Merge #2523

2523: Improve the tasks error reporting when processed in batches r=irevoire a=Kerollmops

This fixes #2478 by changing the behavior of the task handler when there is an error in a batch of document addition or update.

What changes is that when there is a user error in a task in a batch we now report this task as failed with the right error message but we continue to process the other tasks. A user error can be when a geo field is invalid, a document id is invalid, or missing.

fixes #2582, #2478

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

Cargo.lock

@@ -644,9 +644,9 @@ dependencies = [
 
 [[package]]
 name = "charabia"
-version = "0.5.0"
+version = "0.5.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4a26a3df4d9c9231eb1e757fe6b1c66c471e0c2cd5410265e7c3109a726663c4"
+checksum = "2ed19edcd98f5bf6572f48d6f5982d595cb8718e47c6f0066d942b280575ff02"
 dependencies = [
  "character_converter",
  "cow-utils",
@@ -1123,8 +1123,8 @@ dependencies = [
 
 [[package]]
 name = "filter-parser"
-version = "0.31.2"
+version = "0.32.0"
-source = "git+https://github.com/meilisearch/milli.git?tag=v0.31.2#132558bf6a4e434de2a48314c4a208dea295a992"
+source = "git+https://github.com/meilisearch/milli.git?tag=v0.32.0#e1bc610d2722a8010216c45d5a32cbe3db18468e"
 dependencies = [
  "nom",
  "nom_locate",
@@ -1148,8 +1148,8 @@ dependencies = [
 
 [[package]]
 name = "flatten-serde-json"
-version = "0.31.2"
+version = "0.32.0"
-source = "git+https://github.com/meilisearch/milli.git?tag=v0.31.2#132558bf6a4e434de2a48314c4a208dea295a992"
+source = "git+https://github.com/meilisearch/milli.git?tag=v0.32.0#e1bc610d2722a8010216c45d5a32cbe3db18468e"
 dependencies = [
  "serde_json",
 ]
@@ -1661,8 +1661,8 @@ dependencies = [
 
 [[package]]
 name = "json-depth-checker"
-version = "0.31.2"
+version = "0.32.0"
-source = "git+https://github.com/meilisearch/milli.git?tag=v0.31.2#132558bf6a4e434de2a48314c4a208dea295a992"
+source = "git+https://github.com/meilisearch/milli.git?tag=v0.32.0#e1bc610d2722a8010216c45d5a32cbe3db18468e"
 dependencies = [
  "serde_json",
 ]
@@ -2013,7 +2013,7 @@ dependencies = [
  "sha2",
  "thiserror",
  "time 0.3.9",
- "uuid 1.1.2",
+ "uuid",
 ]
 
 [[package]]
@@ -2082,7 +2082,7 @@ dependencies = [
  "tokio",
  "tokio-stream",
  "urlencoding",
- "uuid 1.1.2",
+ "uuid",
  "vergen",
  "walkdir",
  "yaup",
@@ -2147,7 +2147,7 @@ dependencies = [
  "thiserror",
  "time 0.3.9",
  "tokio",
- "uuid 1.1.2",
+ "uuid",
  "walkdir",
  "whoami",
 ]
@@ -2189,8 +2189,8 @@ dependencies = [
 
 [[package]]
 name = "milli"
-version = "0.31.2"
+version = "0.32.0"
-source = "git+https://github.com/meilisearch/milli.git?tag=v0.31.2#132558bf6a4e434de2a48314c4a208dea295a992"
+source = "git+https://github.com/meilisearch/milli.git?tag=v0.32.0#e1bc610d2722a8010216c45d5a32cbe3db18468e"
 dependencies = [
  "bimap",
  "bincode",
@@ -2229,7 +2229,7 @@ dependencies = [
  "tempfile",
  "thiserror",
  "time 0.3.9",
- "uuid 0.8.2",
+ "uuid",
 ]
 
 [[package]]
@@ -3671,15 +3671,6 @@ version = "0.1.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5190c9442dcdaf0ddd50f37420417d219ae5261bbf5db120d0f9bab996c9cba1"
 
-[[package]]
-name = "uuid"
-version = "0.8.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bc5cf98d8186244414c848017f0e2676b3fcb46807f6668a97dfe67359a3c4b7"
-dependencies = [
- "getrandom",
-]
-
 [[package]]
 name = "uuid"
 version = "1.1.2"

meilisearch-auth/Cargo.toml

@@ -7,7 +7,7 @@ edition = "2021"
 enum-iterator = "0.7.0"
 hmac = "0.12.1"
 meilisearch-types = { path = "../meilisearch-types" }
-milli = { git = "https://github.com/meilisearch/milli.git", tag = "v0.31.2" }
+milli = { git = "https://github.com/meilisearch/milli.git", tag = "v0.32.0" }
 rand = "0.8.4"
 serde = { version = "1.0.136", features = ["derive"] }
 serde_json = { version = "1.0.79", features = ["preserve_order"] }

meilisearch-http/src/helpers/env.rs

@@ -1,17 +0,0 @@
-use meilisearch_lib::heed::Env;
-use walkdir::WalkDir;
-
-pub trait EnvSizer {
-    fn size(&self) -> u64;
-}
-
-impl EnvSizer for Env {
-    fn size(&self) -> u64 {
-        WalkDir::new(self.path())
-            .into_iter()
-            .filter_map(|entry| entry.ok())
-            .filter_map(|entry| entry.metadata().ok())
-            .filter(|metadata| metadata.is_file())
-            .fold(0, |acc, m| acc + m.len())
-    }
-}

meilisearch-http/src/helpers/mod.rs

@@ -1,3 +0,0 @@
-mod env;
-
-pub use env::EnvSizer;

meilisearch-http/src/lib.rs

@@ -5,7 +5,6 @@ pub mod analytics;
 pub mod task;
 #[macro_use]
 pub mod extractors;
-pub mod helpers;
 pub mod option;
 pub mod routes;
 
@@ -30,9 +29,9 @@ pub static AUTOBATCHING_ENABLED: AtomicBool = AtomicBool::new(false);
 pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<MeiliSearch> {
     let mut meilisearch = MeiliSearch::builder();
 
-    // enable autobatching?
+    // disable autobatching?
     AUTOBATCHING_ENABLED.store(
-        opt.scheduler_options.enable_auto_batching,
+        !opt.scheduler_options.disable_auto_batching,
         std::sync::atomic::Ordering::Relaxed,
     );
 

meilisearch-http/src/task.rs

@@ -231,7 +231,7 @@ pub struct TaskView {
     #[serde(serialize_with = "time::serde::rfc3339::option::serialize")]
     finished_at: Option<OffsetDateTime>,
     #[serde(skip_serializing_if = "Option::is_none")]
-    batch_uid: Option<Option<BatchId>>,
+    batch_uid: Option<BatchId>,
 }
 
 impl From<Task> for TaskView {
@@ -380,15 +380,15 @@ impl From<Task> for TaskView {
 
         let duration = finished_at.zip(started_at).map(|(tf, ts)| (tf - ts));
 
-        let batch_uid = if AUTOBATCHING_ENABLED.load(std::sync::atomic::Ordering::Relaxed) {
+        let batch_uid = AUTOBATCHING_ENABLED
-            let id = events.iter().find_map(|e| match e {
+            .load(std::sync::atomic::Ordering::Relaxed)
-                TaskEvent::Batched { batch_id, .. } => Some(*batch_id),
+            .then(|| {
-                _ => None,
+                events.iter().find_map(|e| match e {
-            });
+                    TaskEvent::Batched { batch_id, .. } => Some(*batch_id),
-            Some(id)
+                    _ => None,
-        } else {
+                })
-            None
+            })
-        };
+            .flatten();
 
         Self {
             uid: id,

meilisearch-http/tests/documents/add_documents.rs

@@ -1,5 +1,6 @@
 use crate::common::{GetAllDocumentsOptions, Server};
 use actix_web::test;
+
 use meilisearch_http::{analytics, create_app};
 use serde_json::{json, Value};
 use time::{format_description::well_known::Rfc3339, OffsetDateTime};
@@ -326,7 +327,7 @@ async fn error_add_malformed_json_documents() {
     assert_eq!(
         response["message"],
         json!(
-            r#"The `json` payload provided is malformed. `Couldn't serialize document value: invalid type: string "0123456789012345678901234567...890123456789", expected a documents, or a sequence of documents. at line 1 column 102`."#
+            r#"The `json` payload provided is malformed. `Couldn't serialize document value: invalid type: string "0123456789012345678901234567...890123456789012345678901234567890123456789", expected a sequence at line 1 column 102`."#
         )
     );
     assert_eq!(response["code"], json!("malformed_payload"));
@@ -349,9 +350,7 @@ async fn error_add_malformed_json_documents() {
     assert_eq!(status_code, 400);
     assert_eq!(
         response["message"],
-        json!(
+        json!("The `json` payload provided is malformed. `Couldn't serialize document value: invalid type: string \"0123456789012345678901234567...90123456789012345678901234567890123456789m\", expected a sequence at line 1 column 103`.")
-            r#"The `json` payload provided is malformed. `Couldn't serialize document value: invalid type: string "0123456789012345678901234567...90123456789m", expected a documents, or a sequence of documents. at line 1 column 103`."#
-        )
     );
     assert_eq!(response["code"], json!("malformed_payload"));
     assert_eq!(response["type"], json!("invalid_request"));
@@ -388,7 +387,7 @@ async fn error_add_malformed_ndjson_documents() {
     assert_eq!(
         response["message"],
         json!(
-            r#"The `ndjson` payload provided is malformed. `Couldn't serialize document value: key must be a string at line 1 column 2`."#
+            r#"The `ndjson` payload provided is malformed. `Couldn't serialize document value: key must be a string at line 2 column 2`."#
         )
     );
     assert_eq!(response["code"], json!("malformed_payload"));
@@ -411,9 +410,7 @@ async fn error_add_malformed_ndjson_documents() {
     assert_eq!(status_code, 400);
     assert_eq!(
         response["message"],
-        json!(
+        json!("The `ndjson` payload provided is malformed. `Couldn't serialize document value: key must be a string at line 2 column 2`.")
-            r#"The `ndjson` payload provided is malformed. `Couldn't serialize document value: key must be a string at line 1 column 2`."#
-        )
     );
     assert_eq!(response["code"], json!("malformed_payload"));
     assert_eq!(response["type"], json!("invalid_request"));
@@ -1020,7 +1017,7 @@ async fn add_documents_invalid_geo_field() {
     index.wait_task(2).await;
     let (response, code) = index.get_task(2).await;
     assert_eq!(code, 200);
-    assert_eq!(response["status"], "succeeded");
+    assert_eq!(response["status"], "failed");
 }
 
 #[actix_rt::test]
@@ -1099,3 +1096,62 @@ async fn add_documents_with_primary_key_twice() {
     let (response, _code) = index.get_task(1).await;
     assert_eq!(response["status"], "succeeded");
 }
+
+#[actix_rt::test]
+async fn batch_several_documents_addition() {
+    let server = Server::new().await;
+    let index = server.index("test");
+
+    let mut documents: Vec<_> = (0..150usize)
+        .into_iter()
+        .map(|id| {
+            json!(
+                {
+                    "id": id,
+                    "title": "foo",
+                    "desc": "bar"
+                }
+            )
+        })
+        .collect();
+
+    documents[100] = json!({"title": "error", "desc": "error"});
+
+    // enqueue batch of documents
+    let mut waiter = Vec::new();
+    for chunk in documents.chunks(30) {
+        waiter.push(index.add_documents(json!(chunk), Some("id")));
+    }
+
+    // wait first batch of documents to finish
+    futures::future::join_all(waiter).await;
+    index.wait_task(4).await;
+
+    // run a second completely failing batch
+    documents[40] = json!({"title": "error", "desc": "error"});
+    documents[70] = json!({"title": "error", "desc": "error"});
+    documents[130] = json!({"title": "error", "desc": "error"});
+    let mut waiter = Vec::new();
+    for chunk in documents.chunks(30) {
+        waiter.push(index.add_documents(json!(chunk), Some("id")));
+    }
+    // wait second batch of documents to finish
+    futures::future::join_all(waiter).await;
+    index.wait_task(9).await;
+
+    let (response, _code) = index.filtered_tasks(&[], &["failed"]).await;
+
+    // Check if only the 6th task failed
+    println!("{}", &response);
+    assert_eq!(response["results"].as_array().unwrap().len(), 5);
+
+    // Check if there are exactly 120 documents (150 - 30) in the index;
+    let (response, code) = index
+        .get_all_documents(GetAllDocumentsOptions {
+            limit: Some(200),
+            ..Default::default()
+        })
+        .await;
+    assert_eq!(code, 200, "failed with `{}`", response);
+    assert_eq!(response["results"].as_array().unwrap().len(), 120);
+}

meilisearch-http/tests/search/mod.rs

@@ -708,9 +708,7 @@ async fn faceting_max_values_per_facet() {
             }),
             |response, code| {
                 assert_eq!(code, 200, "{}", response);
-                let numbers = dbg!(&response)["facetDistribution"]["number"]
+                let numbers = &response["facetDistribution"]["number"].as_object().unwrap();
-                    .as_object()
-                    .unwrap();
                 assert_eq!(numbers.len(), 10_000);
             },
         )

meilisearch-lib/Cargo.toml

@@ -28,7 +28,7 @@ lazy_static = "1.4.0"
 log = "0.4.14"
 meilisearch-auth = { path = "../meilisearch-auth" }
 meilisearch-types = { path = "../meilisearch-types" }
-milli = { git = "https://github.com/meilisearch/milli.git", tag = "v0.31.2" }
+milli = { git = "https://github.com/meilisearch/milli.git", tag = "v0.32.0" }
 mime = "0.3.16"
 num_cpus = "1.13.1"
 obkv = "0.2.0"

meilisearch-lib/src/document_formats.rs

@@ -1,10 +1,10 @@
 use std::borrow::Borrow;
 use std::fmt::{self, Debug, Display};
-use std::io::{self, BufRead, BufReader, BufWriter, Cursor, Read, Seek, Write};
+use std::io::{self, BufReader, Read, Seek, Write};
 
 use meilisearch_types::error::{Code, ErrorCode};
 use meilisearch_types::internal_error;
-use milli::documents::DocumentBatchBuilder;
+use milli::documents::{DocumentsBatchBuilder, Error};
 
 type Result<T> = std::result::Result<T, DocumentFormatError>;
 
@@ -18,9 +18,9 @@ pub enum PayloadType {
 impl fmt::Display for PayloadType {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         match self {
-            PayloadType::Ndjson => write!(f, "ndjson"),
+            PayloadType::Ndjson => f.write_str("ndjson"),
-            PayloadType::Json => write!(f, "json"),
+            PayloadType::Json => f.write_str("json"),
-            PayloadType::Csv => write!(f, "csv"),
+            PayloadType::Csv => f.write_str("csv"),
         }
     }
 }
@@ -28,7 +28,7 @@ impl fmt::Display for PayloadType {
 #[derive(Debug)]
 pub enum DocumentFormatError {
     Internal(Box<dyn std::error::Error + Send + Sync + 'static>),
-    MalformedPayload(Box<milli::documents::Error>, PayloadType),
+    MalformedPayload(Error, PayloadType),
 }
 
 impl Display for DocumentFormatError {
@@ -36,7 +36,7 @@ impl Display for DocumentFormatError {
         match self {
             Self::Internal(e) => write!(f, "An internal error has occurred: `{}`.", e),
             Self::MalformedPayload(me, b) => match me.borrow() {
-                milli::documents::Error::JsonError(se) => {
+                Error::Json(se) => {
                     // https://github.com/meilisearch/meilisearch/issues/2107
                     // The user input maybe insanely long. We need to truncate it.
                     let mut serde_msg = se.to_string();
@@ -59,11 +59,11 @@ impl Display for DocumentFormatError {
 
 impl std::error::Error for DocumentFormatError {}
 
-impl From<(PayloadType, milli::documents::Error)> for DocumentFormatError {
+impl From<(PayloadType, Error)> for DocumentFormatError {
-    fn from((ty, error): (PayloadType, milli::documents::Error)) -> Self {
+    fn from((ty, error): (PayloadType, Error)) -> Self {
         match error {
-            milli::documents::Error::Io(e) => Self::Internal(Box::new(e)),
+            Error::Io(e) => Self::Internal(Box::new(e)),
-            e => Self::MalformedPayload(Box::new(e), ty),
+            e => Self::MalformedPayload(e, ty),
         }
     }
 }
@@ -79,51 +79,67 @@ impl ErrorCode for DocumentFormatError {
 
 internal_error!(DocumentFormatError: io::Error);
 
-/// reads csv from input and write an obkv batch to writer.
+/// Reads CSV from input and write an obkv batch to writer.
 pub fn read_csv(input: impl Read, writer: impl Write + Seek) -> Result<usize> {
-    let writer = BufWriter::new(writer);
+    let mut builder = DocumentsBatchBuilder::new(writer);
-    let builder =
-        DocumentBatchBuilder::from_csv(input, writer).map_err(|e| (PayloadType::Csv, e))?;
 
-    let count = builder.finish().map_err(|e| (PayloadType::Csv, e))?;
+    let csv = csv::Reader::from_reader(input);
+    builder.append_csv(csv).map_err(|e| (PayloadType::Csv, e))?;
 
-    Ok(count)
+    let count = builder.documents_count();
+    let _ = builder
+        .into_inner()
+        .map_err(Into::into)
+        .map_err(DocumentFormatError::Internal)?;
+
+    Ok(count as usize)
 }
 
-/// reads jsonl from input and write an obkv batch to writer.
+/// Reads JSON Lines from input and write an obkv batch to writer.
 pub fn read_ndjson(input: impl Read, writer: impl Write + Seek) -> Result<usize> {
-    let mut reader = BufReader::new(input);
+    let mut builder = DocumentsBatchBuilder::new(writer);
-    let writer = BufWriter::new(writer);
+    let reader = BufReader::new(input);
 
-    let mut builder = DocumentBatchBuilder::new(writer).map_err(|e| (PayloadType::Ndjson, e))?;
+    for result in serde_json::Deserializer::from_reader(reader).into_iter() {
-    let mut buf = String::new();
+        let object = result
-
+            .map_err(Error::Json)
-    while reader.read_line(&mut buf)? > 0 {
-        // skip empty lines
-        if buf == "\n" {
-            buf.clear();
-            continue;
-        }
-        builder
-            .extend_from_json(Cursor::new(&buf.as_bytes()))
             .map_err(|e| (PayloadType::Ndjson, e))?;
-        buf.clear();
+        builder
+            .append_json_object(&object)
+            .map_err(Into::into)
+            .map_err(DocumentFormatError::Internal)?;
     }
 
-    let count = builder.finish().map_err(|e| (PayloadType::Ndjson, e))?;
+    let count = builder.documents_count();
+    let _ = builder
+        .into_inner()
+        .map_err(Into::into)
+        .map_err(DocumentFormatError::Internal)?;
 
-    Ok(count)
+    Ok(count as usize)
 }
 
-/// reads json from input and write an obkv batch to writer.
+/// Reads JSON from input and write an obkv batch to writer.
 pub fn read_json(input: impl Read, writer: impl Write + Seek) -> Result<usize> {
-    let writer = BufWriter::new(writer);
+    let mut builder = DocumentsBatchBuilder::new(writer);
-    let mut builder = DocumentBatchBuilder::new(writer).map_err(|e| (PayloadType::Json, e))?;
+    let reader = BufReader::new(input);
-    builder
+
-        .extend_from_json(input)
+    let objects: Vec<_> = serde_json::from_reader(reader)
+        .map_err(Error::Json)
         .map_err(|e| (PayloadType::Json, e))?;
 
-    let count = builder.finish().map_err(|e| (PayloadType::Json, e))?;
+    for object in objects {
+        builder
+            .append_json_object(&object)
+            .map_err(Into::into)
+            .map_err(DocumentFormatError::Internal)?;
+    }
 
-    Ok(count)
+    let count = builder.documents_count();
+    let _ = builder
+        .into_inner()
+        .map_err(Into::into)
+        .map_err(DocumentFormatError::Internal)?;
+
+    Ok(count as usize)
 }

meilisearch-lib/src/dump/error.rs

@@ -11,7 +11,7 @@ pub enum DumpError {
     #[error("An internal error has occurred. `{0}`.")]
     Internal(Box<dyn std::error::Error + Send + Sync + 'static>),
     #[error("{0}")]
-    IndexResolver(#[from] IndexResolverError),
+    IndexResolver(Box<IndexResolverError>),
 }
 
 internal_error!(
@@ -26,6 +26,12 @@ internal_error!(
     TaskError
 );
 
+impl From<IndexResolverError> for DumpError {
+    fn from(e: IndexResolverError) -> Self {
+        Self::IndexResolver(Box::new(e))
+    }
+}
+
 impl ErrorCode for DumpError {
     fn error_code(&self) -> Code {
         match self {

meilisearch-lib/src/error.rs

@@ -25,6 +25,7 @@ impl ErrorCode for MilliError<'_> {
                     // TODO: wait for spec for new error codes.
                     UserError::SerdeJson(_)
                     | UserError::DocumentLimitReached
+                    | UserError::AccessingSoftDeletedDocument { .. }
                     | UserError::UnknownInternalDocumentId { .. } => Code::Internal,
                     UserError::InvalidStoreFile => Code::InvalidStore,
                     UserError::NoSpaceLeftOnDevice => Code::NoSpaceLeftOnDevice,
@@ -32,7 +33,9 @@ impl ErrorCode for MilliError<'_> {
                     UserError::AttributeLimitReached => Code::MaxFieldsLimitExceeded,
                     UserError::InvalidFilter(_) => Code::Filter,
                     UserError::MissingDocumentId { .. } => Code::MissingDocumentId,
-                    UserError::InvalidDocumentId { .. } => Code::InvalidDocumentId,
+                    UserError::InvalidDocumentId { .. } | UserError::TooManyDocumentIds { .. } => {
+                        Code::InvalidDocumentId
+                    }
                     UserError::MissingPrimaryKey => Code::MissingPrimaryKey,
                     UserError::PrimaryKeyCannotBeChanged(_) => Code::PrimaryKeyAlreadyPresent,
                     UserError::SortRankingRuleMissing => Code::Sort,

meilisearch-lib/src/index/dump.rs

@@ -4,7 +4,7 @@ use std::path::Path;
 
 use anyhow::Context;
 use indexmap::IndexMap;
-use milli::documents::DocumentBatchReader;
+use milli::documents::DocumentsBatchReader;
 use milli::heed::{EnvOpenOptions, RoTxn};
 use milli::update::{IndexDocumentsConfig, IndexerConfig};
 use serde::{Deserialize, Serialize};
@@ -135,19 +135,20 @@ impl Index {
         if !empty {
             tmp_doc_file.seek(SeekFrom::Start(0))?;
 
-            let documents_reader = DocumentBatchReader::from_reader(tmp_doc_file)?;
+            let documents_reader = DocumentsBatchReader::from_reader(tmp_doc_file)?;
 
             //If the document file is empty, we don't perform the document addition, to prevent
             //a primary key error to be thrown.
             let config = IndexDocumentsConfig::default();
-            let mut builder = milli::update::IndexDocuments::new(
+            let builder = milli::update::IndexDocuments::new(
                 &mut txn,
                 &index,
                 indexer_config,
                 config,
                 |_| (),
             )?;
-            builder.add_documents(documents_reader)?;
+            let (builder, user_error) = builder.add_documents(documents_reader)?;
+            user_error?;
             builder.execute()?;
         }
 

meilisearch-lib/src/index/error.rs

@@ -40,6 +40,12 @@ impl ErrorCode for IndexError {
     }
 }
 
+impl From<milli::UserError> for IndexError {
+    fn from(error: milli::UserError) -> IndexError {
+        IndexError::Milli(error.into())
+    }
+}
+
 #[derive(Debug, thiserror::Error)]
 pub enum FacetError {
     #[error("Invalid syntax for the filter parameter: `expected {}, found: {1}`.", .0.join(", "))]

meilisearch-lib/src/index/index.rs

@@ -4,7 +4,6 @@ use std::marker::PhantomData;
 use std::ops::Deref;
 use std::path::Path;
 use std::sync::Arc;
-use walkdir::WalkDir;
 
 use fst::IntoStreamer;
 use milli::heed::{CompactionOption, EnvOpenOptions, RoTxn};
@@ -14,6 +13,7 @@ use serde::{Deserialize, Serialize};
 use serde_json::{Map, Value};
 use time::OffsetDateTime;
 use uuid::Uuid;
+use walkdir::WalkDir;
 
 use crate::index::search::DEFAULT_PAGINATION_MAX_TOTAL_HITS;
 
@@ -245,11 +245,8 @@ impl Index {
         let fields_ids_map = self.fields_ids_map(&txn)?;
         let all_fields: Vec<_> = fields_ids_map.iter().map(|(id, _)| id).collect();
 
-        let iter = self.all_documents(&txn)?.skip(offset).take(limit);
-
         let mut documents = Vec::new();
-
+        for entry in self.all_documents(&txn)?.skip(offset).take(limit) {
-        for entry in iter {
             let (_id, obkv) = entry?;
             let document = obkv_to_json(&all_fields, &fields_ids_map, obkv)?;
             let document = match &attributes_to_retrieve {
@@ -302,7 +299,7 @@ impl Index {
     }
 
     pub fn size(&self) -> u64 {
-        WalkDir::new(self.inner.path())
+        WalkDir::new(self.path())
             .into_iter()
             .filter_map(|entry| entry.ok())
             .filter_map(|entry| entry.metadata().ok())

meilisearch-lib/src/index/mod.rs

@@ -24,12 +24,12 @@ pub use test::MockIndex as Index;
 /// code for unit testing, in places where an index would normally be used.
 #[cfg(test)]
 pub mod test {
-    use std::path::Path;
+    use std::path::{Path, PathBuf};
-    use std::path::PathBuf;
     use std::sync::Arc;
 
-    use milli::update::IndexerConfig;
+    use milli::update::{
-    use milli::update::{DocumentAdditionResult, DocumentDeletionResult, IndexDocumentsMethod};
+        DocumentAdditionResult, DocumentDeletionResult, IndexDocumentsMethod, IndexerConfig,
+    };
     use nelson::Mocker;
     use uuid::Uuid;
 
@@ -162,7 +162,7 @@ pub mod test {
             primary_key: Option<String>,
             file_store: UpdateFileStore,
             contents: impl Iterator<Item = Uuid>,
-        ) -> Result<DocumentAdditionResult> {
+        ) -> Result<Vec<Result<DocumentAdditionResult>>> {
             match self {
                 MockIndex::Real(index) => {
                     index.update_documents(method, primary_key, file_store, contents)

meilisearch-lib/src/index/updates.rs

@@ -3,7 +3,7 @@ use std::marker::PhantomData;
 use std::num::NonZeroUsize;
 
 use log::{debug, info, trace};
-use milli::documents::DocumentBatchReader;
+use milli::documents::DocumentsBatchReader;
 use milli::update::{
     DocumentAdditionResult, DocumentDeletionResult, IndexDocumentsConfig, IndexDocumentsMethod,
     Setting,
@@ -11,7 +11,7 @@ use milli::update::{
 use serde::{Deserialize, Serialize, Serializer};
 use uuid::Uuid;
 
-use super::error::Result;
+use super::error::{IndexError, Result};
 use super::index::{Index, IndexMeta};
 use crate::update_file_store::UpdateFileStore;
 
@@ -299,7 +299,7 @@ impl Index {
         primary_key: Option<String>,
         file_store: UpdateFileStore,
         contents: impl IntoIterator<Item = Uuid>,
-    ) -> Result<DocumentAdditionResult> {
+    ) -> Result<Vec<Result<DocumentAdditionResult>>> {
         trace!("performing document addition");
         let mut txn = self.write_txn()?;
 
@@ -323,19 +323,34 @@ impl Index {
             indexing_callback,
         )?;
 
+        let mut results = Vec::new();
         for content_uuid in contents.into_iter() {
             let content_file = file_store.get_update(content_uuid)?;
-            let reader = DocumentBatchReader::from_reader(content_file)?;
+            let reader = DocumentsBatchReader::from_reader(content_file)?;
-            builder.add_documents(reader)?;
+            let (new_builder, user_result) = builder.add_documents(reader)?;
+            builder = new_builder;
+
+            let user_result = match user_result {
+                Ok(count) => {
+                    let addition = DocumentAdditionResult {
+                        indexed_documents: count,
+                        number_of_documents: count,
+                    };
+                    info!("document addition done: {:?}", addition);
+                    Ok(addition)
+                }
+                Err(e) => Err(IndexError::from(e)),
+            };
+
+            results.push(user_result);
         }
 
-        let addition = builder.execute()?;
+        if results.iter().any(Result::is_ok) {
+            let _addition = builder.execute()?;
+            txn.commit()?;
+        }
 
-        txn.commit()?;
+        Ok(results)
-
-        info!("document addition done: {:?}", addition);
-
-        Ok(addition)
     }
 
     pub fn update_settings(&self, settings: &Settings<Checked>) -> Result<()> {

meilisearch-lib/src/index_resolver/mod.rs

@@ -150,25 +150,34 @@ mod real {
                     })
                     .await;
 
-                    let event = match result {
+                    match result {
-                        Ok(Ok(result)) => TaskEvent::Succeeded {
+                        Ok(Ok(results)) => {
-                            timestamp: OffsetDateTime::now_utc(),
+                            for (task, result) in tasks.iter_mut().zip(results) {
-                            result: TaskResult::DocumentAddition {
+                                let event = match result {
-                                indexed_documents: result.indexed_documents,
+                                    Ok(addition) => {
-                            },
+                                        TaskEvent::succeeded(TaskResult::DocumentAddition {
-                        },
+                                            indexed_documents: addition.indexed_documents,
-                        Ok(Err(e)) => TaskEvent::Failed {
+                                        })
-                            timestamp: OffsetDateTime::now_utc(),
+                                    }
-                            error: e.into(),
+                                    Err(error) => {
-                        },
+                                        TaskEvent::failed(IndexResolverError::from(error))
-                        Err(e) => TaskEvent::Failed {
+                                    }
-                            timestamp: OffsetDateTime::now_utc(),
+                                };
-                            error: IndexResolverError::from(e).into(),
+                                task.events.push(event);
-                        },
+                            }
-                    };
+                        }
-
+                        Ok(Err(e)) => {
-                    for task in tasks.iter_mut() {
+                            let event = TaskEvent::failed(e);
-                        task.events.push(event.clone());
+                            for task in tasks.iter_mut() {
+                                task.events.push(event.clone());
+                            }
+                        }
+                        Err(e) => {
+                            let event = TaskEvent::failed(IndexResolverError::from(e));
+                            for task in tasks.iter_mut() {
+                                task.events.push(event.clone());
+                            }
+                        }
                     }
                 }
                 _ => panic!("invalid batch!"),

meilisearch-lib/src/options.rs

@@ -41,27 +41,10 @@ pub struct IndexerOpts {
 
 #[derive(Debug, Clone, Parser, Default, Serialize)]
 pub struct SchedulerConfig {
-    /// enable the autobatching experimental feature
+    /// The engine will disable task auto-batching,
-    #[clap(long, hide = true)]
+    /// and will sequencialy compute each task one by one.
-    pub enable_auto_batching: bool,
+    #[clap(long, env = "DISABLE_AUTO_BATCHING")]
-
+    pub disable_auto_batching: bool,
-    // The maximum number of updates of the same type that can be batched together.
-    // If unspecified, this is unlimited. A value of 0 is interpreted as 1.
-    #[clap(long, requires = "enable-auto-batching", hide = true)]
-    pub max_batch_size: Option<usize>,
-
-    // The maximum number of documents in a document batch. Since batches must contain at least one
-    // update for the scheduler to make progress, the number of documents in a batch will be at
-    // least the number of documents of its first update.
-    #[clap(long, requires = "enable-auto-batching", hide = true)]
-    pub max_documents_per_batch: Option<usize>,
-
-    /// Debounce duration in seconds
-    ///
-    /// When a new task is enqueued, the scheduler waits for `debounce_duration_sec` seconds for new updates before
-    /// starting to process a batch of updates.
-    #[clap(long, requires = "enable-auto-batching", hide = true)]
-    pub debounce_duration_sec: Option<u64>,
 }
 
 impl TryFrom<&IndexerOpts> for IndexerConfig {

meilisearch-lib/src/tasks/scheduler.rs

@@ -3,7 +3,6 @@ use std::collections::{hash_map::Entry, BinaryHeap, HashMap, VecDeque};
 use std::ops::{Deref, DerefMut};
 use std::slice;
 use std::sync::Arc;
-use std::time::Duration;
 
 use atomic_refcell::AtomicRefCell;
 use milli::update::IndexDocumentsMethod;
@@ -248,17 +247,10 @@ impl Scheduler {
     pub fn new(
         store: TaskStore,
         performers: Vec<Arc<dyn BatchHandler + Sync + Send + 'static>>,
-        mut config: SchedulerConfig,
+        config: SchedulerConfig,
     ) -> Result<Arc<RwLock<Self>>> {
         let (notifier, rcv) = watch::channel(());
 
-        let debounce_time = config.debounce_duration_sec;
-
-        // Disable autobatching
-        if !config.enable_auto_batching {
-            config.max_batch_size = Some(1);
-        }
-
         let this = Self {
             snapshots: VecDeque::new(),
             tasks: TaskQueue::default(),
@@ -275,12 +267,7 @@ impl Scheduler {
 
         let this = Arc::new(RwLock::new(this));
 
-        let update_loop = UpdateLoop::new(
+        let update_loop = UpdateLoop::new(this.clone(), performers, rcv);
-            this.clone(),
-            performers,
-            debounce_time.filter(|&v| v > 0).map(Duration::from_secs),
-            rcv,
-        );
 
         tokio::task::spawn_local(update_loop.run());
 
@@ -497,27 +484,17 @@ fn make_batch(tasks: &mut TaskQueue, config: &SchedulerConfig) -> Processing {
                     match list.peek() {
                         Some(pending) if pending.kind == kind => {
                             // We always need to process at least one task for the scheduler to make progress.
-                            if task_list.len() >= config.max_batch_size.unwrap_or(usize::MAX).max(1)
+                            if config.disable_auto_batching && !task_list.is_empty() {
-                            {
                                 break;
                             }
                             let pending = list.pop().unwrap();
                             task_list.push(pending.id);
 
-                            // We add the number of documents to the count if we are scheduling document additions and
+                            // We add the number of documents to the count if we are scheduling document additions.
-                            // stop adding if we already have enough.
-                            //
-                            // We check that bound only after adding the current task to the batch, so that a batch contains at least one task.
                             match pending.kind {
                                 TaskType::DocumentUpdate { number }
                                 | TaskType::DocumentAddition { number } => {
                                     doc_count += number;
-
-                                    if doc_count
-                                        >= config.max_documents_per_batch.unwrap_or(usize::MAX)
-                                    {
-                                        break;
-                                    }
                                 }
                                 _ => (),
                             }

meilisearch-lib/src/tasks/update_loop.rs

@@ -1,9 +1,7 @@
 use std::sync::Arc;
-use std::time::Duration;
 
 use time::OffsetDateTime;
 use tokio::sync::{watch, RwLock};
-use tokio::time::interval_at;
 
 use super::batch::Batch;
 use super::error::Result;
@@ -17,20 +15,17 @@ pub struct UpdateLoop {
     performers: Vec<Arc<dyn BatchHandler + Send + Sync + 'static>>,
 
     notifier: Option<watch::Receiver<()>>,
-    debounce_duration: Option<Duration>,
 }
 
 impl UpdateLoop {
     pub fn new(
         scheduler: Arc<RwLock<Scheduler>>,
         performers: Vec<Arc<dyn BatchHandler + Send + Sync + 'static>>,
-        debuf_duration: Option<Duration>,
         notifier: watch::Receiver<()>,
     ) -> Self {
         Self {
             scheduler,
             performers,
-            debounce_duration: debuf_duration,
             notifier: Some(notifier),
         }
     }
@@ -43,11 +38,6 @@ impl UpdateLoop {
                 break;
             }
 
-            if let Some(t) = self.debounce_duration {
-                let mut interval = interval_at(tokio::time::Instant::now() + t, t);
-                interval.tick().await;
-            };
-
             if let Err(e) = self.process_next_batch().await {
                 log::error!("an error occurred while processing an update batch: {}", e);
             }

meilisearch-lib/src/update_file_store.rs

@@ -3,7 +3,7 @@ use std::io::{self, BufReader, BufWriter, Write};
 use std::ops::{Deref, DerefMut};
 use std::path::{Path, PathBuf};
 
-use milli::documents::DocumentBatchReader;
+use milli::documents::DocumentsBatchReader;
 use serde_json::Map;
 use tempfile::{NamedTempFile, PersistError};
 use uuid::Uuid;
@@ -44,7 +44,8 @@ into_update_store_error!(
     PersistError,
     io::Error,
     serde_json::Error,
-    milli::documents::Error
+    milli::documents::Error,
+    milli::documents::DocumentsBatchCursorError
 );
 
 impl UpdateFile {
@@ -149,12 +150,13 @@ mod store {
 
             let update_file = File::open(update_file_path)?;
             let mut dst_file = NamedTempFile::new_in(&dump_path)?;
-            let mut document_reader = DocumentBatchReader::from_reader(update_file)?;
+            let (mut document_cursor, index) =
+                DocumentsBatchReader::from_reader(update_file)?.into_cursor_and_fields_index();
 
             let mut document_buffer = Map::new();
             // TODO: we need to find a way to do this more efficiently. (create a custom serializer
             // for jsonl for example...)
-            while let Some((index, document)) = document_reader.next_document_with_index()? {
+            while let Some(document) = document_cursor.next_document()? {
                 for (field_id, content) in document.iter() {
                     if let Some(field_name) = index.name(field_id) {
                         let content = serde_json::from_slice(content)?;

permissive-json-pointer/src/lib.rs

@@ -49,7 +49,7 @@ fn contained_in(selector: &str, key: &str) -> bool {
 /// map_leaf_values(
 ///     value.as_object_mut().unwrap(),
 ///     ["jean.race.name"],
-///     |key, value| match (value, dbg!(key)) {
+///     |key, value| match (value, key) {
 ///         (Value::String(name), "jean.race.name") => *name = "patou".to_string(),
 ///         _ => unreachable!(),
 ///     },
@@ -729,7 +729,7 @@ mod tests {
         map_leaf_values(
             value.as_object_mut().unwrap(),
             ["jean.race.name"],
-            |key, value| match (value, dbg!(key)) {
+            |key, value| match (value, key) {
                 (Value::String(name), "jean.race.name") => *name = S("patou"),
                 _ => unreachable!(),
             },