Merge pull request #5509 from meilisearch/release-v1.14.0-tmp

Bring back changes from v1.14.0 to main

crates/index-scheduler/Cargo.toml

@@ -13,6 +13,7 @@ license.workspace = true
 [dependencies]
 anyhow = "1.0.95"
 bincode = "1.3.3"
+byte-unit = "5.1.6"
 bumpalo = "3.16.0"
 bumparaw-collections = "0.1.4"
 convert_case = "0.6.0"
@@ -22,6 +23,7 @@ dump = { path = "../dump" }
 enum-iterator = "2.1.0"
 file-store = { path = "../file-store" }
 flate2 = "1.0.35"
+indexmap = "2.7.0"
 meilisearch-auth = { path = "../meilisearch-auth" }
 meilisearch-types = { path = "../meilisearch-types" }
 memmap2 = "0.9.5"

crates/index-scheduler/src/insta_snapshot.rs

@@ -344,6 +344,7 @@ pub fn snapshot_batch(batch: &Batch) -> String {
     let Batch { uid, details, stats, started_at, finished_at, progress: _, enqueued_at } = batch;
     let stats = BatchStats {
         progress_trace: Default::default(),
+        internal_database_sizes: Default::default(),
         write_channel_congestion: None,
         ..stats.clone()
     };

crates/index-scheduler/src/lib.rs

@@ -625,8 +625,8 @@ impl IndexScheduler {
         task_id: Option<TaskId>,
         dry_run: bool,
     ) -> Result<Task> {
-        // 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())
+        // if the task doesn't delete or cancel anything and 40% of the task queue is full, we must refuse to enqueue the incoming task
+        if !matches!(&kind, KindWithContent::TaskDeletion { tasks, .. } | KindWithContent::TaskCancelation { tasks, .. } if !tasks.is_empty())
             && (self.env.non_free_pages_size()? * 100) / self.env.info().map_size as u64 > 40
         {
             return Err(Error::NoSpaceLeftInTaskQueue);

crates/index-scheduler/src/processing.rs

@@ -64,6 +64,13 @@ make_enum_progress! {
     }
 }
 
+make_enum_progress! {
+    pub enum FinalizingIndexStep {
+        Committing,
+        ComputingStats,
+    }
+}
+
 make_enum_progress! {
     pub enum TaskCancelationProgress {
         RetrievingTasks,

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

@@ -292,8 +292,6 @@ impl Queue {
             return Ok(task);
         }
 
-        // Get rid of the mutability.
-        let task = task;
         self.tasks.register(wtxn, &task)?;
 
         Ok(task)

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

@@ -364,7 +364,7 @@ fn test_task_queue_is_full() {
     // we won't be able to test this error in an integration test thus as a best effort test I still ensure the error return the expected error code
     snapshot!(format!("{:?}", result.error_code()), @"NoSpaceLeftOnDevice");
 
-    // Even the task deletion that doesn't delete anything shouldn't be accepted
+    // Even the task deletion and cancelation that don't delete anything should be refused
     let result = index_scheduler
         .register(
             KindWithContent::TaskDeletion { query: S("test"), tasks: RoaringBitmap::new() },
@@ -373,10 +373,39 @@ fn test_task_queue_is_full() {
         )
         .unwrap_err();
     snapshot!(result, @"Meilisearch cannot receive write operations because the limit of the task database has been reached. Please delete tasks to continue performing write operations.");
+    let result = index_scheduler
+        .register(
+            KindWithContent::TaskCancelation { query: S("test"), tasks: RoaringBitmap::new() },
+            None,
+            false,
+        )
+        .unwrap_err();
+    snapshot!(result, @"Meilisearch cannot receive write operations because the limit of the task database has been reached. Please delete tasks to continue performing write operations.");
+
     // we won't be able to test this error in an integration test thus as a best effort test I still ensure the error return the expected error code
     snapshot!(format!("{:?}", result.error_code()), @"NoSpaceLeftOnDevice");
 
-    // But a task deletion that delete something should works
+    // But a task cancelation that cancel something should work
+    index_scheduler
+        .register(
+            KindWithContent::TaskCancelation { query: S("test"), tasks: (0..100).collect() },
+            None,
+            false,
+        )
+        .unwrap();
+    handle.advance_one_successful_batch();
+
+    // But we should still be forbidden from enqueuing new tasks
+    let result = index_scheduler
+        .register(
+            KindWithContent::IndexCreation { index_uid: S("doggo"), primary_key: None },
+            None,
+            false,
+        )
+        .unwrap_err();
+    snapshot!(result, @"Meilisearch cannot receive write operations because the limit of the task database has been reached. Please delete tasks to continue performing write operations.");
+
+    // And a task deletion that delete something should works
     index_scheduler
         .register(
             KindWithContent::TaskDeletion { query: S("test"), tasks: (0..100).collect() },

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

@@ -20,10 +20,12 @@ use std::path::PathBuf;
 use std::sync::atomic::{AtomicBool, AtomicU32, Ordering};
 use std::sync::Arc;
 
+use convert_case::{Case, Casing as _};
 use meilisearch_types::error::ResponseError;
 use meilisearch_types::heed::{Env, WithoutTls};
 use meilisearch_types::milli;
 use meilisearch_types::tasks::Status;
+use process_batch::ProcessBatchInfo;
 use rayon::current_num_threads;
 use rayon::iter::{IntoParallelIterator, ParallelIterator};
 use roaring::RoaringBitmap;
@@ -223,16 +225,16 @@ impl IndexScheduler {
         let mut stop_scheduler_forever = false;
         let mut wtxn = self.env.write_txn().map_err(Error::HeedTransaction)?;
         let mut canceled = RoaringBitmap::new();
-        let mut congestion = None;
+        let mut process_batch_info = ProcessBatchInfo::default();
 
         match res {
-            Ok((tasks, cong)) => {
+            Ok((tasks, info)) => {
                 #[cfg(test)]
                 self.breakpoint(crate::test_utils::Breakpoint::ProcessBatchSucceeded);
 
                 let (task_progress, task_progress_obj) = AtomicTaskStep::new(tasks.len() as u32);
                 progress.update_progress(task_progress_obj);
-                congestion = cong;
+                process_batch_info = info;
                 let mut success = 0;
                 let mut failure = 0;
                 let mut canceled_by = None;
@@ -350,6 +352,9 @@ impl IndexScheduler {
         // We must re-add the canceled task so they're part of the same batch.
         ids |= canceled;
 
+        let ProcessBatchInfo { congestion, pre_commit_dabases_sizes, post_commit_dabases_sizes } =
+            process_batch_info;
+
         processing_batch.stats.progress_trace =
             progress.accumulated_durations().into_iter().map(|(k, v)| (k, v.into())).collect();
         processing_batch.stats.write_channel_congestion = congestion.map(|congestion| {
@@ -359,6 +364,33 @@ impl IndexScheduler {
             congestion_info.insert("blocking_ratio".into(), congestion.congestion_ratio().into());
             congestion_info
         });
+        processing_batch.stats.internal_database_sizes = pre_commit_dabases_sizes
+            .iter()
+            .flat_map(|(dbname, pre_size)| {
+                post_commit_dabases_sizes
+                    .get(dbname)
+                    .map(|post_size| {
+                        use byte_unit::{Byte, UnitType::Binary};
+                        use std::cmp::Ordering::{Equal, Greater, Less};
+
+                        let post = Byte::from_u64(*post_size as u64).get_appropriate_unit(Binary);
+                        let diff_size = post_size.abs_diff(*pre_size) as u64;
+                        let diff = Byte::from_u64(diff_size).get_appropriate_unit(Binary);
+                        let sign = match post_size.cmp(pre_size) {
+                            Equal => return None,
+                            Greater => "+",
+                            Less => "-",
+                        };
+
+                        Some((
+                            dbname.to_case(Case::Camel),
+                            format!("{post:#.2} ({sign}{diff:#.2})").into(),
+                        ))
+                    })
+                    .into_iter()
+                    .flatten()
+            })
+            .collect();
 
         if let Some(congestion) = congestion {
             tracing::debug!(

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

@@ -12,7 +12,7 @@ use roaring::RoaringBitmap;
 
 use super::create_batch::Batch;
 use crate::processing::{
-    AtomicBatchStep, AtomicTaskStep, CreateIndexProgress, DeleteIndexProgress,
+    AtomicBatchStep, AtomicTaskStep, CreateIndexProgress, DeleteIndexProgress, FinalizingIndexStep,
     InnerSwappingTwoIndexes, SwappingTheIndexes, TaskCancelationProgress, TaskDeletionProgress,
     UpdateIndexProgress,
 };
@@ -22,6 +22,16 @@ use crate::utils::{
 };
 use crate::{Error, IndexScheduler, Result, TaskId};
 
+#[derive(Debug, Default)]
+pub struct ProcessBatchInfo {
+    /// The write channel congestion. None when unavailable: settings update.
+    pub congestion: Option<ChannelCongestion>,
+    /// The sizes of the different databases before starting the indexation.
+    pub pre_commit_dabases_sizes: indexmap::IndexMap<&'static str, usize>,
+    /// The sizes of the different databases after commiting the indexation.
+    pub post_commit_dabases_sizes: indexmap::IndexMap<&'static str, usize>,
+}
+
 impl IndexScheduler {
     /// Apply the operation associated with the given batch.
     ///
@@ -35,7 +45,7 @@ impl IndexScheduler {
         batch: Batch,
         current_batch: &mut ProcessingBatch,
         progress: Progress,
-    ) -> Result<(Vec<Task>, Option<ChannelCongestion>)> {
+    ) -> Result<(Vec<Task>, ProcessBatchInfo)> {
         #[cfg(test)]
         {
             self.maybe_fail(crate::test_utils::FailureLocation::InsideProcessBatch)?;
@@ -76,7 +86,7 @@ impl IndexScheduler {
 
                 canceled_tasks.push(task);
 
-                Ok((canceled_tasks, None))
+                Ok((canceled_tasks, ProcessBatchInfo::default()))
             }
             Batch::TaskDeletions(mut tasks) => {
                 // 1. Retrieve the tasks that matched the query at enqueue-time.
@@ -115,14 +125,14 @@ impl IndexScheduler {
                         _ => unreachable!(),
                     }
                 }
-                Ok((tasks, None))
-            }
-            Batch::SnapshotCreation(tasks) => {
-                self.process_snapshot(progress, tasks).map(|tasks| (tasks, None))
-            }
-            Batch::Dump(task) => {
-                self.process_dump_creation(progress, task).map(|tasks| (tasks, None))
+                Ok((tasks, ProcessBatchInfo::default()))
             }
+            Batch::SnapshotCreation(tasks) => self
+                .process_snapshot(progress, tasks)
+                .map(|tasks| (tasks, ProcessBatchInfo::default())),
+            Batch::Dump(task) => self
+                .process_dump_creation(progress, task)
+                .map(|tasks| (tasks, ProcessBatchInfo::default())),
             Batch::IndexOperation { op, must_create_index } => {
                 let index_uid = op.index_uid().to_string();
                 let index = if must_create_index {
@@ -139,10 +149,12 @@ impl IndexScheduler {
                     .set_currently_updating_index(Some((index_uid.clone(), index.clone())));
 
                 let mut index_wtxn = index.write_txn()?;
+                let pre_commit_dabases_sizes = index.database_sizes(&index_wtxn)?;
                 let (tasks, congestion) =
-                    self.apply_index_operation(&mut index_wtxn, &index, op, progress)?;
+                    self.apply_index_operation(&mut index_wtxn, &index, op, &progress)?;
 
                 {
+                    progress.update_progress(FinalizingIndexStep::Committing);
                     let span = tracing::trace_span!(target: "indexing::scheduler", "commit");
                     let _entered = span.enter();
 
@@ -153,12 +165,15 @@ impl IndexScheduler {
                 // 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 mut post_commit_dabases_sizes = None;
                 let res = || -> Result<()> {
+                    progress.update_progress(FinalizingIndexStep::ComputingStats);
                     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())))?;
                     let mut wtxn = self.env.write_txn()?;
                     self.index_mapper.store_stats_of(&mut wtxn, &index_uid, &stats)?;
+                    post_commit_dabases_sizes = Some(index.database_sizes(&index_rtxn)?);
                     wtxn.commit()?;
                     Ok(())
                 }();
@@ -171,7 +186,16 @@ impl IndexScheduler {
                     ),
                 }
 
-                Ok((tasks, congestion))
+                let info = ProcessBatchInfo {
+                    congestion,
+                    // In case we fail to the get post-commit sizes we decide
+                    // that nothing changed and use the pre-commit sizes.
+                    post_commit_dabases_sizes: post_commit_dabases_sizes
+                        .unwrap_or_else(|| pre_commit_dabases_sizes.clone()),
+                    pre_commit_dabases_sizes,
+                };
+
+                Ok((tasks, info))
             }
             Batch::IndexCreation { index_uid, primary_key, task } => {
                 progress.update_progress(CreateIndexProgress::CreatingTheIndex);
@@ -239,7 +263,7 @@ impl IndexScheduler {
                     ),
                 }
 
-                Ok((vec![task], None))
+                Ok((vec![task], ProcessBatchInfo::default()))
             }
             Batch::IndexDeletion { index_uid, index_has_been_created, mut tasks } => {
                 progress.update_progress(DeleteIndexProgress::DeletingTheIndex);
@@ -273,7 +297,9 @@ impl IndexScheduler {
                     };
                 }
 
-                Ok((tasks, None))
+                // Here we could also show that all the internal database sizes goes to 0
+                // but it would mean opening the index and that's costly.
+                Ok((tasks, ProcessBatchInfo::default()))
             }
             Batch::IndexSwap { mut task } => {
                 progress.update_progress(SwappingTheIndexes::EnsuringCorrectnessOfTheSwap);
@@ -321,7 +347,7 @@ impl IndexScheduler {
                 }
                 wtxn.commit()?;
                 task.status = Status::Succeeded;
-                Ok((vec![task], None))
+                Ok((vec![task], ProcessBatchInfo::default()))
             }
             Batch::UpgradeDatabase { mut tasks } => {
                 let KindWithContent::UpgradeDatabase { from } = tasks.last().unwrap().kind else {
@@ -351,7 +377,7 @@ impl IndexScheduler {
                     task.error = None;
                 }
 
-                Ok((tasks, None))
+                Ok((tasks, ProcessBatchInfo::default()))
             }
         }
     }

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

@@ -32,7 +32,7 @@ impl IndexScheduler {
         index_wtxn: &mut RwTxn<'i>,
         index: &'i Index,
         operation: IndexOperation,
-        progress: Progress,
+        progress: &Progress,
     ) -> Result<(Vec<Task>, Option<ChannelCongestion>)> {
         let indexer_alloc = Bump::new();
         let started_processing_at = std::time::Instant::now();
@@ -186,7 +186,7 @@ impl IndexScheduler {
                             &document_changes,
                             embedders,
                             &|| must_stop_processing.get(),
-                            &progress,
+                            progress,
                         )
                         .map_err(|e| Error::from_milli(e, Some(index_uid.clone())))?,
                     );
@@ -307,7 +307,7 @@ impl IndexScheduler {
                             &document_changes,
                             embedders,
                             &|| must_stop_processing.get(),
-                            &progress,
+                            progress,
                         )
                         .map_err(|err| Error::from_milli(err, Some(index_uid.clone())))?,
                     );
@@ -465,7 +465,7 @@ impl IndexScheduler {
                             &document_changes,
                             embedders,
                             &|| must_stop_processing.get(),
-                            &progress,
+                            progress,
                         )
                         .map_err(|err| Error::from_milli(err, Some(index_uid.clone())))?,
                     );
@@ -520,7 +520,7 @@ impl IndexScheduler {
                         index_uid: index_uid.clone(),
                         tasks: cleared_tasks,
                     },
-                    progress.clone(),
+                    progress,
                 )?;
 
                 let (settings_tasks, _congestion) = self.apply_index_operation(