Merge #364

364: Fix all the benchmarks  r=Kerollmops a=irevoire

#324 broke all benchmarks.
I fixed everything and noticed that `cargo check --all` was insufficient to check the bench in multiple workspaces, so I also updated the CI to use `cargo check --workspace --all-targets`.

Co-authored-by: Tamo <tamo@meilisearch.com>

.github/workflows/rust.yml

@@ -33,7 +33,7 @@ jobs:
       uses: actions-rs/cargo@v1
       with:
         command: check
-        args: --all
+        args: --workspace --all-targets
     - name: Run cargo test
       uses: actions-rs/cargo@v1
       with:

benchmarks/Cargo.toml

@@ -6,6 +6,9 @@ publish = false
 
 [dependencies]
 milli = { path = "../milli" }
+anyhow = "1.0"
+serde_json = { version = "1.0.62", features = ["preserve_order"] }
+csv = "1.1.6"
 
 [target.'cfg(target_os = "linux")'.dependencies]
 jemallocator = "0.3.2"

benchmarks/benches/indexing.rs

@@ -1,11 +1,12 @@
 mod datasets_paths;
+mod utils;
 
-use std::fs::{create_dir_all, remove_dir_all, File};
+use std::fs::{create_dir_all, remove_dir_all};
 use std::path::Path;
 
 use criterion::{criterion_group, criterion_main, Criterion};
 use heed::EnvOpenOptions;
-use milli::update::{IndexDocumentsMethod, UpdateBuilder, UpdateFormat};
+use milli::update::UpdateBuilder;
 use milli::Index;
 
 #[cfg(target_os = "linux")]
@@ -67,15 +68,10 @@ fn indexing_songs_default(c: &mut Criterion) {
             move |index| {
                 let update_builder = UpdateBuilder::new(0);
                 let mut wtxn = index.write_txn().unwrap();
-                let mut builder = update_builder.index_documents(&mut wtxn, &index);
+                let builder = update_builder.index_documents(&mut wtxn, &index);
 
-                builder.update_format(UpdateFormat::Csv);
-                builder.index_documents_method(IndexDocumentsMethod::ReplaceDocuments);
-                let reader = File::open(datasets_paths::SMOL_SONGS).expect(&format!(
-                    "could not find the dataset in: {}",
-                    datasets_paths::SMOL_SONGS
-                ));
-                builder.execute(reader, |_, _| ()).unwrap();
+                let documents = utils::documents_from(datasets_paths::SMOL_SONGS, "csv");
+                builder.execute(documents, |_, _| ()).unwrap();
                 wtxn.commit().unwrap();
 
                 index.prepare_for_closing().wait();
@@ -118,15 +114,10 @@ fn indexing_songs_without_faceted_numbers(c: &mut Criterion) {
             move |index| {
                 let update_builder = UpdateBuilder::new(0);
                 let mut wtxn = index.write_txn().unwrap();
-                let mut builder = update_builder.index_documents(&mut wtxn, &index);
+                let builder = update_builder.index_documents(&mut wtxn, &index);
 
-                builder.update_format(UpdateFormat::Csv);
-                builder.index_documents_method(IndexDocumentsMethod::ReplaceDocuments);
-                let reader = File::open(datasets_paths::SMOL_SONGS).expect(&format!(
-                    "could not find the dataset in: {}",
-                    datasets_paths::SMOL_SONGS
-                ));
-                builder.execute(reader, |_, _| ()).unwrap();
+                let documents = utils::documents_from(datasets_paths::SMOL_SONGS, "csv");
+                builder.execute(documents, |_, _| ()).unwrap();
                 wtxn.commit().unwrap();
 
                 index.prepare_for_closing().wait();
@@ -165,15 +156,10 @@ fn indexing_songs_without_faceted_fields(c: &mut Criterion) {
             move |index| {
                 let update_builder = UpdateBuilder::new(0);
                 let mut wtxn = index.write_txn().unwrap();
-                let mut builder = update_builder.index_documents(&mut wtxn, &index);
+                let builder = update_builder.index_documents(&mut wtxn, &index);
 
-                builder.update_format(UpdateFormat::Csv);
-                builder.index_documents_method(IndexDocumentsMethod::ReplaceDocuments);
-                let reader = File::open(datasets_paths::SMOL_SONGS).expect(&format!(
-                    "could not find the dataset in: {}",
-                    datasets_paths::SMOL_SONGS
-                ));
-                builder.execute(reader, |_, _| ()).unwrap();
+                let documents = utils::documents_from(datasets_paths::SMOL_SONGS, "csv");
+                builder.execute(documents, |_, _| ()).unwrap();
                 wtxn.commit().unwrap();
 
                 index.prepare_for_closing().wait();
@@ -211,15 +197,10 @@ fn indexing_wiki(c: &mut Criterion) {
             move |index| {
                 let update_builder = UpdateBuilder::new(0);
                 let mut wtxn = index.write_txn().unwrap();
-                let mut builder = update_builder.index_documents(&mut wtxn, &index);
+                let builder = update_builder.index_documents(&mut wtxn, &index);
 
-                builder.update_format(UpdateFormat::Csv);
-                builder.index_documents_method(IndexDocumentsMethod::ReplaceDocuments);
-                let reader = File::open(datasets_paths::SMOL_WIKI_ARTICLES).expect(&format!(
-                    "could not find the dataset in: {}",
-                    datasets_paths::SMOL_SONGS
-                ));
-                builder.execute(reader, |_, _| ()).unwrap();
+                let documents = utils::documents_from(datasets_paths::SMOL_WIKI_ARTICLES, "csv");
+                builder.execute(documents, |_, _| ()).unwrap();
                 wtxn.commit().unwrap();
 
                 index.prepare_for_closing().wait();
@@ -262,13 +243,10 @@ fn indexing_movies_default(c: &mut Criterion) {
             move |index| {
                 let update_builder = UpdateBuilder::new(0);
                 let mut wtxn = index.write_txn().unwrap();
-                let mut builder = update_builder.index_documents(&mut wtxn, &index);
+                let builder = update_builder.index_documents(&mut wtxn, &index);
 
-                builder.update_format(UpdateFormat::Json);
-                builder.index_documents_method(IndexDocumentsMethod::ReplaceDocuments);
-                let reader = File::open(datasets_paths::MOVIES)
-                    .expect(&format!("could not find the dataset in: {}", datasets_paths::MOVIES));
-                builder.execute(reader, |_, _| ()).unwrap();
+                let documents = utils::documents_from(datasets_paths::MOVIES, "json");
+                builder.execute(documents, |_, _| ()).unwrap();
                 wtxn.commit().unwrap();
 
                 index.prepare_for_closing().wait();
@@ -316,15 +294,11 @@ fn indexing_geo(c: &mut Criterion) {
             move |index| {
                 let update_builder = UpdateBuilder::new(0);
                 let mut wtxn = index.write_txn().unwrap();
-                let mut builder = update_builder.index_documents(&mut wtxn, &index);
+                let builder = update_builder.index_documents(&mut wtxn, &index);
+
+                let documents = utils::documents_from(datasets_paths::SMOL_ALL_COUNTRIES, "jsonl");
+                builder.execute(documents, |_, _| ()).unwrap();
 
-                builder.update_format(UpdateFormat::JsonStream);
-                builder.index_documents_method(IndexDocumentsMethod::ReplaceDocuments);
-                let reader = File::open(datasets_paths::SMOL_ALL_COUNTRIES).expect(&format!(
-                    "could not find the dataset in: {}",
-                    datasets_paths::SMOL_ALL_COUNTRIES
-                ));
-                builder.execute(reader, |_, _| ()).unwrap();
                 wtxn.commit().unwrap();
 
                 index.prepare_for_closing().wait();

benchmarks/benches/search_geo.rs

@@ -2,7 +2,7 @@ mod datasets_paths;
 mod utils;
 
 use criterion::{criterion_group, criterion_main};
-use milli::update::{Settings, UpdateFormat};
+use milli::update::Settings;
 use utils::Conf;
 
 #[cfg(target_os = "linux")]
@@ -33,7 +33,7 @@ fn base_conf(builder: &mut Settings) {
 #[rustfmt::skip]
 const BASE_CONF: Conf = Conf {
     dataset: datasets_paths::SMOL_ALL_COUNTRIES,
-    dataset_format: UpdateFormat::JsonStream,
+    dataset_format: "jsonl",
     queries: &[
         "",
     ],

benchmarks/benches/utils.rs

@@ -1,10 +1,15 @@
+#![allow(dead_code)]
+
 use std::fs::{create_dir_all, remove_dir_all, File};
+use std::io::{self, Cursor, Read, Seek};
 use std::path::Path;
 
 use criterion::BenchmarkId;
 use heed::EnvOpenOptions;
-use milli::update::{IndexDocumentsMethod, Settings, UpdateBuilder, UpdateFormat};
+use milli::documents::DocumentBatchReader;
+use milli::update::{IndexDocumentsMethod, Settings, UpdateBuilder};
 use milli::{FilterCondition, Index};
+use serde_json::{Map, Value};
 
 pub struct Conf<'a> {
     /// where we are going to create our database.mmdb directory
@@ -13,7 +18,7 @@ pub struct Conf<'a> {
     /// the dataset to be used, it must be an uncompressed csv
     pub dataset: &'a str,
     /// The format of the dataset
-    pub dataset_format: UpdateFormat,
+    pub dataset_format: &'a str,
     pub group_name: &'a str,
     pub queries: &'a [&'a str],
     /// here you can change which criterion are used and in which order.
@@ -33,7 +38,7 @@ pub struct Conf<'a> {
 impl Conf<'_> {
     pub const BASE: Self = Conf {
         database_name: "benches.mmdb",
-        dataset_format: UpdateFormat::Csv,
+        dataset_format: "csv",
         dataset: "",
         group_name: "",
         queries: &[],
@@ -87,11 +92,10 @@ pub fn base_setup(conf: &Conf) -> Index {
     if let None = conf.primary_key {
         builder.enable_autogenerate_docids();
     }
-    builder.update_format(conf.dataset_format);
+    let documents = documents_from(conf.dataset, conf.dataset_format);
+
     builder.index_documents_method(IndexDocumentsMethod::ReplaceDocuments);
-    let reader = File::open(conf.dataset)
-        .expect(&format!("could not find the dataset in: {}", conf.dataset));
-    builder.execute(reader, |_, _| ()).unwrap();
+    builder.execute(documents, |_, _| ()).unwrap();
     wtxn.commit().unwrap();
 
     index
@@ -128,3 +132,58 @@ pub fn run_benches(c: &mut criterion::Criterion, confs: &[Conf]) {
         index.prepare_for_closing().wait();
     }
 }
+
+pub fn documents_from(filename: &str, filetype: &str) -> DocumentBatchReader<impl Read + Seek> {
+    let reader =
+        File::open(filename).expect(&format!("could not find the dataset in: {}", filename));
+    let documents = match filetype {
+        "csv" => documents_from_csv(reader).unwrap(),
+        "json" => documents_from_json(reader).unwrap(),
+        "jsonl" => documents_from_jsonl(reader).unwrap(),
+        otherwise => panic!("invalid update format {:?}", otherwise),
+    };
+    DocumentBatchReader::from_reader(Cursor::new(documents)).unwrap()
+}
+
+fn documents_from_jsonl(reader: impl io::Read) -> anyhow::Result<Vec<u8>> {
+    let mut writer = Cursor::new(Vec::new());
+    let mut documents = milli::documents::DocumentBatchBuilder::new(&mut writer)?;
+
+    let values = serde_json::Deserializer::from_reader(reader)
+        .into_iter::<serde_json::Map<String, serde_json::Value>>();
+    for document in values {
+        let document = document?;
+        documents.add_documents(document)?;
+    }
+    documents.finish()?;
+
+    Ok(writer.into_inner())
+}
+
+fn documents_from_json(reader: impl io::Read) -> anyhow::Result<Vec<u8>> {
+    let mut writer = Cursor::new(Vec::new());
+    let mut documents = milli::documents::DocumentBatchBuilder::new(&mut writer)?;
+
+    let json: serde_json::Value = serde_json::from_reader(reader)?;
+    documents.add_documents(json)?;
+    documents.finish()?;
+
+    Ok(writer.into_inner())
+}
+
+fn documents_from_csv(reader: impl io::Read) -> anyhow::Result<Vec<u8>> {
+    let mut writer = Cursor::new(Vec::new());
+    let mut documents = milli::documents::DocumentBatchBuilder::new(&mut writer)?;
+
+    let mut records = csv::Reader::from_reader(reader);
+    let iter = records.deserialize::<Map<String, Value>>();
+
+    for doc in iter {
+        let doc = doc?;
+        documents.add_documents(doc)?;
+    }
+
+    documents.finish()?;
+
+    Ok(writer.into_inner())
+}