Merge #4646

4646: Reduce `Transform`'s disk usage r=Kerollmops a=Kerollmops

This PR implements what is described in #4485. It reduces the number of disk writes and disk usage.

Co-authored-by: Clément Renault <clement@meilisearch.com>

meilisearch/tests/search/mod.rs

@@ -680,6 +680,26 @@ async fn search_facet_distribution() {
             },
         )
         .await;
+
+    index.update_settings(json!({"filterableAttributes": ["doggos.name"]})).await;
+    index.wait_task(5).await;
+
+    index
+        .search(
+            json!({
+                "facets": ["doggos.name"]
+            }),
+            |response, code| {
+                assert_eq!(code, 200, "{}", response);
+                let dist = response["facetDistribution"].as_object().unwrap();
+                assert_eq!(dist.len(), 1);
+                assert_eq!(
+                    dist["doggos.name"],
+                    json!({ "bobby": 1, "buddy": 1, "gros bill": 1, "turbo": 1, "fast": 1})
+                );
+            },
+        )
+        .await;
 }
 
 #[actix_rt::test]

milli/src/heed_codec/roaring_bitmap/cbo_roaring_bitmap_codec.rs

@@ -195,7 +195,7 @@ mod tests {
     fn merge_cbo_roaring_bitmaps() {
         let mut buffer = Vec::new();
 
-        let small_data = vec![
+        let small_data = [
             RoaringBitmap::from_sorted_iter(1..4).unwrap(),
             RoaringBitmap::from_sorted_iter(2..5).unwrap(),
             RoaringBitmap::from_sorted_iter(4..6).unwrap(),
@@ -209,7 +209,7 @@ mod tests {
         let expected = RoaringBitmap::from_sorted_iter(1..6).unwrap();
         assert_eq!(bitmap, expected);
 
-        let medium_data = vec![
+        let medium_data = [
             RoaringBitmap::from_sorted_iter(1..4).unwrap(),
             RoaringBitmap::from_sorted_iter(2..5).unwrap(),
             RoaringBitmap::from_sorted_iter(4..8).unwrap(),

milli/src/lib.rs

@@ -354,8 +354,7 @@ pub fn is_faceted(field: &str, faceted_fields: impl IntoIterator<Item = impl AsR
 /// assert!(!is_faceted_by("animaux.chien", "animaux.chie"));
 /// ```
 pub fn is_faceted_by(field: &str, facet: &str) -> bool {
-    field.starts_with(facet)
+    field.starts_with(facet) && field[facet.len()..].chars().next().map_or(true, |c| c == '.')
-        && field[facet.len()..].chars().next().map(|c| c == '.').unwrap_or(true)
 }
 
 pub fn normalize_facet(original: &str) -> String {

milli/src/update/index_documents/mod.rs

@@ -6,6 +6,7 @@ mod typed_chunk;
 
 use std::collections::{HashMap, HashSet};
 use std::io::{Read, Seek};
+use std::iter;
 use std::num::NonZeroU32;
 use std::result::Result as StdResult;
 use std::sync::Arc;
@@ -359,7 +360,10 @@ where
                 let min_chunk_size = 1024 * 512; // 512KiB
 
                 // compute the chunk size from the number of available threads and the inputed data size.
-                let total_size = flattened_documents.metadata().map(|m| m.len());
+                let total_size = match flattened_documents.as_ref() {
+                    Some(flattened_documents) => flattened_documents.metadata().map(|m| m.len()),
+                    None => Ok(default_chunk_size as u64),
+                };
                 let current_num_threads = pool.current_num_threads();
                 // if we have more than 2 thread, create a number of chunk equal to 3/4 threads count
                 let chunk_count = if current_num_threads > 2 {
@@ -373,8 +377,14 @@ where
             }
         };
 
-        let original_documents = grenad::Reader::new(original_documents)?;
+        let original_documents = match original_documents {
-        let flattened_documents = grenad::Reader::new(flattened_documents)?;
+            Some(original_documents) => Some(grenad::Reader::new(original_documents)?),
+            None => None,
+        };
+        let flattened_documents = match flattened_documents {
+            Some(flattened_documents) => Some(grenad::Reader::new(flattened_documents)?),
+            None => None,
+        };
 
         let max_positions_per_attributes = self.indexer_config.max_positions_per_attributes;
 
@@ -396,12 +406,20 @@ where
                 let _enter = child_span.enter();
                 puffin::profile_scope!("extract_and_send_grenad_chunks");
                 // split obkv file into several chunks
-                let original_chunk_iter =
+                let original_chunk_iter = match original_documents {
-                    grenad_obkv_into_chunks(original_documents, pool_params, documents_chunk_size);
+                    Some(original_documents) => {
+                        grenad_obkv_into_chunks(original_documents,pool_params,documents_chunk_size).map(either::Left)
+                    },
+                    None => Ok(either::Right(iter::empty())),
+                };
 
                 // split obkv file into several chunks
-                let flattened_chunk_iter =
+                let flattened_chunk_iter = match flattened_documents {
-                    grenad_obkv_into_chunks(flattened_documents, pool_params, documents_chunk_size);
+                    Some(flattened_documents) => {
+                        grenad_obkv_into_chunks(flattened_documents, pool_params, documents_chunk_size).map(either::Left)
+                    },
+                    None => Ok(either::Right(iter::empty())),
+                };
 
                 let result = original_chunk_iter.and_then(|original_chunk| {
                     let flattened_chunk = flattened_chunk_iter?;

milli/src/update/index_documents/transform.rs

@@ -1,7 +1,7 @@
 use std::borrow::Cow;
 use std::collections::btree_map::Entry as BEntry;
 use std::collections::hash_map::Entry as HEntry;
-use std::collections::HashMap;
+use std::collections::{HashMap, HashSet};
 use std::fs::File;
 use std::io::{Read, Seek};
 
@@ -20,21 +20,21 @@ use super::{IndexDocumentsMethod, IndexerConfig};
 use crate::documents::{DocumentsBatchIndex, EnrichedDocument, EnrichedDocumentsBatchReader};
 use crate::error::{Error, InternalError, UserError};
 use crate::index::{db_name, main_key};
-use crate::update::del_add::{
+use crate::update::del_add::{into_del_add_obkv, DelAdd, DelAddOperation, KvReaderDelAdd};
-    del_add_from_two_obkvs, into_del_add_obkv, DelAdd, DelAddOperation, KvReaderDelAdd,
-};
 use crate::update::index_documents::GrenadParameters;
 use crate::update::settings::{InnerIndexSettings, InnerIndexSettingsDiff};
 use crate::update::{AvailableDocumentsIds, UpdateIndexingStep};
-use crate::{FieldDistribution, FieldId, FieldIdMapMissingEntry, FieldsIdsMap, Index, Result};
+use crate::{
+    is_faceted_by, FieldDistribution, FieldId, FieldIdMapMissingEntry, FieldsIdsMap, Index, Result,
+};
 
 pub struct TransformOutput {
     pub primary_key: String,
     pub settings_diff: InnerIndexSettingsDiff,
     pub field_distribution: FieldDistribution,
     pub documents_count: usize,
-    pub original_documents: File,
+    pub original_documents: Option<File>,
-    pub flattened_documents: File,
+    pub flattened_documents: Option<File>,
 }
 
 /// Extract the external ids, deduplicate and compute the new internal documents ids
@@ -808,11 +808,15 @@ impl<'a, 'i> Transform<'a, 'i> {
         })?;
 
         let old_inner_settings = InnerIndexSettings::from_index(self.index, wtxn)?;
+        let fields_ids_map = self.fields_ids_map;
+        let primary_key_id = self.index.primary_key(wtxn)?.and_then(|name| fields_ids_map.id(name));
         let mut new_inner_settings = old_inner_settings.clone();
-        new_inner_settings.fields_ids_map = self.fields_ids_map;
+        new_inner_settings.fields_ids_map = fields_ids_map;
+
         let settings_diff = InnerIndexSettingsDiff {
             old: old_inner_settings,
             new: new_inner_settings,
+            primary_key_id,
             embedding_configs_updated: false,
             settings_update_only: false,
         };
@@ -822,10 +826,12 @@ impl<'a, 'i> Transform<'a, 'i> {
             settings_diff,
             field_distribution,
             documents_count: self.documents_count,
-            original_documents: original_documents.into_inner().map_err(|err| err.into_error())?,
+            original_documents: Some(
-            flattened_documents: flattened_documents
+                original_documents.into_inner().map_err(|err| err.into_error())?,
-                .into_inner()
+            ),
-                .map_err(|err| err.into_error())?,
+            flattened_documents: Some(
+                flattened_documents.into_inner().map_err(|err| err.into_error())?,
+            ),
         })
     }
 
@@ -835,34 +841,66 @@ impl<'a, 'i> Transform<'a, 'i> {
     fn rebind_existing_document(
         old_obkv: KvReader<FieldId>,
         settings_diff: &InnerIndexSettingsDiff,
-        original_obkv_buffer: &mut Vec<u8>,
+        modified_faceted_fields: &HashSet<String>,
-        flattened_obkv_buffer: &mut Vec<u8>,
+        original_obkv_buffer: Option<&mut Vec<u8>>,
+        flattened_obkv_buffer: Option<&mut Vec<u8>>,
     ) -> Result<()> {
-        let mut old_fields_ids_map = settings_diff.old.fields_ids_map.clone();
+        // Always keep the primary key.
-        let mut new_fields_ids_map = settings_diff.new.fields_ids_map.clone();
+        let is_primary_key = |id: FieldId| -> bool { settings_diff.primary_key_id == Some(id) };
+
+        // If only the `searchableAttributes` has been changed, keep only the searchable fields.
+        let must_reindex_searchables = settings_diff.reindex_searchable();
+        let necessary_searchable_field = |id: FieldId| -> bool {
+            must_reindex_searchables
+                && (settings_diff.old.searchable_fields_ids.contains(&id)
+                    || settings_diff.new.searchable_fields_ids.contains(&id))
+        };
+
+        // If only a faceted field has been added, keep only this field.
+        let must_reindex_facets = settings_diff.reindex_facets();
+        let necessary_faceted_field = |id: FieldId| -> bool {
+            let field_name = settings_diff.new.fields_ids_map.name(id).unwrap();
+            must_reindex_facets
+                && modified_faceted_fields
+                    .iter()
+                    .any(|long| is_faceted_by(long, field_name) || is_faceted_by(field_name, long))
+        };
+
+        // Alway provide all fields when vectors are involved because
+        // we need the fields for the prompt/templating.
+        let reindex_vectors = settings_diff.reindex_vectors();
+
         let mut obkv_writer = KvWriter::<_, FieldId>::memory();
-        // We iterate over the new `FieldsIdsMap` ids in order and construct the new obkv.
+        for (id, val) in old_obkv.iter() {
-        for (id, name) in new_fields_ids_map.iter() {
+            if is_primary_key(id)
-            if let Some(val) = old_fields_ids_map.id(name).and_then(|id| old_obkv.get(id)) {
+                || necessary_searchable_field(id)
+                || necessary_faceted_field(id)
+                || reindex_vectors
+            {
                 obkv_writer.insert(id, val)?;
             }
         }
         let data = obkv_writer.into_inner()?;
-        let new_obkv = KvReader::<FieldId>::new(&data);
+        let obkv = KvReader::<FieldId>::new(&data);
-
-        // take the non-flattened version if flatten_from_fields_ids_map returns None.
-        let old_flattened = Self::flatten_from_fields_ids_map(&old_obkv, &mut old_fields_ids_map)?;
-        let old_flattened =
-            old_flattened.as_deref().map_or_else(|| old_obkv, KvReader::<FieldId>::new);
-        let new_flattened = Self::flatten_from_fields_ids_map(&new_obkv, &mut new_fields_ids_map)?;
-        let new_flattened =
-            new_flattened.as_deref().map_or_else(|| new_obkv, KvReader::<FieldId>::new);
 
+        if let Some(original_obkv_buffer) = original_obkv_buffer {
             original_obkv_buffer.clear();
-        flattened_obkv_buffer.clear();
+            into_del_add_obkv(obkv, DelAddOperation::DeletionAndAddition, original_obkv_buffer)?;
+        }
 
-        del_add_from_two_obkvs(&old_obkv, &new_obkv, original_obkv_buffer)?;
+        if let Some(flattened_obkv_buffer) = flattened_obkv_buffer {
-        del_add_from_two_obkvs(&old_flattened, &new_flattened, flattened_obkv_buffer)?;
+            // take the non-flattened version if flatten_from_fields_ids_map returns None.
+            let mut fields_ids_map = settings_diff.new.fields_ids_map.clone();
+            let flattened = Self::flatten_from_fields_ids_map(&obkv, &mut fields_ids_map)?;
+            let flattened = flattened.as_deref().map_or(obkv, KvReader::new);
+
+            flattened_obkv_buffer.clear();
+            into_del_add_obkv(
+                flattened,
+                DelAddOperation::DeletionAndAddition,
+                flattened_obkv_buffer,
+            )?;
+        }
 
         Ok(())
     }
@@ -891,25 +929,36 @@ impl<'a, 'i> Transform<'a, 'i> {
         let documents_count = documents_ids.len() as usize;
 
         // We initialize the sorter with the user indexing settings.
-        let mut original_sorter = create_sorter(
+        let mut original_sorter = if settings_diff.reindex_vectors() {
+            Some(create_sorter(
                 grenad::SortAlgorithm::Stable,
                 keep_first,
                 self.indexer_settings.chunk_compression_type,
                 self.indexer_settings.chunk_compression_level,
                 self.indexer_settings.max_nb_chunks,
                 self.indexer_settings.max_memory.map(|mem| mem / 2),
-        );
+            ))
+        } else {
+            None
+        };
 
         // We initialize the sorter with the user indexing settings.
-        let mut flattened_sorter = create_sorter(
+        let mut flattened_sorter =
+            if settings_diff.reindex_searchable() || settings_diff.reindex_facets() {
+                Some(create_sorter(
                     grenad::SortAlgorithm::Stable,
                     keep_first,
                     self.indexer_settings.chunk_compression_type,
                     self.indexer_settings.chunk_compression_level,
                     self.indexer_settings.max_nb_chunks,
                     self.indexer_settings.max_memory.map(|mem| mem / 2),
-        );
+                ))
+            } else {
+                None
+            };
 
+        if original_sorter.is_some() || flattened_sorter.is_some() {
+            let modified_faceted_fields = settings_diff.modified_faceted_fields();
             let mut original_obkv_buffer = Vec::new();
             let mut flattened_obkv_buffer = Vec::new();
             let mut document_sorter_key_buffer = Vec::new();
@@ -922,16 +971,22 @@ impl<'a, 'i> Transform<'a, 'i> {
                 Self::rebind_existing_document(
                     old_obkv,
                     &settings_diff,
-                &mut original_obkv_buffer,
+                    &modified_faceted_fields,
-                &mut flattened_obkv_buffer,
+                    Some(&mut original_obkv_buffer).filter(|_| original_sorter.is_some()),
+                    Some(&mut flattened_obkv_buffer).filter(|_| flattened_sorter.is_some()),
                 )?;
 
+                if let Some(original_sorter) = original_sorter.as_mut() {
                     document_sorter_key_buffer.clear();
                     document_sorter_key_buffer.extend_from_slice(&docid.to_be_bytes());
                     document_sorter_key_buffer.extend_from_slice(external_id.as_bytes());
                     original_sorter.insert(&document_sorter_key_buffer, &original_obkv_buffer)?;
+                }
+                if let Some(flattened_sorter) = flattened_sorter.as_mut() {
                     flattened_sorter.insert(docid.to_be_bytes(), &flattened_obkv_buffer)?;
                 }
+            }
+        }
 
         let grenad_params = GrenadParameters {
             chunk_compression_type: self.indexer_settings.chunk_compression_type,
@@ -941,17 +996,22 @@ impl<'a, 'i> Transform<'a, 'i> {
         };
 
         // Once we have written all the documents, we merge everything into a Reader.
-        let original_documents = sorter_into_reader(original_sorter, grenad_params)?;
+        let flattened_documents = match flattened_sorter {
-
+            Some(flattened_sorter) => Some(sorter_into_reader(flattened_sorter, grenad_params)?),
-        let flattened_documents = sorter_into_reader(flattened_sorter, grenad_params)?;
+            None => None,
+        };
+        let original_documents = match original_sorter {
+            Some(original_sorter) => Some(sorter_into_reader(original_sorter, grenad_params)?),
+            None => None,
+        };
 
         Ok(TransformOutput {
             primary_key,
             field_distribution,
             settings_diff,
             documents_count,
-            original_documents: original_documents.into_inner().into_inner(),
+            original_documents: original_documents.map(|od| od.into_inner().into_inner()),
-            flattened_documents: flattened_documents.into_inner().into_inner(),
+            flattened_documents: flattened_documents.map(|fd| fd.into_inner().into_inner()),
         })
     }
 }

milli/src/update/settings.rs

@@ -1067,10 +1067,17 @@ impl<'a, 't, 'i> Settings<'a, 't, 'i> {
         // 3. Keep the old vectors but reattempt indexing on a prompt change: only actually changed prompt will need embedding + storage
         let embedding_configs_updated = self.update_embedding_configs()?;
 
-        let new_inner_settings = InnerIndexSettings::from_index(self.index, self.wtxn)?;
+        let mut new_inner_settings = InnerIndexSettings::from_index(self.index, self.wtxn)?;
+        new_inner_settings.recompute_facets(self.wtxn, self.index)?;
+
+        let primary_key_id = self
+            .index
+            .primary_key(self.wtxn)?
+            .and_then(|name| new_inner_settings.fields_ids_map.id(name));
         let inner_settings_diff = InnerIndexSettingsDiff {
             old: old_inner_settings,
             new: new_inner_settings,
+            primary_key_id,
             embedding_configs_updated,
             settings_update_only: true,
         };
@@ -1086,10 +1093,9 @@ impl<'a, 't, 'i> Settings<'a, 't, 'i> {
 pub struct InnerIndexSettingsDiff {
     pub(crate) old: InnerIndexSettings,
     pub(crate) new: InnerIndexSettings,
-
+    pub(crate) primary_key_id: Option<FieldId>,
     // TODO: compare directly the embedders.
     pub(crate) embedding_configs_updated: bool,
-
     pub(crate) settings_update_only: bool,
 }
 
@@ -1099,12 +1105,7 @@ impl InnerIndexSettingsDiff {
     }
 
     pub fn reindex_searchable(&self) -> bool {
-        self.old
+        self.old.stop_words.as_ref().map(|set| set.as_fst().as_bytes())
-            .fields_ids_map
-            .iter()
-            .zip(self.new.fields_ids_map.iter())
-            .any(|(old, new)| old != new)
-            || self.old.stop_words.as_ref().map(|set| set.as_fst().as_bytes())
             != self.new.stop_words.as_ref().map(|set| set.as_fst().as_bytes())
             || self.old.allowed_separators != self.new.allowed_separators
             || self.old.dictionary != self.new.dictionary
@@ -1132,15 +1133,7 @@ impl InnerIndexSettingsDiff {
             return true;
         }
 
-        let faceted_updated =
+        (existing_fields - old_faceted_fields) != (existing_fields - new_faceted_fields)
-            (existing_fields - old_faceted_fields) != (existing_fields - new_faceted_fields);
-
-        self.old
-            .fields_ids_map
-            .iter()
-            .zip(self.new.fields_ids_map.iter())
-            .any(|(old, new)| old != new)
-            || faceted_updated
     }
 
     pub fn reindex_vectors(&self) -> bool {
@@ -1150,6 +1143,10 @@ impl InnerIndexSettingsDiff {
     pub fn settings_update_only(&self) -> bool {
         self.settings_update_only
     }
+
+    pub fn modified_faceted_fields(&self) -> HashSet<String> {
+        &self.old.user_defined_faceted_fields ^ &self.new.user_defined_faceted_fields
+    }
 }
 
 #[derive(Clone)]