Log when an entry is too large to fit in the BBQueue

5124: Optimize Prefixes and Merges r=ManyTheFish a=Kerollmops

In this PR, we plan to optimize the read of LMDB to use read the entries in lexicographic order and better use the memory-mapping OS cache:

 - Optimize the prefix generation for word position docids (`@manythefish)`
 - Optimize the parallel merging of the caches to sort entries before merging the caches (`@kerollmops)`
 
## Benchmarks on 1cpu 2gb gpo3 (5k IOps)
 
Before on the tag meilisearch-v1.12.0-rc.3.

```
word_position_docids:merge_and_send_docids: 988s
compute_word_fst: 23.3s
word_pair_proximity_docids:merge_and_send_docids: 428s
compute_word_prefix_fid_docids:recompute_modified_prefixes: 76.3s
compute_word_prefix_position_docids:recompute_modified_prefixes:from_prefixes: 429s
```

After sorting the whole `HashMap`s in a `Vec` on this branch.

```
word_position_docids:merge_and_send_docids: 202s
compute_word_fst: 20.4s
word_pair_proximity_docids:merge_and_send_docids: 427s
compute_word_prefix_fid_docids:recompute_modified_prefixes: 65.5s
compute_word_prefix_position_docids:recompute_modified_prefixes:from_prefixes: 62.5s
```

Co-authored-by: ManyTheFish <many@meilisearch.com>
Co-authored-by: Kerollmops <clement@meilisearch.com>

Cargo.lock

@@ -492,7 +492,7 @@ checksum = "72b3254f16251a8381aa12e40e3c4d2f0199f8c6508fbecb9d91f575e0fbb8c6"
 [[package]]
 name = "bbqueue"
 version = "0.5.1"
-source = "git+https://github.com/kerollmops/bbqueue#cbb87cc707b5af415ef203bdaf2443e06ba0d6d4"
+source = "git+https://github.com/meilisearch/bbqueue#cbb87cc707b5af415ef203bdaf2443e06ba0d6d4"
 
 [[package]]
 name = "benchmarks"

crates/index-scheduler/src/lib.rs

@@ -1440,7 +1440,7 @@ impl IndexScheduler {
 
         // if the task doesn't delete anything and 50% of the task queue is full, we must refuse to enqueue the incomming task
         if !matches!(&kind, KindWithContent::TaskDeletion { tasks, .. } if !tasks.is_empty())
-            && (self.env.non_free_pages_size()? * 100) / self.env.info().map_size as u64 > 50
+            && (self.env.non_free_pages_size()? * 100) / self.env.info().map_size as u64 > 40
         {
             return Err(Error::NoSpaceLeftInTaskQueue);
         }

crates/milli/Cargo.toml

@@ -98,7 +98,7 @@ allocator-api2 = "0.2.18"
 rustc-hash = "2.0.0"
 uell = "0.1.0"
 enum-iterator = "2.1.0"
-bbqueue = { git = "https://github.com/kerollmops/bbqueue" }
+bbqueue = { git = "https://github.com/meilisearch/bbqueue" }
 flume = { version = "0.11.1", default-features = false }
 
 [dev-dependencies]

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

@@ -7,7 +7,7 @@ use std::num::NonZeroU16;
 use std::ops::Range;
 use std::time::Duration;
 
-use bbqueue::framed::{FrameGrantR, FrameGrantW, FrameProducer};
+use bbqueue::framed::{FrameGrantR, FrameProducer};
 use bbqueue::BBBuffer;
 use bytemuck::{checked, CheckedBitPattern, NoUninit};
 use flume::{RecvTimeoutError, SendError};
@@ -454,14 +454,10 @@ impl<'b> ExtractorBbqueueSender<'b> {
         }
 
         // Spin loop to have a frame the size we requested.
-        let mut grant = reserve_grant(&mut producer, total_length, &self.sender)?;
+        reserve_and_write_grant(&mut producer, total_length, &self.sender, |grant| {
-        payload_header.serialize_into(&mut grant);
+            payload_header.serialize_into(grant);
-
+            Ok(())
-        // We only send a wake up message when the channel is empty
+        })?;
-        // so that we don't fill the channel with too many WakeUps.
-        if self.sender.is_empty() {
-            self.sender.send(ReceiverAction::WakeUp).unwrap();
-        }
 
         Ok(())
     }
@@ -484,6 +480,8 @@ impl<'b> ExtractorBbqueueSender<'b> {
         let payload_header = EntryHeader::ArroySetVectors(arroy_set_vector);
         let total_length = EntryHeader::total_set_vectors_size(embeddings.len(), dimensions);
         if total_length > capacity {
+            tracing::trace!("We are spilling a large vector that is {total_length} bytes which is larger than the capacity of {capacity} bytes");
+
             let mut value_file = tempfile::tempfile().map(BufWriter::new)?;
             for embedding in embeddings {
                 let mut embedding_bytes = bytemuck::cast_slice(embedding);
@@ -500,8 +498,7 @@ impl<'b> ExtractorBbqueueSender<'b> {
         }
 
         // Spin loop to have a frame the size we requested.
-        let mut grant = reserve_grant(&mut producer, total_length, &self.sender)?;
+        reserve_and_write_grant(&mut producer, total_length, &self.sender, |grant| {
-
             let header_size = payload_header.header_size();
             let (header_bytes, remaining) = grant.split_at_mut(header_size);
             payload_header.serialize_into(header_bytes);
@@ -513,11 +510,8 @@ impl<'b> ExtractorBbqueueSender<'b> {
                 }
             }
 
-        // We only send a wake up message when the channel is empty
+            Ok(())
-        // so that we don't fill the channel with too many WakeUps.
+        })?;
-        if self.sender.is_empty() {
-            self.sender.send(ReceiverAction::WakeUp).unwrap();
-        }
 
         Ok(())
     }
@@ -556,6 +550,8 @@ impl<'b> ExtractorBbqueueSender<'b> {
         let payload_header = EntryHeader::DbOperation(operation);
         let total_length = EntryHeader::total_key_value_size(key_length, value_length);
         if total_length > capacity {
+            tracing::trace!("We are spilling a large entry that is {total_length} bytes which is larger than the capacity of {capacity} bytes");
+
             let mut key_buffer = vec![0; key_length.get() as usize].into_boxed_slice();
             let value_file = tempfile::tempfile()?;
             value_file.set_len(value_length.try_into().unwrap())?;
@@ -575,19 +571,13 @@ impl<'b> ExtractorBbqueueSender<'b> {
         }
 
         // Spin loop to have a frame the size we requested.
-        let mut grant = reserve_grant(&mut producer, total_length, &self.sender)?;
+        reserve_and_write_grant(&mut producer, total_length, &self.sender, |grant| {
-
             let header_size = payload_header.header_size();
             let (header_bytes, remaining) = grant.split_at_mut(header_size);
             payload_header.serialize_into(header_bytes);
             let (key_buffer, value_buffer) = remaining.split_at_mut(key_length.get() as usize);
-        key_value_writer(key_buffer, value_buffer)?;
+            key_value_writer(key_buffer, value_buffer)
-
+        })?;
-        // We only send a wake up message when the channel is empty
-        // so that we don't fill the channel with too many WakeUps.
-        if self.sender.is_empty() {
-            self.sender.send(ReceiverAction::WakeUp).unwrap();
-        }
 
         Ok(())
     }
@@ -629,37 +619,44 @@ impl<'b> ExtractorBbqueueSender<'b> {
         }
 
         // Spin loop to have a frame the size we requested.
-        let mut grant = reserve_grant(&mut producer, total_length, &self.sender)?;
+        reserve_and_write_grant(&mut producer, total_length, &self.sender, |grant| {
-
             let header_size = payload_header.header_size();
             let (header_bytes, remaining) = grant.split_at_mut(header_size);
             payload_header.serialize_into(header_bytes);
-        key_writer(remaining)?;
+            key_writer(remaining)
-
+        })?;
-        // We only send a wake up message when the channel is empty
-        // so that we don't fill the channel with too many WakeUps.
-        if self.sender.is_empty() {
-            self.sender.send(ReceiverAction::WakeUp).unwrap();
-        }
 
         Ok(())
     }
 }
 
-/// Try to reserve a frame grant of `total_length` by spin looping
+/// Try to reserve a frame grant of `total_length` by spin
-/// on the BBQueue buffer and panics if the receiver has been disconnected.
+/// looping on the BBQueue buffer, panics if the receiver
-fn reserve_grant<'b>(
+/// has been disconnected or send a WakeUp message if necessary.
-    producer: &mut FrameProducer<'b>,
+fn reserve_and_write_grant<F>(
+    producer: &mut FrameProducer,
     total_length: usize,
     sender: &flume::Sender<ReceiverAction>,
-) -> crate::Result<FrameGrantW<'b>> {
+    f: F,
+) -> crate::Result<()>
+where
+    F: FnOnce(&mut [u8]) -> crate::Result<()>,
+{
     loop {
         for _ in 0..10_000 {
             match producer.grant(total_length) {
                 Ok(mut grant) => {
                     // We could commit only the used memory.
-                    grant.to_commit(total_length);
+                    f(&mut grant)?;
-                    return Ok(grant);
+                    grant.commit(total_length);
+
+                    // We only send a wake up message when the channel is empty
+                    // so that we don't fill the channel with too many WakeUps.
+                    if sender.is_empty() {
+                        sender.send(ReceiverAction::WakeUp).unwrap();
+                    }
+
+                    return Ok(());
                 }
                 Err(bbqueue::Error::InsufficientSize) => continue,
                 Err(e) => unreachable!("{e:?}"),
@@ -668,6 +665,11 @@ fn reserve_grant<'b>(
         if sender.is_disconnected() {
             return Err(Error::InternalError(InternalError::AbortedIndexation));
         }
+
+        // We prefer to yield and allow the writing thread
+        // to do its job, especially beneficial when there
+        // is only one CPU core available.
+        std::thread::yield_now();
     }
 }
 

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

@@ -466,12 +466,13 @@ pub fn transpose_and_freeze_caches<'a, 'extractor>(
     Ok(bucket_caches)
 }
 
-/// Merges the caches that must be all associated to the same bucket.
+/// Merges the caches that must be all associated to the same bucket
+/// but make sure to sort the different buckets before performing the merges.
 ///
 /// # 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<()>
+pub fn merge_caches_sorted<F>(frozen: Vec<FrozenCache>, mut f: F) -> Result<()>
 where
     F: for<'a> FnMut(&'a [u8], DelAddRoaringBitmap) -> Result<()>,
 {
@@ -543,12 +544,12 @@ where
 
     // Then manage the content on the HashMap entries that weren't taken (mem::take).
     while let Some(mut map) = maps.pop() {
-        for (key, bbbul) in map.iter_mut() {
         // Make sure we don't try to work with entries already managed by the spilled
-            if bbbul.is_empty() {
+        let mut ordered_entries: Vec<_> =
-                continue;
+            map.iter_mut().filter(|(_, bbbul)| !bbbul.is_empty()).collect();
-            }
+        ordered_entries.sort_unstable_by_key(|(key, _)| *key);
 
+        for (key, bbbul) in ordered_entries {
             let mut output = DelAddRoaringBitmap::empty();
             output.union_and_clear_bbbul(bbbul);
 

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

@@ -1,6 +1,6 @@
 use std::cell::RefCell;
 use std::fs::File;
-use std::io::{self, BufReader, BufWriter, ErrorKind, Read, Write as _};
+use std::io::{self, BufReader, BufWriter, ErrorKind, Read, Seek as _, Write as _};
 use std::{iter, mem, result};
 
 use bumpalo::Bump;
@@ -97,30 +97,34 @@ pub struct FrozenGeoExtractorData<'extractor> {
 impl<'extractor> FrozenGeoExtractorData<'extractor> {
     pub fn iter_and_clear_removed(
         &mut self,
-    ) -> impl IntoIterator<Item = io::Result<ExtractedGeoPoint>> + '_ {
+    ) -> io::Result<impl IntoIterator<Item = io::Result<ExtractedGeoPoint>> + '_> {
-        mem::take(&mut self.removed)
+        Ok(mem::take(&mut self.removed)
             .iter()
             .copied()
             .map(Ok)
-            .chain(iterator_over_spilled_geopoints(&mut self.spilled_removed))
+            .chain(iterator_over_spilled_geopoints(&mut self.spilled_removed)?))
     }
 
     pub fn iter_and_clear_inserted(
         &mut self,
-    ) -> impl IntoIterator<Item = io::Result<ExtractedGeoPoint>> + '_ {
+    ) -> io::Result<impl IntoIterator<Item = io::Result<ExtractedGeoPoint>> + '_> {
-        mem::take(&mut self.inserted)
+        Ok(mem::take(&mut self.inserted)
             .iter()
             .copied()
             .map(Ok)
-            .chain(iterator_over_spilled_geopoints(&mut self.spilled_inserted))
+            .chain(iterator_over_spilled_geopoints(&mut self.spilled_inserted)?))
     }
 }
 
 fn iterator_over_spilled_geopoints(
     spilled: &mut Option<BufReader<File>>,
-) -> impl IntoIterator<Item = io::Result<ExtractedGeoPoint>> + '_ {
+) -> io::Result<impl IntoIterator<Item = io::Result<ExtractedGeoPoint>> + '_> {
     let mut spilled = spilled.take();
-    iter::from_fn(move || match &mut spilled {
+    if let Some(spilled) = &mut spilled {
+        spilled.rewind()?;
+    }
+
+    Ok(iter::from_fn(move || match &mut spilled {
         Some(file) => {
             let geopoint_bytes = &mut [0u8; mem::size_of::<ExtractedGeoPoint>()];
             match file.read_exact(geopoint_bytes) {
@@ -130,7 +134,7 @@ fn iterator_over_spilled_geopoints(
             }
         }
         None => None,
-    })
+    }))
 }
 
 impl<'extractor> Extractor<'extractor> for GeoExtractor {
@@ -157,7 +161,9 @@ impl<'extractor> Extractor<'extractor> for GeoExtractor {
         let mut data_ref = context.data.borrow_mut_or_yield();
 
         for change in changes {
-            if max_memory.map_or(false, |mm| context.extractor_alloc.allocated_bytes() >= mm) {
+            if data_ref.spilled_removed.is_none()
+                && max_memory.map_or(false, |mm| context.extractor_alloc.allocated_bytes() >= mm)
+            {
                 // We must spill as we allocated too much memory
                 data_ref.spilled_removed = tempfile::tempfile().map(BufWriter::new).map(Some)?;
                 data_ref.spilled_inserted = tempfile::tempfile().map(BufWriter::new).map(Some)?;

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

@@ -6,7 +6,9 @@ mod searchable;
 mod vectors;
 
 use bumpalo::Bump;
-pub use cache::{merge_caches, transpose_and_freeze_caches, BalancedCaches, DelAddRoaringBitmap};
+pub use cache::{
+    merge_caches_sorted, transpose_and_freeze_caches, BalancedCaches, DelAddRoaringBitmap,
+};
 pub use documents::*;
 pub use faceted::*;
 pub use geo::*;

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

@@ -86,9 +86,11 @@ where
         (grenad_parameters, 2 * minimum_capacity), // 100 MiB by thread by default
         |max_memory| {
             // 2% of the indexing memory
-            let total_bbbuffer_capacity = (max_memory / 100 / 2).min(minimum_capacity);
+            let total_bbbuffer_capacity = (max_memory / 100 / 2).max(minimum_capacity);
             let new_grenad_parameters = GrenadParameters {
-                max_memory: Some(max_memory - total_bbbuffer_capacity),
+                max_memory: Some(
+                    max_memory.saturating_sub(total_bbbuffer_capacity).max(100 * 1024 * 1024),
+                ),
                 ..grenad_parameters
             };
             (new_grenad_parameters, total_bbbuffer_capacity)

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

@@ -9,8 +9,8 @@ use roaring::RoaringBitmap;
 
 use super::channel::*;
 use super::extract::{
-    merge_caches, transpose_and_freeze_caches, BalancedCaches, DelAddRoaringBitmap, FacetKind,
+    merge_caches_sorted, transpose_and_freeze_caches, BalancedCaches, DelAddRoaringBitmap,
-    GeoExtractorData,
+    FacetKind, GeoExtractorData,
 };
 use crate::{CboRoaringBitmapCodec, FieldId, GeoPoint, Index, InternalError, Result};
 
@@ -34,7 +34,7 @@ where
         }
 
         let mut frozen = data.into_inner().freeze()?;
-        for result in frozen.iter_and_clear_removed() {
+        for result in frozen.iter_and_clear_removed()? {
             let extracted_geo_point = result?;
             let removed = rtree.remove(&GeoPoint::from(extracted_geo_point));
             debug_assert!(removed.is_some());
@@ -42,7 +42,7 @@ where
             debug_assert!(removed);
         }
 
-        for result in frozen.iter_and_clear_inserted() {
+        for result in frozen.iter_and_clear_inserted()? {
             let extracted_geo_point = result?;
             rtree.insert(GeoPoint::from(extracted_geo_point));
             let inserted = faceted.insert(extracted_geo_point.docid);
@@ -78,7 +78,7 @@ where
         if must_stop_processing() {
             return Err(InternalError::AbortedIndexation.into());
         }
-        merge_caches(frozen, |key, DelAddRoaringBitmap { del, add }| {
+        merge_caches_sorted(frozen, |key, DelAddRoaringBitmap { del, add }| {
             let current = database.get(&rtxn, key)?;
             match merge_cbo_bitmaps(current, del, add)? {
                 Operation::Write(bitmap) => {
@@ -107,7 +107,7 @@ pub fn merge_and_send_facet_docids<'extractor>(
         .map(|frozen| {
             let mut facet_field_ids_delta = FacetFieldIdsDelta::default();
             let rtxn = index.read_txn()?;
-            merge_caches(frozen, |key, DelAddRoaringBitmap { del, add }| {
+            merge_caches_sorted(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) => {

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

@@ -1,4 +1,4 @@
-use std::collections::HashSet;
+use std::collections::BTreeSet;
 use std::io::BufWriter;
 
 use fst::{Set, SetBuilder, Streamer};
@@ -75,8 +75,8 @@ pub struct PrefixData {
 
 #[derive(Debug)]
 pub struct PrefixDelta {
-    pub modified: HashSet<Prefix>,
+    pub modified: BTreeSet<Prefix>,
-    pub deleted: HashSet<Prefix>,
+    pub deleted: BTreeSet<Prefix>,
 }
 
 struct PrefixFstBuilder {
@@ -86,7 +86,7 @@ struct PrefixFstBuilder {
     prefix_fst_builders: Vec<SetBuilder<Vec<u8>>>,
     current_prefix: Vec<Prefix>,
     current_prefix_count: Vec<usize>,
-    modified_prefixes: HashSet<Prefix>,
+    modified_prefixes: BTreeSet<Prefix>,
     current_prefix_is_modified: Vec<bool>,
 }
 
@@ -110,7 +110,7 @@ impl PrefixFstBuilder {
             prefix_fst_builders,
             current_prefix: vec![Prefix::new(); max_prefix_length],
             current_prefix_count: vec![0; max_prefix_length],
-            modified_prefixes: HashSet::new(),
+            modified_prefixes: BTreeSet::new(),
             current_prefix_is_modified: vec![false; max_prefix_length],
         })
     }
@@ -180,7 +180,7 @@ impl PrefixFstBuilder {
         let prefix_fst_mmap = unsafe { Mmap::map(&prefix_fst_file)? };
         let new_prefix_fst = Set::new(&prefix_fst_mmap)?;
         let old_prefix_fst = index.words_prefixes_fst(rtxn)?;
-        let mut deleted_prefixes = HashSet::new();
+        let mut deleted_prefixes = BTreeSet::new();
         {
             let mut deleted_prefixes_stream = old_prefix_fst.op().add(&new_prefix_fst).difference();
             while let Some(prefix) = deleted_prefixes_stream.next() {

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

@@ -1,5 +1,5 @@
 use std::cell::RefCell;
-use std::collections::HashSet;
+use std::collections::BTreeSet;
 use std::io::{BufReader, BufWriter, Read, Seek, Write};
 
 use hashbrown::HashMap;
@@ -37,8 +37,8 @@ impl WordPrefixDocids {
     fn execute(
         self,
         wtxn: &mut heed::RwTxn,
-        prefix_to_compute: &HashSet<Prefix>,
+        prefix_to_compute: &BTreeSet<Prefix>,
-        prefix_to_delete: &HashSet<Prefix>,
+        prefix_to_delete: &BTreeSet<Prefix>,
     ) -> Result<()> {
         delete_prefixes(wtxn, &self.prefix_database, prefix_to_delete)?;
         self.recompute_modified_prefixes(wtxn, prefix_to_compute)
@@ -48,7 +48,7 @@ impl WordPrefixDocids {
     fn recompute_modified_prefixes(
         &self,
         wtxn: &mut RwTxn,
-        prefixes: &HashSet<Prefix>,
+        prefixes: &BTreeSet<Prefix>,
     ) -> Result<()> {
         // We fetch the docids associated to the newly added word prefix fst only.
         // And collect the CboRoaringBitmaps pointers in an HashMap.
@@ -127,7 +127,7 @@ impl<'a, 'rtxn> FrozenPrefixBitmaps<'a, 'rtxn> {
     pub fn from_prefixes(
         database: Database<Bytes, CboRoaringBitmapCodec>,
         rtxn: &'rtxn RoTxn,
-        prefixes: &'a HashSet<Prefix>,
+        prefixes: &'a BTreeSet<Prefix>,
     ) -> heed::Result<Self> {
         let database = database.remap_data_type::<Bytes>();
 
@@ -173,8 +173,8 @@ impl WordPrefixIntegerDocids {
     fn execute(
         self,
         wtxn: &mut heed::RwTxn,
-        prefix_to_compute: &HashSet<Prefix>,
+        prefix_to_compute: &BTreeSet<Prefix>,
-        prefix_to_delete: &HashSet<Prefix>,
+        prefix_to_delete: &BTreeSet<Prefix>,
     ) -> Result<()> {
         delete_prefixes(wtxn, &self.prefix_database, prefix_to_delete)?;
         self.recompute_modified_prefixes(wtxn, prefix_to_compute)
@@ -184,7 +184,7 @@ impl WordPrefixIntegerDocids {
     fn recompute_modified_prefixes(
         &self,
         wtxn: &mut RwTxn,
-        prefixes: &HashSet<Prefix>,
+        prefixes: &BTreeSet<Prefix>,
     ) -> Result<()> {
         // We fetch the docids associated to the newly added word prefix fst only.
         // And collect the CboRoaringBitmaps pointers in an HashMap.
@@ -262,7 +262,7 @@ impl<'a, 'rtxn> FrozenPrefixIntegerBitmaps<'a, 'rtxn> {
     pub fn from_prefixes(
         database: Database<Bytes, CboRoaringBitmapCodec>,
         rtxn: &'rtxn RoTxn,
-        prefixes: &'a HashSet<Prefix>,
+        prefixes: &'a BTreeSet<Prefix>,
     ) -> heed::Result<Self> {
         let database = database.remap_data_type::<Bytes>();
 
@@ -291,7 +291,7 @@ unsafe impl<'a, 'rtxn> Sync for FrozenPrefixIntegerBitmaps<'a, 'rtxn> {}
 fn delete_prefixes(
     wtxn: &mut RwTxn,
     prefix_database: &Database<Bytes, CboRoaringBitmapCodec>,
-    prefixes: &HashSet<Prefix>,
+    prefixes: &BTreeSet<Prefix>,
 ) -> Result<()> {
     // We remove all the entries that are no more required in this word prefix docids database.
     for prefix in prefixes {
@@ -309,8 +309,8 @@ fn delete_prefixes(
 pub fn compute_word_prefix_docids(
     wtxn: &mut RwTxn,
     index: &Index,
-    prefix_to_compute: &HashSet<Prefix>,
+    prefix_to_compute: &BTreeSet<Prefix>,
-    prefix_to_delete: &HashSet<Prefix>,
+    prefix_to_delete: &BTreeSet<Prefix>,
     grenad_parameters: GrenadParameters,
 ) -> Result<()> {
     WordPrefixDocids::new(
@@ -325,8 +325,8 @@ pub fn compute_word_prefix_docids(
 pub fn compute_exact_word_prefix_docids(
     wtxn: &mut RwTxn,
     index: &Index,
-    prefix_to_compute: &HashSet<Prefix>,
+    prefix_to_compute: &BTreeSet<Prefix>,
-    prefix_to_delete: &HashSet<Prefix>,
+    prefix_to_delete: &BTreeSet<Prefix>,
     grenad_parameters: GrenadParameters,
 ) -> Result<()> {
     WordPrefixDocids::new(
@@ -341,8 +341,8 @@ pub fn compute_exact_word_prefix_docids(
 pub fn compute_word_prefix_fid_docids(
     wtxn: &mut RwTxn,
     index: &Index,
-    prefix_to_compute: &HashSet<Prefix>,
+    prefix_to_compute: &BTreeSet<Prefix>,
-    prefix_to_delete: &HashSet<Prefix>,
+    prefix_to_delete: &BTreeSet<Prefix>,
     grenad_parameters: GrenadParameters,
 ) -> Result<()> {
     WordPrefixIntegerDocids::new(
@@ -357,8 +357,8 @@ pub fn compute_word_prefix_fid_docids(
 pub fn compute_word_prefix_position_docids(
     wtxn: &mut RwTxn,
     index: &Index,
-    prefix_to_compute: &HashSet<Prefix>,
+    prefix_to_compute: &BTreeSet<Prefix>,
-    prefix_to_delete: &HashSet<Prefix>,
+    prefix_to_delete: &BTreeSet<Prefix>,
     grenad_parameters: GrenadParameters,
 ) -> Result<()> {
     WordPrefixIntegerDocids::new(

crates/xtask/src/bench/mod.rs

@@ -82,6 +82,10 @@ pub struct BenchDeriveArgs {
     /// Reason for the benchmark invocation
     #[arg(short, long)]
     reason: Option<String>,
+
+    /// The maximum time in seconds we allow for fetching the task queue before timing out.
+    #[arg(long, default_value_t = 60)]
+    tasks_queue_timeout_secs: u64,
 }
 
 pub fn run(args: BenchDeriveArgs) -> anyhow::Result<()> {
@@ -127,7 +131,7 @@ pub fn run(args: BenchDeriveArgs) -> anyhow::Result<()> {
     let meili_client = Client::new(
         Some("http://127.0.0.1:7700".into()),
         args.master_key.as_deref(),
-        Some(std::time::Duration::from_secs(60)),
+        Some(std::time::Duration::from_secs(args.tasks_queue_timeout_secs)),
     )?;
 
     // enter runtime

crates/xtask/src/main.rs

@@ -16,6 +16,7 @@ struct ListFeaturesDeriveArgs {
 #[command(author, version, about, long_about)]
 #[command(name = "cargo xtask")]
 #[command(bin_name = "cargo xtask")]
+#[allow(clippy::large_enum_variant)] // please, that's enough...
 enum Command {
     ListFeatures(ListFeaturesDeriveArgs),
     Bench(BenchDeriveArgs),

workloads/hackernews-add-new-documents.json

@@ -0,0 +1,105 @@
+{
+  "name": "hackernews.add_new_documents",
+  "run_count": 3,
+  "extra_cli_args": [],
+  "assets": {
+      "hackernews-01.ndjson": {
+        "local_location": null,
+        "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/01.ndjson",
+        "sha256": "cd3627b86c064d865b6754848ed0e73ef1d8142752a25e5f0765c3a1296dd3ae"
+      },
+      "hackernews-02.ndjson": {
+        "local_location": null,
+        "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/02.ndjson",
+        "sha256": "5d533b83bcf992201dace88b4d0c0be8b4df5225c6c4b763582d986844bcc23b"
+      },
+      "hackernews-03.ndjson": {
+        "local_location": null,
+        "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/03.ndjson",
+        "sha256": "f5f351a0d04a8a83643ace12cafa2b7ec8ca8cb7d46fd268e5126492a6c66f2a"
+      },
+      "hackernews-04.ndjson": {
+        "local_location": null,
+        "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/04.ndjson",
+        "sha256": "ac1915ee7ce53a6718548c255a6cc59969784b2570745dc5b739f714beda291a"
+      },
+      "hackernews-05.ndjson": {
+        "local_location": null,
+        "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/05.ndjson",
+        "sha256": "be31d5632602f798e62d1c10c83bdfda2b4deaa068477eacde05fdd247572b82"
+      }
+  },
+  "precommands": [
+    {
+      "route": "indexes/movies/settings",
+      "method": "PATCH",
+      "body": {
+        "inline": {
+          "displayedAttributes": [
+            "title",
+            "by",
+            "score",
+            "time",
+            "text"
+          ],
+          "searchableAttributes": [
+            "title",
+            "text"
+          ],
+          "filterableAttributes": [
+            "by",
+            "kids",
+            "parent"
+          ],
+          "sortableAttributes": [
+            "score",
+            "time"
+          ]
+        }
+      },
+      "synchronous": "WaitForTask"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-01.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-02.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-03.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-04.ndjson"
+      },
+      "synchronous": "WaitForTask"
+    }
+  ],
+  "commands": [
+      {
+        "route": "indexes/movies/documents",
+        "method": "POST",
+        "body": {
+          "asset": "hackernews-05.ndjson"
+        },
+        "synchronous": "WaitForTask"
+      }
+  ]
+}

workloads/hackernews-modify-facet-numbers.json

@@ -0,0 +1,111 @@
+{
+  "name": "hackernews.modify_facet_numbers",
+  "run_count": 3,
+  "extra_cli_args": [],
+  "assets": {
+    "hackernews-01.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/01.ndjson",
+      "sha256": "cd3627b86c064d865b6754848ed0e73ef1d8142752a25e5f0765c3a1296dd3ae"
+    },
+    "hackernews-02.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/02.ndjson",
+      "sha256": "5d533b83bcf992201dace88b4d0c0be8b4df5225c6c4b763582d986844bcc23b"
+    },
+    "hackernews-03.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/03.ndjson",
+      "sha256": "f5f351a0d04a8a83643ace12cafa2b7ec8ca8cb7d46fd268e5126492a6c66f2a"
+    },
+    "hackernews-04.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/04.ndjson",
+      "sha256": "ac1915ee7ce53a6718548c255a6cc59969784b2570745dc5b739f714beda291a"
+    },
+    "hackernews-05.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/05.ndjson",
+      "sha256": "be31d5632602f798e62d1c10c83bdfda2b4deaa068477eacde05fdd247572b82"
+    },
+    "hackernews-02-modified-filters.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/02-modified-filters.ndjson",
+      "sha256": "7272cbfd41110d32d7fe168424a0000f07589bfe40f664652b34f4f20aaf3802"
+    }
+  },
+  "precommands": [
+    {
+      "route": "indexes/movies/settings",
+      "method": "PATCH",
+      "body": {
+        "inline": {
+          "displayedAttributes": [
+            "title",
+            "by",
+            "score",
+            "time",
+            "text"
+          ],
+          "searchableAttributes": [
+            "title",
+            "text"
+          ],
+          "filterableAttributes": [
+            "by",
+            "kids",
+            "parent"
+          ],
+          "sortableAttributes": [
+            "score",
+            "time"
+          ]
+        }
+      },
+      "synchronous": "WaitForTask"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-01.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-02.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-03.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-04.ndjson"
+      },
+      "synchronous": "WaitForTask"
+    }
+  ],
+  "commands": [
+      {
+        "route": "indexes/movies/documents",
+        "method": "POST",
+        "body": {
+          "asset": "hackernews-02-modified-filters.ndjson"
+        },
+        "synchronous": "WaitForTask"
+      }
+  ]
+}
+  

workloads/hackernews-modify-facet-strings.json

@@ -0,0 +1,111 @@
+{
+  "name": "hackernews.modify_facet_strings",
+  "run_count": 3,
+  "extra_cli_args": [],
+  "assets": {
+    "hackernews-01.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/01.ndjson",
+      "sha256": "cd3627b86c064d865b6754848ed0e73ef1d8142752a25e5f0765c3a1296dd3ae"
+    },
+    "hackernews-02.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/02.ndjson",
+      "sha256": "5d533b83bcf992201dace88b4d0c0be8b4df5225c6c4b763582d986844bcc23b"
+    },
+    "hackernews-03.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/03.ndjson",
+      "sha256": "f5f351a0d04a8a83643ace12cafa2b7ec8ca8cb7d46fd268e5126492a6c66f2a"
+    },
+    "hackernews-04.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/04.ndjson",
+      "sha256": "ac1915ee7ce53a6718548c255a6cc59969784b2570745dc5b739f714beda291a"
+    },
+    "hackernews-05.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/05.ndjson",
+      "sha256": "be31d5632602f798e62d1c10c83bdfda2b4deaa068477eacde05fdd247572b82"
+    },
+    "hackernews-01-modified-filters.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/01-modified-filters.ndjson",
+      "sha256": "b80c245ce1b1df80b9b38800f677f3bd11947ebc62716fb108269d50e796c35c"
+    }
+  },
+  "precommands": [
+    {
+      "route": "indexes/movies/settings",
+      "method": "PATCH",
+      "body": {
+        "inline": {
+          "displayedAttributes": [
+            "title",
+            "by",
+            "score",
+            "time",
+            "text"
+          ],
+          "searchableAttributes": [
+            "title",
+            "text"
+          ],
+          "filterableAttributes": [
+            "by",
+            "kids",
+            "parent"
+          ],
+          "sortableAttributes": [
+            "score",
+            "time"
+          ]
+        }
+      },
+      "synchronous": "WaitForTask"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-01.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-02.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-03.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-04.ndjson"
+      },
+      "synchronous": "WaitForTask"
+    }
+  ],
+  "commands": [
+      {
+        "route": "indexes/movies/documents",
+        "method": "POST",
+        "body": {
+          "asset": "hackernews-01-modified-filters.ndjson"
+        },
+        "synchronous": "WaitForTask"
+      }
+  ]
+}
+ 

workloads/hackernews-modify-searchables.json

@@ -0,0 +1,123 @@
+{
+  "name": "hackernews.modify_searchables",
+  "run_count": 3,
+  "extra_cli_args": [],
+  "assets": {
+    "hackernews-01.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/01.ndjson",
+      "sha256": "cd3627b86c064d865b6754848ed0e73ef1d8142752a25e5f0765c3a1296dd3ae"
+    },
+    "hackernews-02.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/02.ndjson",
+      "sha256": "5d533b83bcf992201dace88b4d0c0be8b4df5225c6c4b763582d986844bcc23b"
+    },
+    "hackernews-03.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/03.ndjson",
+      "sha256": "f5f351a0d04a8a83643ace12cafa2b7ec8ca8cb7d46fd268e5126492a6c66f2a"
+    },
+    "hackernews-04.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/04.ndjson",
+      "sha256": "ac1915ee7ce53a6718548c255a6cc59969784b2570745dc5b739f714beda291a"
+    },
+    "hackernews-05.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/05.ndjson",
+      "sha256": "be31d5632602f798e62d1c10c83bdfda2b4deaa068477eacde05fdd247572b82"
+    },
+    "hackernews-01-modified-searchables.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/01-modified-searchables.ndjson",
+      "sha256": "e5c08710c6af70031ac7212e0ba242c72ef29c8d4e1fce66c789544641452a7c"
+    },
+    "hackernews-02-modified-searchables.ndjson": {
+      "local_location": null,
+      "remote_location": "https://milli-benchmarks.fra1.digitaloceanspaces.com/bench/datasets/hackernews/modification/02-modified-searchables.ndjson",
+      "sha256": "098b029851117087b1e26ccb7ac408eda9bba54c3008213a2880d6fab607346e"
+    }
+  },
+  "precommands": [
+    {
+      "route": "indexes/movies/settings",
+      "method": "PATCH",
+      "body": {
+        "inline": {
+          "displayedAttributes": [
+            "title",
+            "by",
+            "score",
+            "time",
+            "text"
+          ],
+          "searchableAttributes": [
+            "title",
+            "text"
+          ],
+          "filterableAttributes": [
+            "by",
+            "kids",
+            "parent"
+          ],
+          "sortableAttributes": [
+            "score",
+            "time"
+          ]
+        }
+      },
+      "synchronous": "WaitForTask"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-01.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-02.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-03.ndjson"
+      },
+      "synchronous": "WaitForResponse"
+    },
+    {
+      "route": "indexes/movies/documents",
+      "method": "POST",
+      "body": {
+        "asset": "hackernews-04.ndjson"
+      },
+      "synchronous": "WaitForTask"
+    }
+  ],
+  "commands": [
+      {
+        "route": "indexes/movies/documents",
+        "method": "POST",
+        "body": {
+          "asset": "hackernews-01-modified-searchables.ndjson"
+        },
+        "synchronous": "WaitForTask"
+      },
+      {
+        "route": "indexes/movies/documents",
+        "method": "POST",
+        "body": {
+          "asset": "hackernews-02-modified-searchables.ndjson"
+        },
+        "synchronous": "WaitForTask"
+      }
+  ]
+}