Merge #4953

4953: Move the multi arroy index logic to the arroy wrapper r=irevoire a=irevoire

# Pull Request

## Related issue
Fixes https://github.com/meilisearch/meilisearch/issues/4948

## What does this PR do?
- Make the `ArroyWrapper` we introduced in the last PR handle all the embedded for a specific docid itself.


Co-authored-by: Tamo <tamo@meilisearch.com>

milli/src/index.rs

@@ -1610,24 +1610,6 @@ impl Index {
             .unwrap_or_default())
     }
 
-    pub fn arroy_readers<'a>(
-        &'a self,
-        rtxn: &'a RoTxn<'a>,
-        embedder_id: u8,
-        quantized: bool,
-    ) -> impl Iterator<Item = Result<ArroyWrapper>> + 'a {
-        crate::vector::arroy_db_range_for_embedder(embedder_id).map_while(move |k| {
-            let reader = ArroyWrapper::new(self.vector_arroy, k, quantized);
-            // Here we don't care about the dimensions, but we want to know if we can read
-            // in the database or if its metadata are missing because there is no document with that many vectors.
-            match reader.dimensions(rtxn) {
-                Ok(_) => Some(Ok(reader)),
-                Err(arroy::Error::MissingMetadata(_)) => None,
-                Err(e) => Some(Err(e.into())),
-            }
-        })
-    }
-
     pub(crate) fn put_search_cutoff(&self, wtxn: &mut RwTxn<'_>, cutoff: u64) -> heed::Result<()> {
         self.main.remap_types::<Str, BEU64>().put(wtxn, main_key::SEARCH_CUTOFF, &cutoff)
     }
@@ -1649,14 +1631,9 @@ impl Index {
         let embedding_configs = self.embedding_configs(rtxn)?;
         for config in embedding_configs {
             let embedder_id = self.embedder_category_id.get(rtxn, &config.name)?.unwrap();
-            let embeddings = self
+            let reader =
-                .arroy_readers(rtxn, embedder_id, config.config.quantized())
+                ArroyWrapper::new(self.vector_arroy, embedder_id, config.config.quantized());
-                .map_while(|reader| {
+            let embeddings = reader.item_vectors(rtxn, docid)?;
-                    reader
-                        .and_then(|r| r.item_vector(rtxn, docid).map_err(|e| e.into()))
-                        .transpose()
-                })
-                .collect::<Result<Vec<_>>>()?;
             res.insert(config.name.to_owned(), embeddings);
         }
         Ok(res)

milli/src/search/new/vector_sort.rs

@@ -1,11 +1,10 @@
 use std::iter::FromIterator;
 
-use ordered_float::OrderedFloat;
 use roaring::RoaringBitmap;
 
 use super::ranking_rules::{RankingRule, RankingRuleOutput, RankingRuleQueryTrait};
 use crate::score_details::{self, ScoreDetails};
-use crate::vector::{DistributionShift, Embedder};
+use crate::vector::{ArroyWrapper, DistributionShift, Embedder};
 use crate::{DocumentId, Result, SearchContext, SearchLogger};
 
 pub struct VectorSort<Q: RankingRuleQueryTrait> {
@@ -53,14 +52,9 @@ impl<Q: RankingRuleQueryTrait> VectorSort<Q> {
         vector_candidates: &RoaringBitmap,
     ) -> Result<()> {
         let target = &self.target;
-        let mut results = Vec::new();
 
-        for reader in ctx.index.arroy_readers(ctx.txn, self.embedder_index, self.quantized) {
+        let reader = ArroyWrapper::new(ctx.index.vector_arroy, self.embedder_index, self.quantized);
-            let nns_by_vector =
+        let results = reader.nns_by_vector(ctx.txn, target, self.limit, Some(vector_candidates))?;
-                reader?.nns_by_vector(ctx.txn, target, self.limit, Some(vector_candidates))?;
-            results.extend(nns_by_vector.into_iter());
-        }
-        results.sort_unstable_by_key(|(_, distance)| OrderedFloat(*distance));
         self.cached_sorted_docids = results.into_iter();
 
         Ok(())

milli/src/search/similar.rs

@@ -1,10 +1,9 @@
 use std::sync::Arc;
 
-use ordered_float::OrderedFloat;
 use roaring::RoaringBitmap;
 
 use crate::score_details::{self, ScoreDetails};
-use crate::vector::Embedder;
+use crate::vector::{ArroyWrapper, Embedder};
 use crate::{filtered_universe, DocumentId, Filter, Index, Result, SearchResult};
 
 pub struct Similar<'a> {
@@ -71,23 +70,13 @@ impl<'a> Similar<'a> {
                 .get(self.rtxn, &self.embedder_name)?
                 .ok_or_else(|| crate::UserError::InvalidEmbedder(self.embedder_name.to_owned()))?;
 
-        let mut results = Vec::new();
+        let reader = ArroyWrapper::new(self.index.vector_arroy, embedder_index, self.quantized);
-
+        let results = reader.nns_by_item(
-        for reader in self.index.arroy_readers(self.rtxn, embedder_index, self.quantized) {
-            let nns_by_item = reader?.nns_by_item(
             self.rtxn,
             self.id,
             self.limit + self.offset + 1,
             Some(&universe),
         )?;
-            if let Some(mut nns_by_item) = nns_by_item {
-                results.append(&mut nns_by_item);
-            } else {
-                break;
-            }
-        }
-
-        results.sort_unstable_by_key(|(_, distance)| OrderedFloat(*distance));
 
         let mut documents_ids = Vec::with_capacity(self.limit);
         let mut document_scores = Vec::with_capacity(self.limit);

milli/src/update/index_documents/mod.rs

@@ -689,9 +689,8 @@ where
                         key: None,
                     },
                 )?;
-                let first_id = crate::vector::arroy_db_range_for_embedder(index).next().unwrap();
                 let reader =
-                    ArroyWrapper::new(self.index.vector_arroy, first_id, action.was_quantized);
+                    ArroyWrapper::new(self.index.vector_arroy, index, action.was_quantized);
                 let dim = reader.dimensions(self.wtxn)?;
                 dimension.insert(name.to_string(), dim);
             }
@@ -713,17 +712,8 @@ where
             let is_quantizing = embedder_config.map_or(false, |action| action.is_being_quantized);
 
             pool.install(|| {
-                for k in crate::vector::arroy_db_range_for_embedder(embedder_index) {
+                let mut writer = ArroyWrapper::new(vector_arroy, embedder_index, was_quantized);
-                    let mut writer = ArroyWrapper::new(vector_arroy, k, was_quantized);
+                writer.build_and_quantize(wtxn, &mut rng, dimension, is_quantizing)?;
-                    if is_quantizing {
-                        writer.quantize(wtxn, k, dimension)?;
-                    }
-                    if writer.need_build(wtxn, dimension)? {
-                        writer.build(wtxn, &mut rng, dimension)?;
-                    } else if writer.is_empty(wtxn, dimension)? {
-                        break;
-                    }
-                }
                 Result::Ok(())
             })
             .map_err(InternalError::from)??;

milli/src/update/index_documents/transform.rs

@@ -990,27 +990,24 @@ impl<'a, 'i> Transform<'a, 'i> {
             None
         };
 
-        let readers: Result<BTreeMap<&str, (Vec<ArroyWrapper>, &RoaringBitmap)>> = settings_diff
+        let readers: BTreeMap<&str, (ArroyWrapper, &RoaringBitmap)> = settings_diff
             .embedding_config_updates
             .iter()
             .filter_map(|(name, action)| {
                 if let Some(WriteBackToDocuments { embedder_id, user_provided }) =
                     action.write_back()
                 {
-                    let readers: Result<Vec<_>> = self
+                    let reader = ArroyWrapper::new(
-                        .index
+                        self.index.vector_arroy,
-                        .arroy_readers(wtxn, *embedder_id, action.was_quantized)
+                        *embedder_id,
-                        .collect();
+                        action.was_quantized,
-                    match readers {
+                    );
-                        Ok(readers) => Some(Ok((name.as_str(), (readers, user_provided)))),
+                    Some((name.as_str(), (reader, user_provided)))
-                        Err(error) => Some(Err(error)),
-                    }
                 } else {
                     None
                 }
             })
             .collect();
-        let readers = readers?;
 
         let old_vectors_fid = settings_diff
             .old
@@ -1048,34 +1045,24 @@ impl<'a, 'i> Transform<'a, 'i> {
                     arroy::Error,
                 > = readers
                     .iter()
-                    .filter_map(|(name, (readers, user_provided))| {
+                    .filter_map(|(name, (reader, user_provided))| {
                         if !user_provided.contains(docid) {
                             return None;
                         }
-                        let mut vectors = Vec::new();
+                        match reader.item_vectors(wtxn, docid) {
-                        for reader in readers {
+                            Ok(vectors) if vectors.is_empty() => None,
-                            let Some(vector) = reader.item_vector(wtxn, docid).transpose() else {
+                            Ok(vectors) => Some(Ok((
-                                break;
-                            };
-
-                            match vector {
-                                Ok(vector) => vectors.push(vector),
-                                Err(error) => return Some(Err(error)),
-                            }
-                        }
-                        if vectors.is_empty() {
-                            return None;
-                        }
-                        Some(Ok((
                                 name.to_string(),
                                 serde_json::to_value(ExplicitVectors {
-                                embeddings: Some(VectorOrArrayOfVectors::from_array_of_vectors(
+                                    embeddings: Some(
-                                    vectors,
+                                        VectorOrArrayOfVectors::from_array_of_vectors(vectors),
-                                )),
+                                    ),
                                     regenerate: false,
                                 })
                                 .unwrap(),
-                        )))
+                            ))),
+                            Err(e) => Some(Err(e)),
+                        }
                     })
                     .collect();
 
@@ -1104,12 +1091,10 @@ impl<'a, 'i> Transform<'a, 'i> {
         }
 
         // delete all vectors from the embedders that need removal
-        for (_, (readers, _)) in readers {
+        for (_, (reader, _)) in readers {
-            for reader in readers {
             let dimensions = reader.dimensions(wtxn)?;
             reader.clear(wtxn, dimensions)?;
         }
-        }
 
         let grenad_params = GrenadParameters {
             chunk_compression_type: self.indexer_settings.chunk_compression_type,

milli/src/update/index_documents/typed_chunk.rs

@@ -673,22 +673,14 @@ pub(crate) fn write_typed_chunk_into_index(
                 .get(&embedder_name)
                 .map_or(false, |conf| conf.2);
             // FIXME: allow customizing distance
-            let writers: Vec<_> = crate::vector::arroy_db_range_for_embedder(embedder_index)
+            let writer = ArroyWrapper::new(index.vector_arroy, embedder_index, binary_quantized);
-                .map(|k| ArroyWrapper::new(index.vector_arroy, k, binary_quantized))
-                .collect();
 
             // remove vectors for docids we want them removed
             let merger = remove_vectors_builder.build();
             let mut iter = merger.into_stream_merger_iter()?;
             while let Some((key, _)) = iter.next()? {
                 let docid = key.try_into().map(DocumentId::from_be_bytes).unwrap();
-
+                writer.del_items(wtxn, expected_dimension, docid)?;
-                for writer in &writers {
-                    // Uses invariant: vectors are packed in the first writers.
-                    if !writer.del_item(wtxn, expected_dimension, docid)? {
-                        break;
-                    }
-                }
             }
 
             // add generated embeddings
@@ -716,9 +708,7 @@ pub(crate) fn write_typed_chunk_into_index(
                         embeddings.embedding_count(),
                     )));
                 }
-                for (embedding, writer) in embeddings.iter().zip(&writers) {
+                writer.add_items(wtxn, docid, &embeddings)?;
-                    writer.add_item(wtxn, expected_dimension, docid, embedding)?;
-                }
             }
 
             // perform the manual diff
@@ -733,51 +723,14 @@ pub(crate) fn write_typed_chunk_into_index(
                 if let Some(value) = vector_deladd_obkv.get(DelAdd::Deletion) {
                     let vector: Vec<f32> = pod_collect_to_vec(value);
 
-                    let mut deleted_index = None;
+                    writer.del_item(wtxn, docid, &vector)?;
-                    for (index, writer) in writers.iter().enumerate() {
-                        let Some(candidate) = writer.item_vector(wtxn, docid)? else {
-                            // uses invariant: vectors are packed in the first writers.
-                            break;
-                        };
-                        if candidate == vector {
-                            writer.del_item(wtxn, expected_dimension, docid)?;
-                            deleted_index = Some(index);
-                        }
-                    }
-
-                    // 🥲 enforce invariant: vectors are packed in the first writers.
-                    if let Some(deleted_index) = deleted_index {
-                        let mut last_index_with_a_vector = None;
-                        for (index, writer) in writers.iter().enumerate().skip(deleted_index) {
-                            let Some(candidate) = writer.item_vector(wtxn, docid)? else {
-                                break;
-                            };
-                            last_index_with_a_vector = Some((index, candidate));
-                        }
-                        if let Some((last_index, vector)) = last_index_with_a_vector {
-                            // unwrap: computed the index from the list of writers
-                            let writer = writers.get(last_index).unwrap();
-                            writer.del_item(wtxn, expected_dimension, docid)?;
-                            writers.get(deleted_index).unwrap().add_item(
-                                wtxn,
-                                expected_dimension,
-                                docid,
-                                &vector,
-                            )?;
-                        }
-                    }
                 }
 
                 if let Some(value) = vector_deladd_obkv.get(DelAdd::Addition) {
                     let vector = pod_collect_to_vec(value);
 
                     // overflow was detected during vector extraction.
-                    for writer in &writers {
+                    writer.add_item(wtxn, docid, &vector)?;
-                        if !writer.contains_item(wtxn, expected_dimension, docid)? {
-                            writer.add_item(wtxn, expected_dimension, docid, &vector)?;
-                            break;
-                        }
-                    }
                 }
             }
 

milli/src/vector/mod.rs

@@ -32,105 +32,243 @@ pub const REQUEST_PARALLELISM: usize = 40;
 
 pub struct ArroyWrapper {
     quantized: bool,
-    index: u16,
+    embedder_index: u8,
     database: arroy::Database<Unspecified>,
 }
 
 impl ArroyWrapper {
-    pub fn new(database: arroy::Database<Unspecified>, index: u16, quantized: bool) -> Self {
+    pub fn new(
-        Self { database, index, quantized }
+        database: arroy::Database<Unspecified>,
+        embedder_index: u8,
+        quantized: bool,
+    ) -> Self {
+        Self { database, embedder_index, quantized }
     }
 
-    pub fn index(&self) -> u16 {
+    pub fn embedder_index(&self) -> u8 {
-        self.index
+        self.embedder_index
+    }
+
+    fn readers<'a, D: arroy::Distance>(
+        &'a self,
+        rtxn: &'a RoTxn<'a>,
+        db: arroy::Database<D>,
+    ) -> impl Iterator<Item = Result<arroy::Reader<D>, arroy::Error>> + 'a {
+        arroy_db_range_for_embedder(self.embedder_index).map_while(move |index| {
+            match arroy::Reader::open(rtxn, index, db) {
+                Ok(reader) => match reader.is_empty(rtxn) {
+                    Ok(false) => Some(Ok(reader)),
+                    Ok(true) => None,
+                    Err(e) => Some(Err(e)),
+                },
+                Err(arroy::Error::MissingMetadata(_)) => None,
+                Err(e) => Some(Err(e)),
+            }
+        })
     }
 
     pub fn dimensions(&self, rtxn: &RoTxn) -> Result<usize, arroy::Error> {
+        let first_id = arroy_db_range_for_embedder(self.embedder_index).next().unwrap();
         if self.quantized {
-            Ok(arroy::Reader::open(rtxn, self.index, self.quantized_db())?.dimensions())
+            Ok(arroy::Reader::open(rtxn, first_id, self.quantized_db())?.dimensions())
         } else {
-            Ok(arroy::Reader::open(rtxn, self.index, self.angular_db())?.dimensions())
+            Ok(arroy::Reader::open(rtxn, first_id, self.angular_db())?.dimensions())
         }
     }
 
-    pub fn quantize(
+    pub fn build_and_quantize<R: rand::Rng + rand::SeedableRng>(
         &mut self,
         wtxn: &mut RwTxn,
-        index: u16,
+        rng: &mut R,
         dimension: usize,
+        quantizing: bool,
     ) -> Result<(), arroy::Error> {
-        if !self.quantized {
+        for index in arroy_db_range_for_embedder(self.embedder_index) {
+            if self.quantized {
+                let writer = arroy::Writer::new(self.quantized_db(), index, dimension);
+                if writer.need_build(wtxn)? {
+                    writer.build(wtxn, rng, None)?
+                } else if writer.is_empty(wtxn)? {
+                    break;
+                }
+            } else {
                 let writer = arroy::Writer::new(self.angular_db(), index, dimension);
+                // If we are quantizing the databases, we can't know from meilisearch
+                // if the db was empty but still contained the wrong metadata, thus we need
+                // to quantize everything and can't stop early. Since this operation can
+                // only happens once in the life of an embedder, it's not very performances
+                // sensitive.
+                if quantizing && !self.quantized {
+                    let writer =
                         writer.prepare_changing_distance::<BinaryQuantizedAngular>(wtxn)?;
-            self.quantized = true;
+                    writer.build(wtxn, rng, None)?
+                } else if writer.need_build(wtxn)? {
+                    writer.build(wtxn, rng, None)?
+                } else if writer.is_empty(wtxn)? {
+                    break;
+                }
+            }
         }
         Ok(())
     }
 
-    pub fn need_build(&self, rtxn: &RoTxn, dimension: usize) -> Result<bool, arroy::Error> {
+    /// Overwrite all the embeddings associated with the index and item ID.
-        if self.quantized {
+    /// /!\ It won't remove embeddings after the last passed embedding, which can leave stale embeddings.
-            arroy::Writer::new(self.quantized_db(), self.index, dimension).need_build(rtxn)
+    ///     You should call `del_items` on the `item_id` before calling this method.
-        } else {
+    /// /!\ Cannot insert more than u8::MAX embeddings; after inserting u8::MAX embeddings, all the remaining ones will be silently ignored.
-            arroy::Writer::new(self.angular_db(), self.index, dimension).need_build(rtxn)
+    pub fn add_items(
-        }
-    }
-
-    pub fn build<R: rand::Rng + rand::SeedableRng>(
         &self,
         wtxn: &mut RwTxn,
-        rng: &mut R,
+        item_id: arroy::ItemId,
-        dimension: usize,
+        embeddings: &Embeddings<f32>,
     ) -> Result<(), arroy::Error> {
+        let dimension = embeddings.dimension();
+        for (index, vector) in
+            arroy_db_range_for_embedder(self.embedder_index).zip(embeddings.iter())
+        {
             if self.quantized {
-            arroy::Writer::new(self.quantized_db(), self.index, dimension).build(wtxn, rng, None)
+                arroy::Writer::new(self.quantized_db(), index, dimension)
+                    .add_item(wtxn, item_id, vector)?
             } else {
-            arroy::Writer::new(self.angular_db(), self.index, dimension).build(wtxn, rng, None)
+                arroy::Writer::new(self.angular_db(), index, dimension)
+                    .add_item(wtxn, item_id, vector)?
             }
         }
+        Ok(())
+    }
 
+    /// Add one document int for this index where we can find an empty spot.
     pub fn add_item(
         &self,
         wtxn: &mut RwTxn,
-        dimension: usize,
         item_id: arroy::ItemId,
         vector: &[f32],
     ) -> Result<(), arroy::Error> {
         if self.quantized {
-            arroy::Writer::new(self.quantized_db(), self.index, dimension)
+            self._add_item(wtxn, self.quantized_db(), item_id, vector)
-                .add_item(wtxn, item_id, vector)
         } else {
-            arroy::Writer::new(self.angular_db(), self.index, dimension)
+            self._add_item(wtxn, self.angular_db(), item_id, vector)
-                .add_item(wtxn, item_id, vector)
         }
     }
 
-    pub fn del_item(
+    fn _add_item<D: arroy::Distance>(
+        &self,
+        wtxn: &mut RwTxn,
+        db: arroy::Database<D>,
+        item_id: arroy::ItemId,
+        vector: &[f32],
+    ) -> Result<(), arroy::Error> {
+        let dimension = vector.len();
+
+        for index in arroy_db_range_for_embedder(self.embedder_index) {
+            let writer = arroy::Writer::new(db, index, dimension);
+            if !writer.contains_item(wtxn, item_id)? {
+                writer.add_item(wtxn, item_id, vector)?;
+                break;
+            }
+        }
+        Ok(())
+    }
+
+    /// Delete all embeddings from a specific `item_id`
+    pub fn del_items(
         &self,
         wtxn: &mut RwTxn,
         dimension: usize,
         item_id: arroy::ItemId,
+    ) -> Result<(), arroy::Error> {
+        for index in arroy_db_range_for_embedder(self.embedder_index) {
+            if self.quantized {
+                let writer = arroy::Writer::new(self.quantized_db(), index, dimension);
+                if !writer.del_item(wtxn, item_id)? {
+                    break;
+                }
+            } else {
+                let writer = arroy::Writer::new(self.angular_db(), index, dimension);
+                if !writer.del_item(wtxn, item_id)? {
+                    break;
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    /// Delete one item.
+    pub fn del_item(
+        &self,
+        wtxn: &mut RwTxn,
+        item_id: arroy::ItemId,
+        vector: &[f32],
     ) -> Result<bool, arroy::Error> {
         if self.quantized {
-            arroy::Writer::new(self.quantized_db(), self.index, dimension).del_item(wtxn, item_id)
+            self._del_item(wtxn, self.quantized_db(), item_id, vector)
         } else {
-            arroy::Writer::new(self.angular_db(), self.index, dimension).del_item(wtxn, item_id)
+            self._del_item(wtxn, self.angular_db(), item_id, vector)
         }
     }
 
+    fn _del_item<D: arroy::Distance>(
+        &self,
+        wtxn: &mut RwTxn,
+        db: arroy::Database<D>,
+        item_id: arroy::ItemId,
+        vector: &[f32],
+    ) -> Result<bool, arroy::Error> {
+        let dimension = vector.len();
+        let mut deleted_index = None;
+
+        for index in arroy_db_range_for_embedder(self.embedder_index) {
+            let writer = arroy::Writer::new(db, index, dimension);
+            let Some(candidate) = writer.item_vector(wtxn, item_id)? else {
+                // uses invariant: vectors are packed in the first writers.
+                break;
+            };
+            if candidate == vector {
+                writer.del_item(wtxn, item_id)?;
+                deleted_index = Some(index);
+            }
+        }
+
+        // 🥲 enforce invariant: vectors are packed in the first writers.
+        if let Some(deleted_index) = deleted_index {
+            let mut last_index_with_a_vector = None;
+            for index in
+                arroy_db_range_for_embedder(self.embedder_index).skip(deleted_index as usize)
+            {
+                let writer = arroy::Writer::new(db, index, dimension);
+                let Some(candidate) = writer.item_vector(wtxn, item_id)? else {
+                    break;
+                };
+                last_index_with_a_vector = Some((index, candidate));
+            }
+            if let Some((last_index, vector)) = last_index_with_a_vector {
+                let writer = arroy::Writer::new(db, last_index, dimension);
+                writer.del_item(wtxn, item_id)?;
+                let writer = arroy::Writer::new(db, deleted_index, dimension);
+                writer.add_item(wtxn, item_id, &vector)?;
+            }
+        }
+        Ok(deleted_index.is_some())
+    }
+
     pub fn clear(&self, wtxn: &mut RwTxn, dimension: usize) -> Result<(), arroy::Error> {
+        for index in arroy_db_range_for_embedder(self.embedder_index) {
             if self.quantized {
-            arroy::Writer::new(self.quantized_db(), self.index, dimension).clear(wtxn)
+                let writer = arroy::Writer::new(self.quantized_db(), index, dimension);
+                if writer.is_empty(wtxn)? {
+                    break;
+                }
+                writer.clear(wtxn)?;
             } else {
-            arroy::Writer::new(self.angular_db(), self.index, dimension).clear(wtxn)
+                let writer = arroy::Writer::new(self.angular_db(), index, dimension);
+                if writer.is_empty(wtxn)? {
+                    break;
+                }
+                writer.clear(wtxn)?;
             }
         }
-
+        Ok(())
-    pub fn is_empty(&self, rtxn: &RoTxn, dimension: usize) -> Result<bool, arroy::Error> {
-        if self.quantized {
-            arroy::Writer::new(self.quantized_db(), self.index, dimension).is_empty(rtxn)
-        } else {
-            arroy::Writer::new(self.angular_db(), self.index, dimension).is_empty(rtxn)
-        }
     }
 
     pub fn contains_item(
@@ -139,11 +277,25 @@ impl ArroyWrapper {
         dimension: usize,
         item: arroy::ItemId,
     ) -> Result<bool, arroy::Error> {
-        if self.quantized {
+        for index in arroy_db_range_for_embedder(self.embedder_index) {
-            arroy::Writer::new(self.quantized_db(), self.index, dimension).contains_item(rtxn, item)
+            let contains = if self.quantized {
-        } else {
+                let writer = arroy::Writer::new(self.quantized_db(), index, dimension);
-            arroy::Writer::new(self.angular_db(), self.index, dimension).contains_item(rtxn, item)
+                if writer.is_empty(rtxn)? {
+                    break;
                 }
+                writer.contains_item(rtxn, item)?
+            } else {
+                let writer = arroy::Writer::new(self.angular_db(), index, dimension);
+                if writer.is_empty(rtxn)? {
+                    break;
+                }
+                writer.contains_item(rtxn, item)?
+            };
+            if contains {
+                return Ok(contains);
+            }
+        }
+        Ok(false)
     }
 
     pub fn nns_by_item(
@@ -152,38 +304,91 @@ impl ArroyWrapper {
         item: ItemId,
         limit: usize,
         filter: Option<&RoaringBitmap>,
-    ) -> Result<Option<Vec<(ItemId, f32)>>, arroy::Error> {
+    ) -> Result<Vec<(ItemId, f32)>, arroy::Error> {
         if self.quantized {
-            arroy::Reader::open(rtxn, self.index, self.quantized_db())?
+            self._nns_by_item(rtxn, self.quantized_db(), item, limit, filter)
-                .nns_by_item(rtxn, item, limit, None, None, filter)
         } else {
-            arroy::Reader::open(rtxn, self.index, self.angular_db())?
+            self._nns_by_item(rtxn, self.angular_db(), item, limit, filter)
-                .nns_by_item(rtxn, item, limit, None, None, filter)
         }
     }
 
+    fn _nns_by_item<D: arroy::Distance>(
+        &self,
+        rtxn: &RoTxn,
+        db: arroy::Database<D>,
+        item: ItemId,
+        limit: usize,
+        filter: Option<&RoaringBitmap>,
+    ) -> Result<Vec<(ItemId, f32)>, arroy::Error> {
+        let mut results = Vec::new();
+
+        for reader in self.readers(rtxn, db) {
+            let ret = reader?.nns_by_item(rtxn, item, limit, None, None, filter)?;
+            if let Some(mut ret) = ret {
+                results.append(&mut ret);
+            } else {
+                break;
+            }
+        }
+        results.sort_unstable_by_key(|(_, distance)| OrderedFloat(*distance));
+        Ok(results)
+    }
+
     pub fn nns_by_vector(
         &self,
-        txn: &RoTxn,
+        rtxn: &RoTxn,
-        item: &[f32],
+        vector: &[f32],
         limit: usize,
         filter: Option<&RoaringBitmap>,
     ) -> Result<Vec<(ItemId, f32)>, arroy::Error> {
         if self.quantized {
-            arroy::Reader::open(txn, self.index, self.quantized_db())?
+            self._nns_by_vector(rtxn, self.quantized_db(), vector, limit, filter)
-                .nns_by_vector(txn, item, limit, None, None, filter)
         } else {
-            arroy::Reader::open(txn, self.index, self.angular_db())?
+            self._nns_by_vector(rtxn, self.angular_db(), vector, limit, filter)
-                .nns_by_vector(txn, item, limit, None, None, filter)
         }
     }
 
-    pub fn item_vector(&self, rtxn: &RoTxn, docid: u32) -> Result<Option<Vec<f32>>, arroy::Error> {
+    fn _nns_by_vector<D: arroy::Distance>(
-        if self.quantized {
+        &self,
-            arroy::Reader::open(rtxn, self.index, self.quantized_db())?.item_vector(rtxn, docid)
+        rtxn: &RoTxn,
-        } else {
+        db: arroy::Database<D>,
-            arroy::Reader::open(rtxn, self.index, self.angular_db())?.item_vector(rtxn, docid)
+        vector: &[f32],
+        limit: usize,
+        filter: Option<&RoaringBitmap>,
+    ) -> Result<Vec<(ItemId, f32)>, arroy::Error> {
+        let mut results = Vec::new();
+
+        for reader in self.readers(rtxn, db) {
+            let mut ret = reader?.nns_by_vector(rtxn, vector, limit, None, None, filter)?;
+            results.append(&mut ret);
         }
+
+        results.sort_unstable_by_key(|(_, distance)| OrderedFloat(*distance));
+
+        Ok(results)
+    }
+
+    pub fn item_vectors(&self, rtxn: &RoTxn, item_id: u32) -> Result<Vec<Vec<f32>>, arroy::Error> {
+        let mut vectors = Vec::new();
+
+        if self.quantized {
+            for reader in self.readers(rtxn, self.quantized_db()) {
+                if let Some(vec) = reader?.item_vector(rtxn, item_id)? {
+                    vectors.push(vec);
+                } else {
+                    break;
+                }
+            }
+        } else {
+            for reader in self.readers(rtxn, self.angular_db()) {
+                if let Some(vec) = reader?.item_vector(rtxn, item_id)? {
+                    vectors.push(vec);
+                } else {
+                    break;
+                }
+            }
+        }
+        Ok(vectors)
     }
 
     fn angular_db(&self) -> arroy::Database<Angular> {