TMP: remove optimization where later ranking rules are not applied on buckets of a single document

.dockerignore

@@ -2,3 +2,4 @@ target
 Dockerfile
 .dockerignore
 .gitignore
+**/.git

.github/workflows/publish-apt-brew-pkg.yml

@@ -35,7 +35,7 @@ jobs:
     - name: Build deb package
       run: cargo deb -p meilisearch -o target/debian/meilisearch.deb
     - name: Upload debian pkg to release
-      uses: svenstaro/upload-release-action@2.6.1
+      uses: svenstaro/upload-release-action@2.5.0
       with:
         repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
         file: target/debian/meilisearch.deb

.github/workflows/publish-binaries.yml

@@ -54,7 +54,7 @@ jobs:
     # No need to upload binaries for dry run (cron)
     - name: Upload binaries to release
       if: github.event_name == 'release'
-      uses: svenstaro/upload-release-action@2.6.1
+      uses: svenstaro/upload-release-action@2.5.0
       with:
         repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
         file: target/release/meilisearch
@@ -87,7 +87,7 @@ jobs:
     # No need to upload binaries for dry run (cron)
     - name: Upload binaries to release
       if: github.event_name == 'release'
-      uses: svenstaro/upload-release-action@2.6.1
+      uses: svenstaro/upload-release-action@2.5.0
       with:
         repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
         file: target/release/${{ matrix.artifact_name }}
@@ -121,7 +121,7 @@ jobs:
       - name: Upload the binary to release
         # No need to upload binaries for dry run (cron)
         if: github.event_name == 'release'
-        uses: svenstaro/upload-release-action@2.6.1
+        uses: svenstaro/upload-release-action@2.5.0
         with:
           repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
           file: target/${{ matrix.target }}/release/meilisearch
@@ -183,7 +183,7 @@ jobs:
       - name: Upload the binary to release
         # No need to upload binaries for dry run (cron)
         if: github.event_name == 'release'
-        uses: svenstaro/upload-release-action@2.6.1
+        uses: svenstaro/upload-release-action@2.5.0
         with:
           repo_token: ${{ secrets.MEILI_BOT_GH_PAT }}
           file: target/${{ matrix.target }}/release/meilisearch

.github/workflows/publish-docker-images.yml

@@ -58,9 +58,13 @@ jobs:
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v2
+        with:
+          platforms: linux/amd64,linux/arm64
 
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v2
+        with:
+          platforms: linux/amd64,linux/arm64
 
       - name: Login to Docker Hub
         uses: docker/login-action@v2
@@ -88,10 +92,13 @@ jobs:
           push: true
           platforms: linux/amd64,linux/arm64
           tags: ${{ steps.meta.outputs.tags }}
+          builder: ${{ steps.buildx.outputs.name }}
           build-args: |
             COMMIT_SHA=${{ github.sha }}
             COMMIT_DATE=${{ steps.build-metadata.outputs.date }}
             GIT_TAG=${{ github.ref_name }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
 
       # /!\ Don't touch this without checking with Cloud team
       - name: Send CI information to Cloud team

.github/workflows/sdks-tests.yml

@@ -3,11 +3,6 @@ name: SDKs tests
 
 on:
   workflow_dispatch:
-    inputs:
-      docker_image:
-        description: 'The Meilisearch Docker image used'
-        required: false
-        default: nightly
   schedule:
     - cron: "0 6 * * MON" # Every Monday at 6:00AM
 
@@ -16,28 +11,13 @@ env:
   MEILI_NO_ANALYTICS: 'true'
 
 jobs:
-  define-docker-image:
-    runs-on: ubuntu-latest
-    outputs:
-      docker-image: ${{ steps.define-image.outputs.docker-image }}
-    steps:
-      - uses: actions/checkout@v3
-      - name: Define the Docker image we need to use
-        id: define-image
-        run: |
-          event=${{ github.event.action }}
-          echo "docker-image=nightly" >> $GITHUB_OUTPUT
-          if [[ $event == 'workflow_dispatch' ]]; then
-            echo "docker-image=${{ github.event.inputs.docker_image }}" >> $GITHUB_OUTPUT
-          fi
 
   meilisearch-js-tests:
-    needs: define-docker-image
     name: JS SDK tests
     runs-on: ubuntu-latest
     services:
       meilisearch:
-        image: getmeili/meilisearch:${{ github.event.inputs.docker_image }}
+        image: getmeili/meilisearch:nightly
         env:
           MEILI_MASTER_KEY: ${{ env.MEILI_MASTER_KEY }}
           MEILI_NO_ANALYTICS: ${{ env.MEILI_NO_ANALYTICS }}
@@ -67,12 +47,11 @@ jobs:
         run: yarn test:env:browser
 
   instant-meilisearch-tests:
-    needs: define-docker-image
     name: instant-meilisearch tests
     runs-on: ubuntu-latest
     services:
       meilisearch:
-        image: getmeili/meilisearch:${{ github.event.inputs.docker_image }}
+        image: getmeili/meilisearch:nightly
         env:
           MEILI_MASTER_KEY: ${{ env.MEILI_MASTER_KEY }}
           MEILI_NO_ANALYTICS: ${{ env.MEILI_NO_ANALYTICS }}
@@ -94,12 +73,11 @@ jobs:
         run: yarn build
 
   meilisearch-php-tests:
-    needs: define-docker-image
     name: PHP SDK tests
     runs-on: ubuntu-latest
     services:
       meilisearch:
-        image: getmeili/meilisearch:${{ github.event.inputs.docker_image }}
+        image: getmeili/meilisearch:nightly
         env:
           MEILI_MASTER_KEY: ${{ env.MEILI_MASTER_KEY }}
           MEILI_NO_ANALYTICS: ${{ env.MEILI_NO_ANALYTICS }}
@@ -125,12 +103,11 @@ jobs:
           composer remove --dev guzzlehttp/guzzle http-interop/http-factory-guzzle
 
   meilisearch-python-tests:
-    needs: define-docker-image
     name: Python SDK tests
     runs-on: ubuntu-latest
     services:
       meilisearch:
-        image: getmeili/meilisearch:${{ github.event.inputs.docker_image }}
+        image: getmeili/meilisearch:nightly
         env:
           MEILI_MASTER_KEY: ${{ env.MEILI_MASTER_KEY }}
           MEILI_NO_ANALYTICS: ${{ env.MEILI_NO_ANALYTICS }}
@@ -150,12 +127,11 @@ jobs:
         run: pipenv run pytest
 
   meilisearch-go-tests:
-    needs: define-docker-image
     name: Go SDK tests
     runs-on: ubuntu-latest
     services:
       meilisearch:
-        image: getmeili/meilisearch:${{ github.event.inputs.docker_image }}
+        image: getmeili/meilisearch:nightly
         env:
           MEILI_MASTER_KEY: ${{ env.MEILI_MASTER_KEY }}
           MEILI_NO_ANALYTICS: ${{ env.MEILI_NO_ANALYTICS }}
@@ -163,7 +139,7 @@ jobs:
           - '7700:7700'
     steps:
       - name: Set up Go
-        uses: actions/setup-go@v4
+        uses: actions/setup-go@v3
         with:
           go-version: stable
       - uses: actions/checkout@v3
@@ -180,12 +156,11 @@ jobs:
         run: go test -v ./...
 
   meilisearch-ruby-tests:
-    needs: define-docker-image
     name: Ruby SDK tests
     runs-on: ubuntu-latest
     services:
       meilisearch:
-        image: getmeili/meilisearch:${{ github.event.inputs.docker_image }}
+        image: getmeili/meilisearch:nightly
         env:
           MEILI_MASTER_KEY: ${{ env.MEILI_MASTER_KEY }}
           MEILI_NO_ANALYTICS: ${{ env.MEILI_NO_ANALYTICS }}
@@ -205,12 +180,11 @@ jobs:
         run: bundle exec rspec
 
   meilisearch-rust-tests:
-    needs: define-docker-image
     name: Rust SDK tests
     runs-on: ubuntu-latest
     services:
       meilisearch:
-        image: getmeili/meilisearch:${{ github.event.inputs.docker_image }}
+        image: getmeili/meilisearch:nightly
         env:
           MEILI_MASTER_KEY: ${{ env.MEILI_MASTER_KEY }}
           MEILI_NO_ANALYTICS: ${{ env.MEILI_NO_ANALYTICS }}

.github/workflows/test-suite.yml

@@ -43,7 +43,7 @@ jobs:
           toolchain: nightly
           override: true
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.4.0
+        uses: Swatinem/rust-cache@v2.2.1
       - name: Run cargo check without any default features
         uses: actions-rs/cargo@v1
         with:
@@ -65,7 +65,7 @@ jobs:
     steps:
       - uses: actions/checkout@v3
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.4.0
+        uses: Swatinem/rust-cache@v2.2.1
       - name: Run cargo check without any default features
         uses: actions-rs/cargo@v1
         with:
@@ -105,29 +105,6 @@ jobs:
           command: test
           args: --workspace --locked --release --all-features
 
-  test-disabled-tokenization:
-    name: Test disabled tokenization
-    runs-on: ubuntu-latest
-    container:
-      image: ubuntu:18.04
-    if: github.event_name == 'schedule'
-    steps:
-      - uses: actions/checkout@v3
-      - name: Install needed dependencies
-        run: |
-          apt-get update
-          apt-get install --assume-yes build-essential curl
-      - uses: actions-rs/toolchain@v1 
-        with: 
-          toolchain: stable 
-          override: true 
-      - name: Run cargo tree without default features and check lindera is not present
-        run: |
-          cargo tree -f '{p} {f}' -e normal --no-default-features | grep lindera -vqz
-      - name: Run cargo tree with default features and check lindera is pressent
-        run: |
-          cargo tree -f '{p} {f}' -e normal | grep lindera -qz
-                
   # We run tests in debug also, to make sure that the debug_assertions are hit
   test-debug:
     name: Run tests in debug
@@ -146,7 +123,7 @@ jobs:
           toolchain: stable
           override: true
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.4.0
+        uses: Swatinem/rust-cache@v2.2.1
       - name: Run tests in debug
         uses: actions-rs/cargo@v1
         with:
@@ -165,7 +142,7 @@ jobs:
           override: true
           components: clippy
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.4.0
+        uses: Swatinem/rust-cache@v2.2.1
       - name: Run cargo clippy
         uses: actions-rs/cargo@v1
         with:
@@ -184,7 +161,7 @@ jobs:
           override: true
           components: rustfmt
       - name: Cache dependencies
-        uses: Swatinem/rust-cache@v2.4.0
+        uses: Swatinem/rust-cache@v2.2.1
       - name: Run cargo fmt
         # Since we never ran the `build.rs` script in the benchmark directory we are missing one auto-generated import file.
         # Since we want to trigger (and fail) this action as fast as possible, instead of building the benchmark crate

Cargo.lock

@@ -1207,12 +1207,6 @@ dependencies = [
  "winapi",
 ]
 
-[[package]]
-name = "doc-comment"
-version = "0.3.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fea41bba32d969b513997752735605054bc0dfa92b4c56bf1189f2e174be7a10"
-
 [[package]]
 name = "dump"
 version = "1.2.0"
@@ -1769,15 +1763,6 @@ dependencies = [
  "byteorder",
 ]
 
-[[package]]
-name = "hashbrown"
-version = "0.11.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ab5ef0d4909ef3724cc8cce6ccc8572c5c817592e9285f5464f8e86f8bd3726e"
-dependencies = [
- "ahash 0.7.6",
-]
-
 [[package]]
 name = "hashbrown"
 version = "0.12.3"
@@ -1879,22 +1864,6 @@ dependencies = [
  "digest",
 ]
 
-[[package]]
-name = "hnsw"
-version = "0.11.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2b9740ebf8769ec4ad6762cc951ba18f39bba6dfbc2fbbe46285f7539af79752"
-dependencies = [
- "ahash 0.7.6",
- "hashbrown 0.11.2",
- "libm",
- "num-traits",
- "rand_core",
- "serde",
- "smallvec",
- "space",
-]
-
 [[package]]
 name = "http"
 version = "0.2.9"
@@ -2025,7 +1994,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "bd070e393353796e801d209ad339e89596eb4c8d430d18ede6a1cced8fafbd99"
 dependencies = [
  "autocfg",
- "hashbrown 0.12.3",
+ "hashbrown",
  "serde",
 ]
 
@@ -2117,7 +2086,7 @@ checksum = "37228e06c75842d1097432d94d02f37fe3ebfca9791c2e8fef6e9db17ed128c1"
 dependencies = [
  "cedarwood",
  "fxhash",
- "hashbrown 0.12.3",
+ "hashbrown",
  "lazy_static",
  "phf",
  "phf_codegen",
@@ -2627,7 +2596,6 @@ dependencies = [
  "num_cpus",
  "obkv",
  "once_cell",
- "ordered-float",
  "parking_lot",
  "permissive-json-pointer",
  "pin-project-lite",
@@ -2747,7 +2715,6 @@ dependencies = [
  "bimap",
  "bincode",
  "bstr",
- "bytemuck",
  "byteorder",
  "charabia",
  "concat-arrays",
@@ -2763,7 +2730,6 @@ dependencies = [
  "geoutils",
  "grenad",
  "heed",
- "hnsw",
  "insta",
  "itertools",
  "json-depth-checker",
@@ -2778,7 +2744,6 @@ dependencies = [
  "once_cell",
  "ordered-float",
  "rand",
- "rand_pcg",
  "rayon",
  "roaring",
  "rstar",
@@ -2788,7 +2753,6 @@ dependencies = [
  "smallstr",
  "smallvec",
  "smartstring",
- "space",
  "tempfile",
  "thiserror",
  "time",
@@ -2977,9 +2941,9 @@ checksum = "624a8340c38c1b80fd549087862da4ba43e08858af025b236e509b6649fc13d5"
 
 [[package]]
 name = "ordered-float"
-version = "3.7.0"
+version = "3.6.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2fc2dbde8f8a79f2102cc474ceb0ad68e3b80b85289ea62389b60e66777e4213"
+checksum = "13a384337e997e6860ffbaa83708b2ef329fd8c54cb67a5f64d421e0f943254f"
 dependencies = [
  "num-traits",
 ]
@@ -3363,16 +3327,6 @@ dependencies = [
  "getrandom",
 ]
 
-[[package]]
-name = "rand_pcg"
-version = "0.3.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "59cad018caf63deb318e5a4586d99a24424a364f40f1e5778c29aca23f4fc73e"
-dependencies = [
- "rand_core",
- "serde",
-]
-
 [[package]]
 name = "rayon"
 version = "1.7.0"
@@ -3810,9 +3764,6 @@ name = "smallvec"
 version = "1.10.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a507befe795404456341dfab10cef66ead4c041f62b8b11bbb92bffe5d0953e0"
-dependencies = [
- "serde",
-]
 
 [[package]]
 name = "smartstring"
@@ -3835,16 +3786,6 @@ dependencies = [
  "winapi",
 ]
 
-[[package]]
-name = "space"
-version = "0.17.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "c5ab9701ae895386d13db622abf411989deff7109b13b46b6173bb4ce5c1d123"
-dependencies = [
- "doc-comment",
- "num-traits",
-]
-
 [[package]]
 name = "spin"
 version = "0.5.2"
@@ -4492,7 +4433,7 @@ version = "0.16.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9c531a2dc4c462b833788be2c07eef4e621d0e9edbd55bf280cc164c1c1aa043"
 dependencies = [
- "hashbrown 0.12.3",
+ "hashbrown",
  "once_cell",
 ]
 

Dockerfile

@@ -1,3 +1,4 @@
+# syntax=docker/dockerfile:1.4
 # Compile
 FROM    rust:alpine3.16 AS compiler
 
@@ -11,7 +12,7 @@ ARG     GIT_TAG
 ENV     VERGEN_GIT_SHA=${COMMIT_SHA} VERGEN_GIT_COMMIT_TIMESTAMP=${COMMIT_DATE} VERGEN_GIT_SEMVER_LIGHTWEIGHT=${GIT_TAG}
 ENV     RUSTFLAGS="-C target-feature=-crt-static"
 
-COPY    . .
+COPY    --link . .
 RUN     set -eux; \
         apkArch="$(apk --print-arch)"; \
         if [ "$apkArch" = "aarch64" ]; then \
@@ -30,7 +31,7 @@ RUN     apk update --quiet \
 
 # add meilisearch to the `/bin` so you can run it from anywhere and it's easy
 # to find.
-COPY    --from=compiler /meilisearch/target/release/meilisearch /bin/meilisearch
+COPY    --from=compiler --link /meilisearch/target/release/meilisearch /bin/meilisearch
 # To stay compatible with the older version of the container (pre v0.27.0) we're
 # going to symlink the meilisearch binary in the path to `/meilisearch`
 RUN     ln -s /bin/meilisearch /meilisearch

assets/prometheus-basic-scraper.yml

@@ -1,19 +0,0 @@
-global:
-  scrape_interval:     15s # By default, scrape targets every 15 seconds.
-
-  # Attach these labels to any time series or alerts when communicating with
-  # external systems (federation, remote storage, Alertmanager).
-  external_labels:
-    monitor: 'codelab-monitor'
-
-# A scrape configuration containing exactly one endpoint to scrape:
-# Here it's Prometheus itself.
-scrape_configs:
-  # The job name is added as a label `job=<job_name>` to any timeseries scraped from this config.
-  - job_name: 'meilisearch'
-
-    # Override the global default and scrape targets from this job every 5 seconds.
-    scrape_interval: 5s
-
-    static_configs:
-      - targets: ['localhost:7700']

config.toml

@@ -1,131 +1,131 @@
 # This file shows the default configuration of Meilisearch.
 # All variables are defined here: https://www.meilisearch.com/docs/learn/configuration/instance_options#environment-variables
 
+db_path = "./data.ms"
 # Designates the location where database files will be created and retrieved.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#database-path
-db_path = "./data.ms"
 
+env = "development"
 # Configures the instance's environment. Value must be either `production` or `development`.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#environment
-env = "development"
 
-# The address on which the HTTP server will listen.
 http_addr = "localhost:7700"
+# The address on which the HTTP server will listen.
 
+# master_key = "YOUR_MASTER_KEY_VALUE"
 # Sets the instance's master key, automatically protecting all routes except GET /health.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#master-key
-# master_key = "YOUR_MASTER_KEY_VALUE"
 
+# no_analytics = true
 # Deactivates Meilisearch's built-in telemetry when provided.
 # Meilisearch automatically collects data from all instances that do not opt out using this flag.
 # All gathered data is used solely for the purpose of improving Meilisearch, and can be deleted at any time.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#disable-analytics
-# no_analytics = true
 
+http_payload_size_limit = "100 MB"
 # Sets the maximum size of accepted payloads.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#payload-limit-size
-http_payload_size_limit = "100 MB"
 
+log_level = "INFO"
 # Defines how much detail should be present in Meilisearch's logs.
 # Meilisearch currently supports six log levels, listed in order of increasing verbosity:  `OFF`, `ERROR`, `WARN`, `INFO`, `DEBUG`, `TRACE`
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#log-level
-log_level = "INFO"
 
+# max_indexing_memory = "2 GiB"
 # Sets the maximum amount of RAM Meilisearch can use when indexing.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#max-indexing-memory
-# max_indexing_memory = "2 GiB"
 
+# max_indexing_threads = 4
 # Sets the maximum number of threads Meilisearch can use during indexing.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#max-indexing-threads
-# max_indexing_threads = 4
 
 #############
 ### DUMPS ###
 #############
 
+dump_dir = "dumps/"
 # Sets the directory where Meilisearch will create dump files.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#dump-directory
-dump_dir = "dumps/"
 
+# import_dump = "./path/to/my/file.dump"
 # Imports the dump file located at the specified path. Path must point to a .dump file.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#import-dump
-# import_dump = "./path/to/my/file.dump"
 
+ignore_missing_dump = false
 # Prevents Meilisearch from throwing an error when `import_dump` does not point to a valid dump file.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ignore-missing-dump
-ignore_missing_dump = false
 
+ignore_dump_if_db_exists = false
 # Prevents a Meilisearch instance with an existing database from throwing an error when using `import_dump`.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ignore-dump-if-db-exists
-ignore_dump_if_db_exists = false
 
 
 #################
 ### SNAPSHOTS ###
 #################
 
+schedule_snapshot = false
 # Enables scheduled snapshots when true, disable when false (the default).
 # If the value is given as an integer, then enables the scheduled snapshot with the passed value as the interval
 # between each snapshot, in seconds.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#schedule-snapshot-creation
-schedule_snapshot = false
 
+snapshot_dir = "snapshots/"
 # Sets the directory where Meilisearch will store snapshots.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#snapshot-destination
-snapshot_dir = "snapshots/"
 
+# import_snapshot = "./path/to/my/snapshot"
 # Launches Meilisearch after importing a previously-generated snapshot at the given filepath.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#import-snapshot
-# import_snapshot = "./path/to/my/snapshot"
 
+ignore_missing_snapshot = false
 # Prevents a Meilisearch instance from throwing an error when `import_snapshot` does not point to a valid snapshot file.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ignore-missing-snapshot
-ignore_missing_snapshot = false
 
+ignore_snapshot_if_db_exists = false
 # Prevents a Meilisearch instance with an existing database from throwing an error when using `import_snapshot`.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ignore-snapshot-if-db-exists
-ignore_snapshot_if_db_exists = false
 
 
 ###########
 ### SSL ###
 ###########
 
+# ssl_auth_path = "./path/to/root"
 # Enables client authentication in the specified path.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ssl-authentication-path
-# ssl_auth_path = "./path/to/root"
 
+# ssl_cert_path = "./path/to/certfile"
 # Sets the server's SSL certificates.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ssl-certificates-path
-# ssl_cert_path = "./path/to/certfile"
 
+# ssl_key_path = "./path/to/private-key"
 # Sets the server's SSL key files.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ssl-key-path
-# ssl_key_path = "./path/to/private-key"
 
+# ssl_ocsp_path = "./path/to/ocsp-file"
 # Sets the server's OCSP file.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ssl-ocsp-path
-# ssl_ocsp_path = "./path/to/ocsp-file"
 
+ssl_require_auth = false
 # Makes SSL authentication mandatory.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ssl-require-auth
-ssl_require_auth = false
 
+ssl_resumption = false
 # Activates SSL session resumption.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ssl-resumption
-ssl_resumption = false
 
+ssl_tickets = false
 # Activates SSL tickets.
 # https://www.meilisearch.com/docs/learn/configuration/instance_options#ssl-tickets
-ssl_tickets = false
 
 #############################
 ### Experimental features ###
 #############################
 
+experimental_enable_metrics = false
 # Experimental metrics feature. For more information, see: <https://github.com/meilisearch/meilisearch/discussions/3518>
 # Enables the Prometheus metrics on the `GET /metrics` endpoint.
-experimental_enable_metrics = false
 
-# Experimental RAM reduction during indexing, do not use in production, see: <https://github.com/meilisearch/product/discussions/652>
 experimental_reduce_indexing_memory_usage = false
+# Experimental RAM reduction during indexing, do not use in production, see: <https://github.com/meilisearch/product/discussions/652>

index-scheduler/src/index_mapper/mod.rs

@@ -90,17 +90,8 @@ pub enum IndexStatus {
 pub struct IndexStats {
     /// Number of documents in the index.
     pub number_of_documents: u64,
-    /// Size taken up by the index' DB, in bytes.
-    ///
-    /// This includes the size taken by both the used and free pages of the DB, and as the free pages
-    /// are not returned to the disk after a deletion, this number is typically larger than
-    /// `used_database_size` that only includes the size of the used pages.
+    /// Size of the index' DB, in bytes.
     pub database_size: u64,
-    /// Size taken by the used pages of the index' DB, in bytes.
-    ///
-    /// As the DB backend does not return to the disk the pages that are not currently used by the DB,
-    /// this value is typically smaller than `database_size`.
-    pub used_database_size: u64,
     /// Association of every field name with the number of times it occurs in the documents.
     pub field_distribution: FieldDistribution,
     /// Creation date of the index.
@@ -116,10 +107,10 @@ impl IndexStats {
     ///
     /// - rtxn: a RO transaction for the index, obtained from `Index::read_txn()`.
     pub fn new(index: &Index, rtxn: &RoTxn) -> Result<Self> {
+        let database_size = index.on_disk_size()?;
         Ok(IndexStats {
             number_of_documents: index.number_of_documents(rtxn)?,
-            database_size: index.on_disk_size()?,
-            used_database_size: index.used_size()?,
+            database_size,
             field_distribution: index.field_distribution(rtxn)?,
             created_at: index.created_at(rtxn)?,
             updated_at: index.updated_at(rtxn)?,

index-scheduler/src/lib.rs

@@ -31,7 +31,7 @@ mod uuid_codec;
 pub type Result<T> = std::result::Result<T, Error>;
 pub type TaskId = u32;
 
-use std::collections::{BTreeMap, HashMap};
+use std::collections::HashMap;
 use std::ops::{Bound, RangeBounds};
 use std::path::{Path, PathBuf};
 use std::sync::atomic::AtomicBool;
@@ -573,16 +573,10 @@ impl IndexScheduler {
         &self.index_mapper.indexer_config
     }
 
-    /// Return the real database size (i.e.: The size **with** the free pages)
     pub fn size(&self) -> Result<u64> {
         Ok(self.env.real_disk_size()?)
     }
 
-    /// Return the used database size (i.e.: The size **without** the free pages)
-    pub fn used_size(&self) -> Result<u64> {
-        Ok(self.env.non_free_pages_size()?)
-    }
-
     /// Return the index corresponding to the name.
     ///
     /// * If the index wasn't opened before, the index will be opened.
@@ -762,38 +756,6 @@ impl IndexScheduler {
         Ok(tasks)
     }
 
-    /// The returned structure contains:
-    /// 1. The name of the property being observed can be `statuses`, `types`, or `indexes`.
-    /// 2. The name of the specific data related to the property can be `enqueued` for the `statuses`, `settingsUpdate` for the `types`, or the name of the index for the `indexes`, for example.
-    /// 3. The number of times the properties appeared.
-    pub fn get_stats(&self) -> Result<BTreeMap<String, BTreeMap<String, u64>>> {
-        let rtxn = self.read_txn()?;
-
-        let mut res = BTreeMap::new();
-
-        res.insert(
-            "statuses".to_string(),
-            enum_iterator::all::<Status>()
-                .map(|s| Ok((s.to_string(), self.get_status(&rtxn, s)?.len())))
-                .collect::<Result<BTreeMap<String, u64>>>()?,
-        );
-        res.insert(
-            "types".to_string(),
-            enum_iterator::all::<Kind>()
-                .map(|s| Ok((s.to_string(), self.get_kind(&rtxn, s)?.len())))
-                .collect::<Result<BTreeMap<String, u64>>>()?,
-        );
-        res.insert(
-            "indexes".to_string(),
-            self.index_tasks
-                .iter(&rtxn)?
-                .map(|res| Ok(res.map(|(name, bitmap)| (name.to_string(), bitmap.len()))?))
-                .collect::<Result<BTreeMap<String, u64>>>()?,
-        );
-
-        Ok(res)
-    }
-
     /// Return true iff there is at least one task associated with this index
     /// that is processing.
     pub fn is_index_processing(&self, index: &str) -> Result<bool> {

index-scheduler/src/utils.rs

@@ -466,7 +466,7 @@ impl IndexScheduler {
                         }
                     }
                     Details::DocumentDeletionByFilter { deleted_documents, original_filter: _ } => {
-                        assert_eq!(kind.as_kind(), Kind::DocumentDeletion);
+                        assert_eq!(kind.as_kind(), Kind::DocumentDeletionByFilter);
                         let (index_uid, _) = if let KindWithContent::DocumentDeletionByFilter {
                             ref index_uid,
                             ref filter_expr,

meilisearch-auth/src/lib.rs

@@ -45,11 +45,6 @@ impl AuthController {
         self.store.size()
     }
 
-    /// Return the used size of the `AuthController` database in bytes.
-    pub fn used_size(&self) -> Result<u64> {
-        self.store.used_size()
-    }
-
     pub fn create_key(&self, create_key: CreateApiKey) -> Result<Key> {
         match self.store.get_api_key(create_key.uid)? {
             Some(_) => Err(AuthControllerError::ApiKeyAlreadyExists(create_key.uid.to_string())),

meilisearch-auth/src/store.rs

@@ -75,11 +75,6 @@ impl HeedAuthStore {
         Ok(self.env.real_disk_size()?)
     }
 
-    /// Return the number of bytes actually used in the database
-    pub fn used_size(&self) -> Result<u64> {
-        Ok(self.env.non_free_pages_size()?)
-    }
-
     pub fn set_drop_on_close(&mut self, v: bool) {
         self.should_close_on_drop = v;
     }

meilisearch-types/src/error.rs

@@ -217,8 +217,6 @@ InvalidDocumentFields                 , InvalidRequest       , BAD_REQUEST ;
 MissingDocumentFilter                 , InvalidRequest       , BAD_REQUEST ;
 InvalidDocumentFilter                 , InvalidRequest       , BAD_REQUEST ;
 InvalidDocumentGeoField               , InvalidRequest       , BAD_REQUEST ;
-InvalidVectorDimensions               , InvalidRequest       , BAD_REQUEST ;
-InvalidVectorsType                    , InvalidRequest       , BAD_REQUEST ;
 InvalidDocumentId                     , InvalidRequest       , BAD_REQUEST ;
 InvalidDocumentLimit                  , InvalidRequest       , BAD_REQUEST ;
 InvalidDocumentOffset                 , InvalidRequest       , BAD_REQUEST ;
@@ -334,8 +332,6 @@ impl ErrorCode for milli::Error {
                     UserError::InvalidSortableAttribute { .. } => Code::InvalidSearchSort,
                     UserError::CriterionError(_) => Code::InvalidSettingsRankingRules,
                     UserError::InvalidGeoField { .. } => Code::InvalidDocumentGeoField,
-                    UserError::InvalidVectorDimensions { .. } => Code::InvalidVectorDimensions,
-                    UserError::InvalidVectorsType { .. } => Code::InvalidVectorsType,
                     UserError::SortError(_) => Code::InvalidSearchSort,
                     UserError::InvalidMinTypoWordLenSetting(_, _) => {
                         Code::InvalidSettingsTypoTolerance

meilisearch-types/src/tasks.rs

@@ -395,6 +395,7 @@ impl std::error::Error for ParseTaskStatusError {}
 pub enum Kind {
     DocumentAdditionOrUpdate,
     DocumentDeletion,
+    DocumentDeletionByFilter,
     SettingsUpdate,
     IndexCreation,
     IndexDeletion,
@@ -411,6 +412,7 @@ impl Kind {
         match self {
             Kind::DocumentAdditionOrUpdate
             | Kind::DocumentDeletion
+            | Kind::DocumentDeletionByFilter
             | Kind::SettingsUpdate
             | Kind::IndexCreation
             | Kind::IndexDeletion
@@ -428,6 +430,7 @@ impl Display for Kind {
         match self {
             Kind::DocumentAdditionOrUpdate => write!(f, "documentAdditionOrUpdate"),
             Kind::DocumentDeletion => write!(f, "documentDeletion"),
+            Kind::DocumentDeletionByFilter => write!(f, "documentDeletionByFilter"),
             Kind::SettingsUpdate => write!(f, "settingsUpdate"),
             Kind::IndexCreation => write!(f, "indexCreation"),
             Kind::IndexDeletion => write!(f, "indexDeletion"),

meilisearch/Cargo.toml

@@ -48,7 +48,6 @@ mime = "0.3.17"
 num_cpus = "1.15.0"
 obkv = "0.2.0"
 once_cell = "1.17.1"
-ordered-float = "3.7.0"
 parking_lot = "0.12.1"
 permissive-json-pointer = { path = "../permissive-json-pointer" }
 pin-project-lite = "0.2.9"

meilisearch/src/metrics.rs

@@ -4,32 +4,20 @@ use prometheus::{
     register_int_gauge_vec, HistogramVec, IntCounterVec, IntGauge, IntGaugeVec,
 };
 
-/// Create evenly distributed buckets
-fn create_buckets() -> [f64; 29] {
-    (0..10)
-        .chain((10..100).step_by(10))
-        .chain((100..=1000).step_by(100))
-        .map(|i| i as f64 / 1000.)
-        .collect::<Vec<_>>()
-        .try_into()
-        .unwrap()
-}
+const HTTP_RESPONSE_TIME_CUSTOM_BUCKETS: &[f64; 14] = &[
+    0.0005, 0.0008, 0.00085, 0.0009, 0.00095, 0.001, 0.00105, 0.0011, 0.00115, 0.0012, 0.0015,
+    0.002, 0.003, 1.0,
+];
 
 lazy_static! {
-    pub static ref HTTP_RESPONSE_TIME_CUSTOM_BUCKETS: [f64; 29] = create_buckets();
-    pub static ref MEILISEARCH_HTTP_REQUESTS_TOTAL: IntCounterVec = register_int_counter_vec!(
-        opts!("meilisearch_http_requests_total", "Meilisearch HTTP requests total"),
+    pub static ref HTTP_REQUESTS_TOTAL: IntCounterVec = register_int_counter_vec!(
+        opts!("http_requests_total", "HTTP requests total"),
         &["method", "path"]
     )
     .expect("Can't create a metric");
     pub static ref MEILISEARCH_DB_SIZE_BYTES: IntGauge =
-        register_int_gauge!(opts!("meilisearch_db_size_bytes", "Meilisearch DB Size In Bytes"))
+        register_int_gauge!(opts!("meilisearch_db_size_bytes", "Meilisearch Db Size In Bytes"))
             .expect("Can't create a metric");
-    pub static ref MEILISEARCH_USED_DB_SIZE_BYTES: IntGauge = register_int_gauge!(opts!(
-        "meilisearch_used_db_size_bytes",
-        "Meilisearch Used DB Size In Bytes"
-    ))
-    .expect("Can't create a metric");
     pub static ref MEILISEARCH_INDEX_COUNT: IntGauge =
         register_int_gauge!(opts!("meilisearch_index_count", "Meilisearch Index Count"))
             .expect("Can't create a metric");
@@ -38,16 +26,11 @@ lazy_static! {
         &["index"]
     )
     .expect("Can't create a metric");
-    pub static ref MEILISEARCH_HTTP_RESPONSE_TIME_SECONDS: HistogramVec = register_histogram_vec!(
+    pub static ref HTTP_RESPONSE_TIME_SECONDS: HistogramVec = register_histogram_vec!(
         "http_response_time_seconds",
         "HTTP response times",
         &["method", "path"],
         HTTP_RESPONSE_TIME_CUSTOM_BUCKETS.to_vec()
     )
     .expect("Can't create a metric");
-    pub static ref MEILISEARCH_NB_TASKS: IntGaugeVec = register_int_gauge_vec!(
-        opts!("meilisearch_nb_tasks", "Meilisearch Number of tasks"),
-        &["kind", "value"]
-    )
-    .expect("Can't create a metric");
 }

meilisearch/src/middleware.rs

@@ -52,11 +52,11 @@ where
         if is_registered_resource {
             let request_method = req.method().to_string();
             histogram_timer = Some(
-                crate::metrics::MEILISEARCH_HTTP_RESPONSE_TIME_SECONDS
+                crate::metrics::HTTP_RESPONSE_TIME_SECONDS
                     .with_label_values(&[&request_method, request_path])
                     .start_timer(),
             );
-            crate::metrics::MEILISEARCH_HTTP_REQUESTS_TOTAL
+            crate::metrics::HTTP_REQUESTS_TOTAL
                 .with_label_values(&[&request_method, request_path])
                 .inc();
         }

meilisearch/src/routes/indexes/search.rs

@@ -34,8 +34,6 @@ pub fn configure(cfg: &mut web::ServiceConfig) {
 pub struct SearchQueryGet {
     #[deserr(default, error = DeserrQueryParamError<InvalidSearchQ>)]
     q: Option<String>,
-    #[deserr(default, error = DeserrQueryParamError<InvalidSearchQ>)]
-    vector: Option<Vec<f32>>,
     #[deserr(default = Param(DEFAULT_SEARCH_OFFSET()), error = DeserrQueryParamError<InvalidSearchOffset>)]
     offset: Param<usize>,
     #[deserr(default = Param(DEFAULT_SEARCH_LIMIT()), error = DeserrQueryParamError<InvalidSearchLimit>)]
@@ -82,7 +80,6 @@ impl From<SearchQueryGet> for SearchQuery {
 
         Self {
             q: other.q,
-            vector: other.vector,
             offset: other.offset.0,
             limit: other.limit.0,
             page: other.page.as_deref().copied(),

meilisearch/src/routes/metrics.rs

@@ -17,7 +17,7 @@ pub fn configure(config: &mut web::ServiceConfig) {
 
 pub async fn get_metrics(
     index_scheduler: GuardedData<ActionPolicy<{ actions::METRICS_GET }>, Data<IndexScheduler>>,
-    auth_controller: Data<AuthController>,
+    auth_controller: GuardedData<ActionPolicy<{ actions::METRICS_GET }>, Data<AuthController>>,
 ) -> Result<HttpResponse, ResponseError> {
     let auth_filters = index_scheduler.filters();
     if !auth_filters.all_indexes_authorized() {
@@ -28,10 +28,10 @@ pub async fn get_metrics(
         return Err(error);
     }
 
-    let response = create_all_stats((*index_scheduler).clone(), auth_controller, auth_filters)?;
+    let response =
+        create_all_stats((*index_scheduler).clone(), (*auth_controller).clone(), auth_filters)?;
 
     crate::metrics::MEILISEARCH_DB_SIZE_BYTES.set(response.database_size as i64);
-    crate::metrics::MEILISEARCH_USED_DB_SIZE_BYTES.set(response.used_database_size as i64);
     crate::metrics::MEILISEARCH_INDEX_COUNT.set(response.indexes.len() as i64);
 
     for (index, value) in response.indexes.iter() {
@@ -40,14 +40,6 @@ pub async fn get_metrics(
             .set(value.number_of_documents as i64);
     }
 
-    for (kind, value) in index_scheduler.get_stats()? {
-        for (value, count) in value {
-            crate::metrics::MEILISEARCH_NB_TASKS
-                .with_label_values(&[&kind, &value])
-                .set(count as i64);
-        }
-    }
-
     let encoder = TextEncoder::new();
     let mut buffer = vec![];
     encoder.encode(&prometheus::gather(), &mut buffer).expect("Failed to encode metrics");

meilisearch/src/routes/mod.rs

@@ -231,8 +231,6 @@ pub async fn running() -> HttpResponse {
 #[serde(rename_all = "camelCase")]
 pub struct Stats {
     pub database_size: u64,
-    #[serde(skip)]
-    pub used_database_size: u64,
     #[serde(serialize_with = "time::serde::rfc3339::option::serialize")]
     pub last_update: Option<OffsetDateTime>,
     pub indexes: BTreeMap<String, indexes::IndexStats>,
@@ -261,7 +259,6 @@ pub fn create_all_stats(
     let mut last_task: Option<OffsetDateTime> = None;
     let mut indexes = BTreeMap::new();
     let mut database_size = 0;
-    let mut used_database_size = 0;
 
     for index_uid in index_scheduler.index_names()? {
         // Accumulate the size of all indexes, even unauthorized ones, so
@@ -269,7 +266,6 @@ pub fn create_all_stats(
         // See <https://github.com/meilisearch/meilisearch/pull/3541#discussion_r1126747643> for context.
         let stats = index_scheduler.index_stats(&index_uid)?;
         database_size += stats.inner_stats.database_size;
-        used_database_size += stats.inner_stats.used_database_size;
 
         if !filters.is_index_authorized(&index_uid) {
             continue;
@@ -282,14 +278,10 @@ pub fn create_all_stats(
     }
 
     database_size += index_scheduler.size()?;
-    used_database_size += index_scheduler.used_size()?;
     database_size += auth_controller.size()?;
-    used_database_size += auth_controller.used_size()?;
-    let update_file_size = index_scheduler.compute_update_file_size()?;
-    database_size += update_file_size;
-    used_database_size += update_file_size;
+    database_size += index_scheduler.compute_update_file_size()?;
 
-    let stats = Stats { database_size, used_database_size, last_update: last_task, indexes };
+    let stats = Stats { database_size, last_update: last_task, indexes };
     Ok(stats)
 }
 

meilisearch/src/routes/tasks.rs

@@ -730,7 +730,7 @@ mod tests {
             let err = deserr_query_params::<TaskDeletionOrCancelationQuery>(params).unwrap_err();
             snapshot!(meili_snap::json_string!(err), @r###"
             {
-              "message": "Invalid value in parameter `types`: `createIndex` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentDeletion`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`.",
+              "message": "Invalid value in parameter `types`: `createIndex` is not a valid task type. Available types are `documentAdditionOrUpdate`, `documentDeletion`, `documentDeletionByFilter`, `settingsUpdate`, `indexCreation`, `indexDeletion`, `indexUpdate`, `indexSwap`, `taskCancelation`, `taskDeletion`, `dumpCreation`, `snapshotCreation`.",
               "code": "invalid_task_types",
               "type": "invalid_request",
               "link": "https://docs.meilisearch.com/errors#invalid_task_types"

meilisearch/src/search.rs

@@ -9,15 +9,13 @@ use meilisearch_auth::IndexSearchRules;
 use meilisearch_types::deserr::DeserrJsonError;
 use meilisearch_types::error::deserr_codes::*;
 use meilisearch_types::index_uid::IndexUid;
-use meilisearch_types::milli::dot_product_similarity;
 use meilisearch_types::settings::DEFAULT_PAGINATION_MAX_TOTAL_HITS;
 use meilisearch_types::{milli, Document};
 use milli::tokenizer::TokenizerBuilder;
 use milli::{
     AscDesc, FieldId, FieldsIdsMap, Filter, FormatOptions, Index, MatchBounds, MatcherBuilder,
-    SortError, TermsMatchingStrategy, VectorOrArrayOfVectors, DEFAULT_VALUES_PER_FACET,
+    SortError, TermsMatchingStrategy, DEFAULT_VALUES_PER_FACET,
 };
-use ordered_float::OrderedFloat;
 use regex::Regex;
 use serde::Serialize;
 use serde_json::{json, Value};
@@ -33,13 +31,11 @@ pub const DEFAULT_CROP_MARKER: fn() -> String = || "…".to_string();
 pub const DEFAULT_HIGHLIGHT_PRE_TAG: fn() -> String = || "<em>".to_string();
 pub const DEFAULT_HIGHLIGHT_POST_TAG: fn() -> String = || "</em>".to_string();
 
-#[derive(Debug, Clone, Default, PartialEq, Deserr)]
+#[derive(Debug, Clone, Default, PartialEq, Eq, Deserr)]
 #[deserr(error = DeserrJsonError, rename_all = camelCase, deny_unknown_fields)]
 pub struct SearchQuery {
     #[deserr(default, error = DeserrJsonError<InvalidSearchQ>)]
     pub q: Option<String>,
-    #[deserr(default, error = DeserrJsonError<InvalidSearchQ>)]
-    pub vector: Option<Vec<f32>>,
     #[deserr(default = DEFAULT_SEARCH_OFFSET(), error = DeserrJsonError<InvalidSearchOffset>)]
     pub offset: usize,
     #[deserr(default = DEFAULT_SEARCH_LIMIT(), error = DeserrJsonError<InvalidSearchLimit>)]
@@ -84,15 +80,13 @@ impl SearchQuery {
 // This struct contains the fields of `SearchQuery` inline.
 // This is because neither deserr nor serde support `flatten` when using `deny_unknown_fields.
 // The `From<SearchQueryWithIndex>` implementation ensures both structs remain up to date.
-#[derive(Debug, Clone, PartialEq, Deserr)]
+#[derive(Debug, Clone, PartialEq, Eq, Deserr)]
 #[deserr(error = DeserrJsonError, rename_all = camelCase, deny_unknown_fields)]
 pub struct SearchQueryWithIndex {
     #[deserr(error = DeserrJsonError<InvalidIndexUid>, missing_field_error = DeserrJsonError::missing_index_uid)]
     pub index_uid: IndexUid,
     #[deserr(default, error = DeserrJsonError<InvalidSearchQ>)]
     pub q: Option<String>,
-    #[deserr(default, error = DeserrJsonError<InvalidSearchQ>)]
-    pub vector: Option<Vec<f32>>,
     #[deserr(default = DEFAULT_SEARCH_OFFSET(), error = DeserrJsonError<InvalidSearchOffset>)]
     pub offset: usize,
     #[deserr(default = DEFAULT_SEARCH_LIMIT(), error = DeserrJsonError<InvalidSearchLimit>)]
@@ -132,7 +126,6 @@ impl SearchQueryWithIndex {
         let SearchQueryWithIndex {
             index_uid,
             q,
-            vector,
             offset,
             limit,
             page,
@@ -154,7 +147,6 @@ impl SearchQueryWithIndex {
             index_uid,
             SearchQuery {
                 q,
-                vector,
                 offset,
                 limit,
                 page,
@@ -217,8 +209,6 @@ pub struct SearchHit {
 pub struct SearchResult {
     pub hits: Vec<SearchHit>,
     pub query: String,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub vector: Option<Vec<f32>>,
     pub processing_time_ms: u128,
     #[serde(flatten)]
     pub hits_info: HitsInfo,
@@ -280,10 +270,6 @@ pub fn perform_search(
 
     let mut search = index.search(&rtxn);
 
-    if let Some(ref vector) = query.vector {
-        search.vector(vector.clone());
-    }
-
     if let Some(ref query) = query.q {
         search.query(query);
     }
@@ -403,6 +389,7 @@ pub fn perform_search(
     formatter_builder.highlight_suffix(query.highlight_post_tag);
 
     let mut documents = Vec::new();
+
     let documents_iter = index.documents(&rtxn, documents_ids)?;
 
     for (_id, obkv) in documents_iter {
@@ -429,12 +416,6 @@ pub fn perform_search(
             insert_geo_distance(sort, &mut document);
         }
 
-        if let Some(vector) = query.vector.as_ref() {
-            if let Some(vectors) = extract_field("_vectors", &fields_ids_map, obkv)? {
-                insert_semantic_similarity(vector, vectors, &mut document);
-            }
-        }
-
         let hit = SearchHit { document, formatted, matches_position };
         documents.push(hit);
     }
@@ -484,8 +465,7 @@ pub fn perform_search(
     let result = SearchResult {
         hits: documents,
         hits_info,
-        query: query.q.unwrap_or_default(),
-        vector: query.vector,
+        query: query.q.clone().unwrap_or_default(),
         processing_time_ms: before_search.elapsed().as_millis(),
         facet_distribution,
         facet_stats,
@@ -509,20 +489,6 @@ fn insert_geo_distance(sorts: &[String], document: &mut Document) {
     }
 }
 
-fn insert_semantic_similarity(query: &[f32], vectors: Value, document: &mut Document) {
-    let vectors =
-        match serde_json::from_value(vectors).map(VectorOrArrayOfVectors::into_array_of_vectors) {
-            Ok(vectors) => vectors,
-            Err(_) => return,
-        };
-    let similarity = vectors
-        .into_iter()
-        .map(|v| OrderedFloat(dot_product_similarity(query, &v)))
-        .max()
-        .map(OrderedFloat::into_inner);
-    document.insert("_semanticSimilarity".to_string(), json!(similarity));
-}
-
 fn compute_formatted_options(
     attr_to_highlight: &HashSet<String>,
     attr_to_crop: &[String],
@@ -650,22 +616,6 @@ fn make_document(
     Ok(document)
 }
 
-/// Extract the JSON value under the field name specified
-/// but doesn't support nested objects.
-fn extract_field(
-    field_name: &str,
-    field_ids_map: &FieldsIdsMap,
-    obkv: obkv::KvReaderU16,
-) -> Result<Option<serde_json::Value>, MeilisearchHttpError> {
-    match field_ids_map.id(field_name) {
-        Some(fid) => match obkv.get(fid) {
-            Some(value) => Ok(serde_json::from_slice(value).map(Some)?),
-            None => Ok(None),
-        },
-        None => Ok(None),
-    }
-}
-
 fn format_fields<A: AsRef<[u8]>>(
     document: &Document,
     field_ids_map: &FieldsIdsMap,

meilisearch/tests/tasks/errors.rs

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

milli/Cargo.toml

@@ -15,7 +15,6 @@ license.workspace = true
 bimap = { version = "0.6.3", features = ["serde"] }
 bincode = "1.3.3"
 bstr = "1.4.0"
-bytemuck = { version = "1.13.1", features = ["extern_crate_alloc"] }
 byteorder = "1.4.3"
 charabia = { version = "0.7.2", default-features = false }
 concat-arrays = "0.1.2"
@@ -33,21 +32,18 @@ heed = { git = "https://github.com/meilisearch/heed", tag = "v0.12.6", default-f
     "lmdb",
     "sync-read-txn",
 ] }
-hnsw = { version = "0.11.0", features = ["serde1"] }
 json-depth-checker = { path = "../json-depth-checker" }
 levenshtein_automata = { version = "0.2.1", features = ["fst_automaton"] }
 memmap2 = "0.5.10"
 obkv = "0.2.0"
 once_cell = "1.17.1"
 ordered-float = "3.6.0"
-rand_pcg = { version = "0.3.1", features = ["serde1"] }
 rayon = "1.7.0"
 roaring = "0.10.1"
 rstar = { version = "0.10.0", features = ["serde"] }
 serde = { version = "1.0.160", features = ["derive"] }
 serde_json = { version = "1.0.95", features = ["preserve_order"] }
 slice-group-by = "0.3.0"
-space = "0.17.0"
 smallstr = { version = "0.3.0", features = ["serde"] }
 smallvec = "1.10.0"
 smartstring = "1.0.1"

milli/examples/search.rs

@@ -52,7 +52,6 @@ fn main() -> Result<(), Box<dyn Error>> {
             let docs = execute_search(
                 &mut ctx,
                 &(!query.trim().is_empty()).then(|| query.trim().to_owned()),
-                &None,
                 TermsMatchingStrategy::Last,
                 false,
                 &None,

milli/src/distance.rs

@@ -1,44 +0,0 @@
-use serde::{Deserialize, Serialize};
-use space::Metric;
-
-#[derive(Debug, Default, Clone, Copy, Serialize, Deserialize)]
-pub struct DotProduct;
-
-impl Metric<Vec<f32>> for DotProduct {
-    type Unit = u32;
-
-    // TODO explain me this function, I don't understand why f32.to_bits is ordered.
-    // I tried to do this and it wasn't OK <https://stackoverflow.com/a/43305015/1941280>
-    //
-    // Following <https://docs.rs/space/0.17.0/space/trait.Metric.html>.
-    fn distance(&self, a: &Vec<f32>, b: &Vec<f32>) -> Self::Unit {
-        let dist = 1.0 - dot_product_similarity(a, b);
-        debug_assert!(!dist.is_nan());
-        dist.to_bits()
-    }
-}
-
-/// Returns the dot product similarity score that will between 0.0 and 1.0
-/// if both vectors are normalized. The higher the more similar the vectors are.
-pub fn dot_product_similarity(a: &[f32], b: &[f32]) -> f32 {
-    a.iter().zip(b).map(|(a, b)| a * b).sum()
-}
-
-#[derive(Debug, Default, Clone, Copy, Serialize, Deserialize)]
-pub struct Euclidean;
-
-impl Metric<Vec<f32>> for Euclidean {
-    type Unit = u32;
-
-    fn distance(&self, a: &Vec<f32>, b: &Vec<f32>) -> Self::Unit {
-        let dist = euclidean_squared_distance(a, b).sqrt();
-        debug_assert!(!dist.is_nan());
-        dist.to_bits()
-    }
-}
-
-/// Return the squared euclidean distance between both vectors that will
-/// between 0.0 and +inf. The smaller the nearer the vectors are.
-pub fn euclidean_squared_distance(a: &[f32], b: &[f32]) -> f32 {
-    a.iter().zip(b).map(|(a, b)| (a - b).powi(2)).sum()
-}

milli/src/error.rs

@@ -110,13 +110,9 @@ only composed of alphanumeric characters (a-z A-Z 0-9), hyphens (-) and undersco
     },
     #[error(transparent)]
     InvalidGeoField(#[from] GeoError),
-    #[error("Invalid vector dimensions: expected: `{}`, found: `{}`.", .expected, .found)]
-    InvalidVectorDimensions { expected: usize, found: usize },
-    #[error("The `_vectors` field in the document with the id: `{document_id}` is not an array. Was expecting an array of floats or an array of arrays of floats but instead got `{value}`.")]
-    InvalidVectorsType { document_id: Value, value: Value },
     #[error("{0}")]
     InvalidFilter(String),
-    #[error("Invalid type for filter subexpression: expected: {}, found: {1}.", .0.join(", "))]
+    #[error("Invalid type for filter subexpression: `expected {}, found: {1}`.", .0.join(", "))]
     InvalidFilterExpression(&'static [&'static str], Value),
     #[error("Attribute `{}` is not sortable. {}",
         .field,

milli/src/heed_codec/roaring_bitmap/cbo_roaring_bitmap_codec.rs

@@ -49,7 +49,7 @@ impl CboRoaringBitmapCodec {
         } else {
             // Otherwise, it means we used the classic RoaringBitmapCodec and
             // that the header takes threshold integers.
-            RoaringBitmap::deserialize_unchecked_from(bytes)
+            RoaringBitmap::deserialize_from(bytes)
         }
     }
 
@@ -69,7 +69,7 @@ impl CboRoaringBitmapCodec {
                     vec.push(integer);
                 }
             } else {
-                roaring |= RoaringBitmap::deserialize_unchecked_from(bytes.as_ref())?;
+                roaring |= RoaringBitmap::deserialize_from(bytes.as_ref())?;
             }
         }
 

milli/src/heed_codec/roaring_bitmap/roaring_bitmap_codec.rs

@@ -8,7 +8,7 @@ impl heed::BytesDecode<'_> for RoaringBitmapCodec {
     type DItem = RoaringBitmap;
 
     fn bytes_decode(bytes: &[u8]) -> Option<Self::DItem> {
-        RoaringBitmap::deserialize_unchecked_from(bytes).ok()
+        RoaringBitmap::deserialize_from(bytes).ok()
     }
 }
 

milli/src/index.rs

@@ -8,12 +8,10 @@ use charabia::{Language, Script};
 use heed::flags::Flags;
 use heed::types::*;
 use heed::{CompactionOption, Database, PolyDatabase, RoTxn, RwTxn};
-use rand_pcg::Pcg32;
 use roaring::RoaringBitmap;
 use rstar::RTree;
 use time::OffsetDateTime;
 
-use crate::distance::DotProduct;
 use crate::error::{InternalError, UserError};
 use crate::facet::FacetType;
 use crate::fields_ids_map::FieldsIdsMap;
@@ -23,14 +21,12 @@ use crate::heed_codec::facet::{
 };
 use crate::heed_codec::{ScriptLanguageCodec, StrBEU16Codec, StrRefCodec};
 use crate::{
-    default_criteria, CboRoaringBitmapCodec, Criterion, DocumentId, ExternalDocumentsIds,
-    FacetDistribution, FieldDistribution, FieldId, FieldIdWordCountCodec, GeoPoint, ObkvCodec,
-    Result, RoaringBitmapCodec, RoaringBitmapLenCodec, Search, U8StrStrCodec, BEU16, BEU32,
+    default_criteria, BEU32StrCodec, BoRoaringBitmapCodec, CboRoaringBitmapCodec, Criterion,
+    DocumentId, ExternalDocumentsIds, FacetDistribution, FieldDistribution, FieldId,
+    FieldIdWordCountCodec, GeoPoint, ObkvCodec, Result, RoaringBitmapCodec, RoaringBitmapLenCodec,
+    Search, U8StrStrCodec, BEU16, BEU32,
 };
 
-/// The HNSW data-structure that we serialize, fill and search in.
-pub type Hnsw = hnsw::Hnsw<DotProduct, Vec<f32>, Pcg32, 12, 24>;
-
 pub const DEFAULT_MIN_WORD_LEN_ONE_TYPO: u8 = 5;
 pub const DEFAULT_MIN_WORD_LEN_TWO_TYPOS: u8 = 9;
 
@@ -47,7 +43,6 @@ pub mod main_key {
     pub const FIELDS_IDS_MAP_KEY: &str = "fields-ids-map";
     pub const GEO_FACETED_DOCUMENTS_IDS_KEY: &str = "geo-faceted-documents-ids";
     pub const GEO_RTREE_KEY: &str = "geo-rtree";
-    pub const VECTOR_HNSW_KEY: &str = "vector-hnsw";
     pub const HARD_EXTERNAL_DOCUMENTS_IDS_KEY: &str = "hard-external-documents-ids";
     pub const NUMBER_FACETED_DOCUMENTS_IDS_PREFIX: &str = "number-faceted-documents-ids";
     pub const PRIMARY_KEY_KEY: &str = "primary-key";
@@ -92,7 +87,6 @@ pub mod db_name {
     pub const FACET_ID_STRING_DOCIDS: &str = "facet-id-string-docids";
     pub const FIELD_ID_DOCID_FACET_F64S: &str = "field-id-docid-facet-f64s";
     pub const FIELD_ID_DOCID_FACET_STRINGS: &str = "field-id-docid-facet-strings";
-    pub const VECTOR_ID_DOCID: &str = "vector-id-docids";
     pub const DOCUMENTS: &str = "documents";
     pub const SCRIPT_LANGUAGE_DOCIDS: &str = "script_language_docids";
 }
@@ -117,6 +111,9 @@ pub struct Index {
     /// A prefix of word and all the documents ids containing this prefix, from attributes for which typos are not allowed.
     pub exact_word_prefix_docids: Database<Str, RoaringBitmapCodec>,
 
+    /// Maps a word and a document id (u32) to all the positions where the given word appears.
+    pub docid_word_positions: Database<BEU32StrCodec, BoRoaringBitmapCodec>,
+
     /// Maps the proximity between a pair of words with all the docids where this relation appears.
     pub word_pair_proximity_docids: Database<U8StrStrCodec, CboRoaringBitmapCodec>,
     /// Maps the proximity between a pair of word and prefix with all the docids where this relation appears.
@@ -156,9 +153,6 @@ pub struct Index {
     /// Maps the document id, the facet field id and the strings.
     pub field_id_docid_facet_strings: Database<FieldDocIdFacetStringCodec, Str>,
 
-    /// Maps a vector id to the document id that have it.
-    pub vector_id_docid: Database<OwnedType<BEU32>, OwnedType<BEU32>>,
-
     /// Maps the document id to the document as an obkv store.
     pub(crate) documents: Database<OwnedType<BEU32>, ObkvCodec>,
 }
@@ -172,7 +166,7 @@ impl Index {
     ) -> Result<Index> {
         use db_name::*;
 
-        options.max_dbs(24);
+        options.max_dbs(23);
         unsafe { options.flag(Flags::MdbAlwaysFreePages) };
 
         let env = options.open(path)?;
@@ -183,6 +177,7 @@ impl Index {
         let word_prefix_docids = env.create_database(&mut wtxn, Some(WORD_PREFIX_DOCIDS))?;
         let exact_word_prefix_docids =
             env.create_database(&mut wtxn, Some(EXACT_WORD_PREFIX_DOCIDS))?;
+        let docid_word_positions = env.create_database(&mut wtxn, Some(DOCID_WORD_POSITIONS))?;
         let word_pair_proximity_docids =
             env.create_database(&mut wtxn, Some(WORD_PAIR_PROXIMITY_DOCIDS))?;
         let script_language_docids =
@@ -208,11 +203,11 @@ impl Index {
             env.create_database(&mut wtxn, Some(FACET_ID_IS_NULL_DOCIDS))?;
         let facet_id_is_empty_docids =
             env.create_database(&mut wtxn, Some(FACET_ID_IS_EMPTY_DOCIDS))?;
+
         let field_id_docid_facet_f64s =
             env.create_database(&mut wtxn, Some(FIELD_ID_DOCID_FACET_F64S))?;
         let field_id_docid_facet_strings =
             env.create_database(&mut wtxn, Some(FIELD_ID_DOCID_FACET_STRINGS))?;
-        let vector_id_docid = env.create_database(&mut wtxn, Some(VECTOR_ID_DOCID))?;
         let documents = env.create_database(&mut wtxn, Some(DOCUMENTS))?;
         wtxn.commit()?;
 
@@ -225,6 +220,7 @@ impl Index {
             exact_word_docids,
             word_prefix_docids,
             exact_word_prefix_docids,
+            docid_word_positions,
             word_pair_proximity_docids,
             script_language_docids,
             word_prefix_pair_proximity_docids,
@@ -241,7 +237,6 @@ impl Index {
             facet_id_is_empty_docids,
             field_id_docid_facet_f64s,
             field_id_docid_facet_strings,
-            vector_id_docid,
             documents,
         })
     }
@@ -513,26 +508,6 @@ impl Index {
         }
     }
 
-    /* vector HNSW */
-
-    /// Writes the provided `hnsw`.
-    pub(crate) fn put_vector_hnsw(&self, wtxn: &mut RwTxn, hnsw: &Hnsw) -> heed::Result<()> {
-        self.main.put::<_, Str, SerdeBincode<Hnsw>>(wtxn, main_key::VECTOR_HNSW_KEY, hnsw)
-    }
-
-    /// Delete the `hnsw`.
-    pub(crate) fn delete_vector_hnsw(&self, wtxn: &mut RwTxn) -> heed::Result<bool> {
-        self.main.delete::<_, Str>(wtxn, main_key::VECTOR_HNSW_KEY)
-    }
-
-    /// Returns the `hnsw`.
-    pub fn vector_hnsw(&self, rtxn: &RoTxn) -> Result<Option<Hnsw>> {
-        match self.main.get::<_, Str, SerdeBincode<Hnsw>>(rtxn, main_key::VECTOR_HNSW_KEY)? {
-            Some(hnsw) => Ok(Some(hnsw)),
-            None => Ok(None),
-        }
-    }
-
     /* field distribution */
 
     /// Writes the field distribution which associates every field name with
@@ -1497,9 +1472,9 @@ pub(crate) mod tests {
 
         db_snap!(index, field_distribution,
             @r###"
-        age              1      |
-        id               2      |
-        name             2      |
+        age              1     
+        id               2     
+        name             2     
         "###
         );
 
@@ -1517,9 +1492,9 @@ pub(crate) mod tests {
 
         db_snap!(index, field_distribution,
             @r###"
-        age              1      |
-        id               2      |
-        name             2      |
+        age              1     
+        id               2     
+        name             2     
         "###
         );
 
@@ -1533,9 +1508,9 @@ pub(crate) mod tests {
 
         db_snap!(index, field_distribution,
             @r###"
-        has_dog          1      |
-        id               2      |
-        name             2      |
+        has_dog          1     
+        id               2     
+        name             2     
         "###
         );
     }

milli/src/lib.rs

@@ -5,12 +5,57 @@
 #[global_allocator]
 pub static ALLOC: mimalloc::MiMalloc = mimalloc::MiMalloc;
 
+// #[cfg(test)]
+// pub mod allocator {
+//     use std::alloc::{GlobalAlloc, System};
+//     use std::sync::atomic::{self, AtomicI64};
+
+//     #[global_allocator]
+//     pub static ALLOC: CountingAlloc = CountingAlloc {
+//         max_resident: AtomicI64::new(0),
+//         resident: AtomicI64::new(0),
+//         allocated: AtomicI64::new(0),
+//     };
+
+//     pub struct CountingAlloc {
+//         pub max_resident: AtomicI64,
+//         pub resident: AtomicI64,
+//         pub allocated: AtomicI64,
+//     }
+//     unsafe impl GlobalAlloc for CountingAlloc {
+//         unsafe fn alloc(&self, layout: std::alloc::Layout) -> *mut u8 {
+//             self.allocated.fetch_add(layout.size() as i64, atomic::Ordering::SeqCst);
+//             let old_resident =
+//                 self.resident.fetch_add(layout.size() as i64, atomic::Ordering::SeqCst);
+
+//             let resident = old_resident + layout.size() as i64;
+//             self.max_resident.fetch_max(resident, atomic::Ordering::SeqCst);
+
+//             // if layout.size() > 1_000_000 {
+//             //     eprintln!(
+//             //         "allocating {} with new resident size: {resident}",
+//             //         layout.size() / 1_000_000
+//             //     );
+//             //     // let trace = std::backtrace::Backtrace::capture();
+//             //     // let t = trace.to_string();
+//             //     // eprintln!("{t}");
+//             // }
+
+//             System.alloc(layout)
+//         }
+
+//         unsafe fn dealloc(&self, ptr: *mut u8, layout: std::alloc::Layout) {
+//             self.resident.fetch_sub(layout.size() as i64, atomic::Ordering::Relaxed);
+//             System.dealloc(ptr, layout)
+//         }
+//     }
+// }
+
 #[macro_use]
 pub mod documents;
 
 mod asc_desc;
 mod criterion;
-pub mod distance;
 mod error;
 mod external_documents_ids;
 pub mod facet;
@@ -30,7 +75,6 @@ use std::convert::{TryFrom, TryInto};
 use std::hash::BuildHasherDefault;
 
 use charabia::normalizer::{CharNormalizer, CompatibilityDecompositionNormalizer};
-pub use distance::{dot_product_similarity, euclidean_squared_distance};
 pub use filter_parser::{Condition, FilterCondition, Span, Token};
 use fxhash::{FxHasher32, FxHasher64};
 pub use grenad::CompressionType;
@@ -285,35 +329,6 @@ pub fn normalize_facet(original: &str) -> String {
     CompatibilityDecompositionNormalizer.normalize_str(original.trim()).to_lowercase()
 }
 
-/// Represents either a vector or an array of multiple vectors.
-#[derive(serde::Serialize, serde::Deserialize, Debug)]
-#[serde(transparent)]
-pub struct VectorOrArrayOfVectors {
-    #[serde(with = "either::serde_untagged")]
-    inner: either::Either<Vec<f32>, Vec<Vec<f32>>>,
-}
-
-impl VectorOrArrayOfVectors {
-    pub fn into_array_of_vectors(self) -> Vec<Vec<f32>> {
-        match self.inner {
-            either::Either::Left(vector) => vec![vector],
-            either::Either::Right(vectors) => vectors,
-        }
-    }
-}
-
-/// Normalize a vector by dividing the dimensions by the lenght of it.
-pub fn normalize_vector(mut vector: Vec<f32>) -> Vec<f32> {
-    let squared: f32 = vector.iter().map(|x| x * x).sum();
-    let length = squared.sqrt();
-    if length <= f32::EPSILON {
-        vector
-    } else {
-        vector.iter_mut().for_each(|x| *x /= length);
-        vector
-    }
-}
-
 #[cfg(test)]
 mod tests {
     use serde_json::json;

milli/src/search/mod.rs

@@ -22,7 +22,6 @@ pub mod new;
 
 pub struct Search<'a> {
     query: Option<String>,
-    vector: Option<Vec<f32>>,
     // this should be linked to the String in the query
     filter: Option<Filter<'a>>,
     offset: usize,
@@ -40,7 +39,6 @@ impl<'a> Search<'a> {
     pub fn new(rtxn: &'a heed::RoTxn, index: &'a Index) -> Search<'a> {
         Search {
             query: None,
-            vector: None,
             filter: None,
             offset: 0,
             limit: 20,
@@ -59,11 +57,6 @@ impl<'a> Search<'a> {
         self
     }
 
-    pub fn vector(&mut self, vector: impl Into<Vec<f32>>) -> &mut Search<'a> {
-        self.vector = Some(vector.into());
-        self
-    }
-
     pub fn offset(&mut self, offset: usize) -> &mut Search<'a> {
         self.offset = offset;
         self
@@ -113,7 +106,6 @@ impl<'a> Search<'a> {
             execute_search(
                 &mut ctx,
                 &self.query,
-                &self.vector,
                 self.terms_matching_strategy,
                 self.exhaustive_number_hits,
                 &self.filter,
@@ -140,7 +132,6 @@ impl fmt::Debug for Search<'_> {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
         let Search {
             query,
-            vector: _,
             filter,
             offset,
             limit,
@@ -154,7 +145,6 @@ impl fmt::Debug for Search<'_> {
         } = self;
         f.debug_struct("Search")
             .field("query", query)
-            .field("vector", &"[...]")
             .field("filter", filter)
             .field("offset", offset)
             .field("limit", limit)

milli/src/search/new/bucket_sort.rs

@@ -116,16 +116,15 @@ pub fn bucket_sort<'ctx, Q: RankingRuleQueryTrait>(
     }
 
     while valid_docids.len() < length {
-        // The universe for this bucket is zero or one element, so we don't need to sort
-        // anything, just extend the results and go back to the parent ranking rule.
-        if ranking_rule_universes[cur_ranking_rule_index].len() <= 1 {
-            let bucket = std::mem::take(&mut ranking_rule_universes[cur_ranking_rule_index]);
-            maybe_add_to_results!(bucket);
+        // The universe for this bucket is zero element, so we don't need to sort
+        // anything, just go back to the parent ranking rule.
+        if ranking_rule_universes[cur_ranking_rule_index].is_empty() {
             back!();
             continue;
         }
 
-        let Some(next_bucket) = ranking_rules[cur_ranking_rule_index].next_bucket(ctx, logger, &ranking_rule_universes[cur_ranking_rule_index])? else {
+        let Some(next_bucket) = ranking_rules[cur_ranking_rule_index].next_bucket(ctx, logger, &ranking_rule_universes[cur_ranking_rule_index])?
+        else {
             back!();
             continue;
         };

milli/src/search/new/graph_based_ranking_rule.rs

@@ -181,9 +181,6 @@ impl<'ctx, G: RankingRuleGraphTrait> RankingRule<'ctx, QueryGraph> for GraphBase
         logger: &mut dyn SearchLogger<QueryGraph>,
         universe: &RoaringBitmap,
     ) -> Result<Option<RankingRuleOutput<QueryGraph>>> {
-        // If universe.len() <= 1, the bucket sort algorithm
-        // should not have called this function.
-        assert!(universe.len() > 1);
         // Will crash if `next_bucket` is called before `start_iteration` or after `end_iteration`,
         // should never happen
         let mut state = self.state.take().unwrap();

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

@@ -509,7 +509,6 @@ mod tests {
             let crate::search::PartialSearchResult { located_query_terms, .. } = execute_search(
                 &mut ctx,
                 &Some(query.to_string()),
-                &None,
                 crate::TermsMatchingStrategy::default(),
                 false,
                 &None,

milli/src/search/new/mod.rs

@@ -28,7 +28,6 @@ use db_cache::DatabaseCache;
 use exact_attribute::ExactAttribute;
 use graph_based_ranking_rule::{Exactness, Fid, Position, Proximity, Typo};
 use heed::RoTxn;
-use hnsw::Searcher;
 use interner::{DedupInterner, Interner};
 pub use logger::visual::VisualSearchLogger;
 pub use logger::{DefaultSearchLogger, SearchLogger};
@@ -40,17 +39,13 @@ use ranking_rules::{
 use resolve_query_graph::{compute_query_graph_docids, PhraseDocIdsCache};
 use roaring::RoaringBitmap;
 use sort::Sort;
-use space::Neighbor;
 
 use self::geo_sort::GeoSort;
 pub use self::geo_sort::Strategy as GeoSortStrategy;
 use self::graph_based_ranking_rule::Words;
 use self::interner::Interned;
 use crate::search::new::distinct::apply_distinct_rule;
-use crate::{
-    normalize_vector, AscDesc, DocumentId, Filter, Index, Member, Result, TermsMatchingStrategy,
-    UserError, BEU32,
-};
+use crate::{AscDesc, DocumentId, Filter, Index, Member, Result, TermsMatchingStrategy, UserError};
 
 /// A structure used throughout the execution of a search query.
 pub struct SearchContext<'ctx> {
@@ -354,7 +349,6 @@ fn resolve_sort_criteria<'ctx, Query: RankingRuleQueryTrait>(
 pub fn execute_search(
     ctx: &mut SearchContext,
     query: &Option<String>,
-    vector: &Option<Vec<f32>>,
     terms_matching_strategy: TermsMatchingStrategy,
     exhaustive_number_hits: bool,
     filters: &Option<Filter>,
@@ -434,34 +428,6 @@ pub fn execute_search(
 
     let BucketSortOutput { docids, mut all_candidates } = bucket_sort_output;
 
-    let docids = match vector {
-        Some(vector) => {
-            // return the nearest documents that are also part of the candidates.
-            let mut searcher = Searcher::new();
-            let hnsw = ctx.index.vector_hnsw(ctx.txn)?.unwrap_or_default();
-            let ef = hnsw.len().min(100);
-            let mut dest = vec![Neighbor { index: 0, distance: 0 }; ef];
-            let vector = normalize_vector(vector.clone());
-            let neighbors = hnsw.nearest(&vector, ef, &mut searcher, &mut dest[..]);
-
-            let mut docids = Vec::new();
-            for Neighbor { index, distance: _ } in neighbors.iter() {
-                let index = BEU32::new(*index as u32);
-                let docid = ctx.index.vector_id_docid.get(ctx.txn, &index)?.unwrap().get();
-                if universe.contains(docid) {
-                    docids.push(docid);
-                    if docids.len() == (from + length) {
-                        break;
-                    }
-                }
-            }
-
-            docids.into_iter().skip(from).take(length).collect()
-        }
-        // return the search docids if the vector field is not specified
-        None => docids,
-    };
-
     // The candidates is the universe unless the exhaustive number of hits
     // is requested and a distinct attribute is set.
     if exhaustive_number_hits {

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

@@ -79,7 +79,7 @@ pub fn located_query_terms_from_tokens(
             TokenKind::Separator(separator_kind) => {
                 // add penalty for hard separators
                 if let SeparatorKind::Hard = separator_kind {
-                    position = position.wrapping_add(7);
+                    position = position.wrapping_add(1);
                 }
 
                 phrase = 'phrase: {

milli/src/snapshot_tests.rs

@@ -89,6 +89,7 @@ Create a snapshot test of the given database.
     - `exact_word_docids`
     - `word_prefix_docids`
     - `exact_word_prefix_docids`
+    - `docid_word_positions`
     - `word_pair_proximity_docids`
     - `word_prefix_pair_proximity_docids`
     - `word_position_docids`
@@ -216,6 +217,11 @@ pub fn snap_exact_word_prefix_docids(index: &Index) -> String {
         &format!("{s:<16} {}", display_bitmap(&b))
     })
 }
+pub fn snap_docid_word_positions(index: &Index) -> String {
+    make_db_snap_from_iter!(index, docid_word_positions, |((idx, s), b)| {
+        &format!("{idx:<6} {s:<16} {}", display_bitmap(&b))
+    })
+}
 pub fn snap_word_pair_proximity_docids(index: &Index) -> String {
     make_db_snap_from_iter!(index, word_pair_proximity_docids, |((proximity, word1, word2), b)| {
         &format!("{proximity:<2} {word1:<16} {word2:<16} {}", display_bitmap(&b))
@@ -318,7 +324,7 @@ pub fn snap_field_distributions(index: &Index) -> String {
     let rtxn = index.read_txn().unwrap();
     let mut snap = String::new();
     for (field, count) in index.field_distribution(&rtxn).unwrap() {
-        writeln!(&mut snap, "{field:<16} {count:<6} |").unwrap();
+        writeln!(&mut snap, "{field:<16} {count:<6}").unwrap();
     }
     snap
 }
@@ -328,7 +334,7 @@ pub fn snap_fields_ids_map(index: &Index) -> String {
     let mut snap = String::new();
     for field_id in fields_ids_map.ids() {
         let name = fields_ids_map.name(field_id).unwrap();
-        writeln!(&mut snap, "{field_id:<3} {name:<16} |").unwrap();
+        writeln!(&mut snap, "{field_id:<3} {name:<16}").unwrap();
     }
     snap
 }
@@ -471,6 +477,9 @@ macro_rules! full_snap_of_db {
     ($index:ident, exact_word_prefix_docids) => {{
         $crate::snapshot_tests::snap_exact_word_prefix_docids(&$index)
     }};
+    ($index:ident, docid_word_positions) => {{
+        $crate::snapshot_tests::snap_docid_word_positions(&$index)
+    }};
     ($index:ident, word_pair_proximity_docids) => {{
         $crate::snapshot_tests::snap_word_pair_proximity_docids(&$index)
     }};

milli/src/snapshots/index.rs/initial_field_distribution/1/field_distribution.snap

@@ -1,7 +1,7 @@
 ---
 source: milli/src/index.rs
 ---
-age              1      |
-id               2      |
-name             2      |
+age              1     
+id               2     
+name             2     
 

milli/src/snapshots/index.rs/initial_field_distribution/field_distribution.snap

@@ -1,7 +1,7 @@
 ---
 source: milli/src/index.rs
 ---
-age              1      |
-id               2      |
-name             2      |
+age              1     
+id               2     
+name             2     
 

milli/src/update/clear_documents.rs

@@ -23,6 +23,7 @@ impl<'t, 'u, 'i> ClearDocuments<'t, 'u, 'i> {
             exact_word_docids,
             word_prefix_docids,
             exact_word_prefix_docids,
+            docid_word_positions,
             word_pair_proximity_docids,
             word_prefix_pair_proximity_docids,
             prefix_word_pair_proximity_docids,
@@ -39,7 +40,6 @@ impl<'t, 'u, 'i> ClearDocuments<'t, 'u, 'i> {
             facet_id_is_empty_docids,
             field_id_docid_facet_f64s,
             field_id_docid_facet_strings,
-            vector_id_docid,
             documents,
         } = self.index;
 
@@ -58,7 +58,6 @@ impl<'t, 'u, 'i> ClearDocuments<'t, 'u, 'i> {
         self.index.put_field_distribution(self.wtxn, &FieldDistribution::default())?;
         self.index.delete_geo_rtree(self.wtxn)?;
         self.index.delete_geo_faceted_documents_ids(self.wtxn)?;
-        self.index.delete_vector_hnsw(self.wtxn)?;
 
         // We clean all the faceted documents ids.
         for field_id in faceted_fields {
@@ -81,6 +80,7 @@ impl<'t, 'u, 'i> ClearDocuments<'t, 'u, 'i> {
         exact_word_docids.clear(self.wtxn)?;
         word_prefix_docids.clear(self.wtxn)?;
         exact_word_prefix_docids.clear(self.wtxn)?;
+        docid_word_positions.clear(self.wtxn)?;
         word_pair_proximity_docids.clear(self.wtxn)?;
         word_prefix_pair_proximity_docids.clear(self.wtxn)?;
         prefix_word_pair_proximity_docids.clear(self.wtxn)?;
@@ -97,7 +97,6 @@ impl<'t, 'u, 'i> ClearDocuments<'t, 'u, 'i> {
         facet_id_string_docids.clear(self.wtxn)?;
         field_id_docid_facet_f64s.clear(self.wtxn)?;
         field_id_docid_facet_strings.clear(self.wtxn)?;
-        vector_id_docid.clear(self.wtxn)?;
         documents.clear(self.wtxn)?;
 
         Ok(number_of_documents)
@@ -142,6 +141,7 @@ mod tests {
 
         assert!(index.word_docids.is_empty(&rtxn).unwrap());
         assert!(index.word_prefix_docids.is_empty(&rtxn).unwrap());
+        assert!(index.docid_word_positions.is_empty(&rtxn).unwrap());
         assert!(index.word_pair_proximity_docids.is_empty(&rtxn).unwrap());
         assert!(index.field_id_word_count_docids.is_empty(&rtxn).unwrap());
         assert!(index.word_prefix_pair_proximity_docids.is_empty(&rtxn).unwrap());

milli/src/update/delete_documents.rs

@@ -1,13 +1,11 @@
 use std::collections::btree_map::Entry;
-use std::collections::{BTreeSet, HashMap, HashSet};
+use std::collections::{HashMap, HashSet};
 
 use fst::IntoStreamer;
 use heed::types::{ByteSlice, DecodeIgnore, Str, UnalignedSlice};
 use heed::{BytesDecode, BytesEncode, Database, RwIter};
-use hnsw::Searcher;
 use roaring::RoaringBitmap;
 use serde::{Deserialize, Serialize};
-use space::KnnPoints;
 use time::OffsetDateTime;
 
 use super::facet::delete::FacetsDelete;
@@ -16,9 +14,9 @@ use crate::error::InternalError;
 use crate::facet::FacetType;
 use crate::heed_codec::facet::FieldDocIdFacetCodec;
 use crate::heed_codec::CboRoaringBitmapCodec;
-use crate::index::Hnsw;
 use crate::{
-    ExternalDocumentsIds, FieldId, FieldIdMapMissingEntry, Index, Result, RoaringBitmapCodec, BEU32,
+    ExternalDocumentsIds, FieldId, FieldIdMapMissingEntry, Index, Result, RoaringBitmapCodec,
+    SmallString32, BEU32,
 };
 
 pub struct DeleteDocuments<'t, 'u, 'i> {
@@ -234,6 +232,7 @@ impl<'t, 'u, 'i> DeleteDocuments<'t, 'u, 'i> {
             exact_word_docids,
             word_prefix_docids,
             exact_word_prefix_docids,
+            docid_word_positions,
             word_pair_proximity_docids,
             field_id_word_count_docids,
             word_prefix_pair_proximity_docids,
@@ -250,12 +249,25 @@ impl<'t, 'u, 'i> DeleteDocuments<'t, 'u, 'i> {
             facet_id_exists_docids,
             facet_id_is_null_docids,
             facet_id_is_empty_docids,
-            vector_id_docid,
             documents,
         } = self.index;
-        // Remove from the documents database
+
+        // Retrieve the words contained in the documents.
+        let mut words = Vec::new();
         for docid in &self.to_delete_docids {
             documents.delete(self.wtxn, &BEU32::new(docid))?;
+
+            // We iterate through the words positions of the document id, retrieve the word and delete the positions.
+            // We create an iterator to be able to get the content and delete the key-value itself.
+            // It's faster to acquire a cursor to get and delete, as we avoid traversing the LMDB B-Tree two times but only once.
+            let mut iter = docid_word_positions.prefix_iter_mut(self.wtxn, &(docid, ""))?;
+            while let Some(result) = iter.next() {
+                let ((_docid, word), _positions) = result?;
+                // This boolean will indicate if we must remove this word from the words FST.
+                words.push((SmallString32::from(word), false));
+                // safety: we don't keep references from inside the LMDB database.
+                unsafe { iter.del_current()? };
+            }
         }
         // We acquire the current external documents ids map...
         // Note that its soft-deleted document ids field will be equal to the `to_delete_docids`
@@ -266,27 +278,42 @@ impl<'t, 'u, 'i> DeleteDocuments<'t, 'u, 'i> {
         let new_external_documents_ids = new_external_documents_ids.into_static();
         self.index.put_external_documents_ids(self.wtxn, &new_external_documents_ids)?;
 
-        let mut words_to_keep = BTreeSet::default();
-        let mut words_to_delete = BTreeSet::default();
+        // Maybe we can improve the get performance of the words
+        // if we sort the words first, keeping the LMDB pages in cache.
+        words.sort_unstable();
+
         // We iterate over the words and delete the documents ids
         // from the word docids database.
-        remove_from_word_docids(
-            self.wtxn,
-            word_docids,
-            &self.to_delete_docids,
-            &mut words_to_keep,
-            &mut words_to_delete,
-        )?;
-        remove_from_word_docids(
-            self.wtxn,
-            exact_word_docids,
-            &self.to_delete_docids,
-            &mut words_to_keep,
-            &mut words_to_delete,
-        )?;
+        for (word, must_remove) in &mut words {
+            remove_from_word_docids(
+                self.wtxn,
+                word_docids,
+                word.as_str(),
+                must_remove,
+                &self.to_delete_docids,
+            )?;
+
+            remove_from_word_docids(
+                self.wtxn,
+                exact_word_docids,
+                word.as_str(),
+                must_remove,
+                &self.to_delete_docids,
+            )?;
+        }
 
         // We construct an FST set that contains the words to delete from the words FST.
-        let words_to_delete = fst::Set::from_iter(words_to_delete.difference(&words_to_keep))?;
+        let words_to_delete =
+            words.iter().filter_map(
+                |(word, must_remove)| {
+                    if *must_remove {
+                        Some(word.as_str())
+                    } else {
+                        None
+                    }
+                },
+            );
+        let words_to_delete = fst::Set::from_iter(words_to_delete)?;
 
         let new_words_fst = {
             // We retrieve the current words FST from the database.
@@ -440,30 +467,6 @@ impl<'t, 'u, 'i> DeleteDocuments<'t, 'u, 'i> {
             &self.to_delete_docids,
         )?;
 
-        // An ugly and slow way to remove the vectors from the HNSW
-        // It basically reconstructs the HNSW from scratch without editing the current one.
-        let current_hnsw = self.index.vector_hnsw(self.wtxn)?.unwrap_or_default();
-        if !current_hnsw.is_empty() {
-            let mut new_hnsw = Hnsw::default();
-            let mut searcher = Searcher::new();
-            let mut new_vector_id_docids = Vec::new();
-
-            for result in vector_id_docid.iter(self.wtxn)? {
-                let (vector_id, docid) = result?;
-                if !self.to_delete_docids.contains(docid.get()) {
-                    let vector = current_hnsw.get_point(vector_id.get() as usize).clone();
-                    let vector_id = new_hnsw.insert(vector, &mut searcher);
-                    new_vector_id_docids.push((vector_id as u32, docid));
-                }
-            }
-
-            vector_id_docid.clear(self.wtxn)?;
-            for (vector_id, docid) in new_vector_id_docids {
-                vector_id_docid.put(self.wtxn, &BEU32::new(vector_id), &docid)?;
-            }
-            self.index.put_vector_hnsw(self.wtxn, &new_hnsw)?;
-        }
-
         self.index.put_soft_deleted_documents_ids(self.wtxn, &RoaringBitmap::new())?;
 
         Ok(DetailedDocumentDeletionResult {
@@ -529,24 +532,23 @@ fn remove_from_word_prefix_docids(
 fn remove_from_word_docids(
     txn: &mut heed::RwTxn,
     db: &heed::Database<Str, RoaringBitmapCodec>,
+    word: &str,
+    must_remove: &mut bool,
     to_remove: &RoaringBitmap,
-    words_to_keep: &mut BTreeSet<String>,
-    words_to_remove: &mut BTreeSet<String>,
 ) -> Result<()> {
     // We create an iterator to be able to get the content and delete the word docids.
     // It's faster to acquire a cursor to get and delete or put, as we avoid traversing
     // the LMDB B-Tree two times but only once.
-    let mut iter = db.iter_mut(txn)?;
-    while let Some((key, mut docids)) = iter.next().transpose()? {
-        let previous_len = docids.len();
-        docids -= to_remove;
-        if docids.is_empty() {
-            // safety: we don't keep references from inside the LMDB database.
-            unsafe { iter.del_current()? };
-            words_to_remove.insert(key.to_owned());
-        } else {
-            words_to_keep.insert(key.to_owned());
-            if docids.len() != previous_len {
+    let mut iter = db.prefix_iter_mut(txn, word)?;
+    if let Some((key, mut docids)) = iter.next().transpose()? {
+        if key == word {
+            let previous_len = docids.len();
+            docids -= to_remove;
+            if docids.is_empty() {
+                // safety: we don't keep references from inside the LMDB database.
+                unsafe { iter.del_current()? };
+                *must_remove = true;
+            } else if docids.len() != previous_len {
                 let key = key.to_owned();
                 // safety: we don't keep references from inside the LMDB database.
                 unsafe { iter.put_current(&key, &docids)? };
@@ -625,7 +627,7 @@ mod tests {
 
     use super::*;
     use crate::index::tests::TempIndex;
-    use crate::{db_snap, Filter, Search};
+    use crate::{db_snap, Filter};
 
     fn delete_documents<'t>(
         wtxn: &mut RwTxn<'t, '_>,
@@ -1197,52 +1199,4 @@ mod tests {
             DeletionStrategy::AlwaysSoft,
         );
     }
-
-    #[test]
-    fn delete_words_exact_attributes() {
-        let index = TempIndex::new();
-
-        index
-            .update_settings(|settings| {
-                settings.set_primary_key(S("id"));
-                settings.set_searchable_fields(vec![S("text"), S("exact")]);
-                settings.set_exact_attributes(vec![S("exact")].into_iter().collect());
-            })
-            .unwrap();
-
-        index
-            .add_documents(documents!([
-                { "id": 0, "text": "hello" },
-                { "id": 1, "exact": "hello"}
-            ]))
-            .unwrap();
-        db_snap!(index, word_docids, 1, @r###"
-        hello            [0, ]
-        "###);
-        db_snap!(index, exact_word_docids, 1, @r###"
-        hello            [1, ]
-        "###);
-        db_snap!(index, words_fst, 1, @"300000000000000001084cfcfc2ce1000000016000000090ea47f");
-
-        let mut wtxn = index.write_txn().unwrap();
-        let deleted_internal_ids =
-            delete_documents(&mut wtxn, &index, &["1"], DeletionStrategy::AlwaysHard);
-        wtxn.commit().unwrap();
-
-        db_snap!(index, word_docids, 2, @r###"
-        hello            [0, ]
-        "###);
-        db_snap!(index, exact_word_docids, 2, @"");
-        db_snap!(index, words_fst, 2, @"300000000000000001084cfcfc2ce1000000016000000090ea47f");
-
-        insta::assert_snapshot!(format!("{deleted_internal_ids:?}"), @"[1]");
-        let txn = index.read_txn().unwrap();
-        let words = index.words_fst(&txn).unwrap().into_stream().into_strs().unwrap();
-        insta::assert_snapshot!(format!("{words:?}"), @r###"["hello"]"###);
-
-        let mut s = Search::new(&txn, &index);
-        s.query("hello");
-        let crate::SearchResult { documents_ids, .. } = s.execute().unwrap();
-        insta::assert_snapshot!(format!("{documents_ids:?}"), @"[0]");
-    }
 }

milli/src/update/index_documents/extract/extract_fid_word_count_docids.rs

@@ -1,6 +1,6 @@
 use std::collections::HashMap;
 use std::fs::File;
-use std::io;
+use std::{cmp, io};
 
 use grenad::Sorter;
 
@@ -54,10 +54,11 @@ pub fn extract_fid_word_count_docids<R: io::Read + io::Seek>(
         }
 
         for position in read_u32_ne_bytes(value) {
-            let (field_id, _) = relative_from_absolute_position(position);
+            let (field_id, position) = relative_from_absolute_position(position);
+            let word_count = position as u32 + 1;
 
             let value = document_fid_wordcount.entry(field_id as FieldId).or_insert(0);
-            *value += 1;
+            *value = cmp::max(*value, word_count);
         }
     }
 
@@ -82,7 +83,7 @@ fn drain_document_fid_wordcount_into_sorter(
     let mut key_buffer = Vec::new();
 
     for (fid, count) in document_fid_wordcount.drain() {
-        if count <= 30 {
+        if count <= 10 {
             key_buffer.clear();
             key_buffer.extend_from_slice(&fid.to_be_bytes());
             key_buffer.push(count as u8);

milli/src/update/index_documents/extract/extract_vector_points.rs

@@ -1,69 +0,0 @@
-use std::convert::TryFrom;
-use std::fs::File;
-use std::io;
-
-use bytemuck::cast_slice;
-use serde_json::{from_slice, Value};
-
-use super::helpers::{create_writer, writer_into_reader, GrenadParameters};
-use crate::error::UserError;
-use crate::{FieldId, InternalError, Result, VectorOrArrayOfVectors};
-
-/// Extracts the embedding vector contained in each document under the `_vectors` field.
-///
-/// Returns the generated grenad reader containing the docid as key associated to the Vec<f32>
-#[logging_timer::time]
-pub fn extract_vector_points<R: io::Read + io::Seek>(
-    obkv_documents: grenad::Reader<R>,
-    indexer: GrenadParameters,
-    primary_key_id: FieldId,
-    vectors_fid: FieldId,
-) -> Result<grenad::Reader<File>> {
-    let mut writer = create_writer(
-        indexer.chunk_compression_type,
-        indexer.chunk_compression_level,
-        tempfile::tempfile()?,
-    );
-
-    let mut cursor = obkv_documents.into_cursor()?;
-    while let Some((docid_bytes, value)) = cursor.move_on_next()? {
-        let obkv = obkv::KvReader::new(value);
-
-        // since we only needs the primary key when we throw an error we create this getter to
-        // lazily get it when needed
-        let document_id = || -> Value {
-            let document_id = obkv.get(primary_key_id).unwrap();
-            serde_json::from_slice(document_id).unwrap()
-        };
-
-        // first we retrieve the _vectors field
-        if let Some(vectors) = obkv.get(vectors_fid) {
-            // extract the vectors
-            let vectors = match from_slice(vectors) {
-                Ok(vectors) => VectorOrArrayOfVectors::into_array_of_vectors(vectors),
-                Err(_) => {
-                    return Err(UserError::InvalidVectorsType {
-                        document_id: document_id(),
-                        value: from_slice(vectors).map_err(InternalError::SerdeJson)?,
-                    }
-                    .into())
-                }
-            };
-
-            for (i, vector) in vectors.into_iter().enumerate() {
-                match u16::try_from(i) {
-                    Ok(i) => {
-                        let mut key = docid_bytes.to_vec();
-                        key.extend_from_slice(&i.to_ne_bytes());
-                        let bytes = cast_slice(&vector);
-                        writer.insert(key, bytes)?;
-                    }
-                    Err(_) => continue,
-                }
-            }
-        }
-        // else => the `_vectors` object was `null`, there is nothing to do
-    }
-
-    writer_into_reader(writer)
-}

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

@@ -4,7 +4,6 @@ mod extract_facet_string_docids;
 mod extract_fid_docid_facet_values;
 mod extract_fid_word_count_docids;
 mod extract_geo_points;
-mod extract_vector_points;
 mod extract_word_docids;
 mod extract_word_fid_docids;
 mod extract_word_pair_proximity_docids;
@@ -23,7 +22,6 @@ use self::extract_facet_string_docids::extract_facet_string_docids;
 use self::extract_fid_docid_facet_values::{extract_fid_docid_facet_values, ExtractedFacetValues};
 use self::extract_fid_word_count_docids::extract_fid_word_count_docids;
 use self::extract_geo_points::extract_geo_points;
-use self::extract_vector_points::extract_vector_points;
 use self::extract_word_docids::extract_word_docids;
 use self::extract_word_fid_docids::extract_word_fid_docids;
 use self::extract_word_pair_proximity_docids::extract_word_pair_proximity_docids;
@@ -47,7 +45,6 @@ pub(crate) fn data_from_obkv_documents(
     faceted_fields: HashSet<FieldId>,
     primary_key_id: FieldId,
     geo_fields_ids: Option<(FieldId, FieldId)>,
-    vectors_field_id: Option<FieldId>,
     stop_words: Option<fst::Set<&[u8]>>,
     max_positions_per_attributes: Option<u32>,
     exact_attributes: HashSet<FieldId>,
@@ -72,7 +69,6 @@ pub(crate) fn data_from_obkv_documents(
                     &faceted_fields,
                     primary_key_id,
                     geo_fields_ids,
-                    vectors_field_id,
                     &stop_words,
                     max_positions_per_attributes,
                 )
@@ -283,7 +279,6 @@ fn send_and_extract_flattened_documents_data(
     faceted_fields: &HashSet<FieldId>,
     primary_key_id: FieldId,
     geo_fields_ids: Option<(FieldId, FieldId)>,
-    vectors_field_id: Option<FieldId>,
     stop_words: &Option<fst::Set<&[u8]>>,
     max_positions_per_attributes: Option<u32>,
 ) -> Result<(
@@ -312,25 +307,6 @@ fn send_and_extract_flattened_documents_data(
         });
     }
 
-    if let Some(vectors_field_id) = vectors_field_id {
-        let documents_chunk_cloned = flattened_documents_chunk.clone();
-        let lmdb_writer_sx_cloned = lmdb_writer_sx.clone();
-        rayon::spawn(move || {
-            let result = extract_vector_points(
-                documents_chunk_cloned,
-                indexer,
-                primary_key_id,
-                vectors_field_id,
-            );
-            let _ = match result {
-                Ok(vector_points) => {
-                    lmdb_writer_sx_cloned.send(Ok(TypedChunk::VectorPoints(vector_points)))
-                }
-                Err(error) => lmdb_writer_sx_cloned.send(Err(error)),
-            };
-        });
-    }
-
     let (docid_word_positions_chunk, docid_fid_facet_values_chunks): (Result<_>, Result<_>) =
         rayon::join(
             || {
@@ -349,6 +325,8 @@ fn send_and_extract_flattened_documents_data(
                 // send docid_word_positions_chunk to DB writer
                 let docid_word_positions_chunk =
                     unsafe { as_cloneable_grenad(&docid_word_positions_chunk)? };
+                let _ = lmdb_writer_sx
+                    .send(Ok(TypedChunk::DocidWordPositions(docid_word_positions_chunk.clone())));
 
                 let _ =
                     lmdb_writer_sx.send(Ok(TypedChunk::ScriptLanguageDocids(script_language_pair)));

milli/src/update/index_documents/helpers/merge_functions.rs

@@ -4,6 +4,7 @@ use std::result::Result as StdResult;
 
 use roaring::RoaringBitmap;
 
+use super::read_u32_ne_bytes;
 use crate::heed_codec::CboRoaringBitmapCodec;
 use crate::update::index_documents::transform::Operation;
 use crate::Result;
@@ -21,6 +22,10 @@ pub fn concat_u32s_array<'a>(_key: &[u8], values: &[Cow<'a, [u8]>]) -> Result<Co
     }
 }
 
+pub fn roaring_bitmap_from_u32s_array(slice: &[u8]) -> RoaringBitmap {
+    read_u32_ne_bytes(slice).collect()
+}
+
 pub fn serialize_roaring_bitmap(bitmap: &RoaringBitmap, buffer: &mut Vec<u8>) -> io::Result<()> {
     buffer.clear();
     buffer.reserve(bitmap.serialized_size());

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

@@ -14,8 +14,8 @@ pub use grenad_helpers::{
 };
 pub use merge_functions::{
     concat_u32s_array, keep_first, keep_latest_obkv, merge_cbo_roaring_bitmaps,
-    merge_obkvs_and_operations, merge_roaring_bitmaps, merge_two_obkvs, serialize_roaring_bitmap,
-    MergeFn,
+    merge_obkvs_and_operations, merge_roaring_bitmaps, merge_two_obkvs,
+    roaring_bitmap_from_u32s_array, serialize_roaring_bitmap, MergeFn,
 };
 
 use crate::MAX_WORD_LENGTH;

milli/src/update/index_documents/mod.rs

@@ -304,8 +304,6 @@ where
             }
             None => None,
         };
-        // get the fid of the `_vectors` field.
-        let vectors_field_id = self.index.fields_ids_map(self.wtxn)?.id("_vectors");
 
         let stop_words = self.index.stop_words(self.wtxn)?;
         let exact_attributes = self.index.exact_attributes_ids(self.wtxn)?;
@@ -342,7 +340,6 @@ where
                     faceted_fields,
                     primary_key_id,
                     geo_fields_ids,
-                    vectors_field_id,
                     stop_words,
                     max_positions_per_attributes,
                     exact_attributes,
@@ -2474,11 +2471,11 @@ mod tests {
               {
                 "id": 3,
                 "text": "a a a a a a a a a a a a a a a a a
-                a a a a a a a a a a a a a a a a a a a a a a a a a a
-                a a a a a a a a a a a a a a a a a a a a a a a a a a
-                a a a a a a a a a a a a a a a a a a a a a a a a a a
-                a a a a a a a a a a a a a a a a a a a a a a a a a a
-                a a a a a a a a a a a a a a a a a a a a a a a a a a
+                a a a a a a a a a a a a a a a a a a a a a a a a a a 
+                a a a a a a a a a a a a a a a a a a a a a a a a a a 
+                a a a a a a a a a a a a a a a a a a a a a a a a a a 
+                a a a a a a a a a a a a a a a a a a a a a a a a a a 
+                a a a a a a a a a a a a a a a a a a a a a a a a a a 
                 a a a a a a a a a a a a a a a a a a a a a "
              }
             ]))
@@ -2516,5 +2513,6 @@ mod tests {
 
         db_snap!(index, word_fid_docids, 3, @"4c2e2a1832e5802796edc1638136d933");
         db_snap!(index, word_position_docids, 3, @"74f556b91d161d997a89468b4da1cb8f");
+        db_snap!(index, docid_word_positions, 3, @"5287245332627675740b28bd46e1cde1");
     }
 }

milli/src/update/index_documents/typed_chunk.rs

@@ -4,29 +4,27 @@ use std::convert::TryInto;
 use std::fs::File;
 use std::io;
 
-use bytemuck::allocation::pod_collect_to_vec;
 use charabia::{Language, Script};
 use grenad::MergerBuilder;
 use heed::types::ByteSlice;
-use heed::RwTxn;
-use hnsw::Searcher;
+use heed::{BytesDecode, RwTxn};
 use roaring::RoaringBitmap;
-use space::KnnPoints;
 
 use super::helpers::{
-    self, merge_ignore_values, serialize_roaring_bitmap, valid_lmdb_key, CursorClonableMmap,
+    self, merge_ignore_values, roaring_bitmap_from_u32s_array, serialize_roaring_bitmap,
+    valid_lmdb_key, CursorClonableMmap,
 };
 use super::{ClonableMmap, MergeFn};
-use crate::error::UserError;
 use crate::facet::FacetType;
 use crate::update::facet::FacetsUpdate;
-use crate::update::index_documents::helpers::{as_cloneable_grenad, try_split_array_at};
+use crate::update::index_documents::helpers::as_cloneable_grenad;
 use crate::{
-    lat_lng_to_xyz, normalize_vector, CboRoaringBitmapCodec, DocumentId, GeoPoint, Index, Result,
-    BEU32,
+    lat_lng_to_xyz, BoRoaringBitmapCodec, CboRoaringBitmapCodec, DocumentId, GeoPoint, Index,
+    Result,
 };
 
 pub(crate) enum TypedChunk {
+    DocidWordPositions(grenad::Reader<CursorClonableMmap>),
     FieldIdDocidFacetStrings(grenad::Reader<CursorClonableMmap>),
     FieldIdDocidFacetNumbers(grenad::Reader<CursorClonableMmap>),
     Documents(grenad::Reader<CursorClonableMmap>),
@@ -45,7 +43,6 @@ pub(crate) enum TypedChunk {
     FieldIdFacetIsNullDocids(grenad::Reader<File>),
     FieldIdFacetIsEmptyDocids(grenad::Reader<File>),
     GeoPoints(grenad::Reader<File>),
-    VectorPoints(grenad::Reader<File>),
     ScriptLanguageDocids(HashMap<(Script, Language), RoaringBitmap>),
 }
 
@@ -59,6 +56,29 @@ pub(crate) fn write_typed_chunk_into_index(
 ) -> Result<(RoaringBitmap, bool)> {
     let mut is_merged_database = false;
     match typed_chunk {
+        TypedChunk::DocidWordPositions(docid_word_positions_iter) => {
+            write_entries_into_database(
+                docid_word_positions_iter,
+                &index.docid_word_positions,
+                wtxn,
+                index_is_empty,
+                |value, buffer| {
+                    // ensure that values are unique and ordered
+                    let positions = roaring_bitmap_from_u32s_array(value);
+                    BoRoaringBitmapCodec::serialize_into(&positions, buffer);
+                    Ok(buffer)
+                },
+                |new_values, db_values, buffer| {
+                    let new_values = roaring_bitmap_from_u32s_array(new_values);
+                    let positions = match BoRoaringBitmapCodec::bytes_decode(db_values) {
+                        Some(db_values) => new_values | db_values,
+                        None => new_values, // should not happen
+                    };
+                    BoRoaringBitmapCodec::serialize_into(&positions, buffer);
+                    Ok(())
+                },
+            )?;
+        }
         TypedChunk::Documents(obkv_documents_iter) => {
             let mut cursor = obkv_documents_iter.into_cursor()?;
             while let Some((key, value)) = cursor.move_on_next()? {
@@ -229,40 +249,6 @@ pub(crate) fn write_typed_chunk_into_index(
             index.put_geo_rtree(wtxn, &rtree)?;
             index.put_geo_faceted_documents_ids(wtxn, &geo_faceted_docids)?;
         }
-        TypedChunk::VectorPoints(vector_points) => {
-            let mut hnsw = index.vector_hnsw(wtxn)?.unwrap_or_default();
-            let mut searcher = Searcher::new();
-
-            let mut expected_dimensions = match index.vector_id_docid.iter(wtxn)?.next() {
-                Some(result) => {
-                    let (vector_id, _) = result?;
-                    Some(hnsw.get_point(vector_id.get() as usize).len())
-                }
-                None => None,
-            };
-
-            let mut cursor = vector_points.into_cursor()?;
-            while let Some((key, value)) = cursor.move_on_next()? {
-                // convert the key back to a u32 (4 bytes)
-                let (left, _index) = try_split_array_at(key).unwrap();
-                let docid = DocumentId::from_be_bytes(left);
-                // convert the vector back to a Vec<f32>
-                let vector: Vec<f32> = pod_collect_to_vec(value);
-
-                // TODO Move this error in the vector extractor
-                let found = vector.len();
-                let expected = *expected_dimensions.get_or_insert(found);
-                if expected != found {
-                    return Err(UserError::InvalidVectorDimensions { expected, found })?;
-                }
-
-                let vector = normalize_vector(vector);
-                let vector_id = hnsw.insert(vector, &mut searcher) as u32;
-                index.vector_id_docid.put(wtxn, &BEU32::new(vector_id), &BEU32::new(docid))?;
-            }
-            log::debug!("There are {} entries in the HNSW so far", hnsw.len());
-            index.put_vector_hnsw(wtxn, &hnsw)?;
-        }
         TypedChunk::ScriptLanguageDocids(hash_pair) => {
             let mut buffer = Vec::new();
             for (key, value) in hash_pair {