Move crates under a sub folder to clean up the code

crates/benchmarks/.gitignore

@@ -0,0 +1 @@
+benches/datasets_paths.rs

crates/benchmarks/Cargo.toml

@@ -0,0 +1,50 @@
+[package]
+name = "benchmarks"
+publish = false
+
+version.workspace = true
+authors.workspace = true
+description.workspace = true
+homepage.workspace = true
+readme.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+anyhow = "1.0.86"
+csv = "1.3.0"
+milli = { path = "../milli" }
+mimalloc = { version = "0.1.43", default-features = false }
+serde_json = { version = "1.0.120", features = ["preserve_order"] }
+
+[dev-dependencies]
+criterion = { version = "0.5.1", features = ["html_reports"] }
+rand = "0.8.5"
+rand_chacha = "0.3.1"
+roaring = "0.10.6"
+
+[build-dependencies]
+anyhow = "1.0.86"
+bytes = "1.6.0"
+convert_case = "0.6.0"
+flate2 = "1.0.30"
+reqwest = { version = "0.12.5", features = ["blocking", "rustls-tls"], default-features = false }
+
+[features]
+default = ["milli/all-tokenizations"]
+
+[[bench]]
+name = "search_songs"
+harness = false
+
+[[bench]]
+name = "search_wiki"
+harness = false
+
+[[bench]]
+name = "search_geo"
+harness = false
+
+[[bench]]
+name = "indexing"
+harness = false

crates/benchmarks/README.md

@@ -0,0 +1,138 @@
+Benchmarks
+==========
+
+## TOC
+
+- [Run the benchmarks](#run-the-benchmarks)
+- [Comparison between benchmarks](#comparison-between-benchmarks)
+- [Datasets](#datasets)
+
+## Run the benchmarks
+
+### On our private server
+
+The Meili team has self-hosted his own GitHub runner to run benchmarks on our dedicated bare metal server.
+
+To trigger the benchmark workflow:
+- Go to the `Actions` tab of this repository.
+- Select the `Benchmarks` workflow on the left.
+- Click on `Run workflow` in the blue banner.
+- Select the branch on which you want to run the benchmarks and select the dataset you want (default: `songs`).
+- Finally, click on `Run workflow`.
+
+This GitHub workflow will run the benchmarks and push the `critcmp` report to a DigitalOcean Space (= S3).
+
+The name of the uploaded file is displayed in the workflow.
+
+_[More about critcmp](https://github.com/BurntSushi/critcmp)._
+
+💡 To compare the just-uploaded benchmark with another one, check out the [next section](#comparison-between-benchmarks).
+
+### On your machine
+
+To run all the benchmarks (~5h):
+
+```bash
+cargo bench
+```
+
+To run only the `search_songs` (~1h), `search_wiki` (~3h), `search_geo` (~20m) or `indexing` (~2h) benchmark:
+
+```bash
+cargo bench --bench <dataset name>
+```
+
+By default, the benchmarks will be downloaded and uncompressed automatically in the target directory.<br>
+If you don't want to download the datasets every time you update something on the code, you can specify a custom directory with the environment variable `MILLI_BENCH_DATASETS_PATH`:
+
+```bash
+mkdir ~/datasets
+MILLI_BENCH_DATASETS_PATH=~/datasets cargo bench --bench search_songs # the four datasets are downloaded
+touch build.rs
+MILLI_BENCH_DATASETS_PATH=~/datasets cargo bench --bench songs # the code is compiled again but the datasets are not downloaded
+```
+
+## Comparison between benchmarks
+
+The benchmark reports we push are generated with `critcmp`. Thus, we use `critcmp` to show the result of a benchmark, or compare results between multiple benchmarks.
+
+We provide a script to download and display the comparison report.
+
+Requirements:
+- `grep`
+- `curl`
+- [`critcmp`](https://github.com/BurntSushi/critcmp)
+
+List the available file in the DO Space:
+
+```bash
+./benchmarks/script/list.sh
+```
+```bash
+songs_main_09a4321.json
+songs_geosearch_24ec456.json
+search_songs_main_cb45a10b.json
+```
+
+Run the comparison script:
+
+```bash
+# we get the result of ONE benchmark, this give you an idea of how much time an operation took
+./benchmarks/scripts/compare.sh son songs_geosearch_24ec456.json
+# we compare two benchmarks
+./benchmarks/scripts/compare.sh songs_main_09a4321.json songs_geosearch_24ec456.json
+# we compare three benchmarks
+./benchmarks/scripts/compare.sh songs_main_09a4321.json songs_geosearch_24ec456.json search_songs_main_cb45a10b.json
+```
+
+## Datasets
+
+The benchmarks uses the following datasets:
+- `smol-songs`
+- `smol-wiki`
+- `movies`
+- `smol-all-countries`
+
+### Songs
+
+`smol-songs` is a subset of the [`songs.csv` dataset](https://milli-benchmarks.fra1.digitaloceanspaces.com/datasets/songs.csv.gz).
+
+It was generated with this command:
+
+```bash
+xsv sample --seed 42 1000000 songs.csv -o smol-songs.csv
+```
+
+_[Download the generated `smol-songs` dataset](https://milli-benchmarks.fra1.digitaloceanspaces.com/datasets/smol-songs.csv.gz)._
+
+### Wiki
+
+`smol-wiki` is a subset of the [`wikipedia-articles.csv` dataset](https://milli-benchmarks.fra1.digitaloceanspaces.com/datasets/wiki-articles.csv.gz).
+
+It was generated with the following command:
+
+```bash
+xsv sample --seed 42 500000 wiki-articles.csv -o smol-wiki-articles.csv
+```
+
+_[Download the `smol-wiki` dataset](https://milli-benchmarks.fra1.digitaloceanspaces.com/datasets/smol-wiki-articles.csv.gz)._
+
+### Movies
+
+`movies` is a really small dataset we uses as our example in the [getting started](https://www.meilisearch.com/docs/learn/getting_started/quick_start)
+
+_[Download the `movies` dataset](https://www.meilisearch.com/movies.json)._
+
+
+### All Countries
+
+`smol-all-countries` is a subset of the [`all-countries.csv` dataset](https://milli-benchmarks.fra1.digitaloceanspaces.com/datasets/all-countries.csv.gz)
+It has been converted to jsonlines and then edited so it matches our format for the `_geo` field.
+
+It was generated with the following command:
+```bash
+bat all-countries.csv.gz | gunzip | xsv sample --seed 42 1000000 | csv2json-lite | sd '"latitude":"(.*?)","longitude":"(.*?)"' '"_geo": { "lat": $1, "lng": $2 }' | sd '\[|\]|,$' '' | gzip > smol-all-countries.jsonl.gz
+```
+
+_[Download the `smol-all-countries` dataset](https://milli-benchmarks.fra1.digitaloceanspaces.com/datasets/smol-all-countries.jsonl.gz)._
+

crates/benchmarks/benches/search_geo.rs

@@ -0,0 +1,122 @@
+mod datasets_paths;
+mod utils;
+
+use criterion::{criterion_group, criterion_main};
+use milli::update::Settings;
+use utils::Conf;
+
+#[global_allocator]
+static ALLOC: mimalloc::MiMalloc = mimalloc::MiMalloc;
+
+fn base_conf(builder: &mut Settings) {
+    let displayed_fields =
+        ["geonameid", "name", "asciiname", "alternatenames", "_geo", "population"]
+            .iter()
+            .map(|s| s.to_string())
+            .collect();
+    builder.set_displayed_fields(displayed_fields);
+
+    let searchable_fields =
+        ["name", "alternatenames", "elevation"].iter().map(|s| s.to_string()).collect();
+    builder.set_searchable_fields(searchable_fields);
+
+    let filterable_fields =
+        ["_geo", "population", "elevation"].iter().map(|s| s.to_string()).collect();
+    builder.set_filterable_fields(filterable_fields);
+
+    let sortable_fields =
+        ["_geo", "population", "elevation"].iter().map(|s| s.to_string()).collect();
+    builder.set_sortable_fields(sortable_fields);
+}
+
+#[rustfmt::skip]
+const BASE_CONF: Conf = Conf {
+    dataset: datasets_paths::SMOL_ALL_COUNTRIES,
+    dataset_format: "jsonl",
+    queries: &[
+        "",
+    ],
+    configure: base_conf,
+    primary_key: Some("geonameid"),
+    ..Conf::BASE
+};
+
+fn bench_geo(c: &mut criterion::Criterion) {
+    #[rustfmt::skip]
+    let confs = &[
+        // A basic placeholder with no geo
+        utils::Conf {
+            group_name: "placeholder with no geo",
+            ..BASE_CONF
+        },
+        // Medium aglomeration: probably the most common usecase
+        utils::Conf {
+            group_name: "asc sort from Lille",
+            sort: Some(vec!["_geoPoint(50.62999333378238, 3.086269263384099):asc"]),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "desc sort from Lille",
+            sort: Some(vec!["_geoPoint(50.62999333378238, 3.086269263384099):desc"]),
+            ..BASE_CONF
+        },
+        // Big agglomeration: a lot of documents close to our point
+        utils::Conf {
+            group_name: "asc sort from Tokyo",
+            sort: Some(vec!["_geoPoint(35.749512532692144, 139.61664952543356):asc"]),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "desc sort from Tokyo",
+            sort: Some(vec!["_geoPoint(35.749512532692144, 139.61664952543356):desc"]),
+            ..BASE_CONF
+        },
+        // The furthest point from any civilization
+        utils::Conf {
+            group_name: "asc sort from Point Nemo",
+            sort: Some(vec!["_geoPoint(-48.87561645055408, -123.39275749319793):asc"]),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "desc sort from Point Nemo",
+            sort: Some(vec!["_geoPoint(-48.87561645055408, -123.39275749319793):desc"]),
+            ..BASE_CONF
+        },
+        // Filters
+        utils::Conf {
+            group_name: "filter of 100km from Lille",
+            filter: Some("_geoRadius(50.62999333378238, 3.086269263384099, 100000)"),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "filter of 1km from Lille",
+            filter: Some("_geoRadius(50.62999333378238, 3.086269263384099, 1000)"),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "filter of 100km from Tokyo",
+            filter: Some("_geoRadius(35.749512532692144, 139.61664952543356, 100000)"),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "filter of 1km from Tokyo",
+            filter: Some("_geoRadius(35.749512532692144, 139.61664952543356, 1000)"),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "filter of 100km from Point Nemo",
+            filter: Some("_geoRadius(-48.87561645055408, -123.39275749319793, 100000)"),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "filter of 1km from Point Nemo",
+            filter: Some("_geoRadius(-48.87561645055408, -123.39275749319793, 1000)"),
+            ..BASE_CONF
+        },
+    ];
+
+    utils::run_benches(c, confs);
+}
+
+criterion_group!(benches, bench_geo);
+criterion_main!(benches);

crates/benchmarks/benches/search_songs.rs

@@ -0,0 +1,196 @@
+mod datasets_paths;
+mod utils;
+
+use criterion::{criterion_group, criterion_main};
+use milli::update::Settings;
+use utils::Conf;
+
+#[global_allocator]
+static ALLOC: mimalloc::MiMalloc = mimalloc::MiMalloc;
+
+fn base_conf(builder: &mut Settings) {
+    let displayed_fields =
+        ["id", "title", "album", "artist", "genre", "country", "released", "duration"]
+            .iter()
+            .map(|s| s.to_string())
+            .collect();
+    builder.set_displayed_fields(displayed_fields);
+
+    let searchable_fields = ["title", "album", "artist"].iter().map(|s| s.to_string()).collect();
+    builder.set_searchable_fields(searchable_fields);
+
+    let faceted_fields = ["released-timestamp", "duration-float", "genre", "country", "artist"]
+        .iter()
+        .map(|s| s.to_string())
+        .collect();
+    builder.set_filterable_fields(faceted_fields);
+}
+
+#[rustfmt::skip]
+const BASE_CONF: Conf = Conf {
+    dataset: datasets_paths::SMOL_SONGS,
+    queries: &[
+        "john ",             // 9097
+        "david ",            // 4794
+        "charles ",          // 1957
+        "david bowie ",      // 1200
+        "michael jackson ",  // 600
+        "thelonious monk ",  // 303
+        "charles mingus ",   // 142
+        "marcus miller ",    // 60
+        "tamo ",             // 13
+        "Notstandskomitee ", // 4
+    ],
+    configure: base_conf,
+    primary_key: Some("id"),
+    ..Conf::BASE
+};
+
+fn bench_songs(c: &mut criterion::Criterion) {
+    let default_criterion: Vec<String> =
+        milli::default_criteria().iter().map(|criteria| criteria.to_string()).collect();
+    let default_criterion = default_criterion.iter().map(|s| s.as_str());
+    let asc_default: Vec<&str> =
+        std::iter::once("released-timestamp:asc").chain(default_criterion.clone()).collect();
+    let desc_default: Vec<&str> =
+        std::iter::once("released-timestamp:desc").chain(default_criterion.clone()).collect();
+
+    let basic_with_quote: Vec<String> = BASE_CONF
+        .queries
+        .iter()
+        .map(|s| {
+            s.trim().split(' ').map(|s| format!(r#""{}""#, s)).collect::<Vec<String>>().join(" ")
+        })
+        .collect();
+    let basic_with_quote: &[&str] =
+        &basic_with_quote.iter().map(|s| s.as_str()).collect::<Vec<&str>>();
+
+    #[rustfmt::skip]
+    let confs = &[
+        /* first we bench each criterion alone */
+        utils::Conf {
+            group_name: "proximity",
+            queries: &[
+                "black saint sinner lady ",
+                "les dangeureuses 1960 ",
+                "The Disneyland Sing-Along Chorus ",
+                "Under Great Northern Lights ",
+                "7000 Danses Un Jour Dans Notre Vie ",
+            ],
+            criterion: Some(&["proximity"]),
+            optional_words: false,
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "typo",
+            queries: &[
+                "mongus ",
+                "thelonius monk ",
+                "Disnaylande ",
+                "the white striper ",
+                "indochie ",
+                "indochien ",
+                "klub des loopers ",
+                "fear of the duck ",
+                "michel depech ",
+                "stromal ",
+                "dire straights ",
+                "Arethla Franklin ",
+            ],
+            criterion: Some(&["typo"]),
+            optional_words: false,
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "words",
+            queries: &[
+                "the black saint and the sinner lady and the good doggo ", // four words to pop
+                "les liaisons dangeureuses 1793 ",                         // one word to pop
+                "The Disneyland Children's Sing-Alone song ",              // two words to pop
+                "seven nation mummy ",                                     // one word to pop
+                "7000 Danses / Le Baiser / je me trompe de mots ",         // four words to pop
+                "Bring Your Daughter To The Slaughter but now this is not part of the title ", // nine words to pop
+                "whathavenotnsuchforth and a good amount of words to pop to match the first one ", // 13
+            ],
+            criterion: Some(&["words"]),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "asc",
+            criterion: Some(&["released-timestamp:desc"]),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "desc",
+            criterion: Some(&["released-timestamp:desc"]),
+            ..BASE_CONF
+        },
+
+        /* then we bench the asc and desc criterion on top of the default criterion */
+        utils::Conf {
+            group_name: "asc + default",
+            criterion: Some(&asc_default[..]),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "desc + default",
+            criterion: Some(&desc_default[..]),
+            ..BASE_CONF
+        },
+
+        /* we bench the filters with the default request */
+        utils::Conf {
+            group_name: "basic filter: <=",
+            filter: Some("released-timestamp <= 946728000"), // year 2000
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "basic filter: TO",
+            filter: Some("released-timestamp 946728000 TO 1262347200"), // year 2000 to 2010
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "big filter",
+            filter: Some("released-timestamp != 1262347200 AND (NOT (released-timestamp = 946728000)) AND (duration-float = 1 OR (duration-float 1.1 TO 1.5 AND released-timestamp > 315576000))"),
+            ..BASE_CONF
+        },
+
+        /* the we bench some global / normal search with all the default criterion in the default
+         * order */
+        utils::Conf {
+            group_name: "basic placeholder",
+            queries: &[""],
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "basic without quote",
+            queries: &BASE_CONF
+                .queries
+                .iter()
+                .map(|s| s.trim()) // we remove the space at the end of each request
+                .collect::<Vec<&str>>(),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "basic with quote",
+            queries: basic_with_quote,
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "prefix search",
+            queries: &[
+                "s", // 500k+ results
+                "a", //
+                "b", //
+                "i", //
+                "x", // only 7k results
+            ],
+            ..BASE_CONF
+        },
+    ];
+
+    utils::run_benches(c, confs);
+}
+
+criterion_group!(benches, bench_songs);
+criterion_main!(benches);

crates/benchmarks/benches/search_wiki.rs

@@ -0,0 +1,129 @@
+mod datasets_paths;
+mod utils;
+
+use criterion::{criterion_group, criterion_main};
+use milli::update::Settings;
+use utils::Conf;
+
+#[global_allocator]
+static ALLOC: mimalloc::MiMalloc = mimalloc::MiMalloc;
+
+fn base_conf(builder: &mut Settings) {
+    let displayed_fields = ["title", "body", "url"].iter().map(|s| s.to_string()).collect();
+    builder.set_displayed_fields(displayed_fields);
+
+    let searchable_fields = ["title", "body"].iter().map(|s| s.to_string()).collect();
+    builder.set_searchable_fields(searchable_fields);
+}
+
+#[rustfmt::skip]
+const BASE_CONF: Conf = Conf {
+    dataset: datasets_paths::SMOL_WIKI_ARTICLES,
+    queries: &[
+        "mingus ",        // 46 candidates
+        "miles davis ",   // 159
+        "rock and roll ", // 1007
+        "machine ",       // 3448
+        "spain ",         // 7002
+        "japan ",         // 10.593
+        "france ",        // 17.616
+        "film ",          // 24.959
+    ],
+    configure: base_conf,
+    ..Conf::BASE
+};
+
+fn bench_songs(c: &mut criterion::Criterion) {
+    let basic_with_quote: Vec<String> = BASE_CONF
+        .queries
+        .iter()
+        .map(|s| {
+            s.trim().split(' ').map(|s| format!(r#""{}""#, s)).collect::<Vec<String>>().join(" ")
+        })
+        .collect();
+    let basic_with_quote: &[&str] =
+        &basic_with_quote.iter().map(|s| s.as_str()).collect::<Vec<&str>>();
+
+    #[rustfmt::skip]
+    let confs = &[
+        /* first we bench each criterion alone */
+        utils::Conf {
+            group_name: "proximity",
+            queries: &[
+                "herald sings ",
+                "april paris ",
+                "tea two ",
+                "diesel engine ",
+            ],
+            criterion: Some(&["proximity"]),
+            optional_words: false,
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "typo",
+            queries: &[
+                "migrosoft ",
+                "linax ",
+                "Disnaylande ",
+                "phytogropher ",
+                "nympalidea ",
+                "aritmetric ",
+                "the fronce ",
+                "sisan ",
+            ],
+            criterion: Some(&["typo"]),
+            optional_words: false,
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "words",
+            queries: &[
+                "the black saint and the sinner lady and the good doggo ", // four words to pop, 27 results
+                "Kameya Tokujirō mingus monk ",                            // two words to pop, 55
+                "Ulrich Hensel meilisearch milli ",                        // two words to pop, 306
+                "Idaho Bellevue pizza ",                                   // one word to pop, 800
+                "Abraham machin ",                                         // one word to pop, 1141
+            ],
+            criterion: Some(&["words"]),
+            ..BASE_CONF
+        },
+        /* the we bench some global / normal search with all the default criterion in the default
+         * order */
+        utils::Conf {
+            group_name: "basic placeholder",
+            queries: &[""],
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "basic without quote",
+            queries: &BASE_CONF
+                .queries
+                .iter()
+                .map(|s| s.trim()) // we remove the space at the end of each request
+                .collect::<Vec<&str>>(),
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "basic with quote",
+            queries: basic_with_quote,
+            ..BASE_CONF
+        },
+        utils::Conf {
+            group_name: "prefix search",
+            queries: &[
+                "t", // 453k results
+                "c", // 405k
+                "g", // 318k
+                "j", // 227k
+                "q", // 71k
+                "x", // 17k
+            ],
+            ..BASE_CONF
+        },
+    ];
+
+    utils::run_benches(c, confs);
+}
+
+criterion_group!(benches, bench_songs);
+criterion_main!(benches);

crates/benchmarks/benches/utils.rs

@@ -0,0 +1,256 @@
+#![allow(dead_code)]
+
+use std::fs::{create_dir_all, remove_dir_all, File};
+use std::io::{self, BufRead, BufReader, Cursor, Read, Seek};
+use std::num::ParseFloatError;
+use std::path::Path;
+use std::str::FromStr;
+
+use criterion::BenchmarkId;
+use milli::documents::{DocumentsBatchBuilder, DocumentsBatchReader};
+use milli::heed::EnvOpenOptions;
+use milli::update::{
+    IndexDocuments, IndexDocumentsConfig, IndexDocumentsMethod, IndexerConfig, Settings,
+};
+use milli::{Criterion, Filter, Index, Object, TermsMatchingStrategy};
+use serde_json::Value;
+
+pub struct Conf<'a> {
+    /// where we are going to create our database.mmdb directory
+    /// each benchmark will first try to delete it and then recreate it
+    pub database_name: &'a str,
+    /// the dataset to be used, it must be an uncompressed csv
+    pub dataset: &'a str,
+    /// The format of the dataset
+    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.
+    /// - if you specify something all the base configuration will be thrown out
+    /// - if you don't specify anything (None) the default configuration will be kept
+    pub criterion: Option<&'a [&'a str]>,
+    /// the last chance to configure your database as you want
+    pub configure: fn(&mut Settings),
+    pub filter: Option<&'a str>,
+    pub sort: Option<Vec<&'a str>>,
+    /// enable or disable the optional words on the query
+    pub optional_words: bool,
+    /// primary key, if there is None we'll auto-generate docids for every documents
+    pub primary_key: Option<&'a str>,
+}
+
+impl Conf<'_> {
+    pub const BASE: Self = Conf {
+        database_name: "benches.mmdb",
+        dataset_format: "csv",
+        dataset: "",
+        group_name: "",
+        queries: &[],
+        criterion: None,
+        configure: |_| (),
+        filter: None,
+        sort: None,
+        optional_words: true,
+        primary_key: None,
+    };
+}
+
+pub fn base_setup(conf: &Conf) -> Index {
+    match remove_dir_all(conf.database_name) {
+        Ok(_) => (),
+        Err(e) if e.kind() == std::io::ErrorKind::NotFound => (),
+        Err(e) => panic!("{}", e),
+    }
+    create_dir_all(conf.database_name).unwrap();
+
+    let mut options = EnvOpenOptions::new();
+    options.map_size(100 * 1024 * 1024 * 1024); // 100 GB
+    options.max_readers(10);
+    let index = Index::new(options, conf.database_name).unwrap();
+
+    let config = IndexerConfig::default();
+    let mut wtxn = index.write_txn().unwrap();
+    let mut builder = Settings::new(&mut wtxn, &index, &config);
+
+    if let Some(primary_key) = conf.primary_key {
+        builder.set_primary_key(primary_key.to_string());
+    }
+
+    if let Some(criterion) = conf.criterion {
+        builder.reset_filterable_fields();
+        builder.reset_criteria();
+        builder.reset_stop_words();
+
+        let criterion = criterion.iter().map(|s| Criterion::from_str(s).unwrap()).collect();
+        builder.set_criteria(criterion);
+    }
+
+    (conf.configure)(&mut builder);
+
+    builder.execute(|_| (), || false).unwrap();
+    wtxn.commit().unwrap();
+
+    let config = IndexerConfig::default();
+    let mut wtxn = index.write_txn().unwrap();
+    let indexing_config = IndexDocumentsConfig {
+        autogenerate_docids: conf.primary_key.is_none(),
+        update_method: IndexDocumentsMethod::ReplaceDocuments,
+        ..Default::default()
+    };
+    let builder =
+        IndexDocuments::new(&mut wtxn, &index, &config, indexing_config, |_| (), || false).unwrap();
+    let documents = documents_from(conf.dataset, conf.dataset_format);
+    let (builder, user_error) = builder.add_documents(documents).unwrap();
+    user_error.unwrap();
+    builder.execute().unwrap();
+    wtxn.commit().unwrap();
+
+    index
+}
+
+pub fn run_benches(c: &mut criterion::Criterion, confs: &[Conf]) {
+    for conf in confs {
+        let index = base_setup(conf);
+
+        let file_name = Path::new(conf.dataset).file_name().and_then(|f| f.to_str()).unwrap();
+        let name = format!("{}: {}", file_name, conf.group_name);
+        let mut group = c.benchmark_group(&name);
+
+        for &query in conf.queries {
+            group.bench_with_input(BenchmarkId::from_parameter(query), &query, |b, &query| {
+                b.iter(|| {
+                    let rtxn = index.read_txn().unwrap();
+                    let mut search = index.search(&rtxn);
+                    search.query(query).terms_matching_strategy(TermsMatchingStrategy::default());
+                    if let Some(filter) = conf.filter {
+                        let filter = Filter::from_str(filter).unwrap().unwrap();
+                        search.filter(filter);
+                    }
+                    if let Some(sort) = &conf.sort {
+                        let sort = sort.iter().map(|sort| sort.parse().unwrap()).collect();
+                        search.sort_criteria(sort);
+                    }
+                    let _ids = search.execute().unwrap();
+                });
+            });
+        }
+        group.finish();
+
+        index.prepare_for_closing().wait();
+    }
+}
+
+pub fn documents_from(filename: &str, filetype: &str) -> DocumentsBatchReader<impl BufRead + Seek> {
+    let reader = File::open(filename)
+        .unwrap_or_else(|_| panic!("could not find the dataset in: {}", filename));
+    let reader = BufReader::new(reader);
+    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),
+    };
+    DocumentsBatchReader::from_reader(Cursor::new(documents)).unwrap()
+}
+
+fn documents_from_jsonl(reader: impl BufRead) -> anyhow::Result<Vec<u8>> {
+    let mut documents = DocumentsBatchBuilder::new(Vec::new());
+
+    for result in serde_json::Deserializer::from_reader(reader).into_iter::<Object>() {
+        let object = result?;
+        documents.append_json_object(&object)?;
+    }
+
+    documents.into_inner().map_err(Into::into)
+}
+
+fn documents_from_json(reader: impl BufRead) -> anyhow::Result<Vec<u8>> {
+    let mut documents = DocumentsBatchBuilder::new(Vec::new());
+
+    documents.append_json_array(reader)?;
+
+    documents.into_inner().map_err(Into::into)
+}
+
+fn documents_from_csv(reader: impl BufRead) -> anyhow::Result<Vec<u8>> {
+    let csv = csv::Reader::from_reader(reader);
+
+    let mut documents = DocumentsBatchBuilder::new(Vec::new());
+    documents.append_csv(csv)?;
+
+    documents.into_inner().map_err(Into::into)
+}
+
+enum AllowedType {
+    String,
+    Number,
+}
+
+fn parse_csv_header(header: &str) -> (String, AllowedType) {
+    // if there are several separators we only split on the last one.
+    match header.rsplit_once(':') {
+        Some((field_name, field_type)) => match field_type {
+            "string" => (field_name.to_string(), AllowedType::String),
+            "number" => (field_name.to_string(), AllowedType::Number),
+            // we may return an error in this case.
+            _otherwise => (header.to_string(), AllowedType::String),
+        },
+        None => (header.to_string(), AllowedType::String),
+    }
+}
+
+struct CSVDocumentDeserializer<R>
+where
+    R: Read,
+{
+    documents: csv::StringRecordsIntoIter<R>,
+    headers: Vec<(String, AllowedType)>,
+}
+
+impl<R: Read> CSVDocumentDeserializer<R> {
+    fn from_reader(reader: R) -> io::Result<Self> {
+        let mut records = csv::Reader::from_reader(reader);
+
+        let headers = records.headers()?.into_iter().map(parse_csv_header).collect();
+
+        Ok(Self { documents: records.into_records(), headers })
+    }
+}
+
+impl<R: Read> Iterator for CSVDocumentDeserializer<R> {
+    type Item = anyhow::Result<Object>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        let csv_document = self.documents.next()?;
+
+        match csv_document {
+            Ok(csv_document) => {
+                let mut document = Object::new();
+
+                for ((field_name, field_type), value) in
+                    self.headers.iter().zip(csv_document.into_iter())
+                {
+                    let parsed_value: Result<Value, ParseFloatError> = match field_type {
+                        AllowedType::Number => {
+                            value.parse::<f64>().map(Value::from).map_err(Into::into)
+                        }
+                        AllowedType::String => Ok(Value::String(value.to_string())),
+                    };
+
+                    match parsed_value {
+                        Ok(value) => drop(document.insert(field_name.to_string(), value)),
+                        Err(_e) => {
+                            return Some(Err(anyhow::anyhow!(
+                                "Value '{}' is not a valid number",
+                                value
+                            )))
+                        }
+                    }
+                }
+
+                Some(Ok(document))
+            }
+            Err(e) => Some(Err(anyhow::anyhow!("Error parsing csv document: {}", e))),
+        }
+    }
+}

crates/benchmarks/build.rs

@@ -0,0 +1,115 @@
+use std::fs::File;
+use std::io::{Cursor, Read, Seek, Write};
+use std::path::{Path, PathBuf};
+use std::{env, fs};
+
+use bytes::Bytes;
+use convert_case::{Case, Casing};
+use flate2::read::GzDecoder;
+use reqwest::IntoUrl;
+
+const BASE_URL: &str = "https://milli-benchmarks.fra1.digitaloceanspaces.com/datasets";
+
+const DATASET_SONGS: (&str, &str) = ("smol-songs", "csv");
+const DATASET_SONGS_1_2: (&str, &str) = ("smol-songs-1_2", "csv");
+const DATASET_SONGS_3_4: (&str, &str) = ("smol-songs-3_4", "csv");
+const DATASET_SONGS_4_4: (&str, &str) = ("smol-songs-4_4", "csv");
+const DATASET_WIKI: (&str, &str) = ("smol-wiki-articles", "csv");
+const DATASET_WIKI_1_2: (&str, &str) = ("smol-wiki-articles-1_2", "csv");
+const DATASET_WIKI_3_4: (&str, &str) = ("smol-wiki-articles-3_4", "csv");
+const DATASET_WIKI_4_4: (&str, &str) = ("smol-wiki-articles-4_4", "csv");
+const DATASET_MOVIES: (&str, &str) = ("movies", "json");
+const DATASET_MOVIES_1_2: (&str, &str) = ("movies-1_2", "json");
+const DATASET_MOVIES_3_4: (&str, &str) = ("movies-3_4", "json");
+const DATASET_MOVIES_4_4: (&str, &str) = ("movies-4_4", "json");
+const DATASET_NESTED_MOVIES: (&str, &str) = ("nested_movies", "json");
+const DATASET_GEO: (&str, &str) = ("smol-all-countries", "jsonl");
+
+const ALL_DATASETS: &[(&str, &str)] = &[
+    DATASET_SONGS,
+    DATASET_SONGS_1_2,
+    DATASET_SONGS_3_4,
+    DATASET_SONGS_4_4,
+    DATASET_WIKI,
+    DATASET_WIKI_1_2,
+    DATASET_WIKI_3_4,
+    DATASET_WIKI_4_4,
+    DATASET_MOVIES,
+    DATASET_MOVIES_1_2,
+    DATASET_MOVIES_3_4,
+    DATASET_MOVIES_4_4,
+    DATASET_NESTED_MOVIES,
+    DATASET_GEO,
+];
+
+/// The name of the environment variable used to select the path
+/// of the directory containing the datasets
+const BASE_DATASETS_PATH_KEY: &str = "MILLI_BENCH_DATASETS_PATH";
+
+fn main() -> anyhow::Result<()> {
+    let out_dir = PathBuf::from(env::var(BASE_DATASETS_PATH_KEY).unwrap_or(env::var("OUT_DIR")?));
+
+    let benches_dir = PathBuf::from(env::var("CARGO_MANIFEST_DIR")?).join("benches");
+    let mut manifest_paths_file = File::create(benches_dir.join("datasets_paths.rs"))?;
+    write!(
+        manifest_paths_file,
+        r#"//! This file is generated by the build script.
+//! Do not modify by hand, use the build.rs file.
+#![allow(dead_code)]
+"#
+    )?;
+    writeln!(manifest_paths_file)?;
+
+    for (dataset, extension) in ALL_DATASETS {
+        let out_path = out_dir.join(dataset);
+        let out_file = out_path.with_extension(extension);
+
+        writeln!(
+            &mut manifest_paths_file,
+            r#"pub const {}: &str = {:?};"#,
+            dataset.to_case(Case::ScreamingSnake),
+            out_file.display(),
+        )?;
+
+        if out_file.exists() {
+            eprintln!(
+                "The dataset {} already exists on the file system and will not be downloaded again",
+                out_path.display(),
+            );
+            continue;
+        }
+        let url = format!("{}/{}.{}.gz", BASE_URL, dataset, extension);
+        eprintln!("downloading: {}", url);
+        let bytes = retry(|| download_dataset(url.clone()), 10)?;
+        eprintln!("{} downloaded successfully", url);
+        eprintln!("uncompressing in {}", out_file.display());
+        uncompress_in_file(bytes, &out_file)?;
+    }
+
+    Ok(())
+}
+
+fn retry<Ok, Err>(fun: impl Fn() -> Result<Ok, Err>, times: usize) -> Result<Ok, Err> {
+    for _ in 0..times {
+        if let ok @ Ok(_) = fun() {
+            return ok;
+        }
+    }
+    fun()
+}
+
+fn download_dataset<U: IntoUrl>(url: U) -> anyhow::Result<Cursor<Bytes>> {
+    let bytes =
+        reqwest::blocking::Client::builder().timeout(None).build()?.get(url).send()?.bytes()?;
+    Ok(Cursor::new(bytes))
+}
+
+fn uncompress_in_file<R: Read + Seek, P: AsRef<Path>>(bytes: R, path: P) -> anyhow::Result<()> {
+    let path = path.as_ref();
+    let mut gz = GzDecoder::new(bytes);
+    let mut dataset = Vec::new();
+    gz.read_to_end(&mut dataset)?;
+
+    fs::write(path, dataset)?;
+    Ok(())
+}

crates/benchmarks/scripts/compare.sh

@@ -0,0 +1,38 @@
+#!/usr/bin/env bash
+
+# Requirements:
+# - critcmp. See: https://github.com/BurntSushi/critcmp
+# - curl
+
+# Usage
+# $ bash compare.sh json_file1 json_file1
+# ex: bash compare.sh songs_main_09a4321.json songs_geosearch_24ec456.json
+
+# Checking that critcmp is installed
+command -v critcmp > /dev/null 2>&1
+if [[ "$?" -ne 0 ]]; then
+    echo 'You must install critcmp to make this script work.'
+    echo 'See: https://github.com/BurntSushi/critcmp'
+    echo '  $ cargo install critcmp'
+    exit 1
+fi
+
+s3_url='https://milli-benchmarks.fra1.digitaloceanspaces.com/critcmp_results'
+
+for file in $@
+do
+    file_s3_url="$s3_url/$file"
+    file_local_path="/tmp/$file"
+
+    if [[ ! -f $file_local_path ]]; then
+        curl $file_s3_url --output $file_local_path --silent
+        if [[ "$?" -ne 0 ]]; then
+            echo 'curl command failed.'
+            exit 1
+        fi
+    fi
+done
+
+path_list=$(echo " $@" | sed 's/ / \/tmp\//g')
+
+critcmp $path_list

crates/benchmarks/scripts/list.sh

@@ -0,0 +1,14 @@
+#!/usr/bin/env bash
+
+# Requirements:
+# - curl
+# - grep
+
+res=$(curl -s https://milli-benchmarks.fra1.digitaloceanspaces.com | grep -o '<Key>[^<]\+' | cut -c 5- | grep critcmp_results/ | cut -c 18-)
+
+for pattern in "$@"
+do
+	res=$(echo "$res" | grep $pattern)
+done
+
+echo "$res"

crates/benchmarks/src/lib.rs

@@ -0,0 +1,5 @@
+//! This library is only used to isolate the benchmarks
+//! from the original milli library.
+//!
+//! It does not include interesting functions for milli library
+//! users only for milli contributors.