stream_merger_iter().unwrap();
+    transpose_and_freeze_caches(&mut caches)?.into_par_iter().try_for_each(|frozen| {
-    while let Some((key, deladd)) = merger_iter.next().unwrap() {
+        let rtxn = index.read_txn()?;
-        let current = database.get(rtxn, key)?;
+        let mut buffer = Vec::new();
-        let deladd: &KvReaderDelAdd = deladd.into();
+        merge_caches(frozen, |key, DelAddRoaringBitmap { del, add }| {
-        let del = deladd.get(DelAdd::Deletion);
+            let current = database.get(&rtxn, key)?;
-        let add = deladd.get(DelAdd::Addition);
+            match merge_cbo_bitmaps(current, del, add)? {
-
+                Operation::Write(bitmap) => {
-        match merge_cbo_bitmaps(current, del, add)? {
+                    let value = cbo_bitmap_serialize_into_vec(&bitmap, &mut buffer);
-            Operation::Write(bitmap) => {
+                    docids_sender.write(key, value).unwrap();
-                let value = cbo_bitmap_serialize_into_vec(&bitmap, buffer);
+                    Ok(())
-                docids_sender.write(key, value).unwrap();
+                }
-                register_key(DelAdd::Addition, key)?;
+                Operation::Delete => {
+                    docids_sender.delete(key).unwrap();
+                    Ok(())
+                }
+                Operation::Ignore => Ok(()),
             }
-            Operation::Delete => {
+        })
-                docids_sender.delete(key).unwrap();
+    })
-                register_key(DelAdd::Deletion, key)?;
-            }
-            Operation::Ignore => (),
-        }
-    }
-
-    Ok(())
 }
 
 #[tracing::instrument(level = "trace", skip_all, target = "indexing::merge")]
-fn merge_and_send_facet_docids(
+pub fn merge_and_send_facet_docids<'indexer, 'extractor>(
-    merger: Merger<File, MergeDeladdCboRoaringBitmaps>,
+    global_fields_ids_map: GlobalFieldsIdsMap<'indexer>,
+    mut caches: Vec<BalancedCaches<'extractor>>,
     database: FacetDatabases,
-    rtxn: &RoTxn<'_>,
+    index: &Index,
-    buffer: &mut Vec<u8>,
+    docids_sender: impl DocidsSender + Sync,
-    docids_sender: impl DocidsSender,
+) -> Result<(FacetFieldIdsDelta, FacetSearchBuilder<'indexer>)> {
-    facet_field_ids_delta: &mut FacetFieldIdsDelta,
+    transpose_and_freeze_caches(&mut caches)?
-    facet_search_builder: &mut FacetSearchBuilder,
+        .into_par_iter()
-) -> Result<()> {
+        .map(|frozen| {
-    let mut merger_iter = merger.into_stream_merger_iter().unwrap();
+            let mut facet_field_ids_delta = FacetFieldIdsDelta::default();
-    while let Some((key, deladd)) = merger_iter.next().unwrap() {
+            let rtxn = index.read_txn()?;
-        let current = database.get_cbo_roaring_bytes_value(rtxn, key)?;
+            let localized_attributes_rules = index.localized_attributes_rules(&rtxn)?;
-        let deladd: &KvReaderDelAdd = deladd.into();
+            let mut facet_search_builder = FacetSearchBuilder::new(
-        let del = deladd.get(DelAdd::Deletion);
+                global_fields_ids_map.clone(),
-        let add = deladd.get(DelAdd::Addition);
+                localized_attributes_rules.unwrap_or_default(),
+            );
+            let mut buffer = Vec::new();
+            merge_caches(frozen, |key, DelAddRoaringBitmap { del, add }| {
+                let current = database.get_cbo_roaring_bytes_value(&rtxn, key)?;
+                match merge_cbo_bitmaps(current, del, add)? {
+                    Operation::Write(bitmap) => {
+                        facet_field_ids_delta.register_from_key(key);
+                        facet_search_builder.register_from_key(DelAdd::Addition, key)?;
+                        let value = cbo_bitmap_serialize_into_vec(&bitmap, &mut buffer);
+                        docids_sender.write(key, value).unwrap();
+                        Ok(())
+                    }
+                    Operation::Delete => {
+                        facet_field_ids_delta.register_from_key(key);
+                        facet_search_builder.register_from_key(DelAdd::Deletion, key)?;
+                        docids_sender.delete(key).unwrap();
+                        Ok(())
+                    }
+                    Operation::Ignore => Ok(()),
+                }
+            })?;
 
-        match merge_cbo_bitmaps(current, del, add)? {
+            Ok((facet_field_ids_delta, facet_search_builder))
-            Operation::Write(bitmap) => {
+        })
-                facet_field_ids_delta.register_from_key(key);
+        .reduce(
-                facet_search_builder.register_from_key(DelAdd::Addition, key)?;
+            || Ok((FacetFieldIdsDelta::default(), todo!())),
-                let value = cbo_bitmap_serialize_into_vec(&bitmap, buffer);
+            |lhs, rhs| {
-                docids_sender.write(key, value).unwrap();
+                let (lhs_ffid, lhs_fsb) = lhs?;
-            }
+                let (rhs_ffid, rhs_fsb) = rhs?;
-            Operation::Delete => {
+                let ffid_merged = lhs_ffid.merge(rhs_ffid);
-                facet_field_ids_delta.register_from_key(key);
+                let fsb_merged = todo!();
-                facet_search_builder.register_from_key(DelAdd::Deletion, key)?;
+                Ok((ffid_merged, fsb_merged))
-                docids_sender.delete(key).unwrap();
+            },
-            }
+        )
-            Operation::Ignore => (),
-        }
-    }
-
-    Ok(())
 }
 
-struct FacetDatabases<'a> {
+pub struct FacetDatabases<'a> {
     index: &'a Index,
 }
 
 impl<'a> FacetDatabases<'a> {
-    fn new(index: &'a Index) -> Self {
+    pub fn new(index: &'a Index) -> Self {
         Self { index }
     }
 
@@ -361,7 +171,7 @@ impl<'a> FacetDatabases<'a> {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, Default)]
 pub struct FacetFieldIdsDelta {
     /// The field ids that have been modified
     modified_facet_string_ids: HashSet<FieldId>,
@@ -369,13 +179,6 @@ pub struct FacetFieldIdsDelta {
 }
 
 impl FacetFieldIdsDelta {
-    fn new() -> Self {
-        Self {
-            modified_facet_string_ids: HashSet::new(),
-            modified_facet_number_ids: HashSet::new(),
-        }
-    }
-
     fn register_facet_string_id(&mut self, field_id: FieldId) {
         self.modified_facet_string_ids.insert(field_id);
     }
@@ -414,6 +217,17 @@ impl FacetFieldIdsDelta {
             Some(self.modified_facet_number_ids.iter().copied().collect())
         }
     }
+
+    pub fn merge(mut self, rhs: Self) -> Self {
+        let Self { modified_facet_number_ids, modified_facet_string_ids } = rhs;
+        modified_facet_number_ids.into_iter().for_each(|fid| {
+            self.modified_facet_number_ids.insert(fid);
+        });
+        modified_facet_string_ids.into_iter().for_each(|fid| {
+            self.modified_facet_string_ids.insert(fid);
+        });
+        self
+    }
 }
 
 enum Operation {
@@ -425,13 +239,10 @@ enum Operation {
 /// A function that merges the DelAdd CboRoaringBitmaps with the current bitmap.
 fn merge_cbo_bitmaps(
     current: Option<&[u8]>,
-    del: Option<&[u8]>,
+    del: Option<RoaringBitmap>,
-    add: Option<&[u8]>,
+    add: Option<RoaringBitmap>,
 ) -> Result<Operation> {
     let current = current.map(CboRoaringBitmapCodec::deserialize_from).transpose()?;
-    let del = del.map(CboRoaringBitmapCodec::deserialize_from).transpose()?;
-    let add = add.map(CboRoaringBitmapCodec::deserialize_from).transpose()?;
-
     match (current, del, add) {
         (None, None, None) => Ok(Operation::Ignore), // but it's strange
         (None, None, Some(add)) => Ok(Operation::Write(add)),

milli/src/update/new/mod.rs

@@ -1,4 +1,7 @@
 pub use document_change::{Deletion, DocumentChange, Insertion, Update};
+pub use merger::{
+    merge_and_send_docids, merge_and_send_facet_docids, FacetDatabases, FacetFieldIdsDelta,
+};
 pub use top_level_map::{CowStr, TopLevelMap};
 
 use super::del_add::DelAdd;

milli/src/update/new/parallel_iterator_ext.rs

@@ -1,38 +1,8 @@
 use std::sync::Arc;
 
-use rayon::iter::{MapInit, ParallelIterator};
+use rayon::iter::ParallelIterator;
 
 pub trait ParallelIteratorExt: ParallelIterator {
-    /// Maps items based on the init function.
-    ///
-    /// The init function is ran only as necessary which is basically once by thread.
-    fn try_map_try_init<F, INIT, T, E, R>(
-        self,
-        init: INIT,
-        map_op: F,
-    ) -> MapInit<
-        Self,
-        impl Fn() -> Result<T, Arc<E>> + Sync + Send + Clone,
-        impl Fn(&mut Result<T, Arc<E>>, Self::Item) -> Result<R, Arc<E>> + Sync + Send + Clone,
-    >
-    where
-        E: Send + Sync,
-        F: Fn(&mut T, Self::Item) -> Result<R, E> + Sync + Send + Clone,
-        INIT: Fn() -> Result<T, E> + Sync + Send + Clone,
-        R: Send,
-    {
-        self.map_init(
-            move || match init() {
-                Ok(t) => Ok(t),
-                Err(err) => Err(Arc::new(err)),
-            },
-            move |result, item| match result {
-                Ok(t) => map_op(t, item).map_err(Arc::new),
-                Err(err) => Err(err.clone()),
-            },
-        )
-    }
-
     /// A method to run a closure of all the items and return an owned error.
     ///
     /// The init function is ran only as necessary which is basically once by thread.
@@ -58,17 +28,6 @@ pub trait ParallelIteratorExt: ParallelIterator {
             Err(err) => Err(Arc::into_inner(err).expect("the error must be only owned by us")),
         }
     }
-
-    fn try_arc_for_each<F, E>(self, op: F) -> Result<(), E>
-    where
-        E: Send + Sync,
-        F: Fn(Self::Item) -> Result<(), Arc<E>> + Sync + Send + Clone,
-    {
-        match self.try_for_each(op) {
-            Ok(()) => Ok(()),
-            Err(err) => Err(Arc::into_inner(err).expect("the error must be only owned by us")),
-        }
-    }
 }
 
 impl<T: ParallelIterator> ParallelIteratorExt for T {}

milli/src/update/new/words_prefix_docids.rs

@@ -1,10 +1,16 @@
+use std::cell::RefCell;
 use std::collections::HashSet;
+use std::io::{BufReader, BufWriter, Read, Seek, Write};
 
 use hashbrown::HashMap;
 use heed::types::Bytes;
-use heed::{BytesDecode, Database, RwTxn};
+use heed::{BytesDecode, Database, RoTxn, RwTxn};
-use roaring::RoaringBitmap;
+use rayon::iter::{IntoParallelIterator, ParallelIterator as _};
+use roaring::MultiOps;
+use tempfile::tempfile;
+use thread_local::ThreadLocal;
 
+use super::indexer::document_changes::RefCellExt;
 use crate::heed_codec::StrBEU16Codec;
 use crate::{CboRoaringBitmapCodec, Index, Prefix, Result};
 
@@ -38,22 +44,103 @@ impl WordPrefixDocids {
         prefixes: &HashSet<Prefix>,
     ) -> Result<()> {
         // We fetch the docids associated to the newly added word prefix fst only.
-        let mut docids = RoaringBitmap::new();
+        // And collect the CboRoaringBitmaps pointers in an HashMap.
-        for prefix in prefixes {
+        let frozen = FrozenPrefixBitmaps::from_prefixes(self.database, wtxn, prefixes)?;
-            docids.clear();
-            let prefix = prefix.as_bytes();
-            for result in self.database.prefix_iter(wtxn, prefix)? {
-                let (_word, data) = result?;
-                docids |= &data;
-            }
 
-            self.prefix_database.put(wtxn, prefix, &docids)?;
+        // We access this HashMap in parallel to compute the *union* of all
+        // of them and *serialize* them into files. There is one file by CPU.
+        let local_entries = ThreadLocal::with_capacity(rayon::current_num_threads());
+        prefixes.into_par_iter().map(AsRef::as_ref).try_for_each(|prefix| {
+            let refcell = local_entries.get_or_try(|| {
+                tempfile().map(BufWriter::new).map(|f| RefCell::new((Vec::new(), f, Vec::new())))
+            })?;
+
+            let mut refmut = refcell.borrow_mut_or_yield();
+            let (ref mut index, ref mut file, ref mut buffer) = *refmut;
+
+            let output = frozen
+                .bitmaps(prefix)
+                .unwrap()
+                .iter()
+                .map(|bytes| CboRoaringBitmapCodec::deserialize_from(bytes))
+                .union()?;
+
+            buffer.clear();
+            CboRoaringBitmapCodec::serialize_into(&output, buffer);
+            index.push(PrefixEntry { prefix, serialized_length: buffer.len() });
+            file.write_all(buffer)
+        })?;
+
+        drop(frozen);
+
+        // We iterate over all the collected and serialized bitmaps through
+        // the files and entries to eventually put them in the final database.
+        for refcell in local_entries {
+            let (index, file, mut buffer) = refcell.into_inner();
+            let mut file = file.into_inner().map_err(|e| e.into_error())?;
+            file.rewind()?;
+            let mut file = BufReader::new(file);
+            for PrefixEntry { prefix, serialized_length } in index {
+                buffer.resize(serialized_length, 0);
+                file.read_exact(&mut buffer)?;
+                self.prefix_database.remap_data_type::<Bytes>().put(
+                    wtxn,
+                    prefix.as_bytes(),
+                    &buffer,
+                )?;
+            }
         }
 
         Ok(())
     }
 }
 
+/// Represents a prefix and the lenght the bitmap takes on disk.
+struct PrefixEntry<'a> {
+    prefix: &'a str,
+    serialized_length: usize,
+}
+
+/// Stores prefixes along with all the pointers to the associated
+/// CBoRoaringBitmaps.
+///
+/// They are collected synchronously and stored into an HashMap. The
+/// Synchronous process is doing a small amount of work by just storing
+/// pointers. It can then be accessed in parallel to get the associated
+/// bitmaps pointers.
+struct FrozenPrefixBitmaps<'a, 'rtxn> {
+    prefixes_bitmaps: HashMap<&'a str, Vec<&'rtxn [u8]>>,
+}
+
+impl<'a, 'rtxn> FrozenPrefixBitmaps<'a, 'rtxn> {
+    #[tracing::instrument(level = "trace", skip_all, target = "indexing::prefix")]
+    pub fn from_prefixes(
+        database: Database<Bytes, CboRoaringBitmapCodec>,
+        rtxn: &'rtxn RoTxn,
+        prefixes: &'a HashSet<Prefix>,
+    ) -> heed::Result<Self> {
+        let database = database.remap_data_type::<Bytes>();
+
+        let mut prefixes_bitmaps = HashMap::new();
+        for prefix in prefixes {
+            let mut bitmap_bytes = Vec::new();
+            for result in database.prefix_iter(rtxn, prefix.as_bytes())? {
+                let (_word, bytes) = result?;
+                bitmap_bytes.push(bytes);
+            }
+            assert!(prefixes_bitmaps.insert(prefix.as_str(), bitmap_bytes).is_none());
+        }
+
+        Ok(Self { prefixes_bitmaps })
+    }
+
+    pub fn bitmaps(&self, key: &str) -> Option<&[&'rtxn [u8]]> {
+        self.prefixes_bitmaps.get(key).map(AsRef::as_ref)
+    }
+}
+
+unsafe impl<'a, 'rtxn> Sync for FrozenPrefixBitmaps<'a, 'rtxn> {}
+
 struct WordPrefixIntegerDocids {
     database: Database<Bytes, CboRoaringBitmapCodec>,
     prefix_database: Database<Bytes, CboRoaringBitmapCodec>,