bump cellulite to address backcompat issue from #5307

Synonym performance fix

Cargo.lock

@@ -453,9 +453,9 @@ checksum = "7c02d123df017efcdfbd739ef81735b36c5ba83ec3c59c80a9d7ecc718f92e50"
 
 [[package]]
 name = "arroy"
-version = "0.6.3"
+version = "0.6.4-nested-rtxns"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8578a72223dfa13dfd9fc144d15260d134361789ebdea9b16e85a511edc73c7d"
+checksum = "fb8b6b34d9d83e3b837cb7f72a439dbd2293b102393c084af5e5b097212e1532"
 dependencies = [
  "bytemuck",
  "byteorder",
@@ -1075,9 +1075,9 @@ dependencies = [
 
 [[package]]
 name = "cellulite"
-version = "0.3.0"
+version = "0.3.1-nested-rtxns-2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "71a41aa2cd021bc3f23d97cc1e645848ca8c279fc757d1570ba7fe7ddc021290"
+checksum = "f22d721963ead1a144f10cb8b53dc9469e760723b069123c7c7bc675c7354270"
 dependencies = [
  "crossbeam",
  "geo",
@@ -2758,9 +2758,9 @@ dependencies = [
 
 [[package]]
 name = "hannoy"
-version = "0.0.8"
+version = "0.0.9-nested-rtxns"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0dba13a271c49a119a97862ebf0a74131d879832868400d9fcd937b790058fdd"
+checksum = "cc5a945b92b063e677d658cfcc7cb6dec2502fe44631f017084938f14d6ce30e"
 dependencies = [
  "bytemuck",
  "byteorder",
@@ -2838,9 +2838,9 @@ checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea"
 
 [[package]]
 name = "heed"
-version = "0.22.0"
+version = "0.22.1-nested-rtxns"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6a56c94661ddfb51aa9cdfbf102cfcc340aa69267f95ebccc4af08d7c530d393"
+checksum = "0ff115ba5712b1f1fc7617b195f5c2f139e29c397ff79da040cd19db75ccc240"
 dependencies = [
  "bitflags 2.9.4",
  "byteorder",
@@ -2850,7 +2850,6 @@ dependencies = [
  "lmdb-master-sys",
  "once_cell",
  "page_size",
- "serde",
  "synchronoise",
  "url",
 ]
@@ -3889,9 +3888,9 @@ checksum = "241eaef5fd12c88705a01fc1066c48c4b36e0dd4377dcdc7ec3942cea7a69956"
 
 [[package]]
 name = "lmdb-master-sys"
-version = "0.2.5"
+version = "0.2.6-nested-rtxns"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "864808e0b19fb6dd3b70ba94ee671b82fce17554cf80aeb0a155c65bb08027df"
+checksum = "f4ff85130e3c994b36877045fbbb138d521dea7197bfc19dc3d5d95101a8e20a"
 dependencies = [
  "cc",
  "doxygen-rs",

crates/dump/src/lib.rs

@@ -158,6 +158,9 @@ pub enum KindDump {
     UpgradeDatabase {
         from: (u32, u32, u32),
     },
+    IndexCompaction {
+        index_uid: String,
+    },
 }
 
 impl From<Task> for TaskDump {
@@ -240,6 +243,9 @@ impl From<KindWithContent> for KindDump {
             KindWithContent::UpgradeDatabase { from: version } => {
                 KindDump::UpgradeDatabase { from: version }
             }
+            KindWithContent::IndexCompaction { index_uid } => {
+                KindDump::IndexCompaction { index_uid }
+            }
         }
     }
 }

crates/index-scheduler/src/dump.rs

@@ -234,6 +234,9 @@ impl<'a> Dump<'a> {
                     }
                 }
                 KindDump::UpgradeDatabase { from } => KindWithContent::UpgradeDatabase { from },
+                KindDump::IndexCompaction { index_uid } => {
+                    KindWithContent::IndexCompaction { index_uid }
+                }
             },
         };
 

crates/index-scheduler/src/index_mapper/mod.rs

@@ -341,6 +341,26 @@ impl IndexMapper {
         Ok(())
     }
 
+    /// Closes the specified index.
+    ///
+    /// This operation involves closing the underlying environment and so can take a long time to complete.
+    ///
+    /// # Panics
+    ///
+    /// - If the Index corresponding to the passed name is concurrently being deleted/resized or cannot be found in the
+    ///   in memory hash map.
+    pub fn close_index(&self, rtxn: &RoTxn, name: &str) -> Result<()> {
+        let uuid = self
+            .index_mapping
+            .get(rtxn, name)?
+            .ok_or_else(|| Error::IndexNotFound(name.to_string()))?;
+
+        // We remove the index from the in-memory index map.
+        self.index_map.write().unwrap().close_for_resize(&uuid, self.enable_mdb_writemap, 0);
+
+        Ok(())
+    }
+
     /// Return an index, may open it if it wasn't already opened.
     pub fn index(&self, rtxn: &RoTxn, name: &str) -> Result<Index> {
         if let Some((current_name, current_index)) =

crates/index-scheduler/src/insta_snapshot.rs

@@ -317,6 +317,9 @@ fn snapshot_details(d: &Details) -> String {
         Details::UpgradeDatabase { from, to } => {
             format!("{{ from: {from:?}, to: {to:?} }}")
         }
+        Details::IndexCompaction { index_uid, pre_compaction_size, post_compaction_size } => {
+            format!("{{ index_uid: {index_uid:?}, pre_compaction_size: {pre_compaction_size:?}, post_compaction_size: {post_compaction_size:?} }}")
+        }
     }
 }
 

crates/index-scheduler/src/processing.rs

@@ -138,6 +138,17 @@ make_enum_progress! {
     }
 }
 
+make_enum_progress! {
+    pub enum IndexCompaction {
+        RetrieveTheIndex,
+        CreateTemporaryFile,
+        CopyAndCompactTheIndex,
+        PersistTheCompactedIndex,
+        CloseTheIndex,
+        ReopenTheIndex,
+    }
+}
+
 make_enum_progress! {
     pub enum InnerSwappingTwoIndexes {
         RetrieveTheTasks,

crates/index-scheduler/src/scheduler/autobatcher.rs

@@ -25,6 +25,7 @@ enum AutobatchKind {
     IndexDeletion,
     IndexUpdate,
     IndexSwap,
+    IndexCompaction,
 }
 
 impl AutobatchKind {
@@ -68,6 +69,7 @@ impl From<KindWithContent> for AutobatchKind {
             KindWithContent::IndexCreation { .. } => AutobatchKind::IndexCreation,
             KindWithContent::IndexUpdate { .. } => AutobatchKind::IndexUpdate,
             KindWithContent::IndexSwap { .. } => AutobatchKind::IndexSwap,
+            KindWithContent::IndexCompaction { .. } => AutobatchKind::IndexCompaction,
             KindWithContent::TaskCancelation { .. }
             | KindWithContent::TaskDeletion { .. }
             | KindWithContent::DumpCreation { .. }
@@ -118,6 +120,9 @@ pub enum BatchKind {
     IndexSwap {
         id: TaskId,
     },
+    IndexCompaction {
+        id: TaskId,
+    },
 }
 
 impl BatchKind {
@@ -183,6 +188,13 @@ impl BatchKind {
                 )),
                 false,
             ),
+            K::IndexCompaction => (
+                Break((
+                    BatchKind::IndexCompaction { id: task_id },
+                    BatchStopReason::TaskCannotBeBatched { kind, id: task_id },
+                )),
+                false,
+            ),
             K::DocumentClear => (Continue(BatchKind::DocumentClear { ids: vec![task_id] }), false),
             K::DocumentImport { allow_index_creation, primary_key: pk }
                 if primary_key.is_none() || pk.is_none() || primary_key == pk.as_deref() =>
@@ -287,8 +299,10 @@ impl BatchKind {
         };
 
         match (self, autobatch_kind) {
-            // We don't batch any of these operations  
-            (this, K::IndexCreation | K::IndexUpdate | K::IndexSwap | K::DocumentEdition) => Break((this, BatchStopReason::TaskCannotBeBatched { kind, id })),
+            // We don't batch any of these operations
+            (this, K::IndexCreation | K::IndexUpdate | K::IndexSwap | K::DocumentEdition | K::IndexCompaction) => {
+                Break((this, BatchStopReason::TaskCannotBeBatched { kind, id }))
+            },
             // We must not batch tasks that don't have the same index creation rights if the index doesn't already exists.
             (this, kind) if !index_already_exists && this.allow_index_creation() == Some(false) && kind.allow_index_creation() == Some(true) => {
                 Break((this, BatchStopReason::IndexCreationMismatch { id }))
@@ -483,6 +497,7 @@ impl BatchKind {
                 | BatchKind::IndexDeletion { .. }
                 | BatchKind::IndexUpdate { .. }
                 | BatchKind::IndexSwap { .. }
+                | BatchKind::IndexCompaction { .. }
                 | BatchKind::DocumentEdition { .. },
                 _,
             ) => {

crates/index-scheduler/src/scheduler/create_batch.rs

@@ -55,6 +55,10 @@ pub(crate) enum Batch {
     UpgradeDatabase {
         tasks: Vec<Task>,
     },
+    IndexCompaction {
+        index_uid: String,
+        task: Task,
+    },
 }
 
 #[derive(Debug)]
@@ -110,7 +114,8 @@ impl Batch {
             | Batch::Dump(task)
             | Batch::IndexCreation { task, .. }
             | Batch::Export { task }
-            | Batch::IndexUpdate { task, .. } => {
+            | Batch::IndexUpdate { task, .. }
+            | Batch::IndexCompaction { task, .. } => {
                 RoaringBitmap::from_sorted_iter(std::iter::once(task.uid)).unwrap()
             }
             Batch::SnapshotCreation(tasks)
@@ -155,7 +160,8 @@ impl Batch {
             IndexOperation { op, .. } => Some(op.index_uid()),
             IndexCreation { index_uid, .. }
             | IndexUpdate { index_uid, .. }
-            | IndexDeletion { index_uid, .. } => Some(index_uid),
+            | IndexDeletion { index_uid, .. }
+            | IndexCompaction { index_uid, .. } => Some(index_uid),
         }
     }
 }
@@ -175,6 +181,7 @@ impl fmt::Display for Batch {
             Batch::IndexUpdate { .. } => f.write_str("IndexUpdate")?,
             Batch::IndexDeletion { .. } => f.write_str("IndexDeletion")?,
             Batch::IndexSwap { .. } => f.write_str("IndexSwap")?,
+            Batch::IndexCompaction { .. } => f.write_str("IndexCompaction")?,
             Batch::Export { .. } => f.write_str("Export")?,
             Batch::UpgradeDatabase { .. } => f.write_str("UpgradeDatabase")?,
         };
@@ -430,6 +437,12 @@ impl IndexScheduler {
                 current_batch.processing(Some(&mut task));
                 Ok(Some(Batch::IndexSwap { task }))
             }
+            BatchKind::IndexCompaction { id } => {
+                let mut task =
+                    self.queue.tasks.get_task(rtxn, id)?.ok_or(Error::CorruptedTaskQueue)?;
+                current_batch.processing(Some(&mut task));
+                Ok(Some(Batch::IndexCompaction { index_uid, task }))
+            }
         }
     }
 

crates/index-scheduler/src/scheduler/process_batch.rs

@@ -1,22 +1,26 @@
 use std::collections::{BTreeSet, HashMap, HashSet};
+use std::io::{Seek, SeekFrom};
 use std::panic::{catch_unwind, AssertUnwindSafe};
 use std::sync::atomic::Ordering;
 
+use byte_unit::Byte;
 use meilisearch_types::batches::{BatchEnqueuedAt, BatchId};
 use meilisearch_types::heed::{RoTxn, RwTxn};
+use meilisearch_types::milli::heed::CompactionOption;
 use meilisearch_types::milli::progress::{Progress, VariableNameStep};
 use meilisearch_types::milli::{self, ChannelCongestion};
 use meilisearch_types::tasks::{Details, IndexSwap, Kind, KindWithContent, Status, Task};
 use meilisearch_types::versioning::{VERSION_MAJOR, VERSION_MINOR, VERSION_PATCH};
 use milli::update::Settings as MilliSettings;
 use roaring::RoaringBitmap;
+use tempfile::PersistError;
 use time::OffsetDateTime;
 
 use super::create_batch::Batch;
 use crate::processing::{
     AtomicBatchStep, AtomicTaskStep, CreateIndexProgress, DeleteIndexProgress, FinalizingIndexStep,
-    InnerSwappingTwoIndexes, SwappingTheIndexes, TaskCancelationProgress, TaskDeletionProgress,
-    UpdateIndexProgress,
+    IndexCompaction, InnerSwappingTwoIndexes, SwappingTheIndexes, TaskCancelationProgress,
+    TaskDeletionProgress, UpdateIndexProgress,
 };
 use crate::utils::{
     self, remove_n_tasks_datetime_earlier_than, remove_task_datetime, swap_index_uid_in_task,
@@ -418,6 +422,47 @@ impl IndexScheduler {
                 task.status = Status::Succeeded;
                 Ok((vec![task], ProcessBatchInfo::default()))
             }
+            Batch::IndexCompaction { index_uid: _, mut task } => {
+                let KindWithContent::IndexCompaction { index_uid } = &task.kind else {
+                    unreachable!()
+                };
+
+                let rtxn = self.env.read_txn()?;
+                let ret = catch_unwind(AssertUnwindSafe(|| {
+                    self.apply_compaction(&rtxn, &progress, index_uid)
+                }));
+
+                let (pre_size, post_size) = match ret {
+                    Ok(Ok(stats)) => stats,
+                    Ok(Err(Error::AbortedTask)) => return Err(Error::AbortedTask),
+                    Ok(Err(e)) => return Err(e),
+                    Err(e) => {
+                        let msg = match e.downcast_ref::<&'static str>() {
+                            Some(s) => *s,
+                            None => match e.downcast_ref::<String>() {
+                                Some(s) => &s[..],
+                                None => "Box<dyn Any>",
+                            },
+                        };
+                        return Err(Error::Export(Box::new(Error::ProcessBatchPanicked(
+                            msg.to_string(),
+                        ))));
+                    }
+                };
+
+                task.status = Status::Succeeded;
+                if let Some(Details::IndexCompaction {
+                    index_uid: _,
+                    pre_compaction_size,
+                    post_compaction_size,
+                }) = task.details.as_mut()
+                {
+                    *pre_compaction_size = Some(Byte::from_u64(pre_size));
+                    *post_compaction_size = Some(Byte::from_u64(post_size));
+                }
+
+                Ok((vec![task], ProcessBatchInfo::default()))
+            }
             Batch::Export { mut task } => {
                 let KindWithContent::Export { url, api_key, payload_size, indexes } = &task.kind
                 else {
@@ -493,6 +538,91 @@ impl IndexScheduler {
         }
     }
 
+    fn apply_compaction(
+        &self,
+        rtxn: &RoTxn,
+        progress: &Progress,
+        index_uid: &str,
+    ) -> Result<(u64, u64)> {
+        // 1. Verify that the index exists
+        if !self.index_mapper.index_exists(rtxn, index_uid)? {
+            return Err(Error::IndexNotFound(index_uid.to_owned()));
+        }
+
+        // 2. We retrieve the index and create a temporary file in the index directory
+        progress.update_progress(IndexCompaction::RetrieveTheIndex);
+        let index = self.index_mapper.index(rtxn, index_uid)?;
+
+        // the index operation can take a long time, so save this handle to make it available to the search for the duration of the tick
+        self.index_mapper
+            .set_currently_updating_index(Some((index_uid.to_string(), index.clone())));
+
+        progress.update_progress(IndexCompaction::CreateTemporaryFile);
+        let pre_size = std::fs::metadata(index.path().join("data.mdb"))?.len();
+        let mut file = tempfile::Builder::new()
+            .suffix("data.")
+            .prefix(".mdb.cpy")
+            .tempfile_in(index.path())?;
+
+        // 3. We copy the index data to the temporary file
+        progress.update_progress(IndexCompaction::CopyAndCompactTheIndex);
+        index
+            .copy_to_file(file.as_file_mut(), CompactionOption::Enabled)
+            .map_err(|error| Error::Milli { error, index_uid: Some(index_uid.to_string()) })?;
+        // ...and reset the file position as specified in the documentation
+        file.seek(SeekFrom::Start(0))?;
+
+        // 4. We replace the index data file with the temporary file
+        progress.update_progress(IndexCompaction::PersistTheCompactedIndex);
+        match file.persist(index.path().join("data.mdb")) {
+            Ok(file) => file.sync_all()?,
+            // TODO see if we have a _resource busy_ error and probably handle this by:
+            //      1. closing the index, 2. replacing and 3. reopening it
+            Err(PersistError { error, file: _ }) => return Err(Error::IoError(error)),
+        };
+
+        // 5. Prepare to close the index
+        progress.update_progress(IndexCompaction::CloseTheIndex);
+
+        // unmark that the index is the processing one so we don't keep a handle to it, preventing its closing
+        self.index_mapper.set_currently_updating_index(None);
+
+        self.index_mapper.close_index(rtxn, index_uid)?;
+        drop(index);
+
+        progress.update_progress(IndexCompaction::ReopenTheIndex);
+        // 6. Reopen the index
+        // The index will use the compacted data file when being reopened
+        let index = self.index_mapper.index(rtxn, index_uid)?;
+
+        // if the update processed successfully, we're going to store the new
+        // stats of the index. Since the tasks have already been processed and
+        // this is a non-critical operation. If it fails, we should not fail
+        // the entire batch.
+        let res = || -> Result<_> {
+            let mut wtxn = self.env.write_txn()?;
+            let index_rtxn = index.read_txn()?;
+            let stats = crate::index_mapper::IndexStats::new(&index, &index_rtxn)
+                .map_err(|e| Error::from_milli(e, Some(index_uid.to_string())))?;
+            self.index_mapper.store_stats_of(&mut wtxn, index_uid, &stats)?;
+            wtxn.commit()?;
+            Ok(stats.database_size)
+        }();
+
+        let post_size = match res {
+            Ok(post_size) => post_size,
+            Err(e) => {
+                tracing::error!(
+                    error = &e as &dyn std::error::Error,
+                    "Could not write the stats of the index"
+                );
+                0
+            }
+        };
+
+        Ok((pre_size, post_size))
+    }
+
     /// Swap the index `lhs` with the index `rhs`.
     fn apply_index_swap(
         &self,

crates/index-scheduler/src/scheduler/test.rs

@@ -722,7 +722,7 @@ fn basic_get_stats() {
     let kind = index_creation_task("whalo", "fish");
     let _task = index_scheduler.register(kind, None, false).unwrap();
 
-    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r#"
+    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r###"
     {
       "indexes": {
         "catto": 1,
@@ -742,6 +742,7 @@ fn basic_get_stats() {
         "documentEdition": 0,
         "dumpCreation": 0,
         "export": 0,
+        "indexCompaction": 0,
         "indexCreation": 3,
         "indexDeletion": 0,
         "indexSwap": 0,
@@ -753,10 +754,10 @@ fn basic_get_stats() {
         "upgradeDatabase": 0
       }
     }
-    "#);
+    "###);
 
     handle.advance_till([Start, BatchCreated]);
-    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r#"
+    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r###"
     {
       "indexes": {
         "catto": 1,
@@ -776,6 +777,7 @@ fn basic_get_stats() {
         "documentEdition": 0,
         "dumpCreation": 0,
         "export": 0,
+        "indexCompaction": 0,
         "indexCreation": 3,
         "indexDeletion": 0,
         "indexSwap": 0,
@@ -787,7 +789,7 @@ fn basic_get_stats() {
         "upgradeDatabase": 0
       }
     }
-    "#);
+    "###);
 
     handle.advance_till([
         InsideProcessBatch,
@@ -797,7 +799,7 @@ fn basic_get_stats() {
         Start,
         BatchCreated,
     ]);
-    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r#"
+    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r###"
     {
       "indexes": {
         "catto": 1,
@@ -817,6 +819,7 @@ fn basic_get_stats() {
         "documentEdition": 0,
         "dumpCreation": 0,
         "export": 0,
+        "indexCompaction": 0,
         "indexCreation": 3,
         "indexDeletion": 0,
         "indexSwap": 0,
@@ -828,7 +831,7 @@ fn basic_get_stats() {
         "upgradeDatabase": 0
       }
     }
-    "#);
+    "###);
 
     // now we make one more batch, the started_at field of the new tasks will be past `second_start_time`
     handle.advance_till([
@@ -839,7 +842,7 @@ fn basic_get_stats() {
         Start,
         BatchCreated,
     ]);
-    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r#"
+    snapshot!(json_string!(index_scheduler.get_stats().unwrap()), @r###"
     {
       "indexes": {
         "catto": 1,
@@ -859,6 +862,7 @@ fn basic_get_stats() {
         "documentEdition": 0,
         "dumpCreation": 0,
         "export": 0,
+        "indexCompaction": 0,
         "indexCreation": 3,
         "indexDeletion": 0,
         "indexSwap": 0,
@@ -870,7 +874,7 @@ fn basic_get_stats() {
         "upgradeDatabase": 0
       }
     }
-    "#);
+    "###);
 }
 
 #[test]

crates/index-scheduler/src/utils.rs

@@ -256,14 +256,15 @@ pub fn swap_index_uid_in_task(task: &mut Task, swap: (&str, &str)) {
     use KindWithContent as K;
     let mut index_uids = vec![];
     match &mut task.kind {
-        K::DocumentAdditionOrUpdate { index_uid, .. } => index_uids.push(index_uid),
-        K::DocumentEdition { index_uid, .. } => index_uids.push(index_uid),
-        K::DocumentDeletion { index_uid, .. } => index_uids.push(index_uid),
-        K::DocumentDeletionByFilter { index_uid, .. } => index_uids.push(index_uid),
-        K::DocumentClear { index_uid } => index_uids.push(index_uid),
-        K::SettingsUpdate { index_uid, .. } => index_uids.push(index_uid),
-        K::IndexDeletion { index_uid } => index_uids.push(index_uid),
-        K::IndexCreation { index_uid, .. } => index_uids.push(index_uid),
+        K::DocumentAdditionOrUpdate { index_uid, .. }
+        | K::DocumentEdition { index_uid, .. }
+        | K::DocumentDeletion { index_uid, .. }
+        | K::DocumentDeletionByFilter { index_uid, .. }
+        | K::DocumentClear { index_uid }
+        | K::SettingsUpdate { index_uid, .. }
+        | K::IndexDeletion { index_uid }
+        | K::IndexCreation { index_uid, .. }
+        | K::IndexCompaction { index_uid, .. } => index_uids.push(index_uid),
         K::IndexUpdate { index_uid, new_index_uid, .. } => {
             index_uids.push(index_uid);
             if let Some(new_uid) = new_index_uid {
@@ -618,6 +619,13 @@ impl crate::IndexScheduler {
                     Details::UpgradeDatabase { from: _, to: _ } => {
                         assert_eq!(kind.as_kind(), Kind::UpgradeDatabase);
                     }
+                    Details::IndexCompaction {
+                        index_uid: _,
+                        pre_compaction_size: _,
+                        post_compaction_size: _,
+                    } => {
+                        assert_eq!(kind.as_kind(), Kind::IndexCompaction);
+                    }
                 }
             }
 

crates/meilisearch-auth/src/store.rs

@@ -109,6 +109,7 @@ impl HeedAuthStore {
                             Action::IndexesGet,
                             Action::IndexesUpdate,
                             Action::IndexesSwap,
+                            Action::IndexesCompact,
                         ]
                         .iter(),
                     );

crates/meilisearch-types/src/keys.rs

@@ -380,6 +380,9 @@ pub enum Action {
     #[serde(rename = "webhooks.*")]
     #[deserr(rename = "webhooks.*")]
     WebhooksAll,
+    #[serde(rename = "indexes.compact")]
+    #[deserr(rename = "indexes.compact")]
+    IndexesCompact,
 }
 
 impl Action {
@@ -398,6 +401,7 @@ impl Action {
             INDEXES_UPDATE => Some(Self::IndexesUpdate),
             INDEXES_DELETE => Some(Self::IndexesDelete),
             INDEXES_SWAP => Some(Self::IndexesSwap),
+            INDEXES_COMPACT => Some(Self::IndexesCompact),
             TASKS_ALL => Some(Self::TasksAll),
             TASKS_CANCEL => Some(Self::TasksCancel),
             TASKS_DELETE => Some(Self::TasksDelete),
@@ -462,6 +466,7 @@ impl Action {
             IndexesUpdate => false,
             IndexesDelete => false,
             IndexesSwap => false,
+            IndexesCompact => false,
             TasksCancel => false,
             TasksDelete => false,
             TasksGet => true,
@@ -513,6 +518,7 @@ pub mod actions {
     pub const INDEXES_UPDATE: u8 = IndexesUpdate.repr();
     pub const INDEXES_DELETE: u8 = IndexesDelete.repr();
     pub const INDEXES_SWAP: u8 = IndexesSwap.repr();
+    pub const INDEXES_COMPACT: u8 = IndexesCompact.repr();
     pub const TASKS_ALL: u8 = TasksAll.repr();
     pub const TASKS_CANCEL: u8 = TasksCancel.repr();
     pub const TASKS_DELETE: u8 = TasksDelete.repr();
@@ -614,6 +620,7 @@ pub(crate) mod test {
         assert!(WebhooksDelete.repr() == 47 && WEBHOOKS_DELETE == 47);
         assert!(WebhooksCreate.repr() == 48 && WEBHOOKS_CREATE == 48);
         assert!(WebhooksAll.repr() == 49 && WEBHOOKS_ALL == 49);
+        assert!(IndexesCompact.repr() == 50 && INDEXES_COMPACT == 50);
     }
 
     #[test]

crates/meilisearch-types/src/task_view.rs

@@ -142,6 +142,11 @@ pub struct DetailsView {
     pub old_index_uid: Option<String>,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub new_index_uid: Option<String>,
+    // index compaction
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub pre_compaction_size: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub post_compaction_size: Option<String>,
 }
 
 impl DetailsView {
@@ -314,6 +319,24 @@ impl DetailsView {
                 // We should never be able to batch multiple renames at the same time.
                 (Some(left), Some(_right)) => Some(left),
             },
+            pre_compaction_size: match (
+                self.pre_compaction_size.clone(),
+                other.pre_compaction_size.clone(),
+            ) {
+                (None, None) => None,
+                (None, Some(size)) | (Some(size), None) => Some(size),
+                // We should never be able to batch multiple compactions at the same time.
+                (Some(left), Some(_right)) => Some(left),
+            },
+            post_compaction_size: match (
+                self.post_compaction_size.clone(),
+                other.post_compaction_size.clone(),
+            ) {
+                (None, None) => None,
+                (None, Some(size)) | (Some(size), None) => Some(size),
+                // We should never be able to batch multiple compactions at the same time.
+                (Some(left), Some(_right)) => Some(left),
+            },
         }
     }
 }
@@ -415,6 +438,15 @@ impl From<Details> for DetailsView {
                 upgrade_to: Some(format!("v{}.{}.{}", to.0, to.1, to.2)),
                 ..Default::default()
             },
+            Details::IndexCompaction { pre_compaction_size, post_compaction_size, .. } => {
+                DetailsView {
+                    pre_compaction_size: pre_compaction_size
+                        .map(|size| size.get_appropriate_unit(UnitType::Both).to_string()),
+                    post_compaction_size: post_compaction_size
+                        .map(|size| size.get_appropriate_unit(UnitType::Both).to_string()),
+                    ..Default::default()
+                }
+            }
         }
     }
 }

crates/meilisearch-types/src/tasks.rs

@@ -67,7 +67,8 @@ impl Task {
             | SettingsUpdate { index_uid, .. }
             | IndexCreation { index_uid, .. }
             | IndexUpdate { index_uid, .. }
-            | IndexDeletion { index_uid } => Some(index_uid),
+            | IndexDeletion { index_uid }
+            | IndexCompaction { index_uid } => Some(index_uid),
         }
     }
 
@@ -94,7 +95,8 @@ impl Task {
             | KindWithContent::DumpCreation { .. }
             | KindWithContent::SnapshotCreation
             | KindWithContent::Export { .. }
-            | KindWithContent::UpgradeDatabase { .. } => None,
+            | KindWithContent::UpgradeDatabase { .. }
+            | KindWithContent::IndexCompaction { .. } => None,
         }
     }
 }
@@ -170,6 +172,9 @@ pub enum KindWithContent {
     UpgradeDatabase {
         from: (u32, u32, u32),
     },
+    IndexCompaction {
+        index_uid: String,
+    },
 }
 
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, ToSchema)]
@@ -206,6 +211,7 @@ impl KindWithContent {
             KindWithContent::SnapshotCreation => Kind::SnapshotCreation,
             KindWithContent::Export { .. } => Kind::Export,
             KindWithContent::UpgradeDatabase { .. } => Kind::UpgradeDatabase,
+            KindWithContent::IndexCompaction { .. } => Kind::IndexCompaction,
         }
     }
 
@@ -226,7 +232,8 @@ impl KindWithContent {
             | DocumentClear { index_uid }
             | SettingsUpdate { index_uid, .. }
             | IndexCreation { index_uid, .. }
-            | IndexDeletion { index_uid } => vec![index_uid],
+            | IndexDeletion { index_uid }
+            | IndexCompaction { index_uid } => vec![index_uid],
             IndexUpdate { index_uid, new_index_uid, .. } => {
                 let mut indexes = vec![index_uid.as_str()];
                 if let Some(new_uid) = new_index_uid {
@@ -325,6 +332,11 @@ impl KindWithContent {
                     versioning::VERSION_PATCH,
                 ),
             }),
+            KindWithContent::IndexCompaction { index_uid } => Some(Details::IndexCompaction {
+                index_uid: index_uid.clone(),
+                pre_compaction_size: None,
+                post_compaction_size: None,
+            }),
         }
     }
 
@@ -407,6 +419,11 @@ impl KindWithContent {
                     versioning::VERSION_PATCH,
                 ),
             }),
+            KindWithContent::IndexCompaction { index_uid } => Some(Details::IndexCompaction {
+                index_uid: index_uid.clone(),
+                pre_compaction_size: None,
+                post_compaction_size: None,
+            }),
         }
     }
 }
@@ -469,6 +486,11 @@ impl From<&KindWithContent> for Option<Details> {
                     versioning::VERSION_PATCH,
                 ),
             }),
+            KindWithContent::IndexCompaction { index_uid } => Some(Details::IndexCompaction {
+                index_uid: index_uid.clone(),
+                pre_compaction_size: None,
+                post_compaction_size: None,
+            }),
         }
     }
 }
@@ -579,6 +601,7 @@ pub enum Kind {
     SnapshotCreation,
     Export,
     UpgradeDatabase,
+    IndexCompaction,
 }
 
 impl Kind {
@@ -590,7 +613,8 @@ impl Kind {
             | Kind::SettingsUpdate
             | Kind::IndexCreation
             | Kind::IndexDeletion
-            | Kind::IndexUpdate => true,
+            | Kind::IndexUpdate
+            | Kind::IndexCompaction => true,
             Kind::IndexSwap
             | Kind::TaskCancelation
             | Kind::TaskDeletion
@@ -618,6 +642,7 @@ impl Display for Kind {
             Kind::SnapshotCreation => write!(f, "snapshotCreation"),
             Kind::Export => write!(f, "export"),
             Kind::UpgradeDatabase => write!(f, "upgradeDatabase"),
+            Kind::IndexCompaction => write!(f, "indexCompaction"),
         }
     }
 }
@@ -653,6 +678,8 @@ impl FromStr for Kind {
             Ok(Kind::Export)
         } else if kind.eq_ignore_ascii_case("upgradeDatabase") {
             Ok(Kind::UpgradeDatabase)
+        } else if kind.eq_ignore_ascii_case("indexCompaction") {
+            Ok(Kind::IndexCompaction)
         } else {
             Err(ParseTaskKindError(kind.to_owned()))
         }
@@ -738,6 +765,11 @@ pub enum Details {
         from: (u32, u32, u32),
         to: (u32, u32, u32),
     },
+    IndexCompaction {
+        index_uid: String,
+        pre_compaction_size: Option<Byte>,
+        post_compaction_size: Option<Byte>,
+    },
 }
 
 #[derive(Debug, PartialEq, Clone, Serialize, Deserialize, ToSchema)]
@@ -800,6 +832,10 @@ impl Details {
             Self::ClearAll { deleted_documents } => *deleted_documents = Some(0),
             Self::TaskCancelation { canceled_tasks, .. } => *canceled_tasks = Some(0),
             Self::TaskDeletion { deleted_tasks, .. } => *deleted_tasks = Some(0),
+            Self::IndexCompaction { pre_compaction_size, post_compaction_size, .. } => {
+                *pre_compaction_size = None;
+                *post_compaction_size = None;
+            }
             Self::SettingsUpdate { .. }
             | Self::IndexInfo { .. }
             | Self::Dump { .. }

crates/meilisearch/src/analytics/segment_analytics.rs

@@ -205,6 +205,8 @@ struct Infos {
     experimental_no_snapshot_compaction: bool,
     experimental_no_edition_2024_for_dumps: bool,
     experimental_no_edition_2024_for_settings: bool,
+    experimental_no_edition_2024_for_prefix_post_processing: bool,
+    experimental_no_edition_2024_for_facet_post_processing: bool,
     experimental_vector_store_setting: bool,
     gpu_enabled: bool,
     db_path: bool,
@@ -296,6 +298,8 @@ impl Infos {
             skip_index_budget: _,
             experimental_no_edition_2024_for_settings,
             experimental_no_edition_2024_for_dumps,
+            experimental_no_edition_2024_for_prefix_post_processing,
+            experimental_no_edition_2024_for_facet_post_processing,
         } = indexer_options;
 
         let RuntimeTogglableFeatures {
@@ -365,6 +369,8 @@ impl Infos {
             ssl_resumption,
             ssl_tickets,
             experimental_no_edition_2024_for_settings,
+            experimental_no_edition_2024_for_prefix_post_processing,
+            experimental_no_edition_2024_for_facet_post_processing,
         }
     }
 }

crates/meilisearch/src/option.rs

@@ -55,6 +55,10 @@ const MEILI_EXPERIMENTAL_ENABLE_LOGS_ROUTE: &str = "MEILI_EXPERIMENTAL_ENABLE_LO
 const MEILI_EXPERIMENTAL_CONTAINS_FILTER: &str = "MEILI_EXPERIMENTAL_CONTAINS_FILTER";
 const MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_SETTINGS: &str =
     "MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_SETTINGS";
+const MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_FACET_POST_PROCESSING: &str =
+    "MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_FACET_POST_PROCESSING";
+const MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_PREFIX_POST_PROCESSING: &str =
+    "MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_PREFIX_POST_PROCESSING";
 const MEILI_EXPERIMENTAL_ENABLE_METRICS: &str = "MEILI_EXPERIMENTAL_ENABLE_METRICS";
 const MEILI_EXPERIMENTAL_SEARCH_QUEUE_SIZE: &str = "MEILI_EXPERIMENTAL_SEARCH_QUEUE_SIZE";
 const MEILI_EXPERIMENTAL_DROP_SEARCH_AFTER: &str = "MEILI_EXPERIMENTAL_DROP_SEARCH_AFTER";
@@ -772,6 +776,22 @@ pub struct IndexerOpts {
     #[clap(long, env = MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_DUMPS)]
     #[serde(default)]
     pub experimental_no_edition_2024_for_dumps: bool,
+
+    /// Experimental no edition 2024 to compute prefixes. For more information,
+    /// see: <https://github.com/orgs/meilisearch/discussions/862>
+    ///
+    /// Enables the experimental no edition 2024 to compute prefixes.
+    #[clap(long, env = MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_PREFIX_POST_PROCESSING)]
+    #[serde(default)]
+    pub experimental_no_edition_2024_for_prefix_post_processing: bool,
+
+    /// Experimental no edition 2024 to compute facets. For more information,
+    /// see: <https://github.com/orgs/meilisearch/discussions/862>
+    ///
+    /// Enables the experimental no edition 2024 to compute facets.
+    #[clap(long, env = MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_FACET_POST_PROCESSING)]
+    #[serde(default)]
+    pub experimental_no_edition_2024_for_facet_post_processing: bool,
 }
 
 impl IndexerOpts {
@@ -783,6 +803,8 @@ impl IndexerOpts {
             skip_index_budget: _,
             experimental_no_edition_2024_for_settings,
             experimental_no_edition_2024_for_dumps,
+            experimental_no_edition_2024_for_prefix_post_processing,
+            experimental_no_edition_2024_for_facet_post_processing,
         } = self;
         if let Some(max_indexing_memory) = max_indexing_memory.0 {
             export_to_env_if_not_present(
@@ -808,6 +830,18 @@ impl IndexerOpts {
                 experimental_no_edition_2024_for_dumps.to_string(),
             );
         }
+        if experimental_no_edition_2024_for_prefix_post_processing {
+            export_to_env_if_not_present(
+                MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_PREFIX_POST_PROCESSING,
+                experimental_no_edition_2024_for_prefix_post_processing.to_string(),
+            );
+        }
+        if experimental_no_edition_2024_for_facet_post_processing {
+            export_to_env_if_not_present(
+                MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_FACET_POST_PROCESSING,
+                experimental_no_edition_2024_for_facet_post_processing.to_string(),
+            );
+        }
     }
 }
 
@@ -833,6 +867,10 @@ impl TryFrom<&IndexerOpts> for IndexerConfig {
             chunk_compression_level: Default::default(),
             documents_chunk_size: Default::default(),
             max_nb_chunks: Default::default(),
+            experimental_no_edition_2024_for_prefix_post_processing: other
+                .experimental_no_edition_2024_for_prefix_post_processing,
+            experimental_no_edition_2024_for_facet_post_processing: other
+                .experimental_no_edition_2024_for_facet_post_processing,
         })
     }
 }

crates/meilisearch/src/routes/indexes/compact.rs

@@ -0,0 +1,84 @@
+use actix_web::web::{self, Data};
+use actix_web::{HttpRequest, HttpResponse};
+use index_scheduler::IndexScheduler;
+use meilisearch_types::error::ResponseError;
+use meilisearch_types::index_uid::IndexUid;
+use meilisearch_types::keys::actions;
+use meilisearch_types::tasks::KindWithContent;
+use tracing::debug;
+use utoipa::OpenApi;
+
+use super::ActionPolicy;
+use crate::analytics::Analytics;
+use crate::extractors::authentication::GuardedData;
+use crate::extractors::sequential_extractor::SeqHandler;
+use crate::routes::SummarizedTaskView;
+
+#[derive(OpenApi)]
+#[openapi(
+    paths(compact),
+    tags(
+        (
+            name = "Compact an index",
+            description = "The /compact route uses compacts the database to reorganize and make it smaller and more efficient.",
+            external_docs(url = "https://www.meilisearch.com/docs/reference/api/compact"),
+        ),
+    ),
+)]
+pub struct CompactApi;
+
+pub fn configure(cfg: &mut web::ServiceConfig) {
+    cfg.service(web::resource("").route(web::post().to(SeqHandler(compact))));
+}
+
+/// Compact an index
+#[utoipa::path(
+    post,
+    path = "{indexUid}/compact",
+    tag = "Compact an index",
+    security(("Bearer" = ["search", "*"])),
+    params(("indexUid" = String, Path, example = "movies", description = "Index Unique Identifier", nullable = false)),
+    responses(
+        (status = ACCEPTED, description = "Task successfully enqueued", body = SummarizedTaskView, content_type = "application/json", example = json!(
+            {
+                "taskUid": 147,
+                "indexUid": null,
+                "status": "enqueued",
+                "type": "documentDeletion",
+                "enqueuedAt": "2024-08-08T17:05:55.791772Z"
+            }
+        )),
+        (status = 401, description = "The authorization header is missing", body = ResponseError, content_type = "application/json", example = json!(
+            {
+                "message": "The Authorization header is missing. It must use the bearer authorization method.",
+                "code": "missing_authorization_header",
+                "type": "auth",
+                "link": "https://docs.meilisearch.com/errors#missing_authorization_header"
+            }
+        )),
+    )
+)]
+pub async fn compact(
+    index_scheduler: GuardedData<ActionPolicy<{ actions::INDEXES_COMPACT }>, Data<IndexScheduler>>,
+    index_uid: web::Path<String>,
+    req: HttpRequest,
+    analytics: web::Data<Analytics>,
+) -> Result<HttpResponse, ResponseError> {
+    let index_uid = IndexUid::try_from(index_uid.into_inner())?;
+
+    analytics.publish(IndexCompacted::default(), &req);
+
+    let task = KindWithContent::IndexCompaction { index_uid: index_uid.to_string() };
+    let task =
+        match tokio::task::spawn_blocking(move || index_scheduler.register(task, None, false))
+            .await?
+        {
+            Ok(task) => task,
+            Err(e) => return Err(e.into()),
+        };
+
+    debug!(returns = ?task, "Compact the {index_uid} index");
+    Ok(HttpResponse::Accepted().json(SummarizedTaskView::from(task)))
+}
+
+crate::empty_analytics!(IndexCompacted, "Index Compacted");

crates/meilisearch/src/routes/indexes/mod.rs

@@ -28,6 +28,7 @@ use crate::extractors::sequential_extractor::SeqHandler;
 use crate::routes::is_dry_run;
 use crate::Opt;
 
+pub mod compact;
 pub mod documents;
 mod enterprise_edition;
 pub mod facet_search;
@@ -49,8 +50,9 @@ pub use enterprise_edition::proxy::{PROXY_ORIGIN_REMOTE_HEADER, PROXY_ORIGIN_TAS
         (path = "/", api = facet_search::FacetSearchApi),
         (path = "/", api = similar::SimilarApi),
         (path = "/", api = settings::SettingsApi),
+        (path = "/", api = compact::CompactApi),
     ),
-    paths(list_indexes, create_index, get_index, update_index, delete_index, get_index_stats),
+    paths(list_indexes, create_index, get_index, update_index, delete_index, get_index_stats, compact::compact),
     tags(
         (
             name = "Indexes",
@@ -80,7 +82,8 @@ pub fn configure(cfg: &mut web::ServiceConfig) {
             .service(web::scope("/search").configure(search::configure))
             .service(web::scope("/facet-search").configure(facet_search::configure))
             .service(web::scope("/similar").configure(similar::configure))
-            .service(web::scope("/settings").configure(settings::configure)),
+            .service(web::scope("/settings").configure(settings::configure))
+            .service(web::scope("/compact").configure(compact::configure)),
     );
 }
 

crates/meilisearch/src/routes/tasks_test.rs

@@ -226,14 +226,14 @@ mod tests {
         {
             let params = "types=createIndex";
             let err = deserr_query_params::<TaskDeletionOrCancelationQuery>(params).unwrap_err();
-            snapshot!(meili_snap::json_string!(err), @r#"
+            snapshot!(meili_snap::json_string!(err), @r###"
             {
-              "message": "Invalid value in parameter `types`: `createIndex` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`.",
+              "message": "Invalid value in parameter `types`: `createIndex` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`, `indexCompaction`.",
               "code": "invalid_task_types",
               "type": "invalid_request",
               "link": "https://docs.meilisearch.com/errors#invalid_task_types"
             }
-            "#);
+            "###);
         }
     }
     #[test]

crates/meilisearch/tests/auth/api_keys.rs

@@ -419,14 +419,14 @@ async fn error_add_api_key_invalid_parameters_actions() {
     let (response, code) = server.add_api_key(content).await;
 
     meili_snap::snapshot!(code, @"400 Bad Request");
-    meili_snap::snapshot!(meili_snap::json_string!(response, { ".createdAt" => "[ignored]", ".updatedAt" => "[ignored]" }), @r#"
+    meili_snap::snapshot!(meili_snap::json_string!(response, { ".createdAt" => "[ignored]", ".updatedAt" => "[ignored]" }), @r###"
     {
-      "message": "Unknown value `doc.add` at `.actions[0]`: expected one of `*`, `search`, `documents.*`, `documents.add`, `documents.get`, `documents.delete`, `indexes.*`, `indexes.create`, `indexes.get`, `indexes.update`, `indexes.delete`, `indexes.swap`, `tasks.*`, `tasks.cancel`, `tasks.delete`, `tasks.get`, `settings.*`, `settings.get`, `settings.update`, `stats.*`, `stats.get`, `metrics.*`, `metrics.get`, `dumps.*`, `dumps.create`, `snapshots.*`, `snapshots.create`, `version`, `keys.create`, `keys.get`, `keys.update`, `keys.delete`, `experimental.get`, `experimental.update`, `export`, `network.get`, `network.update`, `chatCompletions`, `chats.*`, `chats.get`, `chats.delete`, `chatsSettings.*`, `chatsSettings.get`, `chatsSettings.update`, `*.get`, `webhooks.get`, `webhooks.update`, `webhooks.delete`, `webhooks.create`, `webhooks.*`",
+      "message": "Unknown value `doc.add` at `.actions[0]`: expected one of `*`, `search`, `documents.*`, `documents.add`, `documents.get`, `documents.delete`, `indexes.*`, `indexes.create`, `indexes.get`, `indexes.update`, `indexes.delete`, `indexes.swap`, `tasks.*`, `tasks.cancel`, `tasks.delete`, `tasks.get`, `settings.*`, `settings.get`, `settings.update`, `stats.*`, `stats.get`, `metrics.*`, `metrics.get`, `dumps.*`, `dumps.create`, `snapshots.*`, `snapshots.create`, `version`, `keys.create`, `keys.get`, `keys.update`, `keys.delete`, `experimental.get`, `experimental.update`, `export`, `network.get`, `network.update`, `chatCompletions`, `chats.*`, `chats.get`, `chats.delete`, `chatsSettings.*`, `chatsSettings.get`, `chatsSettings.update`, `*.get`, `webhooks.get`, `webhooks.update`, `webhooks.delete`, `webhooks.create`, `webhooks.*`, `indexes.compact`",
       "code": "invalid_api_key_actions",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_api_key_actions"
     }
-    "#);
+    "###);
 }
 
 #[actix_rt::test]

crates/meilisearch/tests/auth/errors.rs

@@ -91,14 +91,14 @@ async fn create_api_key_bad_actions() {
     // can't parse
     let (response, code) = server.add_api_key(json!({ "actions": ["doggo"] })).await;
     snapshot!(code, @"400 Bad Request");
-    snapshot!(json_string!(response), @r#"
+    snapshot!(json_string!(response), @r###"
     {
-      "message": "Unknown value `doggo` at `.actions[0]`: expected one of `*`, `search`, `documents.*`, `documents.add`, `documents.get`, `documents.delete`, `indexes.*`, `indexes.create`, `indexes.get`, `indexes.update`, `indexes.delete`, `indexes.swap`, `tasks.*`, `tasks.cancel`, `tasks.delete`, `tasks.get`, `settings.*`, `settings.get`, `settings.update`, `stats.*`, `stats.get`, `metrics.*`, `metrics.get`, `dumps.*`, `dumps.create`, `snapshots.*`, `snapshots.create`, `version`, `keys.create`, `keys.get`, `keys.update`, `keys.delete`, `experimental.get`, `experimental.update`, `export`, `network.get`, `network.update`, `chatCompletions`, `chats.*`, `chats.get`, `chats.delete`, `chatsSettings.*`, `chatsSettings.get`, `chatsSettings.update`, `*.get`, `webhooks.get`, `webhooks.update`, `webhooks.delete`, `webhooks.create`, `webhooks.*`",
+      "message": "Unknown value `doggo` at `.actions[0]`: expected one of `*`, `search`, `documents.*`, `documents.add`, `documents.get`, `documents.delete`, `indexes.*`, `indexes.create`, `indexes.get`, `indexes.update`, `indexes.delete`, `indexes.swap`, `tasks.*`, `tasks.cancel`, `tasks.delete`, `tasks.get`, `settings.*`, `settings.get`, `settings.update`, `stats.*`, `stats.get`, `metrics.*`, `metrics.get`, `dumps.*`, `dumps.create`, `snapshots.*`, `snapshots.create`, `version`, `keys.create`, `keys.get`, `keys.update`, `keys.delete`, `experimental.get`, `experimental.update`, `export`, `network.get`, `network.update`, `chatCompletions`, `chats.*`, `chats.get`, `chats.delete`, `chatsSettings.*`, `chatsSettings.get`, `chatsSettings.update`, `*.get`, `webhooks.get`, `webhooks.update`, `webhooks.delete`, `webhooks.create`, `webhooks.*`, `indexes.compact`",
       "code": "invalid_api_key_actions",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_api_key_actions"
     }
-    "#);
+    "###);
 }
 
 #[actix_rt::test]

crates/meilisearch/tests/batches/errors.rs

@@ -40,14 +40,14 @@ async fn batch_bad_types() {
 
     let (response, code) = server.batches_filter("types=doggo").await;
     snapshot!(code, @"400 Bad Request");
-    snapshot!(json_string!(response), @r#"
+    snapshot!(json_string!(response), @r###"
     {
-      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`.",
+      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`, `indexCompaction`.",
       "code": "invalid_task_types",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_task_types"
     }
-    "#);
+    "###);
 }
 
 #[actix_rt::test]

crates/meilisearch/tests/common/server.rs

@@ -490,6 +490,8 @@ pub fn default_settings(dir: impl AsRef<Path>) -> Opt {
             max_indexing_threads: MaxThreads::from_str("2").unwrap(),
             experimental_no_edition_2024_for_settings: false,
             experimental_no_edition_2024_for_dumps: false,
+            experimental_no_edition_2024_for_prefix_post_processing: false,
+            experimental_no_edition_2024_for_facet_post_processing: false,
         },
         experimental_enable_metrics: false,
         ..Parser::parse_from(None as Option<&str>)

crates/meilisearch/tests/tasks/errors.rs

@@ -97,7 +97,7 @@ async fn task_bad_types() {
     snapshot!(code, @"400 Bad Request");
     snapshot!(json_string!(response), @r#"
     {
-      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`.",
+      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`, `indexCompaction`.",
       "code": "invalid_task_types",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_task_types"
@@ -108,7 +108,7 @@ async fn task_bad_types() {
     snapshot!(code, @"400 Bad Request");
     snapshot!(json_string!(response), @r#"
     {
-      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`.",
+      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`, `indexCompaction`.",
       "code": "invalid_task_types",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_task_types"
@@ -119,7 +119,7 @@ async fn task_bad_types() {
     snapshot!(code, @"400 Bad Request");
     snapshot!(json_string!(response), @r#"
     {
-      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`.",
+      "message": "Invalid value in parameter `types`: `doggo` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentEdition`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`, `export`, `upgradeDatabase`, `indexCompaction`.",
       "code": "invalid_task_types",
       "type": "invalid_request",
       "link": "https://docs.meilisearch.com/errors#invalid_task_types"

crates/meilitool/src/main.rs

@@ -126,7 +126,7 @@ enum Command {
     /// before running the copy and compaction. This way the current indexation must finish before
     /// the compaction operation can start. Once the compaction is done, the big index is replaced
     /// by the compacted one and the mutable transaction is released.
-    CompactIndex { index_name: String },
+    IndexCompaction { index_name: String },
 
     /// Uses the hair dryer the dedicate pages hot in cache
     ///
@@ -165,7 +165,7 @@ fn main() -> anyhow::Result<()> {
             let target_version = parse_version(&target_version).context("While parsing `--target-version`. Make sure `--target-version` is in the format MAJOR.MINOR.PATCH")?;
             OfflineUpgrade { db_path, current_version: detected_version, target_version }.upgrade()
         }
-        Command::CompactIndex { index_name } => compact_index(db_path, &index_name),
+        Command::IndexCompaction { index_name } => compact_index(db_path, &index_name),
         Command::HairDryer { index_name, index_part } => {
             hair_dryer(db_path, &index_name, &index_part)
         }

crates/milli/Cargo.toml

@@ -19,7 +19,7 @@ bstr = "1.12.0"
 bytemuck = { version = "1.23.1", features = ["extern_crate_alloc"] }
 byteorder = "1.5.0"
 charabia = { version = "0.9.7", default-features = false }
-cellulite = "0.3.0"
+cellulite = "0.3.1-nested-rtxns-2"
 concat-arrays = "0.1.2"
 convert_case = "0.8.0"
 crossbeam-channel = "0.5.15"
@@ -34,7 +34,7 @@ grenad = { version = "0.5.0", default-features = false, features = [
     "rayon",
     "tempfile",
 ] }
-heed = { version = "0.22.0", default-features = false, features = [
+heed = { version = "0.22.1-nested-rtxns", default-features = false, features = [
     "serde-json",
     "serde-bincode",
 ] }
@@ -89,8 +89,8 @@ rhai = { version = "1.22.2", features = [
     "no_time",
     "sync",
 ] }
-arroy = "0.6.3"
-hannoy = { version = "0.0.8", features = ["arroy"] }
+arroy = "0.6.4-nested-rtxns"
+hannoy = { version = "0.0.9-nested-rtxns", features = ["arroy"] }
 rand = "0.8.5"
 tracing = "0.1.41"
 ureq = { version = "2.12.1", features = ["json"] }

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

@@ -21,7 +21,7 @@ mod vector_sort;
 #[cfg(test)]
 mod tests;
 
-use std::collections::HashSet;
+use std::collections::{HashMap, HashSet};
 use std::ops::AddAssign;
 use std::time::Duration;
 
@@ -64,6 +64,12 @@ use crate::{
     UserError, Weight,
 };
 
+/// Cache for synonyms to avoid repeated database access
+#[derive(Default)]
+pub struct SynonymCache {
+    pub cache: Option<HashMap<Vec<String>, Vec<Vec<String>>>>,
+}
+
 /// A structure used throughout the execution of a search query.
 pub struct SearchContext<'ctx> {
     pub index: &'ctx Index,
@@ -73,6 +79,7 @@ pub struct SearchContext<'ctx> {
     pub phrase_interner: DedupInterner<Phrase>,
     pub term_interner: Interner<QueryTerm>,
     pub phrase_docids: PhraseDocIdsCache,
+    pub synonym_cache: SynonymCache,
     pub restricted_fids: Option<RestrictedFids>,
     pub prefix_search: PrefixSearch,
     pub vector_store_stats: Option<VectorStoreStats>,
@@ -103,6 +110,7 @@ impl<'ctx> SearchContext<'ctx> {
             phrase_interner: <_>::default(),
             term_interner: <_>::default(),
             phrase_docids: <_>::default(),
+            synonym_cache: <_>::default(),
             restricted_fids: None,
             prefix_search,
             vector_store_stats: None,
@@ -113,6 +121,17 @@ impl<'ctx> SearchContext<'ctx> {
         self.prefix_search != PrefixSearch::Disabled
     }
 
+    /// Get synonyms with caching to avoid repeated database access
+    pub fn get_synonyms(&mut self) -> Result<&HashMap<Vec<String>, Vec<Vec<String>>>> {
+        match self.synonym_cache.cache {
+            Some(ref synonyms) => Ok(synonyms),
+            None => {
+                let synonyms = self.index.synonyms(self.txn)?;
+                Ok(self.synonym_cache.cache.insert(synonyms))
+            }
+        }
+    }
+
     pub fn attributes_to_search_on(
         &mut self,
         attributes_to_search_on: &'ctx [String],

crates/milli/src/search/new/query_term/compute_derivations.rs

@@ -214,7 +214,7 @@ pub fn partially_initialized_term_from_word(
     if is_prefix && use_prefix_db.is_none() {
         find_zero_typo_prefix_derivations(ctx, word_interned, &mut prefix_of)?;
     }
-    let synonyms = ctx.index.synonyms(ctx.txn)?;
+    let synonyms = ctx.get_synonyms()?;
     let mut synonym_word_count = 0;
     let synonyms = synonyms
         .get(&vec![word.to_owned()])

crates/milli/src/search/new/query_term/parse_query.rs

@@ -258,7 +258,7 @@ pub fn make_ngram(
         partially_initialized_term_from_word(ctx, &ngram_str, max_nbr_typos, is_prefix, true)?;
 
     // Now add the synonyms
-    let index_synonyms = ctx.index.synonyms(ctx.txn)?;
+    let index_synonyms = ctx.get_synonyms()?;
 
     term.zero_typo.synonyms.extend(
         index_synonyms.get(&words).cloned().unwrap_or_default().into_iter().map(|words| {

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

@@ -101,6 +101,8 @@ pub struct GrenadParameters {
     pub chunk_compression_level: Option<u32>,
     pub max_memory: Option<usize>,
     pub max_nb_chunks: Option<usize>,
+    pub experimental_no_edition_2024_for_prefix_post_processing: bool,
+    pub experimental_no_edition_2024_for_facet_post_processing: bool,
 }
 
 impl Default for GrenadParameters {
@@ -110,6 +112,8 @@ impl Default for GrenadParameters {
             chunk_compression_level: None,
             max_memory: None,
             max_nb_chunks: None,
+            experimental_no_edition_2024_for_prefix_post_processing: false,
+            experimental_no_edition_2024_for_facet_post_processing: false,
         }
     }
 }

crates/milli/src/update/index_documents/mod.rs

@@ -254,6 +254,12 @@ where
             chunk_compression_level: self.indexer_config.chunk_compression_level,
             max_memory: self.indexer_config.max_memory,
             max_nb_chunks: self.indexer_config.max_nb_chunks, // default value, may be chosen.
+            experimental_no_edition_2024_for_prefix_post_processing: self
+                .indexer_config
+                .experimental_no_edition_2024_for_prefix_post_processing,
+            experimental_no_edition_2024_for_facet_post_processing: self
+                .indexer_config
+                .experimental_no_edition_2024_for_facet_post_processing,
         };
         let documents_chunk_size = match self.indexer_config.documents_chunk_size {
             Some(chunk_size) => chunk_size,

crates/milli/src/update/index_documents/transform.rs

@@ -983,6 +983,12 @@ impl<'a, 'i> Transform<'a, 'i> {
             chunk_compression_level: self.indexer_settings.chunk_compression_level,
             max_memory: self.indexer_settings.max_memory,
             max_nb_chunks: self.indexer_settings.max_nb_chunks, // default value, may be chosen.
+            experimental_no_edition_2024_for_prefix_post_processing: self
+                .indexer_settings
+                .experimental_no_edition_2024_for_prefix_post_processing,
+            experimental_no_edition_2024_for_facet_post_processing: self
+                .indexer_settings
+                .experimental_no_edition_2024_for_facet_post_processing,
         };
 
         // Once we have written all the documents, we merge everything into a Reader.

crates/milli/src/update/indexer_config.rs

@@ -18,6 +18,8 @@ pub struct IndexerConfig {
     pub skip_index_budget: bool,
     pub experimental_no_edition_2024_for_settings: bool,
     pub experimental_no_edition_2024_for_dumps: bool,
+    pub experimental_no_edition_2024_for_prefix_post_processing: bool,
+    pub experimental_no_edition_2024_for_facet_post_processing: bool,
 }
 
 impl IndexerConfig {
@@ -27,6 +29,10 @@ impl IndexerConfig {
             chunk_compression_level: self.chunk_compression_level,
             max_memory: self.max_memory,
             max_nb_chunks: self.max_nb_chunks,
+            experimental_no_edition_2024_for_prefix_post_processing: self
+                .experimental_no_edition_2024_for_prefix_post_processing,
+            experimental_no_edition_2024_for_facet_post_processing: self
+                .experimental_no_edition_2024_for_facet_post_processing,
         }
     }
 }
@@ -68,6 +74,8 @@ impl Default for IndexerConfig {
             skip_index_budget: false,
             experimental_no_edition_2024_for_settings: false,
             experimental_no_edition_2024_for_dumps: false,
+            experimental_no_edition_2024_for_prefix_post_processing: false,
+            experimental_no_edition_2024_for_facet_post_processing: false,
         }
     }
 }

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

@@ -180,12 +180,15 @@ where
         })
         .unwrap()?;
 
-        post_processing::post_process(
-            indexing_context,
-            wtxn,
-            global_fields_ids_map,
-            facet_field_ids_delta,
-        )?;
+        pool.install(|| {
+            post_processing::post_process(
+                indexing_context,
+                wtxn,
+                global_fields_ids_map,
+                facet_field_ids_delta,
+            )
+        })
+        .unwrap()?;
 
         indexing_context.progress.update_progress(IndexingStep::BuildingGeoJson);
         index.cellulite.build(

crates/milli/src/update/new/indexer/post_processing/facet_bulk.rs

@@ -0,0 +1,164 @@
+use std::fs::File;
+use std::io::BufReader;
+use std::{iter, mem};
+
+use grenad::CompressionType;
+use heed::types::{Bytes, LazyDecode};
+use heed::{Database, RwTxn};
+use rayon::prelude::*;
+use roaring::MultiOps;
+use tempfile::tempfile;
+
+use crate::facet::FacetType;
+use crate::heed_codec::facet::{FacetGroupKey, FacetGroupKeyCodec, FacetGroupValueCodec};
+use crate::heed_codec::BytesRefCodec;
+use crate::update::facet::{FACET_GROUP_SIZE, FACET_MIN_LEVEL_SIZE};
+use crate::update::{create_writer, writer_into_reader};
+use crate::{CboRoaringBitmapCodec, FieldId, Index};
+
+/// Generate the facet level based on the level 0.
+///
+/// The function will generate all the group levels from
+/// the group 1 to the level n until the number of group
+/// is smaller than the minimum required size.
+pub fn generate_facet_levels(
+    index: &Index,
+    wtxn: &mut RwTxn,
+    field_id: FieldId,
+    facet_type: FacetType,
+) -> crate::Result<()> {
+    let db = match facet_type {
+        FacetType::String => index
+            .facet_id_string_docids
+            .remap_key_type::<FacetGroupKeyCodec<BytesRefCodec>>()
+            .lazily_decode_data(),
+        FacetType::Number => index
+            .facet_id_f64_docids
+            .remap_key_type::<FacetGroupKeyCodec<BytesRefCodec>>()
+            .lazily_decode_data(),
+    };
+
+    clear_levels(db, wtxn, field_id)?;
+
+    let mut base_level: u8 = 0;
+    // That's a do-while loop
+    while {
+        let mut level_size = 0;
+        let level = base_level.checked_add(1).unwrap();
+        for reader in compute_level(index, wtxn, db, field_id, base_level)? {
+            let mut cursor = reader.into_cursor()?;
+            while let Some((left_bound, facet_group_value)) = cursor.move_on_next()? {
+                level_size += 1;
+                let key = FacetGroupKey { field_id, level, left_bound };
+                debug_assert!(
+                    db.get(wtxn, &key).transpose().is_none(),
+                    "entry must not be there and must have already been deleted: {key:?}"
+                );
+                db.remap_data_type::<Bytes>().put(wtxn, &key, facet_group_value)?;
+            }
+        }
+
+        base_level = level;
+
+        // If the next level will have the minimum required groups, continue.
+        (level_size / FACET_GROUP_SIZE as usize) >= FACET_MIN_LEVEL_SIZE as usize
+    } {}
+
+    Ok(())
+}
+
+/// Compute the groups of facets from the provided base level
+/// and write the content into different grenad files.
+fn compute_level(
+    index: &Index,
+    wtxn: &heed::RwTxn,
+    db: Database<FacetGroupKeyCodec<BytesRefCodec>, LazyDecode<FacetGroupValueCodec>>,
+    field_id: FieldId,
+    base_level: u8,
+) -> Result<Vec<grenad::Reader<BufReader<File>>>, crate::Error> {
+    let thread_count = rayon::current_num_threads();
+    let rtxns = iter::repeat_with(|| index.env.nested_read_txn(wtxn))
+        .take(thread_count)
+        .collect::<heed::Result<Vec<_>>>()?;
+
+    let range = {
+        // Based on the first possible value for the base level up to
+        // the first possible value for the next level *excluded*.
+        let left = FacetGroupKey::<&[u8]> { field_id, level: base_level, left_bound: &[] };
+        let right = FacetGroupKey::<&[u8]> {
+            field_id,
+            level: base_level.checked_add(1).unwrap(),
+            left_bound: &[],
+        };
+        left..right
+    };
+
+    rtxns
+        .into_par_iter()
+        .enumerate()
+        .map(|(thread_id, rtxn)| {
+            let mut writer = tempfile().map(|f| create_writer(CompressionType::None, None, f))?;
+
+            let mut left_bound = None;
+            let mut group_docids = Vec::new();
+            let mut ser_buffer = Vec::new();
+            for (i, result) in db.range(&rtxn, &range)?.enumerate() {
+                let (key, lazy_value) = result?;
+
+                let start_of_group = i % FACET_GROUP_SIZE as usize == 0;
+                let group_index = i / FACET_GROUP_SIZE as usize;
+                let group_for_thread = group_index % thread_count == thread_id;
+
+                if group_for_thread {
+                    if start_of_group {
+                        if let Some(left_bound) = left_bound.take() {
+                            // We store the bitmaps in a Vec this way we can use
+                            // the MultiOps operations that tends to be more efficient
+                            // for unions. The Vec is empty after the operation.
+                            //
+                            // We also don't forget to store the group size corresponding
+                            // to the number of entries merged in this group.
+                            ser_buffer.clear();
+                            let group_len: u8 = group_docids.len().try_into().unwrap();
+                            ser_buffer.push(group_len);
+                            let group_docids = mem::take(&mut group_docids);
+                            let docids = group_docids.into_iter().union();
+                            CboRoaringBitmapCodec::serialize_into_vec(&docids, &mut ser_buffer);
+                            writer.insert(left_bound, &ser_buffer)?;
+                        }
+                        left_bound = Some(key.left_bound);
+                    }
+
+                    // Lazily decode the bitmaps we are interested in.
+                    let value = lazy_value.decode().map_err(heed::Error::Decoding)?;
+                    group_docids.push(value.bitmap);
+                }
+            }
+
+            if let Some(left_bound) = left_bound.take() {
+                ser_buffer.clear();
+                // We don't forget to store the group size corresponding
+                // to the number of entries merged in this group.
+                let group_len: u8 = group_docids.len().try_into().unwrap();
+                ser_buffer.push(group_len);
+                let group_docids = group_docids.into_iter().union();
+                CboRoaringBitmapCodec::serialize_into_vec(&group_docids, &mut ser_buffer);
+                writer.insert(left_bound, &ser_buffer)?;
+            }
+
+            writer_into_reader(writer)
+        })
+        .collect()
+}
+
+/// Clears all the levels and only keeps the level 0 of the specified field id.
+fn clear_levels(
+    db: Database<FacetGroupKeyCodec<BytesRefCodec>, LazyDecode<FacetGroupValueCodec>>,
+    wtxn: &mut RwTxn<'_>,
+    field_id: FieldId,
+) -> heed::Result<()> {
+    let left = FacetGroupKey::<&[u8]> { field_id, level: 1, left_bound: &[] };
+    let right = FacetGroupKey::<&[u8]> { field_id, level: u8::MAX, left_bound: &[] };
+    let range = left..=right;
+    db.delete_range(wtxn, &range).map(drop)
+}

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

@@ -1,5 +1,6 @@
 use std::cmp::Ordering;
 
+use facet_bulk::generate_facet_levels;
 use heed::types::{Bytes, DecodeIgnore, Str};
 use heed::RwTxn;
 use itertools::{merge_join_by, EitherOrBoth};
@@ -23,6 +24,8 @@ use crate::update::new::FacetFieldIdsDelta;
 use crate::update::{FacetsUpdateBulk, GrenadParameters};
 use crate::{GlobalFieldsIdsMap, Index, Result};
 
+mod facet_bulk;
+
 pub(super) fn post_process<MSP>(
     indexing_context: IndexingContext<MSP>,
     wtxn: &mut RwTxn<'_>,
@@ -39,6 +42,7 @@ where
         wtxn,
         facet_field_ids_delta,
         &mut global_fields_ids_map,
+        indexing_context.grenad_parameters,
         indexing_context.progress,
     )?;
     compute_facet_search_database(index, wtxn, global_fields_ids_map, indexing_context.progress)?;
@@ -216,6 +220,7 @@ fn compute_facet_level_database(
     wtxn: &mut RwTxn,
     mut facet_field_ids_delta: FacetFieldIdsDelta,
     global_fields_ids_map: &mut GlobalFieldsIdsMap,
+    grenad_parameters: &GrenadParameters,
     progress: &Progress,
 ) -> Result<()> {
     let rtxn = index.read_txn()?;
@@ -239,9 +244,14 @@ fn compute_facet_level_database(
         match delta {
             FacetFieldIdDelta::Bulk => {
                 progress.update_progress(PostProcessingFacets::StringsBulk);
-                tracing::debug!(%fid, "bulk string facet processing");
-                FacetsUpdateBulk::new_not_updating_level_0(index, vec![fid], FacetType::String)
-                    .execute(wtxn)?
+                if grenad_parameters.experimental_no_edition_2024_for_facet_post_processing {
+                    tracing::debug!(%fid, "bulk string facet processing");
+                    FacetsUpdateBulk::new_not_updating_level_0(index, vec![fid], FacetType::String)
+                        .execute(wtxn)?
+                } else {
+                    tracing::debug!(%fid, "bulk string facet processing in parallel");
+                    generate_facet_levels(index, wtxn, fid, FacetType::String)?
+                }
             }
             FacetFieldIdDelta::Incremental(delta_data) => {
                 progress.update_progress(PostProcessingFacets::StringsIncremental);

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

@@ -1,11 +1,12 @@
 use std::cell::RefCell;
 use std::collections::BTreeSet;
 use std::io::{BufReader, BufWriter, Read, Seek, Write};
+use std::iter;
 
 use hashbrown::HashMap;
-use heed::types::Bytes;
+use heed::types::{Bytes, DecodeIgnore};
 use heed::{BytesDecode, Database, Error, RoTxn, RwTxn};
-use rayon::iter::{IntoParallelIterator, ParallelIterator as _};
+use rayon::iter::{IndexedParallelIterator as _, IntoParallelIterator, ParallelIterator as _};
 use roaring::MultiOps;
 use tempfile::spooled_tempfile;
 use thread_local::ThreadLocal;
@@ -151,22 +152,29 @@ impl<'a, 'rtxn> FrozenPrefixBitmaps<'a, 'rtxn> {
 
 unsafe impl Sync for FrozenPrefixBitmaps<'_, '_> {}
 
-struct WordPrefixIntegerDocids {
+struct WordPrefixIntegerDocids<'i> {
+    index: &'i Index,
     database: Database<Bytes, CboRoaringBitmapCodec>,
     prefix_database: Database<Bytes, CboRoaringBitmapCodec>,
     max_memory_by_thread: Option<usize>,
+    /// Do not use an experimental LMDB feature to read uncommitted data in parallel.
+    no_experimental_post_processing: bool,
 }
 
-impl WordPrefixIntegerDocids {
+impl<'i> WordPrefixIntegerDocids<'i> {
     fn new(
+        index: &'i Index,
         database: Database<Bytes, CboRoaringBitmapCodec>,
         prefix_database: Database<Bytes, CboRoaringBitmapCodec>,
-        grenad_parameters: &GrenadParameters,
-    ) -> WordPrefixIntegerDocids {
+        grenad_parameters: &'_ GrenadParameters,
+    ) -> WordPrefixIntegerDocids<'i> {
         WordPrefixIntegerDocids {
+            index,
             database,
             prefix_database,
             max_memory_by_thread: grenad_parameters.max_memory_by_thread(),
+            no_experimental_post_processing: grenad_parameters
+                .experimental_no_edition_2024_for_prefix_post_processing,
         }
     }
 
@@ -177,7 +185,131 @@ impl WordPrefixIntegerDocids {
         prefix_to_delete: &BTreeSet<Prefix>,
     ) -> Result<()> {
         delete_prefixes(wtxn, &self.prefix_database, prefix_to_delete)?;
-        self.recompute_modified_prefixes(wtxn, prefix_to_compute)
+        if self.no_experimental_post_processing {
+            self.recompute_modified_prefixes(wtxn, prefix_to_compute)
+        } else {
+            self.recompute_modified_prefixes_no_frozen(wtxn, prefix_to_compute)
+        }
+    }
+
+    /// Computes the same as `recompute_modified_prefixes`.
+    ///
+    /// ...but without aggregating the prefixes mmap pointers into a static HashMap
+    /// beforehand and rather use an experimental LMDB feature to read the subset
+    /// of prefixes in parallel from the uncommitted transaction.
+    #[tracing::instrument(level = "trace", skip_all, target = "indexing::prefix")]
+    fn recompute_modified_prefixes_no_frozen(
+        &self,
+        wtxn: &mut RwTxn,
+        prefixes: &BTreeSet<Prefix>,
+    ) -> Result<()> {
+        let thread_count = rayon::current_num_threads();
+        let rtxns = iter::repeat_with(|| self.index.env.nested_read_txn(wtxn))
+            .take(thread_count)
+            .collect::<heed::Result<Vec<_>>>()?;
+
+        let outputs = rtxns
+            .into_par_iter()
+            .enumerate()
+            .map(|(thread_id, rtxn)| {
+                // `indexes` represent offsets at which prefixes computations were stored in the `file`.
+                let mut indexes = Vec::new();
+                let mut file = BufWriter::new(spooled_tempfile(
+                    self.max_memory_by_thread.unwrap_or(usize::MAX),
+                ));
+
+                let mut buffer = Vec::new();
+                for (prefix_index, prefix) in prefixes.iter().enumerate() {
+                    // Is prefix for another thread?
+                    if prefix_index % thread_count != thread_id {
+                        continue;
+                    }
+
+                    let mut bitmap_bytes_at_positions = HashMap::new();
+                    for result in self
+                        .database
+                        .prefix_iter(&rtxn, prefix.as_bytes())?
+                        .remap_types::<StrBEU16Codec, Bytes>()
+                    {
+                        let ((_word, pos), bitmap_bytes) = result?;
+                        bitmap_bytes_at_positions
+                            .entry(pos)
+                            .or_insert_with(Vec::new)
+                            .push(bitmap_bytes);
+                    }
+
+                    // We track positions with no corresponding bitmap bytes,
+                    // these means that the prefix no longer exists in the database
+                    // and must, therefore, be removed from the index.
+                    for result in self
+                        .prefix_database
+                        .prefix_iter(&rtxn, prefix.as_bytes())?
+                        .remap_types::<StrBEU16Codec, DecodeIgnore>()
+                    {
+                        let ((_word, pos), ()) = result?;
+                        // They are represented by an empty set of bitmaps.
+                        bitmap_bytes_at_positions.entry(pos).or_insert_with(Vec::new);
+                    }
+
+                    for (pos, bitmaps_bytes) in bitmap_bytes_at_positions {
+                        if bitmaps_bytes.is_empty() {
+                            indexes.push(PrefixIntegerEntry {
+                                prefix,
+                                pos,
+                                serialized_length: None,
+                            });
+                        } else {
+                            let output = bitmaps_bytes
+                                .into_iter()
+                                .map(CboRoaringBitmapCodec::deserialize_from)
+                                .union()?;
+                            buffer.clear();
+                            CboRoaringBitmapCodec::serialize_into_vec(&output, &mut buffer);
+                            indexes.push(PrefixIntegerEntry {
+                                prefix,
+                                pos,
+                                serialized_length: Some(buffer.len()),
+                            });
+                            file.write_all(&buffer)?;
+                        }
+                    }
+                }
+
+                Ok((indexes, file))
+            })
+            .collect::<Result<Vec<_>>>()?;
+
+        // We iterate over all the collected and serialized bitmaps through
+        // the files and entries to eventually put them in the final database.
+        let mut key_buffer = Vec::new();
+        let mut buffer = Vec::new();
+        for (index, file) in outputs {
+            let mut file = file.into_inner().map_err(|e| e.into_error())?;
+            file.rewind()?;
+            let mut file = BufReader::new(file);
+            for PrefixIntegerEntry { prefix, pos, serialized_length } in index {
+                key_buffer.clear();
+                key_buffer.extend_from_slice(prefix.as_bytes());
+                key_buffer.push(0);
+                key_buffer.extend_from_slice(&pos.to_be_bytes());
+                match serialized_length {
+                    Some(serialized_length) => {
+                        buffer.resize(serialized_length, 0);
+                        file.read_exact(&mut buffer)?;
+                        self.prefix_database.remap_data_type::<Bytes>().put(
+                            wtxn,
+                            &key_buffer,
+                            &buffer,
+                        )?;
+                    }
+                    None => {
+                        self.prefix_database.delete(wtxn, &key_buffer)?;
+                    }
+                }
+            }
+        }
+
+        Ok(())
     }
 
     #[tracing::instrument(level = "trace", skip_all, target = "indexing::prefix")]
@@ -262,7 +394,7 @@ impl WordPrefixIntegerDocids {
     }
 }
 
-/// Represents a prefix and the lenght the bitmap takes on disk.
+/// Represents a prefix and the length the bitmap takes on disk.
 struct PrefixIntegerEntry<'a> {
     prefix: &'a str,
     pos: u16,
@@ -363,6 +495,7 @@ pub fn compute_word_prefix_fid_docids(
     grenad_parameters: &GrenadParameters,
 ) -> Result<()> {
     WordPrefixIntegerDocids::new(
+        index,
         index.word_fid_docids.remap_key_type(),
         index.word_prefix_fid_docids.remap_key_type(),
         grenad_parameters,
@@ -379,6 +512,7 @@ pub fn compute_word_prefix_position_docids(
     grenad_parameters: &GrenadParameters,
 ) -> Result<()> {
     WordPrefixIntegerDocids::new(
+        index,
         index.word_position_docids.remap_key_type(),
         index.word_prefix_position_docids.remap_key_type(),
         grenad_parameters,